ASUS SK8N + AMD Opteron 144: Uniprocessor Workstation on AMD Opteron and NVIDIA nForce3 Professional

by Ilya Gavrichenkov
07/21/2003 | 05:53 PM

The launch of new Opteron processor family from AMD aroused significant changes in the processor market. Of course, it is still too early to talk about wide spreading of Opteron processors, however, the public has been following very carefully everything that has to do with the new AMD Opteron CPUs.

Dual-processor servers built on AMD Opteron processors have already been tested by a great lot of different web-sites all over the world and many reviewers arrived at the conclusion that the newcomer from AMD suits very well for server applications. However, AMD positions its product not only for server use. They also claim that Opteron can find its way in workstations. Here we have to say that Opteron can boast pretty nice computational power, as it is a direct successor to Athlon architecture. It is exactly due to its computational capacity, that the new Opteron can turn out useful for complex calculations as well as for professional OpenGL applications and rendering tasks. Anyway, we wouldn’t deny the possibility to use Opteron in workstations.

If you take a look at the screenshot taken from AMD’s official materials, you will see that uniprocessor workstations are also considered to be one of the target markets for AMD Opteron processors.

As we see, AMD positions Opteron as a competitor to Pentium 4 in the Low-End uniprocessor workstations. However, we have to point out that right after the processors came into the market, there turned up some problems with the practical use of Opteron CPUs in workstations. The thing is that there appeared to be no mainboards in the market, which would support both: the new AMD Opteron CPUs and AGP graphics cards at a time. The major problem with this shortage was connected with the only chipset for Opteron processors available in the market in those days. It was AMD-8000, which simply didn’t allow installing external AGP graphics cards. To be more exact, this chipset did have an AMD-8151 tunnel, although AMD decided to postpone the supplies of this chip for some reasons. As a result, the mainboards based on AMD-8000 were deprived of the AGP port and hence they couldn’t be used in graphics workstations. This is actually the reason why AMD is now focusing on promotion of Opteron processors for web-servers and database servers, which do not need any high-performance graphics accelerators.

However, NVIDIA decided to help AMD out. NVIDIA has been long known as AMD’s powerful ally. For example, nForce2 chipset for AMD Athlon XP platform is the today’s fastest and most popular solution. So, NVIDIA decided to help AMD to get into the uniprocessor graphics workstations market and introduced their chipset for Opteron processors. This chipset is the new nForce3 professional, which was officially launched together with the one-way AMD Opteron CPU version belonging to 1XX family. In other words, we can state that the new generation uniprocessor graphics systems started coming into the market.

In this article we are going to figure out what the uniprocessor Opteron based systems are capable of, what the peculiarities of the new NVIDIA nForce3 Professional chipset are, what is typical of mainboards based on this chipset and how fast the uniprocessor workstations with AMD Opteron inside will be.

Closer Look: AMD Opteron 1XX Processor

To tell the truth, I don’t think we need to talk a lot about Opteron processors of the one-way 1XX family. These CPUs have no principal differences from their dual-processor fellows. That is why if you would like to learn more about Opteron architecture and AMD64 technology (which is also known as x86-64), go and check our article called A Glance at the Future: AMD Hammer Processors and x86-64 Technology. On my part, I would like to remind you briefly of the main differences between the new Opteron processor and the widely spread Athlon XP.

Opteron architecture is generally based on Athlon architecture. The major difference between Opteron and the predecessors is the support of AMD64 technology, which allows performing 64bit code while retaining full hardware compatibility with the today’s applications. Actually, that is why the new processors are called eighth generation processors. The implementation of 64bit modes in the new Opteron CPUs caused not only the enlargement of the address space, but also doubling of the general purpose registers with the corresponding growth of their width up to 64bit. However, you will be able to take real advantage of all benefits of 64bit modes only if there is appropriate software supporting AMD64 architecture. However, if you use traditional software in an Opteron system, these innovations will not be involved at all and you will see no effect.

Note that the support of AMD64 technology started expanding into the server field very rapidly. There are at least several Linux clones for Opteron processors and some other server software available today. As for the use of Opteron processors in workstations, the situation here is a little more complicated. In fact, no workstation software has been optimized for AMD64 architecture yet. There is even no operation system from Microsoft supporting 64bit mode of Opteron processors. However, the situation is not completely hopeless. For example, the NUMA memory architecture (Non-Uniform Memory Access), used in multi-Opteron systems, is already supported in Windows 2003 Server.

This way, Opteron processors are most likely to work in their regular 32bit mode in top graphics workstations. However, even in this case the new AMD Opteron CPUs can boast a number of enhancements, which will allow them to outperform AMD Athlon XP working at the same core frequency. Among these enhancements we would like to point out the following:

• Memory controller integrated into contemporary Opteron processors supporting dual-channel DDR333 SDRAM;
• Separate Hyper-Transport buses between the CPU and the chipset implementing the AGP 8x support and input/output functions and serving to build up multi-processor configurations;
• Increased 1MB L2 cache featuring higher bandwidth (which might be achieved due to the wider 128bit bus between the cache and the processor core);
• Fully-fledged SSE2 instructions support;
• Longer 12-step integer pipeline and 17-step floating-point pipeline (by Athlon XP these pipelines are 10 and 15 steps long, respectively). This improvement primarily serves better architecture scalability.
• Larger integer scheduler buffers (24 entries against 18 entries by Athlon XP);
• Larger TLB. The TLB of L1 instructions cache has been increased from 24 to 40 entries and the TLB of L2 cache grew up to 512 entries (while Athlon XP has 256 entries);
• Enhanced branch prediction algorithm with the history chart increased up to 1600 entries (Athlon XP has only 400 entries).

As we see, the undertaken architectural changes are quite significant, especially against the background of Athlon XP processor. Moreover, most of these improvements will definitely have a positive effect on the performance of Opteron processors used as a basis for workstations. Here we should first of all stake faster memory subsystem performance together with faster and larger L2 cache, but we shouldn’t also disregard the more in-depth architectural changes. I would like to single out SSE2 instructions support: most today’s 3D rendering applications, such as Lightwave 7.5 are currently optimized for this particular instructions set.

Opteron has also undergone some exterior changes, which made it even more attractive for high-performance workstations. First of all, I have to draw your attention to the package cover, which protects the fragile die against damaging when the coolers are installed on top of it. This time, AMD couldn’t do without this cover, because the Opteron dies are made with SOI technology and are extremely fragile. Moreover, besides the built-in thermal diode, Opteron has also acquired a special anti-burn protection circuit called Thermtrip. Although there is nothing really exclusive behind this name: when the CPU reaches some critical temperature, it simply shuts down.

I also have to stress that in its Opteron processor AMD paid special attention to memory error code correct, which is very important for server use in the first place. The memory controller integrated into the CPU supports parity control and ECC, also ECC is implemented in the L1 and L2 caches. By the way, the ECC algorithm in Athlon XP processors was implemented only for L2 cache, while L1 cache supported only parity control scheme.

According to the marking nomenclature used by AMD for Opteron processor family, there are a few product lines within this family now: Opteron 14x, 24x, 8xx. Note that Opteron 14x doesn’t work in two-way systems, Opteron 24x can be used in max. 2-way systems, and Opteron 8xx is intended for 4- and 8-way configurations. These differences between three Opteron lines are very illustratively marked by their pricing. For instance, a uniprocessor Opteron modification costs about the same as the top Pentium 4 CPU, which once again indicates that uniprocessor Opterons are intended to compete with the top Pentium 4 models in the workstation market.

The architecture of Opteron processors implies the HyperTransport controller supporting three corresponding buses with up to 6.4GB/sec bandwidth each. These buses serve as a system bus and inter-processor bus. However, you should understand that you do not need three HyperTransport buses if there is a uniprocessor Opteron system. Since the configurations with only one Opteron processor do not need any inter-processor connection, only one HyperTransport is involved. This bus connects the CPU and the chipset. Here we should certainly say that Opteron 1XX processors are shipped with only one fully-functional HyperTransport bus, which explains their considerably lower price compared with Opteron 2XX and especially with Opteron 8XX. This is exactly why you will never be able to use low-cost Opteron CPUs in multi-processor systems.

In conclusion to my story about uniprocessor Opteron CPUs, I would like to dwell on the differences between this processor and the upcoming Athlon 64 also based on the 8th generation architecture, which is due in September 2003. In fact, the differences are not so numerous. There will be no architectural differences. In particular, you will not be able to use Athlon 64 in multi-processor configurations, because this processor, like Opteron 1XX, will have only one fully-functional HyperTransport bus. This way, the only distinguishing feature between these two processors will hide in the memory controller. While Opteron 1XX supports dual-channel DDR333 SDRAM with ECC, Athlon 64 will feature simpler memory subsystem. The memory controller of the future desktop processor from AMD will be able to work with only one memory channel and will not support ECC, however, it is very likely to acquire DDR400 SDRAM support.

Closer Look: NVIDIA nForce3 Professional Chipset

Only the announcement of the NVIDIA nForce3 Professional core logic helped uniprocessor systems based on AMD Opteron processor to come into the market. AMD-8000 chipset offered by the company as a solution for the Opteron CPU, is not used in uniprocessor systems for some reason. And the launch of NVIDIA nForce3 Professional, which is targeted exactly at uniprocessor workstations, pushed mainboard guys to start releasing their solutions for Opteron 1XX processor family.

Besides, Opteron based systems owe NVIDIA nForce3 Professional the possibility to use AGP slot for high-performance graphics subsystem. As a result, we got an excellent platform for high-performance graphics workstations based on 8th generation AMD processors.

As a chipset developer, NVIDIA decided to support AMD processors long time ago and has already made good progress. Of course, the company couldn’t miss out the so long-awaited Opteron processors. Moreover, NVIDIA is also very much interested in bigger sales of the graphics workstations, because this company also offers Quadro graphics accelerator family positioned for this particular market. As a result, they rolled out nForce3 Professional chipset, which has every chance to become a perfect basis for the graphics workstations as well as for home systems of computer enthusiasts.

Having applied all the experienced the company acquired when working on the nForce2, NVIDIA managed to implement a bunch of new cool solutions in their nForce3 Professional. First of all, I would like to point out that NVIDIA nForce3 Professional is a single-chip core logic set. Since Opteron features its own memory controller, nForce3 doesn’t have any. As a result, the chipset North Bridge became very simple: the dual-channel memory controller with the DASP, which they used in nForce2 was no longer needed.

Besides, the integrated graphics core also disappeared from the chipset North Bridge, because this component is absolutely unnecessary in the graphics workstations market. As a result, this nForce3 North Bridge contains only one AGP 8x controller and HyperTransport bus controller, and thus occupies very little on-die space. This fact, as well as the innovative 0.15micron production technology used for nForce3 Professional manufacturing allowed NVIDIA to fit the South Bridge onto the same die with the North Bridge. Compared with the South Bridge of nForce2, the South Bridge of nForce3 Professional boasts a few new functions. Note that integrating both chipset Bridges onto a single die allows reducing the latencies during the work of input/output devices with the CPU and the memory. In this case, when the memory controller is integrated into the CPU, removing the bus between the chipset bridges means the system gets rid of a serious bottleneck. You will feel the difference when working with a network and various external devices.

NVIDIA now offers the nForce3 chipset version aka nForce3 Pro 150. The chipset South Bridge (which is located on the same die with the North Bridge) contains NVIDIA 10/100 MAC (which is almost identical to the network controller used in nForce2), three-channel IDE controller, PCI controller, USB 2.0 controller, and supports AC’97 audio (nForce3 doesn’t have NVIDIA’s brand name APU). Of course, the absence of IEEE1394 interface support implemented in MCP-T of the nForce2 chipset catches your eye at once, but NVIDIA decided to give it up this time. Moreover, we were a little bit upset to find out that the PCI controller of nForce3 Professional supports only 33MHz 32bit bus and doesn’t allow using faster PCI devices, which could be of interest to high-performance workstation users.

However, the South Bridge of the nForce3 Professional has its indisputable advantages, too. Its most interesting peculiarity is the IDE controller. In the current chipset version it supports three ATA/133 channels and allows connecting up to 6 IDE devices. Moreover, the third controller channel can also support two SerialATA-150 HDDs if the mainboard is equipped with two IDE-to-SATA Bridge chips from Silicon Image. NVIDIA didn’t simply increase the number of supported hard disk drives. nForce3 Professional supports RAID as well. The current South bridge version supports RAID 0, 1 and 0+1 arrays, which can be built of any ATA HDDs connected to the controller integrated into the chipset.

Here are the formal specifications of the NVIDIA nForce3 Pro 150 chipset:

• HyperTransport bus with 3.6GB/s bandwidth for connection with the CPU;
• AGP 8x;
• Up to six 32bit PCI slots working at 33MHz;
• Triple IDE controller supporting three ATA/133 channels;
• 0, 1, 0+1 RAID arrays;
• Fast Ethernet 10/100Mbit/s  network controller;
• Six USB 2.0 ports;
• AC’97 version 2.1.

This fall NVIDIA is going to enlarge the features list for its nForce3 Professional by enhancing the integrated chipset South Bridge. The new chipset version for Opteron processors will be called nForce3 Pro 250. The enhancements will touch upon the network and IDE controllers. The network controller will support up to 1Gbit/s connections. The IDE controller of nForce3 Pro 250 will boast its own support of four SerialATA-150 channels and two Parallel ATA/133 channels, so that the nForce3 Pro 250 based systems will support up to 8 HDDs. The chipset will certainly retain the support of 0, 1 and 0+1 RAID arrays.

Closer Look: ASUS SK8N Mainboard

The first and the only manufacturer of mainboards based on the new NVIDIA nForce3 150 appeared ASUS Company.  NVIDIA chose ASUS to be their major partner in nForce3 Professional promotion not for nothing. ASUS and NVIDIA have been partnering for a long time, besides, ASUS is a pretty well-known manufacturer of server platforms. Anyway, in the near future ASUS will no longer be a monopolist of the nForce3 based solutions, as other mainboard guys will also start rolling out their nForce3 based mainboards.

So far, the only nForce3 based mainboard available in the market is ASUS SK8N.

Let’s check the product specifications first:

Now let’s discuss the features of ASUS SK8N in a bit more details. First of all, I would like to make a few things clear about the processors supported by the mainboard. Since this mainboard is intended for Opteron CPUs, it is equipped with Socket940. Note that there are three modifications of Opteron processors in the today’s market: Opteron 1XX for uniprocessor systems, Opteron 2XX for dual-processor systems and Opteron 8XX for four- and eight-way systems. All these processors are designed in the same form-factor: Socket940, and any Opteron modification can be installed into ASUS SK8N mainboard. However, it hardly makes any sense to use Opteron processors other than 1XX with this board, because they will offer you the same functionality but for much more money, as these CPUs are targeted for the multi-processor configurations.

For the memory modules there are four 184-pin DDR DIMM slots on ASUS SK8N located in pairs. Since the Opteron memory controller boasts dual-channel architecture each pair of memory slots is assigned to a certain channel. That is why if you want your ASUS SK8N platform show the maximum performance, you’d better use pairs of identical memory modules installed symmetrically into DIMM slots corresponding to different channels. Besides, you should remember about one more thing. Since Opteron processors are optimized for server use, they require Registered memory modules with ECC. ASUS SK8N uses the memory controller integrated into the Opteron processor, That is why only the CPU sets requirements to memory. ASUS SK8N simply will not work with common DDR modules: you have to get Registered DDR333/DDR266 memory modules. So if you decide to build your system on ASUS SK8N mainboard (a hardcore gaming system, for instance), you have to make sure that you have proper memory modules.

Registered DDR333 SDRAM from Transcend

I would also like to point out that the four DIMM slots available onboard allow installing maximum 8GB of memory into ASUS SK8N, although the address space supported by processors based on AMD64 architecture is much bigger.

As for other expansion slots, ASUS SK8N features an AGP 8x slot and 5 absolutely common PCI slots. Although this mainboard is targeted for graphics workstations, ASUS didn’t implement the AGP Pro slot, which is required for some professional graphics cards, such as the 3DLabs solutions, for instance. That is why you will have to choose a graphics accelerator, which doesn’t need an AGP Pro slot, if you decide on an ASUS SK8N based system. Among the available choice will be the solutions from NVIDIA Quadro family. As for the PCI slots, the limitations imposed by the NVIDIA nForce3 Pro 150 chipset make them nothing special: 32bit and 33MHz. that is why you will not be able to use some advanced controller cards on ASUS SK8N mainboard, too. This indicates once again that ASUS SK8N can hardly be regarded as a good choice for the highest-end graphics workstations.

Although NVIDIA nForce3 Pro 150 used in ASUS SK8N supports 3 ATA/133 channels, only two of them are laid out on the board. To implement the support of more hard disk drives, ASUS used an external IDE RAID controller from Promise – Promise PDC20378. This controller features two SerialATA-150 channels and one ATA/133 channel. This way, you can connect maximum 8 hard disk drives to ASUS SK8N mainboard. The onboard Promise PDC20378 controller and the nForce3 Pro 150 chipset allow building 0, 1 and 0+1 RAID arrays from hard disk drives connected to them.

Due to nForce3 Pro 150 chipset, ASUS SK8N features 6 USB 2.0 ports. Four of them are located on the rear panel and two mode ports are implemented via the onboard connector. Besides, ASUS thought it would be also good to have FireWire onboard and equipped its SK8N with TI TSB43AB22A IEEE1394 controller. As a result we can also see an IEEE1394 connector on the rear panel of ASUS SK8N mainboard. one more IEEE1394 port is implemented as an onboard connector.

Despite the fact that ASUS has been equipping all its new mainboards with Gigabit Ethernet controllers lately, ASUS SK8N features only 10/100Mbit Fast Ethernet. They used 10/100 MAC integrated into NVIDIA nForce3 Pro 150 chipset and a physical Realtek RTL8201BL controller. Faster Ethernet will be implemented in the ongoing chipset version aka nForce3 Pro 250 that is why the upcoming ASUS mainboards for Opteron processors will most probably have Gigabit Ethernet as well.

ASUS SK8N is also equipped with AC’97 sound chip. Since this mainboard is targeted for workstations, ASUS decided not to provide it with any advanced sound solutions and used a pretty simple but efficient Realtek ALC650 codec.

The PCB design and functioning of ASUS SK8N didn’t cause us any problems. The electronic components are located pretty cleverly on the mainboard PCB. All the slots available on this board are located in parallel to one another, which is a little unusual for us.

Note that you don’t need any special power supply units for a system built on ASUS SK8N mainboard. It works perfectly fine with the common PSUs complying with the ATX 2.03 specification and featuring an additional 4-pin 12V power cable.

I would like to draw your attention to the heatsink cooling down the chipset on the ASUS SK8N mainboard:

Although nForce3 is a pretty complex chipset containing both: North and South Bridges, it dissipates very little heat. The thing is that nForce3 Pro 150 appeared the first core logic set manufactured with 0.15micron technology. the result of this innovation is evident: a tiny passive aluminum heatsink is more than enough to ensure proper cooling of this chipset.

Opteron Overclocking on ASUS SK8N

Strange as it might seem, but ASUS SK8N mainboard offers some overclocking-friendly functions. Although this product is strictly oriented for workstations, ASUS couldn’t help implementing a few brand name opportunities for CPU overclocking.

BIOS Setup of ASUS SK8N is based on AMI microcode and is very similar to BIOS Setup of the newest ASUS mainboards on i875 and i865 chipsets. The BIOS Setup has a special page containing major overclocking tools and the possibility to change FSB frequency and processor Vcore.

The clock frequency multiplier seems to be locked in Opteron processors. At least, ASUS SK8N doesn’t allow adjusting it. As for the FSB frequency, it can be increased from the nominal 200MHz up to 300MHz with 1MHz increment. Also ASUS SK8N lets you increase Opteron’s Vcore from the normal 1.55V up to 1.65V or even 1.75V.

A separate page of the Setup is devoted to Vdimm adjustment: you can raise this parameter from the nominal 2.5V to 2.6V or 2.7V.

The AGP bus frequency is set separately, and can vary from 66MHz to 77MHz with 1MHz increment. Also BIOS Setup of ASUS SK8N increasing Vagp from 1.5V to 1.6V or 1.7V.

We decided to overclock our Opteron 244 working at the nominal 1.8GHz frequency with the help of the available options (the multiplier was set to 9x, the FSB frequency equaled 200MHz).

Having increased the processor Vcore up to 1.65V we managed to reach 218MHz FSB frequency without any stability losses. However, we couldn’t raise the FSB frequency any higher, because the system turned unstable and the tests run in the OS caused system freezing or BSOD. So, the maximum frequency we managed to squeeze out of our Opteron 244 was 1962MHz.

This result can hardly be called impressive. However, it is pretty logical. Opteron 244 is the top processor model so far, so that further frequency increase is certainly limited by some AMD’s problems connected with the implementation of SOI technology. That is why the today’s Opteron processors are very unlikely to be overclocked to 2GHz. At the same time, we should give proper credit to ASUS and we should admit that ASUS SK8N is the only mainboard, which allows overclocking AMD Opteron processors, and it copes with this task pretty well, I should say.

Also I would like to draw your attention to the monitoring options offered by ASUS SK8N. This mainboard allows not only checking the temperature with the help of the built-in processor diode. It also measures the temperature inside the system case, tracks rotation speeds of three fans and controls major voltages produced by the PSU. Moreover, the mainboard supports ASUS Q-Fan technology reducing the noise level by slowing down the processor fan rotation speed depending on the CPU temperature. As you can see on the screenshot below, ASUS SK8N supports three-level Q-Fan. Moreover, you can manually adjust those critical temperatures, when the cooler shifts from one mode to another.

BIOS Setup of ASUS SK8N helps to manage the memory controller built into AMD Opteron processor. There is a special page with all settings necessary for that. Here you can adjust the memory frequency (DDR333/DDR266/DDR200), enable banks alternation and set the key timings of the memory subsystem. You can also disable ECC, although this will not eliminate the necessity to use Registered memory modules.

Testbed and Methods

In this review we decided to figure out how fast can AMD Opteron run if it is used in a uniprocessor workstation. For our tests we assembled a system with ASUS SK8N mainboard and Opteron 144 processor working at the actual 1.8GHz frequency. Note that 1.8GHz is the today’s top core clock for AMD Opteron processors. That is why we decided to compare the performance of our today’s Opteron processor with that of top Pentium 4 and Athlon XP models.

The tests were run in Windows XP operation system, that is why w can’t talk about the effect achieved due to x86-64 technology of Opteron processors. Unfortunately, most professional applications work in this particular operation system, that is why we decided to refrain from installing any Linux closes on our testbeds.

Together with the tests in professional applications, we decided to check Opteron’s performance in classical tasks: gaming and office applications. In this case, the detailed study of Opteron performance will let us draw some preliminary conclusions about the performance of the upcoming Athlon 64 processor family intended for common desktop systems.

We tested Opteron with two and only one memory channels involved. Single-channel configurations, where Opteron doesn’t show everything it is capable of, are very interesting to us, because this way we can at least evaluate the performance of Athlon 64 processors, which will feature single-channel memory subsystem.

Comments:

• The memory was used in the fastest mode with the most aggressive timings set to 2-2-2-5.
• We ran the tests in Windows XP SP1 operation system with the installed DirectX 9.0a.

Performance: Synthetic Benchmarks and Memory Tests

Before we pass over to fully-fledged testing of a workstation based on AMD Opteron Processor, we decided to evaluate the performance of the dual-channel memory controller integrated into it. I wonder if the memory controller built into the Opteron processor will be able to outperform the latest dual-channel memory controller used in Intel’s i875P chipset. For this purpose we resorted to the means that stood all tests of time: the good old Cachemem benchmark:

As we see from the memory bandwidth numbers during various operations, the leadership here belongs to Intel Pentium 4. Actually, no wonder. The thing is that there is dual-channel DDR400 SDRAM used together with the Pentium 4 processor, while the fastest memory supported by AMD Opteron is DDR333 SDRAM. And it is not AMD engineers who are to blame for any sluggishness here. Opteron processor is intended primarily for server systems and high-performance workstations. That is why AMD decided to make its solution support only Registered memory modules with ECC. These modules are, firstly, more reliable, and secondly, can be of bigger capacity. However, unfortunately, there are simply no Registered DDR400 SDRAM DIMMs in the market. As a result, the maximum theoretical bandwidth of the memory bus in Opteron systems can reach only 5.4GB/sec yielding to Pentium 4 systems with 800MHz bus, which boast up to 6.4GB/sec peak bandwidth. At the same time, we can’t help saying that the use of integrated memory controller did have a positive effect for AMD. The memory latency in Opteron based system appeared the lowest of all.

We would also like to point out very good results obtained when only one memory channel of the Opteron based system was involved. As we see, the operations with the memory are not any slower than by Athlon XP 3200+ with dual-channel DDR400 SDRAM and 400MHz system bus.

Anyway, we can state that when processing large amounts of data a workstation built on Opteron processor from AMD would definitely show its best due to highly efficient memory controller and 1MB big L2 cache with exclusive architecture.

Performance: Data Encoding

In this section we are going to check how well a workstation built of AMD Opteron processor and ASUS SK8N mainboard will cope with data encoding and streaming data processing (audio and video). Although this is not a very typical task for a workstation, we couldn’t leave it out.

We have already pointed out more than once that high-speed ALU is required to ensure high-speed MP3 encoding. That is why Pentium 4 with its powerful Hyper-Threading technology wins this test. Opteron working at a not very high core clock frequency of only 1.8GHz, even falls behind the fastest Athlon XP processors.

During data compression with the help of some ordinary archiving utilities, the memory subsystem performance affects the results most of all. Therefore, Opteron with the built-in dual-channel memory controller and 1MB L2 cache almost catches up with the fastest Pentium 4 models, even despite its relatively low core clock.

Canopus ProCoder used for professional videos digitizing, has always run very well on AMD processors. This proves true in our case, too. Although we have to stress that Opteron performs slower than the top Athlon XP models because of its lower working frequency.

CPUs with NetBurst architecture cope perfectly well with video encoding into MPEG-4 format. Besides, Hyper-Threading technology also improves the result shown by Pentium 4. As a result, AMD processors appear in the very end of the race falling quite noticeably behind the top Pentium 4 models.

We see very similar picture when using Windows Media Encoder 9. A workstation based on AMD Opteron fails to show its best in this application, too.

As a result, we can conclude that Opteron based systems do not cope very well with data encoding tasks, especially against a colorful background of more successful competitors. In fact, the situation has hardly changed compared with what Athlon XP showed us during the tests. This is pretty logical, of course, as both CPUs have very similar architectures. As for Intel, they optimized their CPUs for streaming data processing tasks from the very beginning, that is why we are not at all surprised to see how high the results shown by Pentium 4 processor are.

Performance: Scientific Calculations

To test the performance of an Opteron based system in scientific applications, we resorted to ScienceMark 2.0 test, which shows how quickly the typical mathematics modeling algorithms can be processed.

Athlon XP is fairly considered to be the best processor for scientific calculations due to its powerful three-pipeline FPU unit. By Opteron this unit remained almost unchanged. However, the core clock rate got considerably lower than that of Athlon XP processors. As a result, Opteron falls behind Athlon XP in all tests, and in every two tests out of three it also falls behind the top Pentium 4 processor model. All in all, Opteron appears not the best choice for scientific calculations today. At least in those tasks, which do not take advantage of AMD64 technology.

Besides ScienceMark 2.0 test, we decided to check the performance of the new processor in a distributed computing client of the Russian MD@home project, calculating the dynamic features of olygopeptides (protein ferments) with the help of molecular dynamics methods.

Again there is nothing optimistic about the results of AMD Opteron. This processor definitely lacks core frequency. However, some production problems still do not allow AMD to increase the core clock rate of its Opteron processors at least up to the level of top Athlon XP models.

Besides the mathematical modeling tasks, ScienceMark 2.0 test package also contains one more interesting test aka BLAS. This benchmark, just like Linpack test, measures how quickly large floating-point matrices can be multiplied. However, in this case there are a few code versions compiled with various optimizations, which allows evaluating the efficiency of various processor units for this or that case.

We have already mentioned several times that AMD processors are fast at processing x87 operations. The three-pipeline FPU of Athlon XP and Opteron processors helps them to win in benchmarks dealing with floating-point numbers.

As for the SSE2 unit, which appeared in Opteron processors, it turns out not very fast, especially in vector operations. However, as soon as it comes to scalar operations, SSE2 of the Opteron processor helps it to defeat more than twice even the fastest Intel Pentium 4 CPUs. To be fair, we should say that scalar SSE2 operations do not occur very frequently.

Performance: Photoshop

Since Photoshop is one of the applications used quite often in graphics workstations, we decided to pay special attention to it. For our tests we used a popular Psbench 1.11 benchmark working with an 80MB test image.

The optimization of Photoshop 7.0 for Pentium 4 processors and their Hyper-Threading technology do not leave even a single change for AMD Opteron. Even the SSE2 instructions support implemented in the Opteron CPU and large cache-memory, don’t save it from failure. The detailed results of this benchmark are given in the table below. The table contains the time in seconds it took each testing participant to complete the filter, that is why smaller numerical values indicate better result:

Well, for AMD fans, the results are far from pleasing. In 18 filters out of 21, Pentium 4 CPU appears faster. Opteron processor fails to win even a single time.

Performance: Professional OpenGL Applications

To evaluate the performance of our Opteron based workstation we took a widely spread SPECviewperf 7.1 benchmark showing the performance of our testing participants in professional OpenGL application packages.

Pentium 4 processors working at 3.0GHz and 3.2GHz manage to leave the rivals far behind in these tests. Athlon XP and Opteron sway back and forth struggling for the second position.

Performance: Rendering

As soon as Pentium 4 processors acquired Hyper-Threading support, their performance in 3ds max 5.1 final rendering grew up significantly. Neither Athlon XP, nor the new Opteron can compete with the top Intel Pentium 4 models.

In Lightwave 7.5 the situation is a little different. This application uses SSE2 instructions very actively that is why Opteron processor supporting this instructions set almost catches up with the fastest Pentium 4 processors.

However, SSE2 is not so actively used for rendering of all scenes. Therefore, sometimes Opteron may yield quite tangibly not only to Pentium 4 but also to Athlon XP processor.

We can conclude only one thing here: either the working frequencies of the AMD Opteron processor will grow up, or the corresponding applications will finally get optimized for AMD64 technology. Otherwise, the users will have to decide on other processors for rendering tasks.

Performance: Office and Content Creation Applications

To test our systems in regular office applications we used a standard Winstone test set.

This is where a lot of advantages of AMD Opteron processors, such as integrated memory controller and large L2 cache, finally come to light. As a result, Opteron 1.8GHz outperformed Athlon XP 2.2GHz in both tests. Everything indicates that the upcoming Athlon 64 processor family, which is to come into desktop systems, will become a worthy successor to Athlon XP family.

Performance: 3D Games

Well, I don’t think that many of you will decide to use a graphics workstation based on Opteron processor to play 3D shooter games. Nevertheless, we tested our Opteron system in popular games in order to be able to forecast how well the upcoming Athlon 64 will cope with this task, since this processor will be based on the same architecture as the today’s Opteron CPU.

Strange as it might seem, but Opteron processor proved the best in gaming applications. Large cache-memory and fast system memory allow Opteron to defeat Athlon XP in all tests and even to outperform Intel Pentium 4 3.2GHz in some cases. All this indicates that the future Athlon 64 processors should become perfect gaming solutions, especially since AMD is going to raise the core frequencies of its K8 processors by then.

Conclusion

So, we have just taken a close look at a uniprocessor workstation based on the new AMD Opteron processor and ASUS SK8N mainboard on NVIDIA nForce3 Pro 150 chipset. So far, we would like to refrain from wild enthusiasm. We have to admit that despite all the innovations, this workstation still has a lot of drawbacks, which come not only from the CPU, but also from the mainboard and the chipset. For instance, I was very upset to see no AGP Pro slot or 64bit PCI slot on the mainboard. As for the CPU, even the today’s latest Opteron models are not ideal. The new processor has a few indisputable advantages, such as large L2 cache, high performance memory subsystem and SSE2 instructions support. However, the other scale contains relatively low core clock frequency. As a result, we either see a performance boost or a performance drop, depending on the factor each particular application is critical to. That is why even though Opteron performs very well in server applications, it is still unable to prove up to the mark in workstations.

We should also complain about the fact that Opteron based workstations appeared a little bit too early. Unlike the server market, where 64bit software has already spread pretty widely, the workstation field still doesn’t have any and unfortunately, you will have to use 32bit applications only, because the tasks of the kind do not support AMD64 technology yet. As a result, we can say that uniprocessor Opteron workstations can be demanded only by some enthusiastic users, dedicated fans of AMD platforms.

However, the situation can change any minute. The changes can be provoked either by the release of Windows OS optimized for AMD64 technology, or by the launch of AMD Opteron processors working at higher core clock.

At the same time, we have to admit that these tests give us some hope that the upcoming Athlon 64 processor turns out a success. Look, we have just seen that Opteron 144 working at 1.8GHz proved simply outstanding in gaming applications. That is why Athlon 64 with higher working frequencies may become an ideal gamer’s platform. So, have patience: there is not too much waiting left :)