ATi RV250 + R300 and nVidia NV30 specifications
July 14, 2002 / by Leonidas / page 2 of 2 / English version translated by Zeckensack / updated July 18
The following is a comparison between the four big chips of this fall and the ATi RV250 - we must put emphasis on the uncertainty of some of the information again. Even small changes in some areas may mean a considerably different outcome regarding the balance of power. So please take everything with the mandatory grain of salt, especially the quoted clockspeeds:
Graphic Chips Q3/Q4 2002 | |||||
3Dlabs P10 |
ATi RV250 |
ATi R300 |
Matrox Parhelia |
nVidia NV30 |
|
Technology class |
DX8 | DX8 | DX9 | DX8 | DX9 |
Transistors (Mill.) |
75 | 35-45 | 110 | 80 | 120 |
Process (micron) |
0.15 | 0.15 | 0.15 | 0.15 | 0.13 |
Architecture (bit) |
512 | 256 | 512 | 512 | 256 or 512 |
Rendering pipelines |
4 | 4 | 8 | 4 | 8 |
Texture units (each) |
2 | 1 | 1 | 4 | 2 |
Textures per pass |
8 | 6 | 16 | 4 | 16 |
Core clock (MHz) |
ca 250 | RV250: 275 RV250LE: 250 |
ca 300-330 | Retail: 220 Bulk: 200 |
ca 400 |
Fill rate (GPixel) |
ca 1.0 | RV250: 1.1 RV250LE: 1.0 |
ca 2.4-2.64 | Retail: 0.88 Bulk: 0.8 |
ca 3.2 |
Fill rate (GTexel) |
ca 2.0 (trilinear) |
RV250: 1.1 RV250LE: 1.0 (trilinear) |
ca 2.4-2.64 (trilinear) |
Retail: 3.52 Bulk: 3.2 (bilinear) |
ca 6.4 |
Memory interface (bit) |
256 DDR | 128 DDR | 256 DDR | 256 DDR | 128 or 256 DDR |
Bandwith features |
Virtual Memory System | HyperZ II | HyperZ III | - | Lightspeed Memory Architecture III |
Memory clock (MHz) |
ca 300 | RV250: 275 RV250LE: 200 |
ca 300-330 | Retail: 275 Bulk: 250 |
ca 450 (?) |
Bandwith (GB/sec) * |
ca 17.9 | RV250: 8.2 RV250LE: 6.0 |
ca 17.9-19.7 | Retail: 16.4 Bulk: 14.9 |
? |
Vertex Shader engine |
Quad | Single | Quad | Quad | Quad (?) |
Vertex Shader version |
2.0 | 1.1 | 2.0 | 2.0 | 2.0 |
Pixel Shader version |
1.3 | 1.4 | 2.0 | 1.3 | 2.0 |
DirectX version |
8.1 (partly 9.0) |
8.1 | 9.0 | 8.1 (partly 9.0) |
9.0 |
OpenGL version |
2.0 | 1.3 | 2.0 (?) | 1.3 (partly 2.0) |
2.0 (?) |
Anisotropic filtering (max level) |
? | bilinear: 16x | bilinear: 16x trilinear:16x |
bilinear: 16x trilinear: 8x |
? |
Anti-Aliasing (max samples) |
8x (MS) |
6x SS |
6x MS |
16x (FAA) |
? |
Rendering precision |
minimum 64 bit (integer) |
? (integer) |
128 bit (floating point) |
40 bit (integer) |
128 bit (floating point) |
Displacement Mapping |
yes | no | yes | yes | yes |
AGP specification (max mode) |
2.0 (4x) |
2.0 (4x) |
3.0 (8x) |
2.0 (4x) |
3.0 (8x) |
Onboard memory (max. MB) |
256 | 128 | 256 | 256 | 256 |
MultiChip capable |
no | no | no | no | yes (?) |
* on an 1 GByte = 1024 Byte ^ 3 basis |
Given the uncertain nature of some key information we'd rather not make any forecasts regarding the two main contenders' respective performance - these would be ATi R300 and nVidia NV30 of course. At least both chips will be considerably faster than every other chip on the market, that much is for sure, and they'll be the only ones with complete DirectX9 support. The battle for supremacy will be fought between these two - it remains to be seen how much of a contender the 3DLabs P10 will be.
In closing we'd like to summarize the information already known about the chips coming yet later, after the "hot graphics fall of 2002":
ATi´s RV280 is supposed to be the AGPx8 version of RV250 - which might also be a pipe dream as there's only one little hint towards the bare existance of this chip. If ATi really has plans regarding an RV280, it'll possibly - but only slightly possibly - be an 0.13 micron part. The RV280 is expected on early winter 2002.
ATi´s RV300 & RL300 are both chips mentioned on older ATi roadmaps. These suggested that both would be cut down versions of the full R300, with only two and one vertex shader respectively, most likely lower clock speeds. We don't know whether these old plans of ATi are still current. Neither chip will see the light of day before spring 2003.
ATi´s R400 is the next big thing after R300 and its respins (which include an 0.13 micron version). Specs are unknown. From a schedule point of view it should be ATi's fall part of 2003, but this is unconfirmed.
nVidia´s NV18 & NV28 are AGPx8 variations of the present chips NV17 (GeForce4 MX) and NV25 (GeForce4 Ti). Up to now it is believed that both chips are otherwise unchanged and are still manufactured in a 0.15 micron process. According to nVidia's plans NV18 and NV28 will be launched in August/September.
nVidia´s NV31 will supposedly be a purely clock speed reduced - and otherwise unchanged - version of NV30 for the mainstream market. Having no real information on this highly debatable part we'll just have to wait and see. The supposed launch date is one month after NV30 - still in time for this year's Christmas.
nVidia´s NV35 should be NV30's spring refresh with increased clock speed as the main difference. There are talks about core and memory clocks of 500 MHz. According to the original plans it was slated for introduction in March 2003 but with the current delay of the NV31 it might well be pushed back into summer.
nVidia´s NV40 is only looming on the far horizon, like ATi's R400. Architectural details are unknown, but there are already a few performance figures: 600 million polygons/s and a fill rate of 4 GPix/s. Further assuming eight rendering pipelines this would mean a 500 MHz core clock. Extrapolating from vertex shader performance seen on Geforce 3/4 - but most likely implemented as an external chip, like NV30 - and assuming a quad vertex shader, this unit would run at 600 MHz. Of course this assumption might be completely off the mark, if the vertex shader architecture differs too much. The NV40's release date is commonly assumed to be fall 2003 but as with ATi's R400, there's no confirmation.
Unfortunately, there's zero information about future chips both from 3DLabs and Matrox.