by Ilya Gavrichenkov
06/21/2004 | 06:57 PM
For the past few months we had a great opportunity to see Intel doing something outstanding: they have been copying all most successful AMD’s initiatives, introduced into life since the company launched their processors based on AMD64 architecture. At first Intel revealed its intention to provide x86 processors with the 64bit EMT64 expansion, which looked so very similar to AMD64 and which was compatible with the latter on the software level. Then the company announced its plans about the implementation of the NX bit support in their Pentium 4 processors based on the Prescott core, which are due in the fall. This bit is supposed to help implement additional protection of the system OS against viruses: this technology has actually been also borrowed from AMD Athlon 64 on their launch day.
The third idea Intel borrowed from the competitor processors was the Cool’n’Quiet technology, as Intel is going to implement its vision of this initiative in its upcoming Pentium 4 CPUs. However following in the competitor’s footsteps in terms of progressive new technologies support doesn’t at all mean that Intel decided to give up the laurels of the technology leader. The company engineers simply cannot disregard any smart ideas no matter who suggests them. In fact the today’s announcement is a direct proof of this point and of the fact that Intel keeps acting as the industry locomotive force.
First of all, today Intel officially announced one more CPU in the Pentium 4 processor family based on the 90nm Prescott core. This time it is a 3.6GHz model. The launch of this CPU is evidently none other but Intel’s response to AMD’s 3.6GHz CPU launch, which took place in the beginning of the month. As you remember the company has already announced 3800+ models with the dual channel memory controller and 2.4GHz actual working frequency. Today we are going to find out if Intel responded worthily, and in the meanwhile we will not dwell that much on the new processor, especially since it is not that very different from the predecessor from the architectural point of view. Actually, the today’s second announcement is of much more importance.
Besides the new CPU, Intel is also launching today a completely new platform: a new chipset family also known as Alderwood and Grantsdale. With the launch of these chipsets Intel starts pushing into the market a few innovative technologies, including such really important once as DDR2 SDRAM and the new bus for graphics and peripheral devices called PCI Express. Moreover, the new platforms will also feature a number of other important innovations, such as new LGA775 processor socket form-factor, Intel Graphics Media Accelerator 900, Intel Matrix Storage, Intel High Definition Audio, Intel Wireless Connect technologies, etc. As a result, we can even claim that the Pentium 4 platform is undergoing the most serious changes it has ever gone through.
It is very important to understand what we all need these mass changes for, especially since Intel’s platform has already been very successful all this time. Intel is very unlikely to introduce such serious innovations just for the sake of these mere innovations. Intel should evidently be pursuing some utilitarian goals by launching a completely new architecture, rather than strive for the simple performance increase. Since most of the freshly introduced technologies will not simply speed up the platform. The idea behind many of them has a lot to do with the “Digital home and office” concept. Due to higher bandwidth of the buses, improved multimedia features and increased functionality in terms of new network connections support, new platforms are becoming more attractive for home and office needs, which should be coordinating all home and office appliances together or even replacing some of them completely. However, let’s leave the marketing reasons for Intel employees, as they can definitely explain them much better to you. We are going to take a closer look at the new Intel platforms from the technological point of view.
Since we have already agreed that the Increase in the Pentium 4 clock frequency increase is not the most important thing for today, let’s start with the new chipsets. In fact the major technological innovations Intel revealed today with the launch of the Alderwood and Grantsdale platforms are not so few. Among them are:
All these technologies have been implemented in both new chipset families: Alderwood and Grantsdale. The major difference between these two chipset families lies in their marketing positioning in the first place. The Alderwood chipset is targeted for high-performance computer systems, where Intel expects it to be used together with either the new Pentium 4 3.6GHz or the new Pentium 4 Extreme Edition. And the other one, Grantsdale, is intended for the mass market. According to these positioning differences, Alderwood and Grantsdale got their official codenames. The high-performance Alderwood is called Intel 925X Express, while the mass Grantsdale chipset family is also known as Intel 915 Express. In fact, the codenames differences indicate very clearly that i925X Express and i915 Express do not differ from one another any greater than i875P and i865E.
If we delve a little bit more into details here, we should say that i925X Express features a slightly faster memory controller than i915 Express. However, the way this advantage is actually implemented is different from the way the well-known PAT technology in i875P works. As a result Intel claims that the mainboard makers will not be able to increase the performance of i915 Express up to the level of i925X Express with any of the existing tricks. In other words, we can hardly expect that the mainboard makers will find any undocumented ways of speeding up the boards based on the mass i915 Express, just like they did with i865PE. The advantages of the memory controller built into the i925X Express chipset result from the minimization of the memory latencies during memory addressing.
This minimization is possible because the data saved in the memory is rearranged in the most optimal way and also because the service commands have been integrated into the data stream. As the data is rearranged in the channels and memory banks, the access time during data extraction gets lower. The integration of the service commands into the data stream allows managing the entire memory subsystem in a more flexible way during data transfer, while most sets of core logic (including the new i915 Express) at first send only data managing commands and only in the end get to service commands.
The new chipset family is now represented by four new models:
In order to make it easier for you to read about these new guys I decided to compose a table with all the major features of the newcomers listed accordingly:
Intel Pentium 4
Intel Pentium 4 / Celeron D
800MHz / 533MHz
Memory channels / DIMMs per channel
2 channels / 2 DIMMs per channel
Maximum memory capacity
Supported memory types
DDR2 533/400 or DDR 400/333
Supported FSB modes / Memory frequencies
Intel Graphics Media Accelerator 900
PCI Express x16
PCI Express slots
(1) x16, (4) x1
1 ATA-100 channel
8 USB 2.0 ports
10/100 Mbit LAN MAC
Intel High Definition Audio, 24-bit 192kHz
Supported South Bridges
ICH6, R, W, RW
First of all I would like to say that Intel will very soon introduce one more high-performance core logic set besides the today’s i925X Express. It will be Intel 925XE Express. Unlike the currently available solution, the XE version will also support 1066MHz processor bus. The corresponding CPUs supporting this bus should also be announced in Q3 2004.
Also I would like to say a few words about different South Bridge modifications, which can go with the new chipsets. Besides the ICH6 with pretty standard features listed in the table above, Intel is also planning to ship ICH6R South Bridge supporting Intel Matrix Storage Technology (i.e. SerialATA RAID), ICH6W supporting Intel Wireless Connect Technology (WiFi) and ICH6RW featuring both: WiFi and Serial ATA RAID.
The new i925 and i915 chipsets are not compatible with Intel’s older North Bridges. The thing is that Intel has finally given up the Hub Link bus with pretty low bandwidth of 266MB/sec to connect the chipset Bridges. Instead of this bus they now use a new DMI (Direct Media Interface) for the i925/i915 chipset families. Its bandwidth equals 2GB/sec (1GB in each direction). This bus is built similarly to PCI Express, and should provide sufficient bandwidth for all external devices connected to the chipset South Bridge, which you can actually see from the diagrams above.
As the bandwidth of the bus between the bridges got higher, Intel could avoid connecting the gigabit network controller directly to the chipset North Bridge now. That is why the CSA bus (Communication Streaming Architecture) bus used in the i875/i865 chipsets is no longer necessary, so the new i915/i925 solutions simply do not have it any more. Now all gigabit network controllers should better be connected to the South Bridge of the chipset via the PCI Express x1 bus.
Note that since all the new chipsets have been transferred to the PCI Express interface, Intel made it impossible to use the old AGP graphics cards in the new platforms. The new sets of core logic do not support this bus at all now. However, a few mainboard manufacturers found a way to implement AGP on their products based on i915 chipset, although this solution involves the PCI bus, which limits the bandwidth of such interface telling negatively on the graphics subsystem performance, of course. At the same time, i915 still supports DDR SDRAM side by side with the new DDR2 SDRAM. While the high-performance i925X Express cannot boast DDR SDRAM support.
I would also like to point out that the new Intel chipsets are not claimed to support the 400MHz Quad Pumped Bus. It means that you will not be able to use the old Celeron CPUs on Northwood-128 core in any of the new i915/i925 based mainboards. By the way, Intel 925X Express targeted for the fastest CPUs doesn’t support 533MHz bus. One more curious fact about the budget Celeron D (Prescott-256 based) processors support implies that the new mainboards based on i915 solutions will allow using this processor only with the regular DDR SDRAM, because DDR2 SDRAM is supported only together with the 800MHz Quad Pumped Bus.
Besides the support of DDR2 SDRAM, PCI Express x16 for graphics cards and PCI Express x1 for external devices, I would also like to draw your attention to the IDE controller, which has been changed greatly. Unlike the previous chipsets, the new ICH6 South Bridges support 4 Serial ATA-150 channels instead of 2. At the same time, the number of Parallel ATA-100 channels has been reduced to 1.
Although our today’s review is not going to focus on the details of the new generation integrated graphics core from Intel aka Intel Graphics Media Accelerator 900, which is used in the i915G and i915GV chipsets, it would be unfair if we didn’t mention it at all here. The thing is that this graphics core is dramatically different from all the graphics cores Intel would use in the previous generation products. Intel Graphics Media Accelerator 900 is compatible with DirectX9 and provides hardware acceleration of the pixel shaders version 2.0 and vertex shaders. Moreover, Intel Graphics Media Accelerator 900 works at 333MHz frequency, features 4 pixel pipelines and has the opportunity to allocate up to 224MB RAM for the Video subsystem. Such drastic changes of the features give us some hope that the new integrated chipsets from Intel will look pretty fine even against the background of such serious competitor as ATI RADEON 9100 PRO IGP. In particular, the performance of Intel Graphics Media Accelerator 900 could very nicely be illustrated with the score obtained in 3DMark 2001 SE test package run on a system with Pentium 4 3.0E CPU. We managed to achieve 5600 points. And according to 3DMark03, Intel Graphics Media Accelerator 900 outperforms all low-cost discrete graphics cards, namely NVIDIA GeForce FX 5200 and ATI RADEON 9200.
Having said a few good words about the new chipset families, we suggest going a bit more into details about them. Now we are going to continue our discussion of the major innovations introduced in the new i925/i915 chipsets.
One of the key innovations, which appeared in the new i925/i915 chipsets is the support of the new DDR2 SDRAM memory. This allows Intel to increase the bandwidth of the memory subsystem. For example the bandwidth of the dual-channel memory subsystem where DDR2-533 SDRAM is used makes 8.5GB/sec, which is 33% higher than the bandwidth provided by the previous generation DDR400 SDRAM. However, I cannot say that the previous platforms based on i875/i865 chipsets had insufficient memory bandwidth. The bandwidth of the dual-channel DDR2-533 memory today is higher than that of the bus between the chipset North Bridge and the processor, which at least means that the CPU will not be able to use the bandwidth of the new memory subsystem to the full extent. However, in case of the integrated graphics core, which also uses quite a bit of the memory subsystem bandwidth, the DDR2-533 can be just the right thing. Besides, you should also take into account that the support of DDR2 memory in the i915/i925 platforms is a good “reserve” for the future. In Q3 2004 already Intel Company will please us with the new CPUs featuring 1066MHz bus, which will be bale to use the bandwidth of the DDR2-533 SDRAM in full.
In order to better understand all highs and lows of the DDR2-533 SDRAM compared with the ordinary DDR SDRAM, you should have at least some idea about the new memory architecture. First of all, note that DDR2 SDRAM is hardly that much different from the regular DDR SDRAM at all. However, while DDR SDRAM transfers two data batches per clock along the memory bus, DDR2 SDRAM effects four data transfers like that. At the same time DDR2 memory is built from the same memory cells as DDR SDRAM , and the performance is doubled due to the use of multiplexing technique.
The memory chips core itself works at the same clock frequency as in case of DDR memory and SDR SDRAM. It is only the input buffers frequency that gets higher, and the bus between the memory core and these buffers that gets wider. The input/output buffers perform multiplexing. The data transferred from the memory cells along the wide bus are actually leaving the cells along the bus of the regular bandwidth but at twice the frequency of DDR SDRAM. This simple approach allows increasing the memory bandwidth even higher without speeding up the memory cells at all. In other words, the memory cells of the today’s most advanced DDR2-533 SDRAM work with the same frequency as the memory cells of DDR266 SDRAM or PC133 SDRAM.
However, this simple way of increasing the memory bandwidth is not free from a few negative consequences of course. First of all, this is higher latency. Of course, the latency is determined neither by the buffers working frequency nor by the width of the bus the data from the memory cells is transferred along. The No.1 factor affecting the latency is the actual latency of the memory cells themselves. This way the latency of DDR2-533 is comparable with that of DDR266 or PC133 SDRAM, and evidently yields to the latency of the most advanced DDR SDRAM working at 400MHz frequency.
The examples are right at hand: the table below lists all latencies and bandwidth for the most widely spread memory standards:
Bandwidth in the dual-channel mode
As we see, if the introduction of DDR2 SDRAM provides a significant advantage in terms of memory bus bandwidth compared with the regular DDR SDRAM, then the low latency is definitely not one of its trumps. In fact, we will hardly see DDR2 memory modules with the latency comparable with that of the DDR400 SDRAM in the nearest future. The modern and fast DDR2-533 SDRAM with 4-4-4-12 timings boasts 1.5 times worse latency than DDR SDRAM working with 2-3-2-6 timings.
Does it make any sense to shift to DDR2 SDRAM at all then? The answer to this question is yes. However, it makes sense only for the Pentium 4 platform, because the performance of this platform really depends on the memory bandwidth a lot. As for Athlon64, for instance, it values lower latencies much more than higher bandwidth that is why AMD’s architecture will hardly benefit from the transition to DDR2 available so far. This is actually one of the reasons why AMD is not going to modify the memory controller of its CPUs to support DDR2 memory in the near future.
In fact, this intention to move to DDR2 SDRAM reminds us of Intel’s attempts to transfer its platforms to RDRAM. However, in our today’s case Intel made sure that its new platforms are backward compatible with the DDR400 SDRAM and that the industry supports DDR2 standard: it is an open standard and the production costs of DDR2 memory modules are as high as those of the regular DDR SDRAM, because they are using the same memory cells, as I have already told you. This way, DDR2 SDRAM will eventually settle down in the fastest Pentium 4 platforms and Intel will hardly have any causes for concern as far as the implementation of this new initiative goes.
Besides the increased input/output buffers working frequency and the use of twice as high multiplexing coefficient, DDR2 also has a few other distinguishing features, which are actually of no key importance. Therefore, we will simply list them in the table below for your convenience and then comment briefly:
Data transfer rate
200, 266, 333, 400MHz
400, 533, (667, 800) MHz
TSOP and FBGA
64Mbit – 1Gbit
256Mbit – 4Gbit
4 and 8
Prefetch (MIN Write Burst)
CAS Latency (CL)
2, 2.5, 3
3, 4, 5
Additive Latency (AL)
0, 1, 2, 3, 4
Read latency - 1
Off-Chip Driver (OCD) Calibration
Bidirectional Strobe (single ended)
Bidirectional Strobe (single ended or
On-die bus termination
2, 4, 8
In fact, I have to specifically stress the Additive Latency mechanism and the bus termination feature built into the chips. The Additive Latency mechanism makes data transfer rate somewhat more efficient. This algorithm solves an occasional problem of DDR SDRAM, when the read commands from one initialized memory bank and the initialization of the other memory bank cannot be performed simultaneously. However, this innovation doesn’t affect the real performance that much.
As for the on-die termination, the bus terminating resistors intended for damping the signals reflected from the bus ends are now located not on the mainboard but inside the memory chips. On the one hand, it allows improving the actual termination process, and on the other to reduce the mainboard production cost as there is no need to install a lot of resistors around the DIMM slots any more.
DDR2 DIMM modules look very similar to DDR memory:
However, DDR2 DIMMs are definitely incompatible with the older slots. They differ from the traditional DDR DIMM modules at least by the number of pins. While DDR SDRAM modules feature 184 pins, the DDR2 DIMMs have 240 of them. At the same time note that the physical dimensions of the DDR2 memory modules (height and width) are absolutely identical to those of DDR modules.
DDR SDRAM modules (top) and DDR2 SDRAM modules (bottom).
DDR2 SDRAM chips are designed in FBGA package, which is stated very clearly in the specs. The new chip packaging allows improving heat dissipation and minimizing the EMI imposed by the chips on one another. Besides the new chip packaging (as you know most DDR SDRAM chips were packed in TSOP), DDR2 SDRAM chips feature lower power voltage and as a result dissipate about 30% less heat. In particular, this is one of the reasons there can be designed DDR2 chips of higher capacity than in case of the regular DDR SDRAM.
To conclude out story about the new DDR2 SDRAM, which will now be supported by the new Pentium 4 platforms based on i925 and i915 chipsets, I would like to say a few words about the peculiarities of the new dual-channel memory controller used in these sets of core logic. As we all remember, the memory controller integrated into the previous generation i875 and i865 chipsets had a pretty tricky configuring algorithm, so that it turned out a not very easy task to squeeze the maximum performance out of the boards based on them. The launching of i925 and i915 made things much simpler due to Flex Memory technology support. In fact, the memory controller of the new chipsets can work in three modes for both: DDR and DDR2 SDRAM. Here they are:
I don’t think you seriously assume that the AGP 8x bus doesn’t provide sufficient bandwidth for contemporary graphics cards. According to our experience, all contemporary graphics accelerators store all the data they need in the local graphics memory that is why the data transfer rate along the bus connecting the graphics card with the chipset is not that important. However, Intel gave up AGP 8x bus in the new generation platforms in favor of the new and more promising PCI Express x16.
The thing is that transition to this bus is more likely to be the reflection of the industry tendencies rather than a move caused by some practical issues. For the past few years we have been witnessing the gradual replacement of the parallel PC buses with the serial ones. This way we ca not only simplify the connections configuration but also speed up the data transfer. The transition from AGP 8x to PCI Express x16 is exactly this type of transition from the parallel to serial bus. However, this transition will also bring us a bunch of positive side effects such as higher bandwidth, isolated read and write channels, etc.
Without going too deep into details I would like to note that PCI Express x16 boasts the speed of 2.5 gigatransfers per second in each direction. Depending on the bus width (in our case it is the number of data transfer channels, i.e. there are 16 of them in PCI Express x16) you can transfer from 1 to 32bits of info in each direction within a single transfer operation. Taking into account that the data is transferred along the PCI Express bus with 8/10 compressing coding involved (they use 10 bits to code 8 bits of the initial data), and that the data and instructions are transferred along PCI Express along the same signal lines, the bandwidth of the new PCI Express x16 reaches 4GB/sec in each direction, which makes the total of 8GB/sec. This way, the introduction of the PCI Express x16 bus makes the bandwidth of the bus between the graphics card and the chipset 4 times higher compared with what the AGP 8x provided us with.
Moreover, the shift to PCI express x16 also provides a number of other advantages. First of all I would like to mention the independent channels for transferring the data both ways. PCI Express x16 guarantees the 4GB/sec bandwidth for data transfers in either of the directions or in both of them. The AGP 8x bus didn’t have any dedicated channels that is why the data could be transferred either one way or another only.
PCI Express x16 slot, which will now become a common this for many mainboards based on i925 and i915 chipsets is of the similar physical size than the AGP 8x one.
You will not be able to install AGP 8x graphics cards into the PCI express x16 slot neither mechanically nor logically because these two slots use completely different protocols for data transfers. That is why you will have to get a new graphics card with PCI Express x16 interface for the new i925X and i915 based mainboards.
The major graphics chip developers, ATI and NVIDIA, have already got ready for the transfer to the new bus interface. In the nearest future we will see a lot of solutions for the new bus standard based on the chips from both companies. However, the support of the new PCI Express x16 bus will be of pretty “transitional” type still, because ATI and NVIDIA haven’t yet redesigned their chips completely for the new graphics interface.
However, ATI’s and NVIDIA’s approaches to this matter appeared radically different. NVIDIA actually provides its existing graphics chips supporting AGP interface with a new additional bridge, which will be converting data packs transferred along the PCI Express x16 bus into AGP 8x data format. NVIDIA uses an external HSI (High Speed Interconnect) chip for this purpose, which can be added to any of the already existing solutions.
ATI approached this problem from a different angle and simply replaced the interface part of the already existing chips, having introduced PCI Express x16 instead of the good old AGP 8x.
Here I have to add that in order to get PCI Express x16 bus fully supported, the drivers for the video subsystem as well as for the overall platform need to be revised thoroughly. However, this process hasn’t yet been completed. The fully-fledged support of the PCI Express x16 is most likely to appear only with the Longhorn announcement. All this results into the fact that contemporary graphics cards with PCI Express x16 interface cannot take full advantage of the new graphics bus, so that we will really benefit from this innovation much later.
However, we can notice a few improvements of the practical bandwidth of the bus between the chipset and the graphics card even now. Even PCI Express x16 graphics cards from NVIDIA with HIS bridges and AGP 8x bus “inside” the graphics cards, still can use the higher bandwidth of the new bus. To connect the graphics chip and the HSI bridge in NVIDIA based graphics cards they use AGP interface overclocked to AGP 16x. The bandwidth of this bus is about 4GB/sec, which is exactly the bandwidth of PCI Express x16 in one direction. In other words, this solution may lead to data transfer rate drops only in case of the duplex mode, i.e. when the data is transferred along the PCI Express x16 bus in both directions simultaneously. As for the conversion of the data from PCI Express into AGP format, NVIDIA claims that there is a maximum of 3-5% latency worsening in this case.
However, it is very simple to find out if these theoretical suppositions are actually true or not. We have a special utility at our disposal, which allows measuring how fast the data from the main system memory can be written into the graphics memory. Due to this small software tool developed by Andrew Filimonov, who is also the developer of our brand name Xbitmark test, we can evaluate the efficiency of PCI Express x16 implementation in ATI and NVIDIA based graphics cards. For this test we measured the data transfer rate in the i875 based platform equipped withy an AGP 8x NVIDIA GeForce FX 5900XT, and the data transfer rate along the PCI Express x16 bus of the i925X Express based platform equipped with the ATI and NVIDIA based graphics cards supporting PCI Express x16 interface. In this case we used NVIDIA GeForce PCX 5900 and ATI RADEON X600. The results of this experiment are given on the diagrams below:
As we see, the tremendous growth of the theoretical bus bandwidth between the graphics core and the chipset doesn’t result into a tremendous practical data transfer rate boost. However, I haven’t expected anything else, to tell the truth. The raw software for the PCI Express x16 bus limits the maximum bandwidth increase provided by the new bus to 40% during data transfer towards the graphics card. As for the data transfer rate in the opposite direction, we will see a more significant increase in the practical bandwidth here. Also note that ATI’s solution without the additional bridge-chip looks more attractive. RADEON X600 transfers data along the bus much faster than GeForce PCX 5900.
Here I would like to point out the following issue. Despite the fully-fledged duplex mode supported by the PCI Express x16 architecture, we see that the data transfer rate from the graphics card is considerably lower than the data transfer rate to the graphics memory by both: ATI solution with the native support of this bus as well as by NVIDIA solution with the converter bridge-chip. But this peculiarity is a distinguishing feature of the AGP bus, and it actually shouldn’t occur in PCI Express x16 solution. Therefore this strange observation may give you food for thought especially as far as the “fairness” of the native PCI Express x16 implementation in ATI chips is concerned. However, the most likely explanation in this case is either the issue with the PCI Express x16 implementation or the problem with the drivers.
However, it would still be a way too early to expect that the graphics cards supporting new bus interface would ensure any increase in the gaming performance, even if the PCI Express x16 had been properly implemented from the very beginning. AGP bus is pretty slow compared with the local graphics memory. Besides, it has been out there long enough for the game developers to arrive at an undeclared conclusion that they should avoid transferring data along this slow bus at any rate. Therefore, the graphics accelerators store as much data they might need for building the frame as possible in the local graphics memory. That is why the effect provided by the higher bandwidth of the bus between the graphics accelerator and the chipset will be minimal today. On the other hand the launching of the new graphics bus with high bandwidth and low latency has every chance to ruin the stereotypes so that in the near future software developers will no longer avoid transferring the data along the new PCI Express x16 bus. Then we will probably get a better chance to evaluate the efficiency of the new PCI Express x16 technology.
One more indirect advantage of the new PCI Express x16 bus is a more powerful voltage regulator circuitry implemented in it. It can even accept 12V power lines, and the maximum load it can bear reaches 75W. Due to this outstanding fact, many graphics cards, which have been requiring an additional power supply connector, have every chance to give it up once and for all. For example, the NVIDIA GeForce PCX 5900 and ATI RADEON X600 we tested this time didn’t require any additional power supply.
Having introduced the new PCI Express x16 bus with its new i925 and i915 chipsets, Intel gave up backward compatibility. The new chipsets do not support AGP 8x that is why most mainboards based on these sets of core logic will have no AGP 8x slot onboard and will require new graphics cards. However, some mainboard manufacturers are still going to release solutions based around i925/i915, which will be equipped not only with the new PCI Express x16 but also with the old AGP 8x slots. In this case, you should keep in mind that the AGP slot on these boards is implemented via the PCI bus, which limits its performance dramatically and affects negatively the graphics accelerator performance.
Besides the new PCI Express x16 graphics bus, Intel also introduced a new bus for regular expansion cards: PCI Express x1. However, unlike PCI Express x16, which doesn’t have any alternatives in the new i925/i915 chipsets, the announcement of the PCI Express x1 support doesn’t imply that the current PCI standard will sink into oblivion. ICH6 South Bridges, which will go with the new i925/i915 solutions will still support up to 6 PCI Master devices. They will simply acquire up to PCI Express x1 devices support. As a result, mainboards based on the new chipsets from Intel will offer different combinations of PCI and PCI Express x1 slots at the same time.
PCI Express slots are laid out on the mainboard PCB instead of the regular PCI slots, however, you will be able to distinguish between them easily: the 36-pin connector of the serial PCI Express x1 bus is much shorter than the standard PCI slot.
What are the advantages of the PCI Express x1? First of all, it is higher bus bandwidth. Unlike the regular 32bit 33MHz PCI bus, the bandwidth of the new PCI Express x1 is much higher and reaches 500MB/sec. Moreover, PCI Express x1 features point-to-point topology like any serial bus. As a result, each PCI Express x1 device receives a dedicated bandwidth of 500MB/sec while all devices connected to a parallel bus share the 133MB/sec bandwidth. Besides that, some of the PCI Express x1 advantages are coming from its architecture. Here I am talking about such things as pipelines reading or lower latencies.
Of course, those devices that feel pretty limited by the today’s PCI interface should very soon move to the new bus. Among them are gigabit network controllers, high-performance RAID controllers, etc. However, unlike graphics card manufacturers, the peripherals developers didn’t respond with the same excitement and activity, that is why the only PCI Express x1 device available today is a gigabit network controller – Marvell Yukon 88E8050.
I have to point out that the mainboard manufacturers gave this controller a very warm welcome and today you can see it integrated onto the whole bunch of mainboards based on i925X Express and i915 Express chipsets.
Since we had an i925X Express mainboard with this controller onboard at our disposal, we decided to check how fast it actually was. Let’s find out if connecting this controller to the high-performance PCI Express x1 bus is efficient at all and how the controller performance corresponds to that of the Intel 82547EI controller connected along the CSA bus with 266MB/sec bandwidth in i875/i865 systems. We ran the tests in a system with Intel Pentium 4 3.4E CPU. The performance was measured with the help of PassMark Advanced Network Test utility.
As we see, the use of PCI Express x1 bus does provide certain benefits when we have a gigabit network controller connected to it. At least Marvell Yukon 88E8050 works faster with PCI Express x1 bus than a similar chip with the PCI interface. However, controller for the dedicated CSA bus introduced by Intel for gigabit network implementation in i875/i865 chipsets does work much faster still. Despite this fact Intel gave up the CSA bus in the new i925/i915, because the network controller makers didn’t welcome it that warmly.
Intel replaced the good old AC97 sound standard with the new High Definition Audio concept in the i925/i915 chipset series. This solution is also known as Azalia. The major goal of this announcement is to offer the users a better value alternative to the expensive discrete sound cards. Therefore, the new High Definition Audio implements better quality 192kHz 24bit 8-channel sound with the whole bunch of additional advantages.
Besides the higher overall sound quality and support of 8 channels, Intel High Definition Audio supports all new audio formats including Dolby Digital 5.1/6.1/7.1, DTS ES/Discrete 6/1, DVD-Audio and SACD, etc. Besides, it allows better quality voice recording for burst transfer. However, the most interesting innovation implemented in Intel High Definition Audio appeared multi-streaming mode supporting practice it implies that different audio streams can be sent to different devices simultaneously. For instance, Intel High Definition Audio allows using some of the 8 available channels for sound playback in one application, while all other channels will at the same time be involved by another running application. With the sound system based on Intel High Definition Audio you will be able to watch digital video, while another user of your system will be able to listen to music through the headphones connected to the free jacks. And there could be even more examples like that. Moreover, you can also do it the other way around: with one sound output device involved you can play some game and use a voice chat to talk to your opponent at the same time.
Intel High Definition Audio naturally supports Jack Sensing/Retasking technology: automatic resetting of the audio connector functionality depending on the type of the device connected to it. For instance, when the microphone is connected via the loud-speaker, the system automatically shifts the mic channel to this connector, etc.
Of course, the features and high quality of the Intel High Definition Audio subsystem can become an important concept of the Digital Home initiative. However, the developers of codec used together with the new Intel’s audio system can significantly limit the functionality of Intel High Definition Audio implemented in IHC6 South Bridge. Therefore, the real mainboards may also come equipped with low-cost codecs, which will not support certain features implemented in Intel High Definition Audio in order to reduce the mainboard production costs.
The SerialATA controller integrated into the ICH6 South Bridges has also undergone certain changes. The major and most noticeable modification, which took place on transition from ICH5 to ICH6 is the increase of the number of supported Serial ATA-150 ports. If the previous generation chipset from Intel supported two Serial ATA ports, the current i925/i915 chipsets allow up to 4 Serial ATA ports. At the same time, note that the increase in the number of supported Serial ATA ports automatically led to fewer Parallel ATA ports. There is basically only one of those left now. In other words, the rapidly evolving Serial ATA standard started ousting Parallel ATA little by little, which is actually not at all surprising, keeping in mined that there are ever more storage devices supporting Serial ATA interface in the market every day.
The support of more Serial ATA channels couldn’t help telling on the functionality of ICH6R South Bridge supporting RAID arrays. Just like ICH5R, the latter also supports Raid 0 and 1 arrays, and the available four Serial ATA channels allow ICH6R building two arrays at a time. Despite our expectations, ICH6R doesn’t support RAID 0+1 arrays, because Intel engineers assume that you will hardly ever use 4 hard disk drives in a single PC system. However, Intel suggested a very interesting alternative to RAID 0+1 aka Matrix RAID.
Matrix RAID technology allows creating RAID 1 and RAID 0 simultaneously on only two hard disk drives. The idea behind this technology implies that each drive of the two is split into two parts. The first parts of both drives are used to create RAID 0 array, i.e. are used to store data which should be always available for fast access. The second halves of both disks are mirrored, that is used to build RAID 1 array, where the most valuable data are stored. From Intel’s viewpoint, the data should be stored in Matrix RAID array as follows: the first part of both drives, which is used to build RAID 0 array, should store operation system, applications and swap-file, while the second part of both drives with RAID 1 array on it should be used for user’s own files and data. This way, Matrix RAID technology ensures fast data access as well as higher data security with only two hard disk drives involved. In other words, Matrix RAID can turn out a good alternative to RAID 0+1 especially since you do not have to buy 4 hard drives to take advantage of it.
I would also like to stress that Serial ATA controller of the ICH6 turned into a fully-fledged AHCI device (Advanced Host Controller Interface). It actually appeared a prerequisite for “hot swappable” Serial ATA HDDs and for Native Command Queuing (NCQ) technology, which was borrowed by the ATA drives from the more expensive SCSI analogs. NCQ technology allows the HDD to rearrange the incoming data requests in order to reduce the latencies and increase the performance.
Only the actual device can rearrange the commands in the most optimal way, because it is the only one who actually knows the disk structure and the location of the read/write heads. That is why the implementation of NCQ requires appropriate support by the HDD, controller and driver. ICH6R and the corresponding new Intel Application Accelerator 4.0 driver do have this support. Therefore, Serial ATA hard disk drives supporting NCQ will be able to get a “free” performance boost once used with i925/i915 based mainboards.
To illustrate this discussion I tested one of the first Serial ATA hard drives supporting NCQ. This was Maxtor MaXline III. We measured the HDD performance in “real applications” with the help of PCMark04 test package. The tests were run with the old Serial ATA controller implemented in the ICH5R South Bridge, and with a new controller of the ICH6R South Bridge. All tests were performed in two modes: with a standard driver without NCQ support and with Intel Application Accelerator 4.0 driver with NCQ support.
According to the benchmark results, the Serial ATA controller integrated into the ICH6R South Bridge is faster than the controller of the previous South Bridge version from the very beginning. When we enable NCQ, ICH6 gets even better results, and the boost is quite significant I should say. The disk subsystem got about 7-10% faster in real applications once we enabled NCQ. This way, the use of this technology does really optimize the performance of Serial ATA data storage devices.
In conclusion I would like to mention one more important improvement, which appeared in the new controller of the ICH6 Bridge. Now this controller also supports ATAPI protocol, which allows using Serial ATA optical drives, for instance, with the i925/i915 based mainboards. Since ICH6 has only one Parallel ATA channel now, this innovation is of very big importance.
Together with the new Intel 925X Express and Intel 915 Express chipsets, the company also announced a few new processors from the Pentium 4 and Pentium 4 Extreme Edition families. Although the new CPUs do not differ that greatly from the predecessors from the architectural point of view, they still boast a few innovations, which have mostly to do with the marketing, I believe. The new CPUs are designed in LGA775 form-factor, which should replace the current Socket478, and are marked in a different way. Now the CPUs are no longer marked with their actual working frequency, but feature the so-called “processor number”. We are going to tell you more about these new things later today, and now take a look at the main characters of our today’s story:
Prescott CPUs. Socket 478 on the left, LCG775 on the right.
The shift to a new i925/i915 platform is very closely connected with the new LGA775 socket. According to Intel’s plans, all new mainboards based on the freshly announced chipsets should come equipped with the new processor socket. Although no one can actually prevent the mainboard makers from breaking this undeclared rule, most mainboards based on the new Intel chipsets will come with LGA775 socket. That is why together with the new chipsets announcement, Intel has also released the whole bunch of CPUs designed in LGA775 form-factor. Today this processor family includes a few Pentium 4 5XX models , which are actually none other but the regular Pentium 4 processors on the 90nm Prescott core, and a Pentium 4 Extreme Edition 3.4GHz CPU. Note that the new LGA775 family from Intel includes no budget Celeron CPUs and no CPUs on Northwood core. However, LGA775 Celeron processors are about to come out pretty soon. As for Pentium 4 on Northwood core, we are very unlikely to ever see them in LGA775 form-factor at all.
So, let’s see what Intel CPUs are already available for the LGA775 platform:
Pentium 4 Extreme Edition 3.4GHz
Pentium 4 560
Pentium 4 550
Pentium 4 540
Pentium 4 530
Pentium 4 520
You are already familiar with Prescott based processor from the previous article, which you can find in the CPU section. The new solutions announced today differ only by the form-factor:
However, I should point out that the new Pentium 4 (Prescott) CPUs should be based on the new D0 core stepping, while the Pentium 4 processors we tested a while ago were based on the C0 core stepping. The shift to the new core stepping has nothing to do with the change of the processor form-factor. The migration to the new Prescott core stepping is a pre-planned thing intended to reduce the heat dissipation and increase the frequency potential of the 90nm core. It actually allowed Intel to introduce a 3.6GHz CPU today called Pentium 4 560.
As for the Pentium 4 Extreme Edition 3.4 for LGA775, it is a total analog of the Socket478 processor, we tested not so long ago.
However, note that the use of new processor packaging for this solution did result into a slight heat dissipation increase. All in all, electrical and thermal characteristics of the LGA775 family look as follows (for a more illustrative comparison I also included the data for the Socket478 models):
Pentium 4 2.8E-3.0E
Pentium 4 3.2E
Pentium 4 3.4E
Pentium 4 XE 3.4
Pentium 4 520-540
Pentium 4 550-560
Pentium 4 XE 3.4
This way, despite all Intel’s attempts and the use of new D0 core stepping in Pentium 4 (Prescott), the heat dissipation of the LGA775 models of these CPUs got higher. However, this is not going to lead to any drama this time. Mainboards with LGA775 socket are designed taking into account all specific power consumption and heat dissipation requirements set by the new processors. Moreover, Intel has also designed a new more efficient cooling system for its new CPUs: LGS775 coolers feature new retention mechanism and are of much bigger size.
I would like to particularly dwell on the new LGA775 processor socket or the so-called Socket T. The major difference between the new LGA775 and the previous socket is a considerably bigger number of pins (the new socket has 775 pins while the old one had only 478) as well as a completely new socket design. LGA775 processors will have no common pins. Instead there are flat contact spots, which do not stand out from the bottom of the CPU. The spring contact pins are located in the processor socket. The CPU is installed and fixed in this socket by a special framing and a clip lock pressing and holding the CPU real tight for better contact between the contact spots and socket pins.
LGA775 processor socket.
CPU contact spots.
Socket pins (closeup).
However, it is much more interesting to find out what pushed Intel to switch to the new LGA775 Socket form-factor. Of course, the change made to the retention mechanism is a matter of taste. For instance, Athlon 64 FX and Opteron processors use the regular Socket 940 with 940 pins and do not suffer from any mechanical problems. So the use of the new retention mechanism is probably just Intel’s desire to make it even easier and more comfortable for the users to open and close the clip with the new larger cooling solutions used for the new and upcoming Pentium 4 processors with higher heat dissipation and the upcoming transition to the new BTX case form factor.
As for the significant increase in the number of contact pins and spots (to be more exact the number of pins by the new LGA775 got 62% bigger compared with the Socket 478) within the same processor family and same NetBurst CPU architecture, there are multiple opinions available. However, the most likely explanation is Intel’s desire to spread the electrical workload more evenly due to the possibility to duplicate certain important lines, such as power supply lines, in the first place. In other words, in each particular spot of the processor die we observe the reduction of power losses occurring on the transition from the contact spot to the transistor hidden deep inside the die. The more pins are involved under the same overall workload, the lower gets the specific load onto each particular part of the die next to the corresponding pin. As a result, the inductance and resistance in each transitional spot are lower, and the voltage deviations caused by constant switching of the status of tens of millions transistors get smoother. As a result, transistors can work at a lower nominal voltage. And lower voltage certainly implies lower power consumption.
This way, the bigger number of contact pins should help solve two major problems. First, we save some power, which we have already mentioned as one of the advantages of the new LGA775. Of course, the heat dissipation will also get lower in this case. But do not get too much excited yet: we do not save enough to reduce the heat dissipation of the current Prescott based processors. However, in the future when Prescott II based Pentium 4 processors working at 4.0+GHz will dissipate about 150W of heat, any power saving like that would be valuable. Secondly, more pins ensure higher stability of the CPUs at higher clock frequencies. Therefore, the shift to LGA775 may be regarded as a certain preparation for the Pentium 4 processors to move to a faster system bus. LGA775 CPUs are expected to support 1066MHz bus ensuring up to 8.5GB/sec bandwidth.
As for the life cycle of the new LGA775 socket, it will evidently be not any shorter than that of the Socket 478. Intel is going to release Pentium 4 processors on Prescott core at least for next couple of years. LGA775 is expected to be widely used up to 2006. And only in a little bit over two years, when 65nm Nahalem, Merom and Conroe based CPUs should come out, the desktop systems will acquire new processor socket currently known as Socket C.
Keeping in mind that unlike their Socket 478 predecessors, LGA775 processors will be marked with a new processor rating, I decided to devote a separate chapter to this interesting matter. Intel claims that the major goal of this change is to make it easier for the unsophisticated users to read the processor marking. It is true that Intel is currently offering a few processor families with radically different features. However, the traditional way of marking the CPUs with their working frequency, which is so logical for the professionals, sometimes misleads the unsophisticated users.
So, today Intel is offering four different processor families for desktops:
Of course, since there are several processor models working at the same clock frequencies, which is highlighted by many system builders as the major feature of their products nowadays, most users get really puzzled and lost. Especially, since you can often come across several modifications of one and the same CPU working at the same clock rate, but featuring different specs.
For instance, there are 6 Intel CPUs working at 2.8GHz in the today’s market. They are: Pentium 4 2.8 Northwood with 533MHz bus, Pentium 4 2.8A on Prescott core with 533MHz bus, Pentium 4 2.8C on Northwood core with 800MHz bus and Hyper-Threading support, Pentium 4 2.8E on Prescott core with 800MHz bus and Hyper-Threading support, Celeron 2.8GHz with 400MHz bus and 128KB L2 cache, and Celeron D 2.8 with 533MHz bus and 256KB L2 cache. It is very easy to get lost in this variety of processors, especially taking into account that the processors of the same family working at the same frequency differ only by one single letter after the frequency number.
This is exactly the reason why Intel Company decided to change the marking for its processors from now on in order to make it simple for ordinary users. As a result, all Intel CPUs will be marked differently: with a three-digit number, which will indicate the die architecture, the clock frequency, the FSB frequency, cache sizes and the support of additional technologies. However, this marking will be very simple and intuitive and will be easy to read even for unsophisticated users, thus revealing the CPU actual positioning in the market. You should understand that Intel’s new marking has absolutely nothing to do with the AMD’s performance rating. If the AMD’s rating is a certain reflection of the CPU performance and a few CPUs with different processor architecture can actually have the same performance rating, then Intel’s new marking scheme makes this absolutely impossible: if the CPUs differ from one another in some parameters, they will undoubtedly have different marking. However, note that this “processor number” is no technical characteristic. Also, Intel’s “processor number” has nothing to do with the CPU performance and serves to reflect purely marketing facts.
In particular, Intel processors will form three series: 7XX, 5XX and 3XX. Like in case of BMW car marking, 7XX series will be positioned as high-end most expensive solutions for enthusiast users, 5XX will be a mainstream price range, while 3XX will be targeted for the budget segment.
So far the new marking has been applied only to the relatively new processors. The older CPUs based on 0.13micron cores (such as the LGA775 modification of Pentium 4 XE) will continue with the frequency marking until their last day. It is also very interesting that the processor number will also be used only for Mobile and desktop processors. Server processors from the Xeon and Itanium families will still be marked with their clock frequencies, because Intel considers the people working with server and workstation equipment to be experienced enough and not to need any “simpler” CPU marking scheme.
Despite the fact that the new processors will be marked with the so-called “processor number”, it doesn’t at all imply that they will disregard all objective issues that have been taken into account for the older marking. In other words, besides the processor rating number there will also be its frequency, FSB frequency, cache memory size, etc. However, the rating-type of marking will certainly be the No.1 priority. The table below contains the meanings of the Intel’s processor numbers assigned to the already released and upcoming desktop processors.
2 MB L2
1 MB L2
1 MB L2
1 MB L2
1 MB L2
1 MB L2
1 MB L2
1 MB L2
256 KB L2
256 KB L2
256 KB L2
256 KB L2
256 KB L2
256 KB L2
When you look at the processor numbers and corresponding parameters descriptions, you can see that the new marking can be comparable only within each particular CPU family. It doesn’t make any sense to compare the processors with the same numbers but belonging to different processor families. That is why the CPU will be marked with the brand name and number after it, for instance: Pentium 4 530, or Celeron 335. At the same time if the two CPUs of one family are marked with two different numbers, the one with the bigger number is always better than the one with the smaller number in terms of one or more parameters. However, you shouldn’t base your buying decision on the marking only. Higher rating doesn’t at all mean that the considered CPU would be preferable for any type of tasks.
Note that Intel’s decision to shift to the processor rating marking has been in the air for a long time already. Therefore, today this measure looks pretty logical, I should say. Moreover, we appear to be unwitting accomplices of the fact that CPU clock frequency is no longer the most important thing. The CPU makers use all sorts of other tricks to speed up their babies and to enrich their functionality. You could have noticed that the working frequencies of both: Intel and AMD processors didn’t grow up that greatly during the past year or so. However, it doesn’t at all mean that the systems performance remained on the same level too. The matter is that CPU makers resorted to all sorts of different tricks to increase the performance of their solutions: they increased the FSB frequency, added more cache memory, introduced different technologies like 64bit extensions or Hyper-Threading. Later one this extensive development will continue anew. For example, we should soon see first dual-core processors composed of two processor dies hidden in one and the same package. You should also keep in mind that NetBurst architecture will last for a limited period of time. Next year Intel is going to adapt Pentium M architecture for desktop needs. It will inevitably cause a great reduction of the CPU clock frequencies, and by that time the users should already know very well that frequency is a technical parameter, which is indirectly connected with the actual CPU performance.
During this test session we will study the performance of the new LGA775 platform from Intel and will compare the speed of new LGA775 processors running on i925T Express platform with that of the Socket 478 processors running on an i875P platform. Besides, we will also compare the performance of the new Intel platform with that of the top CPUs from Intel’s No.1 competitor – AMD. For our tests we used the following configurations:
This way LGA775 processors worked in slightly different conditions determined by the peculiarities of the new platform. Thus we used different graphics cards with the PCI Express x16 interface in the LGA775 system. However, the graphics core architecture and the working frequencies of the card remained the same as those of the AGP ones. As a result, we could carry out fair and correct comparison of the performances shown by different platforms.
The test systems were built using the following hardware:
The tests were run in Windows XP SP1 operation system with DirectX 9.0b software installed.
Before we pass over to the actual benchmark results, take a look at Intel D925XCV mainboard, because we haven’t yet reviewed it.
The mainboard is based on Intel 925X Express chipset and supports LGA775 processors with 800MHz bus. Intel D925XCV is equipped with a PCI Express x16 slot, two PCI Express x1 slots and four PCI slots. For the memory modules there are four 240-pin DDR2 DIMM slots laid out in two pairs: a pair for each channel. The mainboard supports Matrix RAID, High Definition Audio and features integrated gigabit network controller onboard connected to the PCI Express bus. This way, due to Intel D925XCV we got a brilliant opportunity to test all the new exciting stuff of the LGA775 platform from Intel.
Since the launching of the i925/i915 platform brought us a completely new memory subsystem based on DDR2 SDRAM, we should first of all study its performance in synthetic benchmarks. At first we will use ScienceMark 2.0 utility with pretty rich opportunities for extensive memory testing. First of all we measured the memory bandwidth and memory latencies obtained on platforms with Pentium 4 processors. We will compare the results for the new DDr2 SDRAM with those for the regular DDR400 SDRAM. The table below contains our measurements taken from Socket 478 and LGA775 platforms running with Pentium 4 CPUs based on different processor cores but working at the same clock frequencies of 3.4GHz. Moreover, we also added here the results for Socket 939 Athlon 64, Socket 940 Athlon 64 FX and Socket 754 Athlon 64. In order to make it a fair comparison we tested AMD64 processor with 2.2GHz core clock.
ScienceMark 2.0, Memory Bandwidth, MB/s
ScienceMark 2.0, Memory Latency, cycles
ScienceMark 2.0, Memory Latency, ns
Athlon 64 3500+
Two DDR400 SDRAM channels
Athlon 64 3400+
One DDR400 SDRAM channel
Athlon 64 FX-51
Two registered DDR400 SDRAM channels
Pentium 4 3.4
Two DDR400 SDRAM channels
Pentium 4 3.4E
Two DDR400 SDRAM channels
Pentium 4 XE 3.4
Two DDR400 SDRAM channels
Pentium 4 550
Two DDR2-533 SDRAM channels
Pentium 4 XE 3.4
Two DDR2-533 SDRAM channels
The obtained results show that as we have expected, DDR2 memory features higher practical latency than DDR400 SDRAM. However, despite its higher theoretical peak bandwidth, DDR2-533 cannot boast higher practical bandwidth than the regular DDR400 SDRAM. The thing is that the 8.5GB/sec bandwidth provided by the dual-channel DDR2-533 cannot be used to the full extent by the today’s Pentium 4 processors with 800MHz bus, because this processor bus features lower bandwidth of only 6.4GB/sec. This way Pentium 4 processor will be able to enjoy all the advantages of the DDR2 memory only when they acquire 1066MHz Quad Pumped Bus support. There is not so much waiting left, actually: Intel is planning to announce these CPUs some time in Q3 this year already.
Now let’s take a look at the results our testing participants showed in the memory subsystem tests of the SiSoftware Sandra 2004 package using Stream algorithm to measure the practical bandwidth of the memory subsystem:
Sandra 2004, Memory Bandwidth Int, MB/s
Sandra 2004, Memory Bandwidth Float, MB/s
Athlon 64 3500+
Two DDR400 SDRAM channels
Athlon 64 3400+
One DDR400 SDRAM channel
Athlon 64 FX-51
Two registered DDR400 SDRAM channels
Pentium 4 3.4
Two DDR400 SDRAM channels
Pentium 4 3.4E
Two DDR400 SDRAM channels
Pentium 4 XE 3.4
Two DDR400 SDRAM channels
Pentium 4 550
Two DDR2-533 SDRAM channels
Pentium 4 XE 3.4
Two DDR2-533 SDRAM channels
Note that the systems working with DDR2-533 SDRAM are again showing lower results than systems working with DDR400 SDRAM. As a result we should only state that Intel’s desire to equip its platforms with the new memory type is more like a good move for the future, which advantages we will feel some time later. So far, the memory subsystem built with DDR2 SDRAM cannot please us with high performance in synthetic benchmarks. At the same time, I should say that it is still too early to conclude whether the use of DDR2 SDRAM slows the contemporary platforms down compared with the regular DDR SDRAM. All in all, the performance depends on the algorithms involved into work with the memory subsystem that is why systems with DDR2 SDRAM can still work faster than those with DDR400 SDRAM in certain specific applications.
I would also like to draw your attention to the fact that DDR2 SDRAM reveals its potential much better when it works with the Prescott based CPUs. It can be probably explained by the hardware and software data prefetch algorithms, which underwent serious modifications in these CPUs. Anyway, you’d better keep this peculiarity in mind. Later on we will see if this tendency is also true in real applications, or if it reveals itself only in synthetic tests.
It is far not an easy task to test the new LGA775 platforms. The thing is that the mainboards based on i925/i915 are equipped with PCI Express x16 graphics bus and are incompatible with AGP 8x. Socket 478 platforms based on i875/i865 chipsets on the contrary support AGP 8x and have no PCI Express x16 bus. That is when we compare the old and new Intel platforms we have to use different graphics cards. We tried to eliminate the influence of this factor on the systems performance by using graphics cards based on the same graphics chip in both platforms. It was NVIDIA GeForce FX 5900 with 128MB of graphics memory and 390MHz/700MHz working frequencies. However, our experiments showed that it was not enough. The thing is that NVIDIA doesn’t have an official driver for PCI Express x16. We had to use a beta driver (we used ForceWare 61.32), which are not always bug-free. Moreover, in some gaming applications these drivers have problems with the AGP 8x cards, and in some other apps – with PCI Express x16 cards. And in both cases the applications are absolutely different. Therefore, we limited the number of gaming applications we are going to run for benchmarking needs today. Here are the results obtained in those games where both: AGP 8x and PCI Express x16 graphics cards worked well:
These results show that the transition to the new platform didn’t result into any performance advantages for the Pentium 4 and Pentium 4 XE processors. If we compare the results shown by Pentium 4 3.4E on Prescott core with those of Pentium 4 550 working at the same 3.4GHz clock frequency, we will see that the CPU working in a system with DDR400 memory performs better than a similar LGA775 CPU. In fact this is not surprising at all. As is known lower memory subsystem latency matters much more for contemporary gaming applications than the high memory bandwidth. Therefore, DDR2-533 cannot yet provide the gamers with any tangible performance advantages. By the way, this is one of the reasons why Athlon 64 processors are so fast in games today. In fact the new Pentium 4 560 with 3.6GHz frequency can hardly compete with Athlo9n 64 3400+ at all.
The situation doesn’t get any better even for the Pentium 4 XE 3.4GHz, which suffered a complete disaster after shifting to the LGA775 platform with DDR2 memory. The new chipsets equipped with the DDr2 controller manage to perform much better with the Prescott based Pentium 4 processors, that is why Pentium 4 XE 3.4 appears faster than Pentium 4 (Prescott) only when it works in an i875P based system supporting DDR400 memory.
All in all, this means that gamers looking for maximum gaming performance should better forget about new Intel platforms. Moreover, processors based on AMD64 architecture can show much better performance in these applications so far.
The new SYSmark 2004 test package developed by a group of companies including Intel and AMD is pretty good at showing the actual systems performance during typical complex tasks processing. Therefore we paid special attention to the tests in this benchmark set.
Before we start commenting on the obtained results let me briefly tell you about the test set we are using:
As you see from the results obtained there are tasks where LGA775 platform with DDr2 memory works faster than Socket 478 platforms using DDR SDRAM. For example, LGA775 Pentium 4 550 outperforms Socket 478 Pentium 4 3.4E in 2D Creation subtest where video and images processing is emulated. However, in general we see that the overall picture we have already seen in gaming application repeats: high DDR2 SDRAM latency results into a significant lag of the LGA775 processors behind their Socket 478 fellows working at the same frequency.
Now let’s sum up the results on a single diagram:
The highest performance belongs to the new Pentium 4 560 processor, which is a Prescott based solution working at 3.6GHz. This CPU manages to outperform all CPUs working in an i875P platform as well as Pentium 4 XE which working frequency doesn’t exceed 3.4GHz. I have to stress that according to this test Athlon 64 processors fall behind Pentium 4 processor family. It can be explained by the fact that due to Hyper-Threading technology support Pentium 4 is a little bit batter at parallel data processing. The user model created in the SYSmark 2004 set implies that the user works in a few applications at a time, which is actually very close to reality.
This is one more test, which allows evaluating the systems performance during work in office applications and digital content creation tasks. As usual, we are offering you the results of this benchmark. Before we start let me point out that the model used in these tests implies that the user doesn’t have any applications running simultaneously, that is why the performance boost from Hyper-Threading is negligible here.
While the i875P based platform is ahead of other testing participants in Business Winstone, Content Creation Winstone pushes i925X Express with DDR2 memory to the top (if the CPUs are working at the same clock frequency, of course). However, this is not a surprise for us, as we have already seen something similar in SYSmark 2004. This way we can only confirm that the new DDR2 SDRAM with the new Prescott based CPUs is a very good solution for video and image processing tasks.
Here come the results obtained in the Winstone tests running for multi-tasking workload.
These benchmarks emulate the following situations:
As for the results, I would like to draw your attention to Test 2 in the first place, where LGA775 platform has a pretty big advantage over the Socket 478 systems. In general, this is a very rare situation and in other two tests the old platform wins the race if the CPUs are running at the same core clock. The performance of Pentium 4 560 working at 3.6GHz allows reaching higher results than in case we have a Pentium 4 3.4E with i875P chipset and DDR400 memory.
Although the shift to Prescott core has significantly increased the Pentium 4 performance in archiving tasks, the introduction of the new platform didn’t continue this tendency. Higher memory latency caused by the use of new DDR2-533 SDRAM resulted into a significant performance drop during data compression. For example, even the fastest in WinRAR LGA755Pentium 4 560 is still slower than the Pentium 4 3.4E working in an i875P based system. In 7-zip the new platform performs a bit better than in WinRAR however it doesn’t save the situation.
The reverse process, data decompression, requires high computing power from any CPU. That is why Athlon 64 is far ahead Pentium 4 in this type of tasks, and among Intel processors the best result belongs to Pentium 4 560, which is the today’s fastest NetBurst CPU as far as the core clock frequency is concerned.
During audio and video content encoding all disadvantages of the DDR2 memory in LGA775 platform become less evident. As a result the performance of LGA775 and Socket 478 platforms in these applications is about the same with a slight advantage in i875P’s favor.
The bandwidth and the memory subsystem latency hardly affect the final rendering speed. So, i925X Express and i875P perform about the same.
At the same time I would like to point out that if we compare the performance of Athlon 64 and Pentium 4 processors then it will be really hard to single out a definite leader. Depending on the type and contents of the rendering object the result can balance to the one side or to the other.
Adobe Photoshop CS 8.0 is a very popular graphics editor, which many people use for 2D graphics processing. Therefore we decided to pay special attention to the performance of our testing participants here. For the tests we used a slightly modified PSBench 7 benchmark with a 100MB image.
As an overall performance index we would like to offer you a geometric mean of the time it took our testing participants to complete a few widespread operations. This way we somehow level out the contribution of the platforms during various image processing tasks to the final performance index. The diagram below contains the results for each tested platform in seconds. Therefore, the lower is the value in the diagram the higher is the actual performance of the platform:
Here are some more detailed results for your reference. The table below shows how the tested systems coped with various Photoshop CS 8.0 filters. The time in the table is also given in seconds:
We have already mentioned multiple times in our articles that AMD Athlon 64 processors are very good at calculations. This test is just another proof to the point. All tested Athlon 64 CPUs are considerably faster than Pentium 4 processors. As for the performance of the new Intel platform, the use of DDR2 SDRAM with high memory latency does tell negatively here as well.
Summing up I would like to say that no definite conclusion can be made about the new i925/i915 platform so far. Of course, the new chipsets feature a number of indisputable advantages, such as High Definition Audio support, Matrix RAID technology and WiFi. Also among the highs of the new i925/i915 solutions I could mention the new PCI Express x1 bus, which should eliminate all bandwidth limitations imposed by the 33MHz 32bit PCI bus upon some devices.
Among the new features of the reviewed platforms are a few arguable issues, too. Among these we should certainly mention the PCI Express x16 bus for graphics cards, which has no alternative and offers the users no choice. Of course, higher bandwidth of this bus is a true plus, but contemporary graphics cards can still be quite happy with the good old AGP 8x. Anyway, the progress keeps going and the graphics cards are very likely to learn to use all the advantages of the new PCI Express x16 in the future. I am only upset that the new chipsets are incompatible with the widely spread cool AGP 8x graphics solutions available in the today’s market.
A definite drawback of the new i925/i915 chipsets is DDR2 SDRAM support. So far this memory with high bandwidth and higher latency than that of DDR400 SDRAM cannot guarantee the new platforms any performance gain at all. In fact, DDR2 support is a serious slow-down for i925/i915, so that these chipsets show pretty low performance compared with their predecessors. I will not deny that DDR2 SDRAM is a promising solution for the future. Of course, later on when the processor bus speeds up and the DDR2 memory specs get improved and polished off, this memory type will become truly demanded. But today using DDr2 memory in Pentium 4 platform doesn’t make much sense.
The graph below shows how greatly the performance drops when we move from i875P to i925X Express in systems with Pentium 4 (Prescott) and Pentium 4 Extreme Edition (the CPUs work at the same frequency).
i925X owes these dramatic results to the DDR2 memory in the first place. Luckily, i915 Express chipset family targeted for the mainstream systems is backward compatible with the DDR400 SDRAM. Hopefully, this fact allows i915 to compete successfully with i865. However, we still have to check if our expectations are justified.
At the same time I would like to point out that Prescott processor cores work with the new chipset much better than Pentium 4 XE based on the Northwood/Gallatin core. Of course, DDR2 SDRAM looks better when there are CPUs on the latest cores involved, as they feature enhanced data prefetch algorithms.
Together with the new chipsets Intel also introduced a new processor socket form-factor: LGA775. On the one hand, this transition allows Intel to keep increasing the clock frequencies of its processors as well as that of the system bus without ay concerns, while on the other hand it means that the new CPUs will not be compatible with the old mainboards, as well as the old CPUs will not be able to run in the new mainboards.
All in all I have to say that such a great lot of innovations Intel introduced in its new i925/i915 chipsets deprive the users of the upgrade opportunities completely. In particular, when the users will go from the old platforms to the new ones, they will have to replace not only the mainboard and the CPU, but also the graphics card and the memory. Besides in many cases the shift to i925/i915 platform will also require new hard disk drives with SerialATA interface and some users may also need to replace a few peripherals with their PCI Express x1 analogies. This way, the launching of the new i925/i915 solutions is not just a technological breakthrough but also a perfect way of getting more money from the users.
Therefore we observe quite a paradox. Having upgraded the platform and added a lot of innovations, some of which are really handy, Intel at the same time pushes users to replace their CPUs, memory and graphics cards, even though replacing all this equipment will hardly have any positive effect in the long run. I believe that there are hardly that many arguments in favor of the new platform this way. Especially since Intel is going to release new CPUs with faster bus, larger L2 cache and 64bit extensions in the nearest future. When this happens maybe i925/i915 platform will start looking more attractive to us. And in the meanwhile the good old i875/i865 solutions, which have already stood the test of time, are perfectly suitable for our systems. The time for i925/i915 hasn’t come yet.