ATi RV250 + R300 and nVidia NV30 specifications

July 14, 2002 / by Leonidas / page 1 of 2 / English version translated by Zeckensack / updated July 18

We've already posted similar articles on May 22 regarding nVidia's chips, and on June 6 regarding ATi's chips. A lot has changed in the meantime. Some points of the original articles are simply outdated by now. That's why today we'll revisit the current state of things to come from both ATi and nVidia.

Sources: AnandTech #1 + AnandTech #2 + The Inquirer #1 + The Inquirer #2 + ATi August 2000 roadmap + ATi August 2001 roadmap + MyHard (ATi June 2002 roadmap) + K-Hardware + 3DChipset + DarkCrow + Beyond3D Forum + nVidia June 2002 shareholders meeting + own sources + RV250 & R300 launch data.

ATi RV250 (LE)

35-45 million transistors
0.15 micron manufacturing process by TSMC
256 bit architecture
4 rendering pipelines with 1 texture unit each
6 textures per pass
Single Vertex Shader engine
complete DirectX 8.1 compatible
Vertex Shader 1.1
Pixel Shader 1.4
128 bit DDR-RAM memory interface
Core clock: RV250 275 MHz, RV250LE 250 MHz
Memory clock: RV250 275 MHz, RV250LE 200 MHz
HyperZ II
HydraVision
AGP 2.0 specification with max AGPx4 mode
max 128 MB onboard memory
Presentation on July 17, 2002
Mass production probably running now, on shops July/August
Retail name Radeon 9000 (RV250LE) and 9000 Pro (RV250)

Short summary: a replacement for the Radeon 8500 without any important changes. There will be a few slight simplifications to make the card cheaper to produce but we've learned that the 3D part won't be affected by this. So the original Radeon 8500 and an RV250 are theoretically to have the exact same performance when running at equal clock speeds.

With this chip, ATi will be targeting the mainstream market starting at prices around 150 $/€. The chip and board will be designed to allow for a price reduction to around 100 $/€ and less over a few months, while still earning ATi a profit. As there are no known significant changes, at least concerning the 3D core, the RV250 should be able to offer an excellent price/performance ratio and likewise challenge nVidia's Geforce4 Ti4200 on own ground.

ATi R300

110 million transistors
0.15 micron manufacturing process by TSMC
512 bit architecture
8 rendering pipelines with 1 texture unit each
16 textures per pass
Quad Vertex Shader engine
complete DirectX 9.0 compatible
Vertex Shader 2.0
Pixel Shader 2.0
Displacement Mapping support
128 bit rendering precision
Floating point rendering
256 bit DDR-RAM memory interface
Core clock: ca 300-330 MHz
Memory clock: ca 300-330 MHz
HyperZ III
HydraVision
AGP 3.0 specification with max AGPx8 mode
max 256 MB onboard memory
Presentation on July 17, 2002
Mass production on early fall, probably August/September
Retail name Radeon 9700

By now we're positive that ATi's R300 will definitely be manufactured on a 0.15 micron process. Still, according to own roadmaps, ATi targeted 350 MHz core clock - a daring goal for 107 million transistors, which probably can't be reached. The switch to 0.13 micron is already planned so a spring refresh is a definite possibility. We still think this switch is highly unlikely to occur before the end of the year, mostly because of currently insufficient capacities at TSMC on this process.

nVidia NV30

120 million transistors, probably divided in a rendering chip and an extra Vertex Shader chip
0.13 micron manufacturing process by TSMC
256 or 512 bit architecture
8 rendering pipelines with 2 texture units each
16 textures per pass
probably Quad Vertex Shader engine, probably as extra chip
complete DirectX 9.0 compatible
Vertex Shader 2.0
Pixel Shader 2.0
Displacement Mapping support
128 bit rendering precision
Floating point rendering
128 or 256 bit DDR-RAM memory interface
Core clock: ca 400 MHz
Memory clock: probably ca 450 MHz
4x Anti-Aliasing with dramatic less performance loss than earlier chips
Lightspeed Memory Architecture III
AGP 3.0 specification with max AGPx8 mode
max 256 MB onboard memory
probably MultiChip-capable
Presentation sometime on the end of summer
Mass production on late fall, probably October/November

nVidia's NV30 is surrounded by even bigger mystery than ATi's future chips - no surprise, nVidia will definitely release their fall part later than ATi, and will probably start shipping much later. 0.13 micron technology can be taken for granted, but chip foundry TSMC is still wrestling with yield issues on this process. That's why the NV30 is not expected to hit the shelves before October/November.

Many of the above NV30 specs are directly related to the minimum requirements of DirectX 9.0, which this chip will support completely in hardware, as will ATI's R300. Whether the NV30 will offer 128 bits or 256 bits here is still one of the biggest mysteries about this high end chip to come.

Worth mentioning regarding NV30 is the noticeable influence of 3dfx technology - in case the chip really has an external Vertex Shader unit and is able of multi-chip layouts. An external Vertex Shader unit was subject to be part of the latest 3dfx chip Rampage, which never reached production, with 3dfx having a long tradition and experience in multi-chip layouts. However, it is still too early to state whether nVidia is going to use the multi-chip capability of the NV30 in the customer market, or withhold it for the professional chip segment.