by Grigoriy Gubankov
06/27/2003 | 11:27 PM
Integrated solutions are historically considered to be a forced step. For example, if you wanted sound on your PC but had no money for an external add-on sound card. Such unwelcome attitude towards them has quite weighty reasons: the production technologies were less advanced, and the companies had to save on transistors, i.e. actually on quality and features, to make integrated solutions cheaper (otherwise they would make no sense). Besides, some time ago the demand for such solutions wasn’t great: large companies which were the main customers of PC makers didn’t save on PCs and purchased them with external graphics and sound cards. Home PCs were bought mostly by computer fans who were ready to pay through the nose for an aggregation of metal, plastic and silicon.
Soon, however, it occurred to large corporations that it would be a good idea to save on PCs; computers were getting closer to ordinary consumer equipment, and people who would like to save on it were getting more numerous, especially because computers were primarily used as a typewriter. First integrated solutions made their way to the market in mid 90s of the last century. Companies producing mainboards mounted graphics chips right on the boards to cut down the overall production cost. A bit later mainboards were also bundled with sound chips, although their quality and functions were somewhat worse than those of discrete solutions, that was enough for undemanding users. The production technologies kept developing, and soon it became possible to integrate more functions into chipsets or chips mounted on mainboards, so that users could do without external add-on video or sound cards.
Audio codecs were more aggressively entering the market and succeeded in replacing add-on solutions in undemanding users’ computers. As for me, I feel OK with an integrated sound solution because I’m not a great fan of music and I don’t have high-quality audio equipment at home. MP3 files are played fairly well, and there is nothing more I need. As far as I understand, this is a very popular standpoint. Office PCs do not need add-on sound cards at all. At the same time, integrated or simply low-cost sound solutions are almost never installed in serious gaming PCs.
Integrated video cores had a more thorny way. Video is more complicated to output onto the monitor than sound; production of video cores needs more transistors, and it’s more difficult to house them within a chipset North Bridge. That is why the first experiments in this sphere weren’t that successful. Take for example, Intel 810 chipset. We can’t say that the graphics quality provided by i810 chipset was good; it was simply impossible to play any games. But the chipset gained some popularity with corporate users thanks to its attractive price. Then there was i815. It had the same core but it was generally improved and it supported external graphics cards, that is why customers could decided on them with an eye to future acquisition of external graphics cards. The following chipset i845G offered decent graphics quality and it was in much greater demand among corporate users. Computers based on the i845G and the Celeron were pretty inexpensive, and an external graphics card was also supported.
Although the corporate market was conquered, integrated video solutions didn’t become common in home PCs. Since most home users play games, such boards were simply a bottleneck. The performance of boards based on Shared Memory Architecture (i.e. when the frame buffer uses a part of the system memory, like in Intel’s solutions, for instance) was low partially because of the low memory bandwidth available for the core. The matter is that until recently the systems supported mostly SDRAM and then single-channel DDR memory, while external graphics cards got the fast DDR memory of a much higher bandwidth. Nonetheless, external frame buffers weren’t of much help for integrated solutions because the cores were very plain and slow.
NVIDIA was the one to promise us a revolution in the use of integrated graphics for home needs. The company introduced the nForce chipset that had dual-channel DDR memory support and sported good bandwidth for an onboard core. The core was actually an altered GeForce2 MX processor, and it wasn’t very efficient but was fast enough for games to be played with acceptable image quality. But the nForce failed. NVIDIA overpriced the integrated chipset (over US$90); besides, this company didn’t have the reputation of a reliable chipsets manufacturer, and finally, the users didn’t trust the integrated solutions that much. The nForce version without the integrated graphics core was launched a bit later and even won some users’ hearts. But it failed to become popular despite its outstandingly high performance and immensely advanced integrated sound solution aka APU (Audio Processor Unit).
But NVIDIA didn’t give up. The company took its mistakes into account and developed the nForce2, which was positioned as the fastest solution for AMD processors. By the way, NVIDIA has no plans to release a chipset for Intel platform, as they consider the license fees to be too high. The company used another approach with the nForce2. First they launched a version without the integrated graphics core (which has been replaced with GeForce4 MX) named nForce2 SPP, and only after that they started supplying the nForce2 IGP (Integrated Graphics Processor). The high integration of the nForce2 made it possible to create micro-ATX mainboards with pretty rich features. One of these boards, Albatron KM18G Pro, as well as the nForce2 IGP in general, will be the topic of our today’s review.
So, NVIDIA nForce2. We will not dwell on the discrete version of this chipset, as it has already been discussed in a huge number of various articles, reviews and news stories. Let me just briefly remind you of its features: it uses a dual-channel memory controller supporting up to PC3200 (the nForce2 IGP supports up to PC2700 to reduce the EMI from the memory controller onto the graphics core). In this case the CPU uses the available memory bandwidth most efficiently when the FSB and memory clock rates coincide. The excessive memory bandwidth compared to the FSB is used for the DASP algorithm (Dynamic Adaptive Speculative Pre-processor) for data pre-fetch or for the integrated graphics core.
The connection with the chipset South Bridge is implemented via HyperTransport bus featuring 800MB/s bandwidth. There are two types of South Bridges: MCP and MCP-T, which is more functional due to the second Ethernet controller, FireWire support and NVIDIA SoundStorm technology that includes hardware decoding of the Dolby Digital data stream. The nForce2 also supports StreamThru Technology similar to the Intel CSA (read more about it here), i.e. it’s developed to boost the network performance by assigning a separate connection of the guaranteed throughput. I should say that NVIDIA’s solution looks better than Intel’s one because it needs no support from the board maker, i.e. no additional components should be laid out on the PCB, while the Intel CSA needs a network controller to be mounted on a separate channel with all the resulting consequences.
The graphics core in the nForce2 IGP is none other but GeForce4 MX adapted for integration into the chipset North Bridge, and all capabilities of the graphics core remain untouched. If you remember, GeForce4 MX has two rendering pipelines with two texturing units in each, Accuview AA technology, 2x anisotropic filtering, memory optimization technologies in 3D modes, vertex shaders unit/T&L, two integrated 350MHz RAMDACs up to 2048x1532x32, an integrated TV-out controller up to 1024x768, and a special hardware unit controlling certain DVD and HDTV decoding stages.
To integrate it into the North Bridge the engineers probably redesigned the memory controller and brought down the core clock to 200MHz. The latter measure was a forced step because the electromagnetic interference from the high-frequency core could affect operation of other North Bridge components and hamper cooling. Moreover, NVIDIA hardly wants the integrated chipset be as efficient as add-on graphics solutions because it can negatively tell on sales of external graphics cards, which NVIDIA and its partners would hardly wish to happen. So, the core clock is 200MHz, 50MHz lower than that of the slowest model of the GeForce4 MX family - GeForce4 MX420. The situation with the memory bandwidth available for the integrated graphics core looks as follows. The chipset can allocate 2.2GB/s for the integrated GPU at worst (if the processor uses 400MHz bus). But it’s not a typical situation because processors with the 400MHz bus belong to the high-end market and their owners would hardly use integrated video. Moreover, NVIDIA didn’t announce the official support of the 400MHz bus for the nForce2 IGP. It also concerns processors with the 333MHz bus, even the slowest Athlon XP 2600+ is not a budget product.
So, we can consider only the processors with the 266MHz bus. In this case the memory bandwidth offered for the GPU is 3.3GB/s for PC2700 and 2.1GB/s for PC2100. While the first figure is good enough (it’s still much higher than what GeForce4 MX420 offers even on external cards), the second one is not that impressive. Anyway, this is much better than other integrated cores based on SMA idea use. The influence of the speed and timings of the system memory on the graphics core performance will be discussed later, and now, please, take a look at the board which we are going to test today.
The board is designed in uATX form-factor as it’s positioned for inexpensive compact systems. Such systems usually do not house a lot of PCI cards and they are not selected for overclocking purposes. In other respects, this board has a lot in common with its larger siblings. The only thing it lacks is SerialATA, but SerialATA hard drives are not budget solutions at all.
Albatron offers the KM18G in two versions: Pro and Pro II. The latter is more functional thanks to a different South Bridge - MCP-T that supports FireWire and SoundStorm technology mentioned above. I’ve got the less functional model, but you know the difference.
Specification of the Albatron KM18G Pro look as follows:
Albatron KM18G Pro | |
CPU | Socket A AMD Athlon XP/Duron |
Chipset | NVIDIA nForce2 IGP |
FSB frequencies | 100-200MHz (with 1MHz increment) |
DDR DIMM slots | 3 DDR DIMM slots |
AGP slot | AGP 8x |
Expansion slots (PCI/ACR/CNR) | 3/0/1 |
USB 2.0 ports | 6 |
IEEE1394 ports | None |
Additional IDE-controllers | None |
Integrated sound | 5.1 sound integrated into the chipset South Bridge |
Integrated network | Fast Ethernet integrated into the chipset South Bridge |
Additional features | Protection against CPU overheating |
BIOS | PhoenixBIOS |
Form-factor | MicroATX |
Although the integrated nForce2 IGP graphics core supports nView, Albatron considers that one VGA connector is enough for a uATX board. By the way, Shuttle has an opposite point of view and lays out two VGA connectors on the MN31N boards. It is difficult to say which approach is more reasonable: on the one hand, a budget system will be hardly used with two monitors; on the other hand, there are not many devices with a COM port either (maybe just external modems), though Albatron mainboard has this port instead of the second VGA.
The board has not many accessories with it, but it’s typical of Albatron. So, it comes with a CD-disk with drivers, FDD and ATA/100 cables and a bracket with two USB ports. There are no panels with other ports because the connectors configuration is pretty standard. Other things are actually not so vitally important and would only increase the product cost.
Features of such board entirely depend on the South Bridge because there are no other controllers there. That is why it supports two Parallel IDE channels without RAID (RAID is not a standard solution in a low-cost system intended especially for home use), and six USB 2.0 ports. A pair of USB 2.0 ports is located on the board’s rear panel, the other pair - on a separate bracket included into the package, and the third pair should be connected to USB ports of the PC case, as Albatron’s engineers and marketing people see it. If you don’t like this idea, you can get one more bracket with USB ports.
Sound on the KM18G Pro is implemented via the popular Realtek ALC650codec. The sound quality it provides has been discussed all over the Net. Anyway, it is more than enough for me, though I’m not a passionate music fan, as I have already said.
The board’s design is generally good. But I have to say here, that I judge it by how it matches my Chieftec DA-01 PC case as I have no other cases to play with. This case is certainly not for uATX boards, so my discussion of PCB design highs and lows will be very subjective. But it’s impossible to cover all types of PC cases including those which are designed for uATX boards, because of some of them may be pretty far from being standard.
The connectors are located pretty conveniently except CD-IN and S/PDIF. The CD-IN is placed close to the back PCB edge while the CD-ROM drive is usually in the front. Secondly, it can be uneasy to reach this connector when AGP video card is installed. On the other hand, the board has integrated video, and an external solution may be unnecessary. Also, an analog CD cable is not used that often. The S/PDIF is also placed in the back, and is tucked between the PCI slots, that is why it can be a problem to reach it. Also, if you try to connect S/PDIF to something located in the front, the cable will stretch along the whole PC case. But these are minor drawbacks.
Another small difficulty may arise if you try to install a “long” AGP card (like GeForce4 Ti4600): it will lock the memory modules, as the DIMM slots clips will be impossible to open. However, it was a problem only with GeForce4 Ti based cards, and with RADEON 9700 Pro or GeForce4 MX440 there were absolutely no problems. After all, not every one will need external graphics.
There are no more problems with the PCB layout. But the above mentioned ones are not that serious either. Also remember that the uATX format doesn’t give the designers too much freedom. Keeping all these facts in mind we can call the design of Albatron KM18G Pro a success.
The Albatron board possesses a dual-channel CPU power supply system. It means that it won’t be of any interest for overclockers. One more piece of evidence is the absence of four holes around the processor socket for massive coolers mounting. Nonetheless, AMD recommended coolers can easily be installed on it, and the lack of overclocking friendly options is not a weak point for a product of such class. The board features anti-burn protection for Athlon XP CPUs. The temperature is monitored with the internal CPU diode. The BIOS, however, doesn’t have any settings for this function, and I sincerely hope that this function works indeed and that the guys at Albatron made correct settings for the temperature. Actually, I have no grounds to doubt it.
The BIOS is quite standard. It’s PHOENIX-AWARD, however, the interface is a little different, though you quickly get used to it. Almost all items are standard, I just want to dwell on only one: Advanced -> Advanced Chipset Features. There are no voltage settings at al, so you will be unable to adjust any voltages. Add the fact that the CPU multiplier is unchangeable, and you will get one more piece of evidence that Albatron KM18G Pro is not targeted for overclockers.
At the same time, the Advanced Chipset Features section has a great deal of settings. At the top you can see the System Performance item which can be switched between Optimal, Aggressive, Turbo and Expert. Each of them fixes the system parameters at certain values, and the last one allows manual adjustment of all parameters. Actually, when we tried to switch System Performance parameter from Optimal to Aggressive, we discovered one interesting peculiarity, which is probably just a small programming error of the BIOS developers. In Optimal mode we can change the DDR:CPU Ratio, and can’t do it in the Aggressive and Turbo modes. I obviously selected Expert mode to be able to play with all available settings.
So, the first item is FSB Frequency. It ranges from 100 to 200MHz with 1-2MHz increments. I wish it could be possible to type in figures manually instead of looking through the list. By the way, this parameter can be altered in all System Performance modes. So, even though Albatron KM18G Pro is not an overclocker’s solution, this way Albatron made a certain curtsy for overclockers.
The CPU Interface item contains two options: Optimal and Aggressive. They change FSB timings, and the variants names speak for themselves. Then comes DDR:CPU Ratio, which I have already mentioned above. There are a lot of values here, as nForce2 is very flexible in this respect. Note that the memory subsystem of nForce2 reaches its maximum performance relative to the CPU when the FSB and memory work synchronously, i.e. when this parameter is set to 1.
A bit lower you can see the Memory Timings line with the identical presets as those of the System Performance. When you change one of the System Performance parameters a corresponding item in the Memory Timings menu will change respectively. The Expert mode in the Memory Timings is the only exception: you should enable it manually. The range of settings is pretty wide thanks to the nForce2 capabilities.
All other parameters are standard except, maybe, Frame Buffer Size, which determines how much memory of the system memory should be allocated for the frame buffer of the integrated graphics core. It’s seen only if an external graphics card is not installed.
Now, when we are finally through with the description of the board and chipset, let’s start our tests.
First of all, a couple of words on the testing methods. Since our system is based on the uATX board with integrated graphics, it should represent a budget solution, i.e. a processor mustn’t be too fast and powerful. Thus, we decided in favor of the Athlon XP 1700+, because this is the very budget solution. Also, it’s useless to compare the integrated core with the GeForce4 Ti or RADEON 9500/9600, or even GeForce FX or RADEON 9700/9800 Pro cards, because the outcome is predictable. The best choice would be the GeForce4 MX420, as its core clock and memory bandwidth are close to those of nForce2 IGP. But at that time we didn’t have this card at our disposal, that is why we will compare our today’s hero with GeForce4 MX440 w/AGP 8x. We will also find out how greatly lower memory bandwidth and core clock may affect the performance of GeForce4 MX. Here is the final testbed configuration:
Integrated graphics core | External graphics card | |
CPUs | AMD Athlon XP 1700+ | |
Mainboards | Albatron KM18G Pro | |
Memory | Corsair XMS3200 v1.1 | |
HDD | Seagate Barracuda ATA IV, 80GB | |
Drivers | nForce drivers 2.03 | Detonator 44.03 |
Note that we used Corsair memory, because we simply didn’t have any slower memory at hand. This is not an appeal to use such expensive memory in budget systems. But still, you should make sure that your nForce2 IGP based system has quality memory.
Now, a few words about the testing conditions. Since the graphics memory of the integrated core is a part of the system memory, it’s interesting to see how the system memory latency affects the performance in graphics applications. We also take into consideration that the memory working at 266MHz (PC2100) with aggressive timings is pretty inexpensive and widely spread, while far not all modules will work as PC2700 with the most aggressive timings.
The integrated graphics core was tested with the following memory settings: PC2100 with the 2-2-2-5 timings, PC2700 with 2.5-3-3-7 and PC2700 with the temporary 2-2-2-5 timings. When the system was tested with the external graphics card in games the PC2700 wasn’t used with the conservative timings as they had almost no effect on the overall performance.
We didn’t switch to high-quality graphics modes by forcing full-screen anti-aliasing and texture anisotropic filtering because it’s obvious that the integrated solution will perform not very fast there. We don’t recommend you to enable them either.
The first part of our tests will be Winstone business benchmarks. The idea is simple: nForce2 based platform can work as an office solution. And we measure the performance to find out how the graphics core can affect the memory subsystem performance. The speed will fall down at least because it takes time to distribute data between the processor and the graphics core.
Well, we were right asserting that the integrated core would lower the system performance in the office tasks. But the drop is not dramatic: in Business Winstone with the memory working as PC2100 the performance falls down by 2% at the most, which is not that noticeable. Note that when the memory is not synchronous to the FSB, the efficiency drops by over 3% in Business Winstone with the external card installed. It’s not much but it demonstrates that fast memory is not always a good thing.
These tests traditionally start with the 3DMark2001 SE.
In contrast to the business tests, the faster memory helps the integrated nForce2 a lot. It is not at all surprising, as the minimal memory bandwidth allocated for the integrated graphics core gets at least 1.5 times higher. Nonetheless, the integrated graphics loses to the external video card by a third at least because of the much lower memory bandwidth and lower core clock. It’s also interesting to know that the system with integrated graphics slows down by 5.5% with the higher memory latency. Here is what we have in the separate tests:
The results are generally similar to the total scores. All the systems have over 60fps, which is acceptable for most gamers.
However, the 3DMark is a synthetic test. Let’s see what our systems can do in real games:
Return To Castle Wolfenstein built on the Quake3 engine demonstrates how the performance correlation depends on the resolution. The graphics card performs equally fast in all resolutions, which indicates that this game is already too simple even for low-end video accelerators like GeForce4 MX440.
The integrated graphics core, however, behaves differently: in 640x480 it comes close to the external card. A minor difference is not that vital because all systems ensure over 100fps. At the higher resolution the difference grows up, but it’s still inconsiderable because the performance still doesn’t fall beyond 94fps. Finally, at the highest resolution the IGP reaches only 72.7fps (with the DDR266), though this is also quite enough to play games comfortably. The PC2700 used instead of the PC2100 boosts it by 15%.
Now comes the most difficult game for graphics accelerators - Serious Sam: The Second Encounter.
The results look similar to the previous game, only the integrated video starts losing to the external card at a lower resolution and by a greater margin. However, at 1024x768 the external card also performs not good enough. But you can play games anyway, even with the integrated video, especially with the PC2700 with lower latencies.
And now it’s high time for the most complicated game: Unreal Tournament 2003 Demo. All quality settings on the Details tab were set to Normal.
In the flyby mode (flight over the game map before the game is started) the systems have acceptable speed of at least 47fps (with the integrated graphics and the slowest memory involved). The scores entirely depend on the graphics cards power, and the gap between these graphics solutions is about 30-35% in all resolutions tested.
The flyby looks good but it’s not playable. In the botmatch the situation changes sharply. In this mode the physics and bots’ artificial intelligence are also calculated, that is why the graphics subsystem is of no primary importance anymore. It makes the difference between the integrated and external graphics negligible at 640x480 and at 800x600 (especially in the first case, of course). Only at 1024x768 the integrated nForce2 is noticeably slower. However, the game sometimes works slower than needed even with the add-on graphics card, though it is not that noticeable and doesn’t actually affect the playing. We can say the same things about playing at 800x600 with the integrated graphics involved. But this is my own opinion. Moreover, I am not an experienced player in Unreal Tournament 2003. We checked that on the Anubis map with the Bombing Run game type. We didn’t notice these effects on the Antalus map.
So, what can we say about nForce2 graphics core performance? It turned out that an undemanding user who is not crazy about games won’t notice a great difference between an external add-on graphics card based on the GeForce4 MX440 and the integrated graphics of nForce2 used with the PC2700 memory in all today’s games except Unreal Tournament 2003. In the RTCW it doesn’t matter whether the speed is 85 or 110fps. In the Serious Sam the minimal fps makes 37.4 at the average value of 47. In the Unreal Tournament the difference implies that with an add-on graphics card you can play at 1024x768, and with the integrated graphics - only at 800x600. And that’s it. It’s quite another question how comfortable you feel playing in resolutions lower than 1024x768. But this is solely up to you.
Speaking of this particular incarnation of the nForce2 IGP platform, Albatron KM18G Pro board, I think that it is a pretty successful product. The absence of overclocking friendly options can be regarded as a weak point, but it’s not dramatic for a product of this type. Some of you may be unhappy about no second monitor support, though the nForce2 IGP allows it. But it is again up to you to decided what is more important: a second monitor or a COM port that is why this peculiarity may be regarded as an advantage or a disadvantage. And if you desperately need a FireWire port, take another board modification aka Albatron KM18G Pro II, featuring FireWire support.
In conclusion, I would like to say that this board is priced at about US$100 (according to PriceWatch). It’s not that bad considering that the minimal price of GeForce4 MX440 cards, which you can do without, is about US$50. But remember that if you want to squeeze the best performance from your nForce2 IGP in games, you should get good PC2700 memory and make sure that you have enough memory for the system and the frame buffer of the integrated graphics core.