[an error occurred while processing this directive]
When ATi released its first really successful desktop chipset – Radeon IGP 9100 – it seemed that the company would focus on products for Pentium 4: the market share is larger, no strong competitors with integrated video (NVIDIA did not have a license to develop products for Intel FSB). However the management of the Canadian manufacturer began to think about the expansion even at that time – surely not to the dying Socket A market but to a promising plot of new AMD sockets. Mass products lacked integrated video here, and so the first company who staked out the claim could count on a great deal. Of course, if you remember 5-year old integrated video, it was an attribute of office computers. But present low end video can manage 2-year old game hits, that is it's sufficient and even excessive for most users. However, ATi ventured to create more than a "standard chipset for AMD64" with integrated video – the company staked on PCI-Express and was the first to start mass production.
We offer a flowchart of new chipsets with their key properties:
You cannot but admit that this time ATi created a truly innovative chipset – all previous generations (IGP 320/340, IGP 9000/9100/Pro) were only more or less good copies of already existing solutions, except for the proprietary Canadian integrated video. Here the introduction of PCI-Express traditionally results in a new video slot PCIEx16, but the four peripheral PCIEx1 ports are connected to the northbridge instead of the southbridge. This solution is quite logical: as the memory controller is integrated into processor anyway, and the northbridge looks "empty" without it. Besides high-speed PCIE periphery now has a shorter way, and correspondingly faster access to memory. Certainly, this bridge of the chipset is also connected to the processor (potentially any AMD64 models) by the HT bus at up to 1 GHz and to the southbridge by PCI-Express bus (parameters of this interface are not provided in the specs). HT at 1 GHz – it's only a neat figure for victorious press releases, applications actually do not need this bus throughput. ATi chipsets have always used standard interfaces to connect to the southbridge (PCI, later HT, now another novelty is selected) – it's easy to implement and allows to use southbridges from other manufacturers (we have seen such an offer from ULi).
Xpress 200 (codenamed RS480) chipset differs from Xpress 200P (codenamed RX480)
in the integrated video available in the former. Note that the motherboard
on Xpress 200 that we tested is actually the first integrated product
for the new AMD platform, though integrated chipsets for K8 had been
announced almost earlier than those processors. But whether 3D graphics
performance of these solutions from VIA and SiS was indecently low
or this integrated video noticeably slowed down the system in common
applications, or because of some other reasons – we have never
seen the production motherboards. As it's claimed in the specs, the
video accelerator in Xpress 200 is a counterpart of Radeon
X300, but exact parameters (like the core frequencies) are not
announced. Remember that X300 is a modification of Radeon 9600 for
PCI-Express, which has hardware support for DirectX 9 (vertex and
pixel Shaders 2.0)
An interesting feature of RS480 is a proprietary technology HyperMemory. Integrated video traditionally uses an allocated region of system memory for a frame buffer. Advantages of this approach are obvious – internal video gets its video memory by means of minor (as a rule) decrease of total available memory without messing with PCB layout and adding additional chips to the motherboard. But a disadvantage of this mode (it's called UMA – Unified Memory Architecture) is the decrease of overall performance – the memory controller has to switch between "important" processor calls and "ordinary" (display refresh) calls of the video chip. ATi offers to equip motherboards with dedicated memory, which will be used only for a frame buffer. There are a lot of implementation options (for example, the buffer size may be up to 128 MB); we'll wait for production samples to see the tendencies. Our engineering sample of the motherboard on Xpress 200 seems to use the minimum option: 16 MB GDDR chip with 2.5 ns access time from Samsung. However, these motherboards can also operate in UMA mode even with available dedicated memory – we shall compare the performance in these two cases when carrying out our tests.
Among additional features of the integrated video RS480 we can note video output to two independent receivers: CRT/LCD or TV/LCD (output to three monitors will be supported in combination with an external ATi video card). D-Sub (analog output) and DVI (digital) will most likely become the standard elements on the rear panel in production motherboards on Xpress 200, like in our engineering sample.
South bridges are the weakest links in the previous generation of ATi chipsets, it was their poor functionality that gave rise to the most unfavourable criticism. This time (the official pair of Xpress 200/P is IXP400) everything is much better. So, it supports:
Everything is up-to-date, there is only no High Definition Audio (high time though, but it's promised only in IXP450(500?)) and integrated LAN. According to ATi, there is no need in integrating a MAC-controller into the chipset due to the appearance of PCI-Express and a large number of gigabit network adapters for this interface (from Marvell, Broadcom?), as one sterling external chip will be cheaper and no slower.
Concluding the descriptive part of the article let me say a few words about the tested engineering samples of motherboards on ATi Xpress 200/P. In our opinion, they do not deserve detailed descriptions, as we don't know what the final motherboards will be (and a motherboard on RS480 was even manufactured by ATi itself and it is probably not intended for mass production). The number of BIOS settings is impressive, but the operating stability is sometimes low – however, we tested the motherboards, "squeezing" maximum from memory, and we used a video card from NVIDIA (ATi does not recommend this combination – allegedly because of potential problems with the PCI-E–AGP bridge in the latter).
We shall use abbreviations in the diagrams for short: "int.video" for cases when integrated video accesses RAM using the UMA mode and "HyperMemory" for cases when integrated frame buffer is used on motherboards (certainly motherboards are also tested with enabled integrated video in the latter case).
Low level analysis of memory speed using RightMark Memory Analyzer (developed by our programmers) answers a lot of questions. There exists some formal difference between the chipsets, but it would be more correct to say that this difference exists between the motherboards (BIOS settings result in minor gains/losses in performance amplified by slight differences in FSB/CPU frequencies). In fact this single test is enough to acknowledge identical performance of the chipsets, but we shall carry out an extensive analysis anyway.
The situation with integrated video is more interesting. The use of UMA slightly decreases the system performance, but dedicated memory (even of a small capacity) in an integrated video accelerator nullifies this performance drop. It's a wonderful result, literally, taking into account that motherboard manufacturers will probably equip their models with frame buffer memory!
Two archivers configured for maximum compression are perhaps the example of applications, which are most critical to memory performance. As you can see here, even with enabled integrated video, ATi Xpress 200/P is practically no different from the VIA chipset. Several more tests to go:
There is no need in comments. These popular media encoders and converters demonstrate only insignificant performance fluctuations in tested systems depending on the real CPU frequency, and the same memory controller in all cases (because it's in the common element of these systems – Athlon 64 3500+ CPU) demonstrates the same performance.
Taking into account the above results, two questions remain vital for the game performance analysis: comparison of VIA K8T800 Pro and ATi Xpress 200P and the difference between memory access modes in the ATi chipset in case of enabled integrated video. In the first case we are interested in the battle of video interfaces rather than in the operating speed with memory (we have already acknowledged it to be the same): ATi Xpress 200 being the first representative of chipsets with PCI Express x16 [in our lab]. In the second case the obvious advantage of the mode with local frame buffer potentially gets balanced by the frame buffer size limit allocated in the main RAM: as ATi recommends, in this mode the UMA buffer size must match the size of the local one (while in the "pure" UMA mode you can allocate up to 128 MB).
Doom 3 demonstrates parity in both pairs compared, with a tiny advantage of ATi Xpress 200/PCIEx16 over VIA K8T800 Pro/AGP 8x. (Remember that the video cards we used for different interfaces are equalized in their frequencies, but the AGP model potentially profits from the lack of an interface bridge [on the card itself].)
The results are more interesting in several FarCry demos (there is some difference between them but the overall picture remains the same): in maximum simplified game modes PCIEx16 and AGP 8x demonstrate parity (also with a slight advantage of the former), but as the video settings get more complex, the AGP modification unexpectedly gains obvious advantage. The same tendency is in the UMA/HyperMemory pair: approximate parity and even slight advantage of the local frame buffer access in the tests with the lowest settings, but the large UMA buffer is an obvious leader when you switch to high resolutions and enable "bells and whistles".
However, Unreal Tournament 2004 always prefers the HyperMemory mode for a change. As a result, the only possible conclusion on the game tests – everything is up to a given game and even a given scene.
When we get hold of production samples of motherboards on Xpress 200, we shall perform additional tests of their integrated video to find their degree of performance conformity to external video cards on X300. Then we'll talk about the use of such video accelerators in modern games. As for now, you can make your own preliminary conclusions by absolute figures demonstrated by Xpress 200, taking into account that the minimum quality mode in 640x480 is purely synthetic, it's not interesting to play in this mode.
ATi managed to create a fast (in other words – not slower than competitors) chipset supporting almost all modern functions. PCIEx16 is not a preferable replacement for AGP 8x, but there is almost no difference for the new systems, while PCIEx1 support is a great advantage of Xpress 200/P for the future. Thus, a motherboard on Xpress 200P is a good if not the best (for the present moment) option for those who are going to buy a new computer with a 64-bit AMD processor. What concerns Xpress 200, this chipset has no competitors: if you want integrated video (for which you won't have to pay by a performance drop), which can at the very least allow you to play games, even if not 2004 hits, then your choice is obvious. We congratulate ATi and look forward to the production samples!
Sergei Pikalov (firstname.lastname@example.org)
December 21, 2004