Graphics Outlook 2003

January 5, 2003 by Leonidas / translated January 10, 2003 by zeckensack & nggalai / page 3 of 4

  Deutsche Version anzeigen

   Second quarter:   ATi R350

Only a few months after the GeForceFX's appearance, ATi will answer with their R350. With the tape out stage already behind them, a market entry in the 2nd quarter is within reach. ATi may even present the new chip at CeBIT, carrying the expected "Radeon 9900" branding, and thus disturb the GeForceFX launch. It's hard to predict whether the R350 will be able to beat the GeForceFX or not, a draw is just as likely.

ATi's options for R350 are limited: Manufactured under the same process rules as the Radeon 9700 Pro, 400MHz is the chip's likely clock speed ceiling and the GeForceFX would still win. Significant improvements can only be made on the memory side: Outfitted with a DDR/II interface and memory chips the R350 could run its bus at 500MHz, yielding simply unbelievable memory bandwidth. But without being able to match GeForceFX's rendering brunt, a draw in performance between the two is the most likely outcome. At least we believe that if the R350 manages to beat the GeForceFX, then it'll only do so by a slim margin.

Another uncertainty about the R350 is whether ATi will improve its programmability up to GeForceFX levels. We really think that it's a bit too early after the Radeon 9700 Pro to expect such drastic changes to the architecture. And increased clock speeds and small tweaks to the memory interface can't justify a new chip, with costs for a production switch in the dozens of million dollars. We'll have to wait, probably we'll still see some minute flexibility improvements.

From a price perspective, the R350 will indeed "be able to" pull even with the GeForceFX. The production costs of 0.15 micron chips running at 400MHz will be even higher than those of the Radeon 9700Pros, so ATi will likely have to charge 500€ a piece to make any profit. It's just as likely, by the way, that we'll see the same cut-down versions that we've become used to with the Radeon 9700 Pro (9700 non-Pro, 9500 /Pro). These serve to keep yields high by still being able to sell chips that don't meet the frequency criteria for the 'big' chips, or chips with defects in some of the rendering pipelines (defective ones can be deactivated).

   Second quarter:   ATi RV350

Most likely on the same day as R350, ATi will also unveil another DirectX9 chip, this one has supposedly already taped out too. We need to mention that on other accounts the RV350 will not surface before the 3rd quarter. The RV350 is supposed to fit the needs of a wholly different market segment, ATi have planned it to be the first really affordable DirectX9 graphics chip. Seeing that the Radeon 9500 / Pro already occupy the DirectX9 segment at around 200€ for ATi, this can only mean that ATi plans to price the RV350 significantly below the 200€ barrier, serving DirectX9 capabilities to the mass market.

To achieve this goal the RV350 will be ATi's first chip to be manufactured on a 0.13 micron process, bringing costs in line with the target market. It's also quite sure that the RV350 will only come with four rendering pipelines to reduce the transistor count and accompanying cost. Judging by the similar architecture it will depend on the clock speeds, where this part will stand performance-wise when compared to the Radeon 9500 Pro. But we think ATi will want to compete well against the NV31 which will be out in the same time frame, and adjust frequencies accordingly.

Because RV350 will deliver better performance than the Radeon 9500 /Pro at the same or lower prices, ATi will phase out the latter. In comparison to RV350 they will be too expensive to make. RV350 however will be with us for a long time and become more and more affordable over time to truly transform into a mass market DirectX9 accelerator. We also expect "Pro" and "non-Pro" variants of RV350 again, seeing that ATi recently made a habit of releasing such different versions.

   Second quarter:   Trident XP4e

The XP4e is a simple AGP8x respin of the original XP4. Current plans don't call for changes in clock speeds, so the performance of this part will be exactly the same as XP4's. Scheduled for mass production in May, this chips will be so late to market that original XP4 pricing won't be sensible any longer, Trident will have to budge, somewhat. Maybe Trident even cancels the project altogether, outdated as it will be by then. The original XP4 was already delayed quite enough.

   Second quarter:   PowerVR Series 5

After the demise of Series 4 (aka Kyro III), PowerVR concentrated their efforts on Series 5, which has been in development since 2001 and is to be released sometime in spring. Unfortunately, technical specifications about this project are extremely scarce, it will be a tile based deferred renderer, that's the only known fact :-). We can only speculate whether it will be a DirectX8 or DirectX9 part.

One might believe that PowerVR will only be able to stand a chance with the "DirectX9" label, if they want to compete in 2003's mass market. But from a usability perspective, DirectX9 isn't really necessary for quite some time yet. PowerVR has traditionally been content with fulfilling just the necessities. Update: It's going to be a DirectX9 part, but please don't ask us about our sources, mkay? ;-)

We can say at least one thing: if the chip's in time and performs sufficiently, it'll be most interesting to watch how Series 5 will fare against ATi's RV350 and nVidia's NV31. As was to be expected, PowerVR is again headed for the sub-200€ market.

But please keep in mind: After losing STMicro as co-developer of the chip, PowerVR can only develop the chip, but won't be able to work through the production details. So, if no willing-and-able partner shows up when Series 5 is finished, it will suffer the same fate as Series 4 and be scrapped without reaching the graphics card market. Without a reliable partner, PowerVR Series 5 must be considered a theoretical exercise only.

   Second quarter:   SiS Xabre 2

It's already been a while since Xabre 2 showed up on roadmaps for the first half of this year, outlined as a DirectX9 chip on a 0.13 micron process. Except for the customary eight rendering pipelines, very little is known about it, but we assume that SiS turned their backs on (slow) software emulated vertex shaders and implemented a hardware vertex unit for the Xabre 2. In DirectX9, vertex shaders are far too important to be left solely to the CPU.

We also assume that SiS will keep the pricing strategy of the first Xabre and try to establish the follow-up part at around 150€. The success of this strategy will largely depend on whether Xabre 2 will again try to grab high benchmark scores by severly reducing texture quality. If that happens, we expect an even larger backlash than with the original Xabre. These kinds of sacrifices won't be as easily tolerated this year.

   Third quarter:   S3 DeltaChrome MS

Being a low-cost derivative of the original DeltaChrome, the DeltaChrome MS will only have four rendering pipelines. The vertex and pixel shader implementations seem to be somewhat cut down too, S3 reduced the "2.0+" specification to "2.0" for DeltaChrome MS. Core and memory clock speeds are supposed to be faster than 300MHz. We only know that DeltaChrome MS should come shortly after DeltaChrome, without having details about the exact time frame. The competitors of this chip probably will be RV350 and NV34 - but it'll likely be priced even lower.