VOGONS


First post, by ratfink

User metadata
Rank Oldbie
Rank
Oldbie

I have a few of these Wildcat VP cards, they use something called a P10 VPU and were aimed at being mid-range workstation cards though supposedly Creative considered releasing something based on them as a consumer card. Found some oddities when I ran 3dmark2001se, thought I'd share - no big deal as these aren't really gaming cards.

These are Dx8.1 and OpenGL cards with driver optimisations/settings for games, so I tried some benchmarking to see whether there was much difference between my cards [3 models], and what difference AA made.

The cards:
VP990 Pro, 512mb memory, 225 vertices/sec
VP870, 128mb memory, 188 vertices/sec
VP760, 64mb memory, 165 vertices/sec

The test: 3dmark2001se on defaults with no AA, 2-sample AA, 4-sample AA, driver set to Direct3d games default [and opengl games default if that matters]

The test system: Soyo 845 board with 2.8ghz P4 and a Santa Cruz for sound; 512mb ram. Windows 2000 with no service packs [the 990 is not compatible with later packs - immediate crash to desktop]. Latest driver used for all cards [same driver file from 3dlabs site].

The results:

No AA: The mid-range 870 beats the 990, with the 760 trailing.
990: 7857 (1.3% lower than 870)
870: 7961
760: 6918 (13% below870)

2-AA: The 990 is the slowest, the 870 leads still
990: 3764 (3.7% lower than 870)
870: 3907
760: 3804 (2.6% lower than 870)

4-AA: same as for 2-AA
990: 1783 (6.3% lower than 870)
870: 1903
760: 1875 (1.5% lower than 870)

Weird? The supposedly faster card is actually slower than the next one down on all tests, and it's the slowest of the three once AA is included.

Assuming the cards do actually meet the reported specs, presumably this means that the bigger memory is wasting time, one way or another. Tempting to suggest it might be a driver issue.

Wondered if there were similar effects on other sorts of graphics cards, where increasing the memory results in a slower card?

Reply 1 of 4, by Old Thrashbarg

User metadata
Rank Oldbie
Rank
Oldbie

Wondered if there were similar effects on other sorts of graphics cards, where increasing the memory results in a slower card?

Yes. I remember the Radeon 8500LE and 9000 (pretty much the same card) 128MB versions being slightly slower than the 64MB versions. Oddly, I think the 'proper' Radeon 8500 was slightly faster with 128MB than 64MB.

The Matrox G200 was also slower when you added the extra SODIMM, but AFAIK that had to do with signaling issues specific to the SODIMM rather than the amount of memory.

Reply 2 of 4, by sliderider

User metadata
Rank l33t++
Rank
l33t++
Old Thrashbarg wrote:

Wondered if there were similar effects on other sorts of graphics cards, where increasing the memory results in a slower card?

Yes. I remember the Radeon 8500LE and 9000 (pretty much the same card) 128MB versions being slightly slower than the 64MB versions. Oddly, I think the 'proper' Radeon 8500 was slightly faster with 128MB than 64MB.

The Matrox G200 was also slower when you added the extra SODIMM, but AFAIK that had to do with signaling issues specific to the SODIMM rather than the amount of memory.

8500LE and 9000 are actually very different internally. The Radeon 9000/9000 Pro/9200/9250 have 4:1:4:4 pixel:vertex:texture:render units while the 8500/8500LE/9100 have 4:2:8:4 pixel:vertex:texture:render units. In AGP an 8500LE would be the preferred card over a 9000. In PCI, it would be the 9100. I went to a lot of trouble a while back to find a Visiontek Xtasy 9100 PCI not only because the 9100 is faster than the 9200/9250 but because the Visiontek card was the only 9100 in PCI that had the fastest spec RAM chips. The few others that were released used slower RAM to keep the price down.

Reply 3 of 4, by Old Thrashbarg

User metadata
Rank Oldbie
Rank
Oldbie

Ah, yeah, I always forget about the 9100, getting it mixed up with the 9000. Now it's starting to come back... how the 9000 was supposed to be the direct replacement for the 8500LE, but didn't quite measure up.

However, I'm still pretty sure the 8500LE and 9000 both saw a performance drop with >64MB, so I wonder if that means the same is true of the 9200/9250. I don't think I've ever seen a comparison done with the different memory configurations on those cards (other than 64-bit vs. 128-bit), but they were available all the way up to 256MB models.