by Tim Tscheblockov
07/05/2004 | 11:09 AM
If we were to compare the evolution rate of discrete graphics processors with that of integrated graphics, we would see that integrated graphics solutions live as if in another world where everything is going on slowly and sedately – quite contrary to the changing and dynamic market of gaming graphics cards.
<%BANNER[article]%>Just consider: we have watched several generations of GPUs passing by in the world of discrete graphics in the last two or three years. The industry greeted and embraced shaders of the eighth and ninth specifications of the DirectX API; the graphics chips have become truly programmable processors; the graphics memory bus expanded to 256 bits in width. DDR2 memory appeared and soon started to give way to GDDR3. I don’t even say by how much the performance of GPUs of all categories has grown in these two-three years!
Integrated graphics chipsets have used this time interval for a sluggish evolution, accumulating the functionality necessary for a modern chipset. They added support of new processors, higher FSB and memory clock rates, various interfaces like USB 2.0 and SerialATA – I could continue this list for long… It seems, however, that the speed and the functionality of the integrated graphics core used to be the last item in the list of intended innovations, and the users have got accustomed to regard integrated graphics disdainfully.
And really, we assume that a system with an integrated chipset should be used as an office “typewriter”, so the only serious requirement to its graphics core is a well-written driver for fault-free work in 2D applications and output of a high-quality image to the monitor. The mainboard should cost as low as possible at that, still providing all the necessary functionality.
At the same time, office is still not the only place for integrated graphics to dwell in. When such a chipset becomes a foundation of a barebone system or a home computer, the requirements to the speed and functionality of the graphics core become stricter.
In a home or “entertainment” system, besides the basic requirement about the high-quality output to the monitor, the integrated graphics core should permit watching videos and, as far as possible, enjoying simplest 3D games, at least.
In our today’s review we will discuss the most advanced of currently-available integrated graphics chipsets exactly from this point of view. We will see how the integrated graphics manages games and video playback and how well it renders 3D scenes in the same games and will also evaluate the quality of output to the monitor, both CRT and LCD. Overall, we’ll check out how well these chipsets would perform in a home computer.
I won’t dwell upon the functionality of the integrated chipsets, besides their graphics capabilities. However, I will be adding links here and there to the reviews of the much-esteemed Mr. Gavric, which deal with these matters.
The RADEON 9100 IGP is not the first attempt of ATI Technologies at developing integrated chipsets: the company has long been shipping such products. We may recall the 320/330/340 IGPs and the mobile 320M/340M IGPs. Still, it is the RADEON 9100 that achieved the highest popularity in the desktop chipset field.
This chipset (see our review of the RADEON 9100 IGP) supports Intel Pentium 4 and Celeron processors. It can clock the system bus at 400/533/800MHz frequencies and thus allows using all the modern CPUs from Intel. What’s most important for integrated graphics, the chipset has a dual-channel memory controller with support of the fastest modules (DDR266/333/400).
The integrated graphics core of the RADEON 9100 IGP, like in all modern integrated chipsets, doesn’t have dedicated graphics memory, but uses a portion of the system RAM. This organization is called Unified Memory Architecture – UMA.
The graphics core and the central processor have to share the memory bus bandwidth and this sharing inevitably leads to performance losses. On the one hand, graphics is poor due to the CPU’s loading the bus and on the other hand the load from the graphics core encumbers the central processor. In other words, if you give one chair to two persons, both would feel uncomfortable.
The dual-channel memory controller improves the situation considerably: using a twice-wider bus, the graphics core and the CPU impede each other less. By the way, the integrated chipsets from ATI Technologies allow us one experiment: we can evaluate the gains from a dual-channel memory controller without artificially lowering the performance by plugging memory into one bank only because the RADEON 9100 IGP has a single-channel analog, the RADEON 9000 Pro IGP, with the same graphics core inside.
Let’s get back to the 9100 IGP, though. The chipset came to us on a P4V800-V Deluxe mainboard from ASUS:
This is the only mainboard among the models included into our today’s tests that is made in the full-size ATX form-factor. The board features the RADEON 9100 IGP chipset with an IXP 150 South Bridge.
Slots and connectors:
The mainboard looks modern and highly functional; evidently, it should cost more than low-end products.
The integrated graphics core implemented in the RADEON 9100 IGP is a close analog of the desktop RADEON 9000/9200 chips. Thus, I can say that this chipset has the most advanced graphics core of all the integrated chipsets now widely available. Right now, ATI Technologies is the only supplier of DirectX 8.1-compliant integrated graphics.
You can learn more about the capabilities of the RADEON 9000/9200 GPU, and, accordingly, of the graphics core of the RADEON 9100 IGP chipset, in our review called ATI RADEON 9000 PRO: TYAN Tachyon G9000 Graphics Card Review; I will only quote the basic facts:
So, here’s the first contender from ATI and ASUS: a dual-channel chipset with a powerful graphics core, excellent functionality, support of modern processors and fast memory.
The RADEON 9000 PRO IGP is a single-channel analog of the RADEON 9100 PRO IGP, which we couldn’t include into our tests because of the absence of such mainboards in retail. The RADEON 9000 PRO IGP has the same graphics core as the RADEON 9100/9100 PRO IGP, but it is “stifled” here by having to access the memory across a narrow 64-bit bus.
ATI claims the memory controller of the RADEON 9000 PRO IGP to be not just “half” of the memory controller of the RADEON 9100 IGP – it has certain improvements for an efficient use of the memory bus, i.e. it is half of the improved controller of the PRO version of the RADEON 9100 IGP. Unfortunately, we don’t have any information about the nature of those improvements yet.
Like the dual-channel version, the RADEON 9000 PRO IGP supports processors with 400/533/800MHz FSB and modules of PC2100/2700/3200 DDR SDRAM – the most progressive processors and modules for today.
The A350-SV mainboard from PowerColor is based on this chipset:
It’s rather strange to see a mainboard under the brand of a graphics card manufacturing company, yes? The board has a RADEON 9000 PRO IGP (RC350) North Bridge coupled with an IXP150 South Bridge. The user manual mentions the SB200 South Bridge, but we wanted to make sure and removed the sticker from the chip – only to see the good old IXP150! The A350-SV complies with the MicroATX form-factor and offers the following slots and connectors:
The RADEON 9000 PRO IGP has exactly the same graphics core as the 9100 IGP, so the chipsets will only differ in the speed of graphics due to the 9000 PRO IGP’s having a single-channel memory controller.
So, ATI Technologies presents two products in our review: one is expensive and, obviously, high-performing dual-channel RADEON 9100 IGP and another is the cheaper single-channel RADEON 9000 PRO IGP. The functionality and the accessories of the mainboards from ASUS and PowerColor fully correspond to the positioning of the chipsets these mainboards are based on.
Before proceeding to other chipsets, I’d like to mention one more curious technology realized in ATI’s chipsets, called SurroundView. The technology permits simultaneous operation of the integrated graphics core and the external graphics card on a chip from ATI that you install into the AGP slot. The result is a multi-display configuration – up to three monitors! This feature may not be too necessary at home, but might be useful in office. The only problem left for us to discuss is the quality of the output to the monitor. We’ll see how well ATI, ASUS and PowerColor are doing here in the appropriate section of the article.
Silicon Integrated Systems has been long known as a chipset manufacturer as well as a developer of standalone graphics chips. You may recall such names as SiS305, SiS315 and SiS Xabre. Earlier graphics chips from SiS had much more popularity than the older seed – well, I remember myself using a SiS6202 graphics card for quite a while.
Time seems to have stopped for SiS with the release of the SiS315 for integrated graphics. Even the most advanced integrated chipset models of the company get this out-dated graphics core in their North Bridge. Notwithstanding the company’s activities in the discrete GPU market and the launch of the Xabre, a not-very-successful GPU against the competitors, but undoubtedly a more up-to-date solution than the SiS315, none of the integrated chipsets from SiS received this new graphics core so far.
The SiS 661FX seems to have been developed from the ordinary single-channel SiS648FX chipset for Pentium 4/Celeron CPUs. It supports Hyper-Threading, 800MHz FSB and PC3200 DDR SDRAM. On the other hand, the SiS661FX may be considered as a result of the evolution of the previous integrated chipset of the company, the SiS651, with added support of those faster processors, FSB and memory clock rates.
The SiS661FX inherited the graphics core from the previous models without any changes. In fact, it is a slightly adjusted – for the purpose of integration into the North Bridge – version of the standalone SiS315 GPU (see our SiS315 review dated November 2001!).
Here’s a list of the basic characteristics of the Real256E core, integrated into the SiS661FX chipset:
The SiS661FX is represented by the P4S800-MX-EAYZ mainboard from ASUS:
You may notice from the screenshot that this model appeals to people who don’t want to spend much money for a mainboard.
The P4S800-MX-EAYZ is made in the MicroATX form-factor. The slots and connectors:
We’ll see shortly if the SiS661FX chipset and the ASUS P4S800-MX-EAYZ mainboard can make a worthy competitor to the other products.
The Intel 865G is a variation of the i865PE, a high-performance chipset with a dual-channel memory controller for Pentium 4/Celeron CPUs, equipped with an integrated graphics core.
The highest characteristics of the i865 as a chipset proper are beyond doubt as its non-integrated prototype, the i865PE, proved its abilities long ago (see our review called Albatron PX865PE Pro Mainboard Review: the First i865PE Based Mainboard in Our Lab). The performance and functionality of the integrated graphics core, however, give fewer reasons for optimism: the i740 discrete graphics chip was the progenitor of Intel’s integrated graphics and it looks like all the subsequent advances of Intel in the graphics field were nothing more but a methodical and slow development of the old architecture. It means – no shaders or T&L, no efficient full-screen antialiasing or fast anisotropic filtering.
On the other hand, just a few days ago, the company announced a bunch of new products, also a new integrated chipset. Among the capabilities of the new graphics core they claim hardware support of DirectX 9 pixel shaders. It is a big leap in the evolution of integrated graphics from Intel. I hope that the performance of the new graphics core is going to match its functionality.
But let’s get back to the i865G and its main characteristics:
The characteristics look very weak, but only compared to modern standalone graphics processors. Our tests will show how well the i865G behaves in contrast to other integrated chipsets. We took an i865G-based D865GLC mainboard from Intel for our tests:
The mainboard is realized in the MicroATX form-factor, in Intel’s natural “laconic” style:
The D865GLC is the last mainboard for Intel processors I took for my tests. Next follow mainboards for AMD’s CPUs.
NVIDIA’s first try at making chipsets was rather far from a success, but the second – the nForce2 – was so good that it became and remains the highest-performing chipset for Socket A processors up to this day (see our review called Contemporary Socket A Chipsets Comparison: NVIDIA nForce2 vs. VIA KT400). The integrated graphics core of the nForce2 IGP is an analog of the discrete GeForce 4 MX processor; its basic characteristics:
The nForce2 with the integrated GeForce4 MX440 graphics core and a dual-channel memory controller is a favorite of our tests for the Socket A platform since VIA and SiS just don’t have logics of that class on their hands now.
Let’s not anticipate, though. The A7N8X-VM mainboard from ASUS represents the class of nForce2 IGP-based products:
The mainboard follows the MicroATX form-factor and has the following slots and connectors:
Although ASUS chose a powerful integrated chipset for this mainboard model, it turned to be less functionality-rich than the one based on the RADEON 9100 PRO IGP. Clearly, the A7N8X-VM is positioned as a relatively inexpensive mainboard, still based on a high-performance integrated chipset.
The SiS 741GX lags behind the top-end products in support of the fastest Socket A processors and DDR SDRAM modules. It is a single-channel chipset for CPUs with 200/266/333MHz FSB; it works with PC2100/2700 memory. The integrated graphics core, the Real256E, doesn’t differ from the one in the SiS661FX: the same modified SiS 315 constrained by a 64-bit memory bus, which it has to share with the CPU. The SiS741GX-based 741GX-M mainboard comes from ECS:
Connectors and slots:
ECS ships inexpensive mainboards and the 741GX-M follows the company’s policy. It doesn’t support topmost processors, fastest memory, new SATA hard disk drives, and is bad at overclocking. But it’s cheap!
S3 Graphics, a daughter company of VIA Technologies, develops graphics cores for VIA’s integrated chipsets. Well, “develops” is too optimistically spoken. VIA obtained integrated graphics quite a long time ago: the various integrated and mobile graphics solutions like ProSavage, Twister and others have all been based around the discrete graphics chips Savage4 and Savage2000, released back when S3 was an independent company. Since then, VIA’s chipsets have been evolving according to the requirements of the times, but the integrated graphics remained nothing more than a hybrid of Savage4 and Savage2000 adapted for integration into system chipsets.
There was some stir following the release of the DeltaChrome, a rather well-made discrete GPU with hardware DirectX 9 support, and VIA Technologies with S3 Graphics started setting up a line of discrete processors (DeltaChrome, GammaChrome, OmniChrome), started making plans for the future and even came up with a new name for their integrated graphics (see our review called The Return of S3: DeltaChrome Graphics Card Review). Regrettably, the point hasn’t changed in the least: the UniChrome is not the result of integration of the new architecture, but only a renamed ProSavage.
The basic characteristics of the UniChrome KM400 graphics core follow:
Considering that this chipset has only a single-channel memory controller, its speed in 3D is rather doubtful. The VIA UniChrome KM400 doesn’t impress as system logic, either. The number “400” in the name is misleading since the chipset doesn’t support the 400MHz FSB or DDR400 SDRAM. And this is the most advanced widely-available integrated chipset from VIA for the Socket A platform!
The A7V8X-MX SE mainboard from ASUS is going to participate in our today’s tests:
Slots and connectors:
This product is obviously a low-end and cheap solution for undemanding users.
So, we’ve taken a look at each of the participants. Let’s get to our tests!
I took two quite fast processors from Intel and AMD for my today’s tests. In fact, this somehow goes against the main line of the review: an integrated chipset is nearly always a compromise between speed, quality and price and few people will prefer to use an expensive and high-performing processor with such a chipset.
Still, the use of the fast processors is good for testing purposes since such CPUs put less restrictions on the integrated graphics, which is initially impeded by the slow memory bus.
So, the testbed was configured as follows:
Software:
I used the latest official versions of the chipset drivers in my tests.
I performed my tests in 3D games using two operational modes offered by the game engines.
The first mode meant medium graphics quality settings and 32-bit color depth of the frame buffer. This is a compromise variant, imposed by the low speed of the integrated graphics, which cannot provide playability at the maximum quality settings.
The second or “speedy” mode used the lowest graphics quality settings and 16-bit color to achieve the maximum fps rates.
So, gaming tests come first.
The chipsets from the respected GPU makers, ATI and NVIDIA, win the test at medium graphics quality settings. The RADEON 9000 IGP and the nForce2 IGP are far ahead of the others due to two factors: first, the integrated graphics cores of these chipsets are more advanced than their competitors and, second, these two chipsets have a dual-channel memory controller, which nearly doubles the speed of the integrated graphics.
Curiously, the advanced architecture shows itself advantageously even with single-channel memory access: the single-channel RADEON 9000 PRO IGP outperforms the dual-channel Intel 865G whose graphics core works in more favorable conditions.
The chipsets from VIA and SiS cannot compete with the leaders.
By switching to 16-bit color, reducing the texture quality, disabling special effects and so on, we reduce the load on the memory bus, the most vulnerable spot of integrated chipsets. The “speedy” settings provide higher results – nearly twice the performance of the first mode with ATI’s and NVIDIA’s chipsets. The products from Intel, VIA and SiS enjoy a triple of quadruple performance growth!
At the medium settings the single-channel RADEON 9000 PRO IGP had twice lower results compared to the RADEON 9100 IGP, but this gap diminishes as we switch to the second test mode: the insufficient memory bus bandwidth affects the performance less.
So, judging from the test results, only the RADEON 9100 IGP and the nForce2 can give you playability at speedy settings in Unreal Tournament 2003 Demo.
The chipsets remained on their respective positions, but this time the single-channel chipset from ATI, the RADEON 9000 PRO IGP, also provides playability in the 800x600 resolution, like the RADEON 9100 IGP and NVIDIA nForce2 IGP do.
The max speed settings in Max Payne 2 use 32-bit rather than 16-bit color – the game engine just doesn’t offer 16-bit-color modes. Anyway, the i865G joined the group of playable chipsets from ATI and NVIDIA at the speedy settings in 800x600– it added in performance considerably, unlike the products from VIA and SiS.
The game engine of C&C Generals: Zero Hour seems to be rendering the landscape by mapping more than two textures and this is a killer case for integrated solutions that can only render two textures per pass at best. The attempt to render additional textures necessitates rendering of the frame in several passes, resulting in a very low performance due to the memory bus bandwidth restrictions.
The ATI RADEON 9100 IGP supports rendering of up to 6 textures per pass and wins the test at the medium quality settings.
The numbers don’t grow much when we switch to the “speedy” settings because C&C Generals: Zero Hour, like Max Payne 2, doesn’t offer 32-bit-color modes. So, making allowances for the fact that C&C Generals: Zero Hour is a strategy game, rather than a shooter, I think the RADEON 9100 IGP and 9000 PRO IGP, the i865G and the nForce2 IGP all provide playability, but the RADEON 9000 IGP and the Intel 865G do so only at the minimum graphics quality settings.
The nForce2 IGP looks very well in this OpenGL test. Up till now, the efficient architecture and HyperZ technology allowed RADEON 9100 IGP to be faster than NVIDIA’s chipset, in spite of having only one TMU per pipeline and thus a lower texturing speed. In this test, however, the nForce2 goes unrivalled.
The results don’t grow too much as we switch to the “speedy” settings. Only the nForce2 IGP and the RADEON 9100 IGP can run Serious Sam fast enough for a comfortable play.
Contrary to my expectations, the Sturmovik is flying fast at the medium graphics-quality settings on the integrated chipsets. In 800x600, all the chipsets, save for the outsiders from VIA and SiS, give out enough fps.
The results grow up as we switch to the “speedy” settings and the SiS741GX may be found playable in 800x600 here. Well, this chipset in fact produces a disgusting picture, but we’ll talk about the image quality later on.
Quake 3: Arena is a crucial test among gaming applications. If a chipset cannot run this game, you can expect it to be better anywhere else, since Quake 3 is a test for which bugs are removed and drivers are optimized in the first place.
So what do we see? At the medium settings, in the 800x600 resolution, the chipsets from VIA and SiS cannot run the game fast enough for you to play comfortably.
The situation improves in the speedy mode: at last we see all the integrated chipsets show the acceptable speed in the 800x600 resolution.
You should only compare the results of the chipsets by the platform: PCMark 04 is not a tool to compare chipsets for different platforms. According to the results, the dual-channel chipsets have the most efficient memory controllers and, quite expectedly, the i865G and nForce2 are the fastest in their categories.
Now let’s watch the integrated chipsets display videos of different formats.
MPEG-4 comes first. I took a copy of the “Shrek” movie (512x384, 130/96kbps video/audio, DIVXMPG4 V3/MPEG Layer-3).
The integrated graphics core doesn’t play any role in decoding MPEG-4 video, so this test characterizes the performance of the chipset proper rather than of the integrated graphics. The results – the CPU load – match the results of PCMark 2004’s memory test.
All the chipsets played the MPEG-4 video without problems, although the CPU load was sometimes very high. As you remember, I used top-end Pentium 4 3000MHz and Athlon XP 3000+ processors in the testbed, so a weaker system may load the CPU to the full, to skipped frames even!
I took a copy of the “What Dreams May Come” movie as an example of a DVD movie (720x576, 8712/448kbps video/audio, Interlaced MPEG-2/Dolby Digital (AC3)).
All the integrated chipsets easily accomplished the task of playing DVD back!
Last goes the hardest version of video, a HDTV “Step into Liquid” clip (1440x1080, 8000/384 kbps video/audio, Windows Media Video/Audio 9 Professional):
HDTV playback may be considered as a good test of the overall performance level of a system since it puts all the system components under a heavy load.
Judging by the CPU load percentages, the difference between chipsets belonging to one platform is small compared to the difference between the two platforms. All is indicative of the Athlon XP 3000+ doing a bad job of decoding this HDTV clip.
Well, we can get much more informative results by measuring the average video playback speed. Windows Media Player 9 outputs the statistics of video playback where it also shows the average playback speed, given in frames per second.
The clip is originally recorded at 24 fps rate, so only the i865G and the RADEON 9100 IGP can provide the required playback speed. Interestingly, none of the Socket A chipsets passed the test of HDTV playback even with an external graphics card. We probably can blame the insufficient optimization of the software decoder for AMD’s processors.
Save for a couple of annoying cases, the integrated chipsets rendered the scene well enough. Only two chipsets acted up. As you may have guessed, they are the ones from VIA and SiS.
First, all chipsets from VIA and SiS didn’t display fog in IL-2. Second, the graphics from SiS has a certain optimization of bi-linear filtering, which makes any smooth gradients into ugly mosaic. This optimization is enabled at multi-texturing in all OpenGL-using games. We covered this matter in more detail and color in our SiS 315 review.
The quality of the image output to the monitor is a sore point of integrated chipsets. They are traditionally inferior to external graphics cards in this respect, mostly due to two reasons: first, integrated graphics processors usually have a lower-frequency RAMDAC than discrete graphics chips. A RAMDAC with a high conversion frequency doesn’t guarantee the quality, but it is a mandatory requirement to reach the highest image precision at high resolutions and high refresh rates. Second, an integrated graphics processor, unlike a discrete GPU, is not a separate unit, but only a functional block of the chipset. Thus it finds itself in an unfavorable environment, next to the other functional units who may bring in noise or cause cross-talk, being located on the same die as the graphics core. Evidently, it takes longer routes to transfer the video signal to the monitor connector, and the lines are laid out among other signal lines.
So, let’s evaluate the onscreen image quality as provided by the mainboards participating in our today’s tests.
I attached a 19” Hitachi CM-661ET CRT monitor to check out the image quality. To support my subjective impressions, I just made snapshots of the onscreen image I got with different mainboards and took them together into a single picture. The original image was a window with Windows’ Explorer:
So, 1024x768x32bit, 85Hz refresh rate:
The image quality is good. A slight fuzziness can be discerned on the VIA KM400, but it is really hard to see.
Next, 1280x1024x32bit, 85Hz refresh rate:
Now we see a noticeable, sometimes strong blurriness. It is the strongest on the SiS661FX and the VIA KM400.
The mainboards based on the i865G and the nForce2 IGP provide the best visual quality.
The next mode: 1600x1200x32bit, 75Hz refresh rate.
It’s quite bad here, especially with the SiS661FX and the VIA KM400.
So, you can expect to see a crisp image attaching a CRT monitor to an integrated chipset and setting the resolution to 1024x768 and the refresh rate to 85Hz.
In the 1280x1024 resolution, all the mainboards, save for the ones based on the SiS 661FX and the VIA KM400, also outputted a sharp picture onto the monitor.
The image quality degenerated in the 1600x1200@75Hz mode, especially with the SiS661FX and the VIA KM400.
Still, you shouldn’t forget that these results only describe the specific mainboard working with the specific monitor. Mainboards from other manufacturers may behave differently, depending on the wiring quality. We have a good example already: the SiS661FX- and the SiS741GX-based mainboards. These chipsets have identical integrated graphics controllers, but the mainboards come from different manufacturers and give out a picture of varying quality: the SiS741GX board, although it comes from ECS, provides a higher image-output quality than ASUS’ one.
Attached to an LCD monitor, an integrated chipset outputs a better picture as a rule. First, there’s no need to set a high refresh rate for an LCD monitor: unlike CRT monitors, LCD panels don’t flicker during refreshes. Second, attached via an analog cable, an LCD monitor digitizes the image before outputting it and thus has an inborn immunity to the distortions of the front of the signal, i.e. to the blur which would be noticeable on a CRT monitor. Of course, this immunity has its limits and cannot help in critical cases.
I proved all these suppositions at practice taking a 17” LCD monitor with a native resolution of 1280x1024 and attached it in the 1280x1024x32bit mode at 60Hz refresh rate. The outcome is visible in the snapshots:
ATI RADEON 9100 IGP
ATI RADEON 9000 PRO IGP
SiS661FX
Intel 865G
NVIDIA nForce2 IGP
SiS741GX
VIA KM400
My expectations came true: all the integrated chipsets, save for the SiS661FX and the VIA KM400, gave out practically perfect images. The SiS661FX and KM400 still produced pictures of a higher quality than those they drew on the CRT monitor in the 1280x1024 resolution.
Let’s do the summing-up from the end.
Irrespective of how you use your system, you want to have the best available image quality from the integrated chipset. My tests on the mainboards included into this review suggest that, save for rare exceptions, integrated chipsets output a good-quality picture to the CRT monitor in the 1280x1024 resolution and lower, at 85Hz refresh rate. When outputting to an LCD monitor, integrated chipsets, again save for rare exceptions, give out nearly a perfect picture.
The rare exceptions in my case are the mainboards based on the SiS661FX and the VIA KM400. In fact, any mainboard may become such an exception – the quality of the output to the monitor depends first of all on the quality of the board itself and only then on the characteristics of the integrated graphics core.
As for the performance in 3D applications, one thing is certain: chipsets from the companies that has a copious experience in 3D graphics are the fastest, and much better than the competitors.
So, if you want to use an integrated chipset in an inexpensive home “entertainment” system and play 3D games from time to time, consider the products from ATI and NVIDIA – only they can provide the bare playability at medium graphics quality settings in “not-very-hard” modern 3D games.
The chipsets from ATI have an additional appeal due to their hardware DirectX 8 support and high-level anisotropic filtering, while the NVIDIA chipsets are the fastest chipsets for the Socket A platform. If the performance of the integrated graphics core becomes insufficient, you may just install an external AGP graphics card.
The existing integrated chipsets from VIA and SiS with their performance in 3D cannot aspire for anything more than a place in an inexpensive office computer or a home “typewriter”.
The i865G looks more advantageous since it is a high-performance dual-channel chipset, interesting to the end-user irrespective of the graphics core. It can form a basement of a powerful workstation with low requirements to 3D graphics or, with an external AGP graphics card, of a full-fledged gaming machine with the graphics core “just in case”.
So, I hope you have formed your opinion about the existing and widely-available integrated chipsets.
The world of integrated graphics is evolving, although slowly. ATI Technologies, a serious new player, appeared recently with its RADEON 9x00 IGP / IGP PRO series. Just a few days ago, Intel announced a new chipset with an integrated graphics core, whose characteristics list hardware support of DirectX 9.
ATI and NVIDIA, having huge experience at developing GPUs, won’t just give up before Intel, but will surely release their own integrated chipsets with hardware DirectX 9 support. There are enough prototypes: ATI has the RV360/380, and NVIDIA can work on the NV34/36 or even some simplified modification of the NV40.
VIA at last seems to be poised to throw the PM800/PM880 chipsets, announced back at the start of the year, into the market. These products feature a new UniChrome Pro graphics controller, based on the discrete AlphaChrome chip, a precursor to the DeltaChrome who never made it into the market. The DeltaChrome itself seems to be waiting its turn to be integrated into chipsets.
SiS also seems to have integrated the Xabre at last. The SiS760, a not-yet-released chipset for the Athlon 64, uses a graphics core with hardware support of DirectX 8.1 shaders. What’s curious, they declare support of the ordinary UMA mode as well as of dedicated DDR SDRAM/SGRAM.
All these facts mean that we will see more new chipsets with integrated graphics in the near future. They will be better and faster and we’ll try to reveal their capabilities to you.
Stay tuned!