X-bit labs - Print version

We are probing a possibility to create overviews of the week’s hottest events featuring commentary from official representatives and industry experts. This is X-bit labs’ pilot issue with the in-depth look into the most significant recent happenings in the personal computer industry in the recent week. Today we are covering the latest stories about AMD Sempron, NVIDIA NV45, dual-core Pentium 4 chips, dispute between NVIDIA and ATI in regards trilinear filtering optimizations, AMD’s plans to supply low-power chips and ABIT’s claims about instability of the NVIDIA nForce3 250-series.

Introduction

Because of Computex Taipei 2004 the week has not been very rich with really important events and, as a result, comments from official and unofficial sources. Still, there have been some very hot topics that definitely worth discussing.

In the particular article X-bit labs does not plan to unveil any astonishingly precise details about the news happened during the week, but is concentrated on brief overview and the review of causes and consequences of everything that happens. We try to offer readers something that is always left behind when reporting the stories by the media – comments from industry experts, officials as well as our own.

Sometimes it is pretty hard or impossible to get comments from officials because of marketing reasons and their resistance to talk about certain products or topics. Nevertheless, end-users should be aware about short-term or long-term directions of the personal computer industry and have a form of discussion prior to official introduction of certain projects. With that said, one of the intentions of X-bit labs’ overviews of the week’s most significant events becomes the initiation of discussions on the level of our readers.

AMD Sempron – The New Duron Hits the Scene

The main headline of the week is indisputably the introduction of AMD Sempron microprocessors that are opposed to Intel Celeron family by Advanced Micro Devices. The main intrigue of the week are the technical specifications of AMD Sempron chips – socket type, cache size, 64-bit support and other important details everyone wants to know about the products that are expected to enter the market from day to day.

AMD vigorously defends its secrets. “More details will be forthcoming in the second half of 2004,” – is the most common answer on all Sempron-related enquiries from AMD’s official representatives.

From unofficial sources it becomes clear that Sempron is not the name of a chip that hits the market just to allow AMD to get rid of its “unconditional” chips, but the name of a new family of chips that will target the entry-level of the market. The new lineup continues the tradition of Duron – the previous-generation processors serving the low-end of the market and discontinued in mid-2003.

Back in 2002 Advanced Micro Devices wanted to kiss the Duron goodbye, but returned it in mid-2003 just in order to satisfy certain orders from some emerging markets, e.g. India, Brazil and China. Eventually the new AMD Duron processors find themselves all around the world, including Europe, Russia and Mid-East, clearly indicating AMD’s desire to sell affordable processors – something for what the company was known for in the early nineties.

With the Sempron, AMD gets back on track of the inexpensive chips, though, this is hardly a negative news, as this is likely to help the company to maintain high prices on its AMD Athlon lineup and achieve excellent gross-margins with its products.

In the early eighties Intel supplied its 8086 and 8088 processors for very expensive and simply expensive desktop computers. In the mid-eighties the industry had 80386DX and 80386SX: one for powerful users and one for mainstream sector. In the early nineties we had 80486DX and 80486SX, again targeting different segments of the market. In mid-nineties we had Pentium processors for the high-end personal computers and 80486DX/Intel DX4 for the others. The latter were not popular at in 1995 and 1996 and they fell into oblivion pretty soon, but already in 1997 we received two lineups back: the Pentium MMX and the Pentium II processors. A year later the former vanished into thin air and the Celeron was born.

AMD itself has never been too devoted to the principles Intel had just because the company’s market share and the number of product families were too small to think about special positioning of different products. Since the 1997 AMD has made a giant leap in the market of central processing unit and since the 1999 the firm has not been considered as a producer of low-end chips, when the original AMD Athlon processors kicked everything out of the Intel Pentium III competitors. The co made a very wise step and introduced the Duron in the year 2000, making it easier to distinguish between the high-end and the low-end CPUs. With the Sempron, the company does the same move once again, though, it is not clear for what AMD needed to discontinue industry-recognized Duron brand-name earlier.

“AMD64 platforms continue to form these days. The introduction of Sempron processor may indicate AMD’s plan to unify its offerings in 2005. However, right now and early next year the number of different processors and platforms – Socket A, Socket 754, Socket 940 and Socket 939 – is just too high and probably may mislead customers,” said X-bit labs’ Editor-in-Chief Anna Filatova.

Fortunately for AMD, it finally gained brand-recognition and with 64-bit trump at hands, the unbelievable product-mix is not likely to slowdown the Sunnyvale, California-based firm’s sales. In contrast, the introduction of reasonably priced offerings for originally pretty luxurious Socket 754 and Socket 939 can boost sales of the latter as quite some customers desire to have some headroom for expansion with their machines.

AMD says that “Sempron” originates from a Latin word “simper” – always. Does it mean that AMD will not discontinue the “durable” chip Duron in future? Well, we will see…

Enthusiasts Confirm: RADEON X800 PRO – Transformable

After the RADEON 9500 and the RADEON 9800 SE were found to be “mod-friendly” into more advanced graphics cards, enthusiasts all around the world started to pay great attention at Powered by ATI graphics products after spending years overclocking NVIDIA GeForce-based graphics cards. The history is repeating today when some hardware maniacs confirm that ATI’s $399 SKUs can be easily modded into $499 graphics cards that have the world’s highest performance these days.

“We do not encourage modifying graphics cards. This is not because it might affect sales, we do not believe that it does, but because if something goes wrong in the modding process, the modder could be left with a dead board. And modding invalidates the warranty, so we would not be able to replace it,” said Chris Evenden, director of public relations for ATI Technologies.

The question that remains unclear is why ATI does not eliminate the possibility to mod the chips – this could give a boost to sales of higher-end graphics cards and also exclude the possibility to damage graphics cards. The matter is truly vague, though, what is absolutely not a secret is that ATI does not want enthusiasts to go the GeForce overclocking route.

Dual-Core Pentium 4 "Prescott" Chips – Something Apple Has Now?

We will never know how Intel saw the future of the microprocessor world when it designed the Pentium 4 processor some 5-7 years ago, but we know for sure that loads of Intel’s plans have been strongly connected with the Pentium 4’s NetBurst micro-architecture for some time now. With aborting of the Tejas and Jayhawk processors earlier this year the question emerged whether Intel will be able to deliver the LaGrande, Vanderpool and other advertised technologies, especially keeping in mind that unofficial sources said Intel canned the Tejas in favour of something Pentium M-like.

The week is also marked with the introduction of Apple’s new Power Mac G5 computers featuring 2.50GHz processors and water-cooling. Apple finally manages to get the speed of its chips to a very high level, however, the company runs into ultimate heat problems that could not be solved using traditional methods. Water-cooling is not considered as cost-effective and mainstream way of cooling chips down, though, with the current trend lead by Intel the situation may change.

“Dual-core microprocessors is a very promising technology that could deliver a totally new level of performance into desktops and laptops, even though, quite a lot of depends on software developers who should optimize their titles to take advantage of two physical central processing units,” said X-bit labs’ platform analyst Ilya Gavrichenkov.

“However, I doubt that dual-core Prescott processors may become a widely-available product even if is made using 65nm fabrication process. Right now heat dissipation of the latest Pentium 4 core is so excessive that a couple of such cores are most probably likely to drive the heat dissipation to an unacceptable level. Well, we have already seen Apple computers with water cooling, Intel may do something like this too, but in that case dual-core Prescott would be something for enthusiasts because of the cost,” continued Ilya Gavrichenkov.

Another possible issue with dual-core Prescott is that the processor’s die is pretty large itself. Produced using 90nm fabrication process the die’s size is 112 square millimeters, which is larger than the company’s previous-generation processor made using 130nm process technology. Intel’s Prescott product with 2MB cache due late this year will have even bigger die and it is likely that the Prescott 2M will eventually make it to dual-core design yielding even larger size of the chip.

Even though Advanced Micro Devices definitely should not be very happy with the size of the AMD Athlon 64 for Socket 939 – 144 square millimeters, even made using 130nm, is pretty large. When transitioned to 90nm, AMD’s chips are likely to have 102 square millimeters die size, nice for single-core chips, but once they get an additional processing engine, the size will be higher than today’s AMD Opteron made using 130nm process technology.

Higher the size of the die is, lower is the yield, higher the final cost for end-users may be.

Intel Pentium M processors are fairly more silicon economical compared to the Pentium 4 and the Athlon 64 chips – with 2MB of L2 cache, the Pentium M “Dothan” chips have only 83.6 millimeters die size. Dual-core Pentium M “Dothan” processors can be the best way out for Intel – even with two processing engines and large level-two cache the chips could be pretty small.

Unlike Advanced Micro Devices, who has to stick to AMD Athlon 64 design, Intel really has options to choose from – Pentium M or Pentium 4. In the first case the company may have difficulties with implementing additional functionality into the processors and achieve high scalability, but in the second case the firm runs into thermal and die-size issues it needs to solve, but, on the other hand, has greater scalability in terms of clock-speeds and “no-hassle” adding of functionality. Which route will be chosen by Intel – it remains to see.

Some companies leak certain pieces of information in order to learn feedback of the industry before making certain decisions. Well, with discussions going around the future of the Pentium M and the Pentium 4 chips at this time Intel definitely has a great chance to find out what is better from customers’ standpoint.

NVIDIA and ATI Fight for Better Trilinear Filtering Quality

The wars for the best image quality in 3D games continued this week when our colleagues from ComputerBase.de web-site published a report concerning a possibility to disable ATI’s optimizations of trilinear filtering from the registry.

Graphics competition these days is more connected with performance numbers for the majority of buyers rather than with image quality for 3D graphics gurus. As a consequence, we should probably expect loads of trick, tweaks and cheats from variety of IHVs in order to pump the speed up.

Getting something of better quality is always a nice option, but with the common approach to simplify everything for the sake of speed it is very questionable whether we should pay attention to the quality of trilinear filtering or we should sacrifice it for higher performance. Or maybe hardware developers should let end-users to decide whether she or he needs the optimizations or not from the driver’s control panel?

“Disabling the trilinear optimization will not enable users to choose between performance and quality. Everyone on the web has been absolutely clear: there is no image quality degradation with this optimization. Therefore, a control panel option would not change image quality, only performance. We do not anticipate adding a checkbox to disable the optimization to future control panels. What would we label it? “Reduce performance”? “Lower frame rates”? Because that is all that switching off the trilinear optimization would do,” commented Chris Evenden for ATI Technologies.

Luciano Alibrandi, European PR Manager at NVIDIA Corp., was more laconic: “Wait few hours… I believe tomorrow morning you will have the response you are waiting for.”

The next morning X-bit labs got a new version of NVIDIA’s soon-to-be-released ForceWare drivers that allowed to turn on and off all filtering optimizations, including trilinear optimization and anisotropic optimization, even though, by default both types of simplifications were enabled.

“If you want to compare red and green performance, please make sure not to compare “full trilinear” with “almost bilinear” (“red driver default”). No matter if someone telling you “patent pending”. In the early nineties, in connection with Winbench, some “driver optimizations” even exceeded the physical memory bandwidth. And even those optimizations have been presented as advantages for the customers,” said Hans-Wolfram Tismer, Gainward’s Managing Director in Europe, who is pretty well-known for making wrong accusations Futuremark’s patches of having negative effect on NVIDIA’s ForceWare shader compiler.

“Modern computer games use extremely complex techniques that are hard to compute in the real-time. In fact, the whole 3D graphics on personal computers is a kind of approximation since it is impossible to get desirable performance amid breath-taking image quality in case we simplify nothing. Therefore, game developers are supposed to approximate certain calculations – including pixel shaders, filtering, etc – in order to make them work quicker and attain high numbers of frames per second. Nowadays games use astonishingly detailed terrains with loads of polygons that do not allow us to see where the full-trilinear filtering and where is the simplified trilinear filtering is,” said Andrew Filimonov, an independent graphics and design expert from Estonia and the developer of Xbitmark benchmark.

“Since the latest controversy hit the web, numerous graphics hobbyists out there have been looking for evidence of image quality degradation. They have found nothing. So I really believe that there are no image quality problems with this [trilinear filtering] optimization. Remember when “optimization” meant “make something better”? This is one of those optimizations, like back in the good old days before the 3DMark03 got compromised,” continued Chris Evenden, ATI Technologies’ Public Relations Director.

“Contemporary graphics cards, such as the NVIDIA GeForce 6800 Ultra and the ATI RADEON X800 XT are so powerful that they hardly require additional performance boosters that degrade image quality. However, for competition reasons both leading developers of graphics products decided to implement optimizations in order to add performance to their new products. Simplifications of trilinear and anisotropic filtering, partial precision of pixel shaders calculations, application specific optimizations, you name it, without letting customers to enable “premier image quality” are the examples of pretty unfair behavior of IHVs. Fortunately, NVIDIA decided to let the customers decide whether they need simplifications of trilinear and anisotropic filtering or not. This is an advantage for the ForceWare, as ATI’s CATALYST drivers still lack this feature,” said Alexey Stepin, a graphics analyst at X-bit labs.

One more problem with allowing customers to disable optimizations is that few end-users have enough qualification to tweak their graphics cards to achieve optimal ration between speed and stability. In case we turn off trilinear, anisotropic filtering optimizations and then disable pixel shader optimizations in FarCry game, we are likely to get slide-show performance even on $499 NVIDIA GeForce 6800 UIltra. Customers will not like low FPS, at the same time, the majority would not see any difference – ATI’s X800 optimizations have been around in the RADEON 9600 for over a year and no one noticed them – between simplified and non-simplified trilinear filtering.

NVIDIA allows its customers to tweak graphics cards according to preferences. ATI does not. Which approach is the right one is something that only time will tell.

ABIT Claims nForce3 250 “Unstable”

ABIT continued this week to demonstrate its position to provide its customers only the best and rock-stable solutions. The company’s anonymous representative told The Inquirer web-site that NVIDIA’s latest core-logic for AMD Athlon 64 processors – the nForce3 250-series – was not enough stable to serve ABIT’s mainboards. Later another source with knowledge of the mainboard maker’s plans told X-bit labs that ABIT still has NVIDIA nForce3 250 mainboards in the roadmap.

“There is a discussion about stability of NVIDIA nForce3 250-series of chipsets…” an anonymous source close to ABIT told X-bit labs.

“Why we are not optimistic about the nForce3 250-series? Well, let me put it this way: we are known for performance, stability, reliability and everything that is involved…” commented a source close to another well-known mainboard maker to X-bit labs.

ABIT KN8, the mainboard that may never become available, photo from Watch Impress web-site

“I can understand ABIT in regards of the nForce3 250. Despite of being very popular among enthusiasts, the previous-generation nForce2 core-logic appeared to be pretty unstable, especially when overclocking. One of the issues with the nForce2 were processor system bus’ frequencies jumping up and down by around 5%, another was lost BIOS settings stored in flash memory. Both troubles probably cost ABIT’s technical support a headache and now engineers from the company want to try everything really thoroughly so that not to run into the same problems again,” said Ilya Gavrichenkov, platform analyst for X-bit labs.

Some believe that NVIDIA will be able to solve all the possible problems with the nForce3 250-series with a revamped revision of the chips. Unfortunately, no one knows when this happens and when mainboard makers are confident about NVIDIA’s latest core-logic.

On the other hand, at least some manufacturers of mainboards, such as Chaintech, ship the nForce3 250-series based platforms now, which means that either they ignore certain possible issues or have ways to overcome the potential problems.

AMD Aims at Centrino’s Laurel Wreath

The trend for low power consuming notebooks with loads of capabilities continues to gather momentum. Customers want to have powerful computers that can play and record DVDs, run computer games and surf the Internet via WLAN networks, but still maintain long battery operation. In order to achieve all such goals, notebook makers have to utilize components with low power consumption, primarily use economical central processing units, hard disk drives and graphics processors.

Last year Intel did what practically revolutionized the market of mobile PCs – it introduced a platform that is fast, reliable and consumes low amount of energy because of using specially-developed microprocessor, which highest TDP is in the range of 25W.

AMD’s positions on the market of mobile products have not been strong historically, but with the introduction of Centrino the company faced even stronger competition from its main rival. All AMD could do is accept the challenge and roll-out powerful processors with extremely low power-consumption. In the Q1 2004 the Sunnyvale, California-based chipmaker already introduced its Mobile Athlon 64 with 35W TDP, but it looks like the company plans chips with 25W thermal design power in the second half of this year.

“A strong side of Intel Centrino is that this is an all-in-one platform for notebook makers: a central processing unit, a core-logic possibly with integrated graphics and a WLAN module. With AMD everything is much more complex for laptop builders: Advanced Micro Devices supplies only microprocessors (even though pretty good one), while notebook designers have to acquire additional components from certain third-party vendors, which is not that comfortable. However, 64-bit capability may push customers and notebook designers to pay more attention to AMD’s chips. Nevertheless, there is always a question about necessity for 64-bit computing – something that has been available on servers historically – on the run,” said X-bit labs’ platform analyst Ilya Gavrichenkov.

Graphics Card Maker Exposes NV45 Product: HSI-Enabled GeForce 6800 Ultra

Rumours about short release of the revamped NV40 graphics processor by NVIDIA Corp. came true this week – the world saw chip code-named NV45, but the product appears to be a bit of disappointment for graphics enthusiasts being the same GeForce 6800 Ultra, but with a bridge to enable PCI Express x16 interface.

NVIDIA declined to discuss its NV45 design saying that the chip is not yet announced, therefore, everything that could be talked about right now is rather untraditional naming pattern of the chip: previously NVIDIA did not introduce the code-name that ends with “5” with chips that do not differ on the micro-architecture level from the “base” chips which code-names end with “0”.

The graphics processor code-named NV45 differs from the product known as the GeForce 6800 Ultra with a special HSI chip placed on the substrate near the graphics processor. This effectively means that the NV45 is still an AGP chip, but since it is to be supplied already equipped with the HSI bridge, it is compatible with PCI Express x16 platforms. Earlier NVIDIA GeForce PCX product-line intended for PCI Express computers had the bridge chip placed onto the PCB of the graphics card, not directly on the substrate of the graphics processor.

The move to install the bridge chip on the graphics processor to enable PCI Express x16 means that NVIDIA could not develop a graphics processor with native support for PCI Express x16 on time or decided not to cope with this possibility for some reason.

Placing the bridge chip on the substrate of graphics processor definitely erodes loads of flexible advantages NVIDIA’s High Speed Interface may have had. On the one hand, NVIDIA still has to make only one type of wafers with the GeForce 6800 Ultra graphics processors for both PCI Express x16 and AGP 8x interfaces, but, on the other hand, when packing the chips, NVIDIA should make a decision whether it needs PEG x16 or AGP 8x products. With the GeForce PCX that required bridge chip to be mounted onto the print circuit board of graphics card NVIDIA had to supply its partners a batch of graphics and HSI chips and let graphics card makers to decide whether they require AGP or PCI Express SKUs.

By contrast, ATI Technologies has two lines of graphics processors – with native support for AGP 8x and with native support for PCI Express x16. This provides relatively low manufacturing flexibility for ATI, but gives loads of options to ATI’s add-in-card partners. Pressure from ATI could lead NVIDIA to offer its customers “all-in-one” PCI Express GeForce 6800 Ultra solution and may force to continue installing the HSI processor on GPU substrate in future.

News Glance: The Essential Commentary on the Week’s Hottest News