ATI's RADEON X600 XT
by Anton Shilov
07/09/2004 | 04:52 PM
Even though the news are rather pale these days, some things are happening and if you can dig deeper, you will certainly find some interesting background and probably will be able to make certain predictions of the future.
<%BANNER[article]%>Today we take a look on the ramp of PCI Express graphics and mainboards. After we started the survey of our sources, we discovered an extremely interesting detail – major OEMs prefer ATI’s PCI Express graphics cards, which may be an indication of NVIDIA’s PCI Express strategy fiasco, at least now. Another two things that concern NVIDIA in this coverage are the SLI technology along with Shader Model 3.0.
We also have something interesting about Intel’s last-minute jump on x86-64 bandwagon that resulted in inability of Microsoft’s operating system for x86-64 systems to function on machines powered by Intel’s latest breed of Xeon processors. In addition, we think again about Kentron’s Quad Band Memory technology to realise whether the technology has future or not.
ATI Technologies, the world’s second largest maker of discrete graphics processors, said it had won roughly 100% of the PCI Express graphics business with OEMs. The company also said it expected to get a large market share of the so-called channel PCI Express business.
“Currently, ATI has about 100% of the OEM PCI Express business, and we expect to get a large proportion of the channel business, much larger than we have achieved in the past, too,” said Chris Evenden, a spokesman for ATI Technologies.
ATI's RADEON X600 XT
In late June ATI said world’s leading OEMs, such as Acer, IBM, Dell, MPC Computers, eMachines, NEC, Fujitsu-Siemens, Packard Bell, Gateway and Sony, along with plethora of second-tier computer makers had chosen PCI Express graphics cards powered by ATI Technologies’ visual processing units to install into new PCs.
ATI’s arch-rival NVIDIA Corporation said that among the first-tier PC makers only Acer and MPC Computers had chosen NVIDIA’s PCI Express offerings. Additionally, a lot of second-tier system integrators unveiled plans to use PCI Express graphics cards based on NVIDIA’s graphics processing units. The company denied to indicate any numbers in regards its PCI Express products’ ramp up.
While the hype about the market of PCI Express x16 graphics cards is pretty strong now, the market share of PCI Express graphics devices is unlikely to surpass 30% - 35% mark this year, some analysts noted.
“We strongly believe in the value PCI Express brings to PCs and that is why we have been working with Intel and the PCI Express SIG for the last 18 months. Since it represents a major transition in the PC industry, it is very difficult to estimate just how long it will take. We believe the transition will happen faster at the high-end than the low-end, but we know there will demand for both PCI-Express and AGP cards at all levels for a long time to come,” said Luciano Alibrandi, NVIDIA’s Director of Product PR – EMEA.
NVIDIA GeForce PCX 5900 from Albatron
“There is going to be a difference in take-up between the channel, such as add-in card vendors, and OEMs. OEM will be in approximately 90% cases at PCI Express by the end of the year; the channel is expected to be at around 30%-40% PCI Express by the end of 2004. So overall, 40% - 50% of new PCs will be with PCI Express by
We decided to ask makers of actual components about the transition and selected ABIT, as this is a company that is praised by computer enthusiasts and that was also recently selected by ATI and Intel to supply graphics cards to their most-important customers among system integrators.
"As the industry is in a major transition period, many mainboard and graphics cards manufacturers are challenged by the new technologies. It will take time and effort to educate the channel and end-users to see true values that the new technologies have to offer. Price and availability are other key-factors that will influence the market. I think we all remember the transition from Parallel ATA to Serial ATA. Serial ATA is pretty much a standard today but it took a long time to get to this point,” said Janet Webskowski, a spokesperson for ABIT
Not very optimistic, is it?
“At ABIT, we are pretty confident about the prospects of PCI Express. We are expecting about 80% of our graphics cards production to be PCI Express cards by early 2005. We have our own unique board and cooling design for current PCI Express cards. Soon users will be able to see mGuru chip on ABIT graphics cards for better overclocking and performance,” Webskowski continued.
ABIT's RADEON X600 XT
“As for mainboards, we’re shipping i925/i915-series chipsets based AA8/AG8 mainboards at full speed. We are going to launch a new Fatal1ty mainboard line for gamers and hard core overclockers in August. It’s a very exciting project as we’re working with the world’s top gamers to develop the boards and include all the newest technologies available on the market. Right now the market is still small and it’s growing slowly, but we might see a huge jump by the end of the third or into the fourth quarters when NVIDIA launches its chipset with PCI Express support,” ABIT’s spokesperson explained.
Being pretty concentrated on gamers, ABIT is forecasting 80% of graphics cards to be produced for PCI Express platforms by early 2005. In case the prediction is not overoptimistic, this is certainly a good news for developers of graphics processors…
PCI Express is a next-generation internal interconnection for personal computers that will exist in certain speed incarnations, e.g., x1, x4, x16. The PCI Express x16 lane is intended primarily for graphics cards, works at about 2.50GHz and is able to pump up to 8GB of data in each direction. Due to totally different transfer protocol from the AGP 8x bus, the PCI Express also has some other advantages over AGP, such as isolated read and write channels, ability to prioritize the data streams and a number of others.
While both NVIDIA and ATI are ready to offer their latest graphics processors to clients who plan to acquire personal computers with PCI Express x16 bus (PEG x16), approaches the companies used to add PCI Express interface into their chips seem to be different. ATI Technologies uses built-in PCI Express x16 controller and is able to offer fully-featured so-called “native PCI Express” solutions. NVIDIA uses a special chip that bridges its AGP graphics processors with PCI Express x16 platforms. NVIDIA claims that PEG x16 bandwidth or bi-directional transfer ability are not sacrificed as a consequence of its “bridge approach”, though, certain PCI Express functionality may not be enabled on NVIDIA’s graphics cards with the HSI bridge, some analysts said.
ATI Technologies currently offers RADEON X300-, RADEON X600- and RADEON X800-series of graphics processors with PCI Express x16 interface targeting entry-level, mainstream and high-end markets. NVIDIA targets the same market segments with its GeForce PCX 5300, GeForce PCX 5750, GeForce PCX 5900, GeForce 6800 GT as well as GeForce 6800 Ultra graphics cards supporting PCI Express x16 interconnection.
Smaller graphics chip designers, such as S3 Graphics, 3Dlabs and XGI Technology, are likely to jump on the PCI Express x16 bandwagon a bit later this Summer.
With the launch of Intel Xeon “Nocona” processors the world’s largest chipmaker admitted: transition of conventional x86 architecture to 64-bit is an inevitable result of technology evolution. However, this evolution may still be a couple of years away from business desktop computers.
Probably for the first time in history Intel Corporation decided to follow its much smaller rival Advanced Micro Devices, who become well-known in the eighties for making clones of
Intel's vision of workstation with x86-64 processor inside
Intel has been downplaying the importance of transition to x86-64 for years and formally launched its Xeon processors with 64-bit capability about 14 months after AMD. While the delay is not that dramatic amid the absence of any mainstream 64-bit operating systems, it turned out that Microsoft intentionally allows its beta version of Windows for x86-64 systems to run only on AMD64-based machines. As a consequence of this, Intel-based x86-64 computers cannot be used for x86-64 software testing purposes, which may generally slowdown transition to 64-bit computing.
While AMD and Intel’s processors with 64-bit extensions are generally compatible, some capabilities of the chips are not equal. For example, AMD sports 3DNow! instructions, while Intel sports Hyper-Threading technology. Furthermore, chips may handle different instructions in different way, therefore, Microsoft will need to do some additional work in order to ensure flawless compatibility with Intel’s microprocessors. Moreover, independent software vendors and driver development teams will have to spend some additional time tweaking their software to run well on Intel’s IA32e Pentium 4 and Xeon microprocessors. With that said we may anticipate that the beginning of transition to 64-bit will be generally delayed because of Intel’s problems, but this is something that we need to take a closer look at, software experts tell us.
“Another situation is development of application software, e.g., antivirus, office, data-base software. In that case developers need software development kit tailored for particular operating system and hardware. In general, SDK requires final version of operating system, which is not present in the particular case. The absence of well-documented and completely tested development tools will not allow developers to produce applications for 64-bit computers, which will indisputably slowdown the transition to x86-64 systems,” Mr. Kirillov continued.
AMD's outdated vision of software transition to 64-bit
It is necessary to point out that AMD still did not deliver its SDK for Windows XP 64-bit Edition. All the demos that ought to showcase tremendous advantage AMD64 gets in 64-bit mode are created using assembler or some beta tools, which cannot be used for mainstream software.
“In the worst case for Microsoft and AMD the release of operating system for 64-bit extended systems will be postponed as a consequence of the fact that IHVs will not be able to certify their drivers with Microsoft because of absence of proper tools for driver development,” the software expert added.
Currently there are three companies to have publicly released drivers for 64-bit systems: ATI, NVIDIA and VIA. Pretty limited support from IHVs does not inspire a lot of optimism, does it?
NVIDIA’s Scalable Link Interface and Symmetric Multi Rendering announcement struck like a storm: nobody had leaked any details concerning this possibility before the launch and nobody could tell anything in addition to what NVIDIA indicated in its statements for media. Since secrets are not kept really well these days, such situation is a bit strange.
NVIDIA’s Scalable Link Interface, dubbed SLI, is a set of hardware and software capabilities that allows two PCI Express x16 graphics cards to work in parallel and deliver higher performance compared to one such graphics card would. With the announcement of the technology that requires a special mainboard and an extremely powerful PSU along with two graphics cards NVIDIA sends a clear message to the world that says “performance at any cost”.
Two GeForce 6800 Ultra graphics cards
Neither NVIDIA, nor ATI Technologies could demonstrated an overwhelming performance advantage with the latest GeForce 6800 Ultra and RADEON X800 XT graphics processors. Sometimes the GeForce 6800 Ultra leads, sometimes, especially in cases when full-scene antialiasing and anisotropic filtering are enabled, ATI’s offering beats the rival. Two NVIDIA GeForce 6800 Ultra graphics cards will deliver unbeatable performance, at least unless ATI announces similar technology. Moreover, dual graphics cards will deliver strong performance in professional applications, something that NVIDIA needs to address the very high-end of the market.
When announced, NVIDIA said the new technology will be available in Fall for desktops and immediately for workstations. For some reasons, no first-tier workstation vendors announced support for Scalable Link Interface and Symmetric Multi Rendering technologies immediately.
Initially power-hungry users will have to buy workstation components or special workstation or gaming machines to enjoy the power of two GeForce 6-series graphics cards. Desktop chipsets available this Summer and Fall from Intel, VIA Technologies and Silicon Integrated Systems are unlikely to support two PCI Express x16 ports for graphics cards. The only core-logic that has support for PCI Express x16 + PCI Express x8 ports is Intel’s E7525 also known as Tumwater designed for workstations. The core-logic itself costs $100, which leads to assume that mainboards based on the chipset will be priced in the range of $350 to $500 or even higher.
Intel E7525 Mainboard
A more user-friendly option from the price point of view will be brought by NVIDIA itself in the fourth quarter of the year. The company will release its NVIDIA nForce4 chipset that will sport two slots for PCI Express graphics cards and will be compatible with the desktop infrastructure, meaning significant cost reduction compared to the Intel Xeon platforms. In case NVIDIA succeeds, end-users will be able to get desktop computer at desktop price-point with one GeForce 6-series graphics card and add another card in future. The only thing end-users should be worried about is powerful and quality PSU.
However, it is pretty interesting to note that mainboard makers got the information about the nForce4 peculiarities at the day of the SLI announcement.
“We do not know exactly what the nForce4 is. NVIDIA wants us to make this, we would like too, but we are not sure about the specs of the product. That was a news for us when we were told that the chipset is set to run two graphics cards,” a spokesperson for a mainboard maker said wishing to remain anonymous.
ATI Technologies, who once failed with MAXX dual-graphics chip approach, was more than quiet with its comments on NVIDIA’s SLI: no messages sent to officials on the matter were answered; still, some representatives for ATI and Alienware, who recently disclosed plans to offer desktops with two graphics cards, commented on NVIDIA’s tech for media.
“Should NVIDIA’s SLI technology ultimately prove faster than Alienware’s Video Array, we will offer it to our customers as soon as it becomes commercially available,” stated Frank Azor, senior vice-president and general manager of Alienware’s Worldwide Product Group.
“Either way, we will continue to develop and optimize Video Array for all non-NVIDIA graphics solutions, such as those from ATI, 3D Labs, Matrox, and any other manufacturers. Our sole interest is in offering technology that provides the very best computing experience possible to Alienware customers,” Mr. Azor added.
Two NVIDIA GPU-based graphics cards work in pair
“ATI could, of course. release its own implementation of SLI, or multi-chip graphics cards, but ATI has no intentions to do this in the near future. The RADEON X800 VPUs support the same scalability as the RADEON 9700 and similar graphics processors, up to 256 VPUs can work in parallel, and graphics cards like SGI’s Quad-RADEON 9800 XT can be made with the RADEON X800 chips. Moreover, technologies that would require more than two PCI Express slots for graphics cards could be made, which is not a problem, as there are chipsets with more than 24 PCI Express lanes coming out soon,” Nils Horstbrink, ATI’s desktop product manager told Golem.de web-site (translation is published by XtremeSystems.org).
“The pairing of ATI’s innovative graphics technology and Alienware’s Video Array has some truly awesome implications for 3D-intensive applications like gaming, animation, and real-time simulation. Alienware’s solution is flexible and powerful, allowing customers to push the performance envelope” said Rick Bergman, Senior Vice President, Marketing, General Manager, Desktop, ATI Technologies.
Well, ATI indicates that it could develop a solution that would allow two consumer graphics cards to work in parallel, but decided not to do this for some reason. The reason for this statement might be very high price of such solution, inability to deliver such product or just unwillingness to disclose its plans…
Kentron does it again: it speaks about advantages its Quad Band Memory brings to users without providing anything real. First announced on
Quad Band DRAM allows to double memory sub-system’s bandwidth using the widely-spread DDR SDRAM chips. Modules, that correspond to QBM standard make use of ordinary memory chips but add a special PLL (switch) to allow half of the chips to work at 90 degrees displaced frequency, hence, data becomes available almost two times more frequently compared to ordinary modules. Basically, such way of functioning resembles a 128-bit memory module with a pair of 64-bit channels that switch, in turns, one over another.
Quad Band memory module
The peak bandwidth of such RAM system can be up to 4.2GB/s when using ordinary 266MHz memory chips (PC2100), the main advantage of this technology is its presumable cost-efficiency. Since QBM technology does not require any additional signaling, QBM memory modules can maintain compatibility with widely available memory slots, such as those used for DDR or DDR2 memory. Still, current version of QBM requires chipset support, which, in the past, lead to Kentron’s flop in the market as VIA Technologies, the only maker of chipsets who had planned to support QBM, dropped the idea of Quad Band Memory-supporting core-logic.
In a recent interview representatives for Kentron Technology indicated plans to release QBM-2 specification along with memory modules featuring DDR2 chips in 1H 2005. This time Kentron does not count on third-party chipset makers, but intends to install a special chip direct onto memory modules.
“Having extra logic to implement into the memory controller limited us to the memory controllers that supported you, which was one. In order to have broad acceptability, we want the user to make the decision rather than the chipset product group,” a Kentron’s spokesman said.
Additional chips and logic have never made products more affordable and from an end-user standpoint it is always more efficient when chipset’s North Bridge supports variety of memory module types, while memory modules themselves carry on memory chips only with no additional logic. However, Kentron believes that additional bandwidth its memory modules will be able to provide will make the chips more valuable than competing products.
The main idea of QBM is providing peak bandwidth of dual-channel memory sub-system by single-channel QBM memory sub-system or providing the same bandwidth using two memory modules featuring lower-speed components. Both cases seem to have their cost advantages, but any new technology has to have something more than just a tiny pricing benefit. Customers should see what they would get with the addition of higher-speed, more expensive and unknown memory modules. Furthermore, under the “customer” term we should consider a system integrator or an actual end-user, two types of clients who are pretty slow when it comes to a new technology that is not pushed in by loads of hardware companies.
Dual-channel QBM DDR2 667MHz sub-system can provide 21.3GB/s bandwidth, a lot more than today’s and tomorrow’s central processing units require, which will hardly make QBM popular in PC segment. Another target market for the QBM memory is communication equipment and consoles, both usually require extreme bandwidth. Rambus succeeds in pushing its memory technologies into comms and consoles, whether Kentron succeeds in this (maybe with Xbox 2?) remains to be seen.
NVIDIA and Crytek recently released the FarCry 1.2 patch to media in order to showcase end-users the advantages the Shader Model 3.0 brings in terms of speed. While custom-made demos in the game did not reveal huge performance benefits, specially-created demos to reflect Shader Model 3.0 (SM 3.0) advantages showcased tangible performance boosts because of geometry instancing and use of pixel shader 3.0 for complex per-pixel lighting.
Despite of ATI’s pretty negative attitude towards SM 3.0 these days, pixel and vertex shaders 3.0 are here and begin to ramp up; they, in spite of some negative predictions, bring performance advantage in current games and are likely to boost speed in future titles.
Unfortunately for both NVIDIA and ATI, the majority of games coming out these days hardly use Shader Model 2.0 seriously, not talking about anything more advanced. However, there are a number of titles that can change the world of gaming and take advantage of the Shader Model 3.0 that is supported now only by NVIDIA. Nevertheless, it looks like we are going to see hardware supporting Shader Model 3.0 from ATI Technologies pretty soon, as Shader Model 4.0 is still too far away and game developers along with marketing teams want to have something to jump on.
Graphics experts consider Shader Model 3.0 as a more advanced superset of the Shader Model 2.0 with more sophisticated approach for programming, which may encourage game-developers to create higher-quality titles in terms of graphics. It should be mentioned that basically all SM 3.0 effects can be made using SM 2.0 and are likely to be made using 2.0 because of broader compatibility. Though, because of performance advantage game-developers are projected to use SM 3.0 for supporting hardware in order to allow their games to perform better. Moreover, eventually those, who have SM 3.0-supporting hardware may see some image quality benefits too.
Back in 2001 NVIDIA downplayed the importance of Pixel Shaders 1.4 developed by ATI saying that game-developers would jump on shaders 1.1 and yet-to-be-release 2.0. In early 2003 NVIDIA encouraged game-developers to use 1.4 instead of 1.1 and 2.0 because of dramatic speed-advantage. What did not bring a boost for the RADEON 8500 brought an advantage for the GeForce FX that came out 1.5 years later.
All in all, Shader Model 3.0 is definitely a feature for considerations now, even though its broad use seems to be 6-12 months away…