GeForce6800 w/ 8 pipes? (no,12)- Full review complete!

Hmmm... after seing this numbers i figure that the 6800 has 8 pipes. :?

 

Perhaps there will be a 6800 XT with only 8 pipes.

 

Something ain't right. Either it's a bum card, which I kind of doubt they'd send out to reviewers unless yields are so bad they're forced to; or there is/was/will be an 8 pipe 6800 and someone screwed up and sent it out.

 

The fact this card seems to be working as an 8x1 makes me think Nvidia had plans that direction from the start.

However, if this is indeed 12-pipe card that's running 8-pipes because Ufon got a bad one, then it's not saying much good about the yields. The 12-pipe R420's seem to be cropping up in decent numbers, and at least the current versions seem to have a high success rate of activating all 16 pipes. Now we're seeing a much smaller NV40 sample and it's running at 8?

One's a lousy number to base statistics on, but it doesn't sound that good.

 

Remember the this thread?

http://www.hkepc.com/hwdb/6800nugt-4.htm <-- the page with benchmark results (remember that the 3dmark03 pixel shader & vertex shader results are swapped in that table)

Well, it looks like tests that we are able to compare seem to match up.

Ufon's 3DMark03 scores taken from this picture he posted

Both used the 61.12 drivers

3DMark03
Ufon - 7565
hkepc - 7912

3DMark03 Pixel Shader
Ufon - 83.3
hkepc - 83.2

3DMark03 Vertex Shader
Ufon - 22.7
hkepc - 22.8

3DMark2001 SE
Ufon - 17556
hkepc - 18124

Aquamark 3
Ufon - 46778
hkepc - 46903

So that's 2 cards, one from Leadtek and the other from Inno3D which would seem to imply that it only has 8 pipelines.

 

I got as far as "¥þ²y¿W®a´ú¸ÕGeforce 6800®a±Ú" in that link before I knew I was out of my element, thanks for explaining that.

So there's "accidently" two 8 pipe 6800s out there, very interesting! Please post up if there are any developments in that thread and thank you again for the info/translated summary!

 

Here's a translated version of that link


A question I had is if there's any chance that it could be so bandwidth limited that it basically looks like a 8 pipe card, even if it really has 12 pipes?

 

...

 

Um if it was only 8 pipes and lower clocked than the 6800 ultra wouldn't it perform at less than 50% of the Ultra?

A 12 pipe 325mhz part would be right around 60% of the ultra, which is where it looks to be in most of the tests.

/shrug

 

double post

 

Yeah, I wasn't thinking about the clock speed difference, mostly the fill rate results he gave.

 

Ufon, thanks for the very interesting post. The comparison with the FX5950U is most intriguing, though it does seem to stack up with some naive theoretical numbers:

Code:

                           Pixel           Texel
Card   | Clocks  | Pipes | Fillrate      | Fillrate        | Bandwidth
6800b2   325/350    12     3900 (12*325)   3900 (12*1*325)   22.4GB/s
6800b1   335/350     8?    2680 (8*335)?   2680 (8*1*335)?   22.4GB/s
5950U    475/475     4     1900 (4*475)    3800 (4*2*475)    30.4GB/s

The 6800's higher 3DM03 score can probably be explained by more capable PSs. The 5950U's higher 16x12 AA+AF UT2K4 score can be explained by its greater bandwidth (edit: right, and/or its extra 128MB). The 6800's Far Cry scores may be explained by still-immature drivers or a suboptimal path. So the game benchmarks seem to make sense given the clock and pipe differences between the 6800U and 6800, as AlphaWolf said.

The fillrate numbers are puzzling, though, especially considering how ridiculously efficient the 6800U is in 3DM03's fillrate test.

Comparing the GL_REME numbers to the 5950U in B3D's 6800U review, for kicks:

Code:

GL_REME:        Overdraw                         % Diff fr B>F
Card          | Factor   | B>F  | F>B  | Random | F>B    Random
6800U (B3D)     3          1293   2987   1789     131%    38%
6800b2 (Ufon)   3           787   1931   1173     145%    49%
6800b1 (Ufon)   3           545   1432    788     163%    45%
5950U (B3D)     3           701   1378    979      97%    40%
6800U (B3D)     8           494   2037    811     312%    64%
6800b2 (Ufon)   8           302   1408    508     366%    68%
6800b1 (Ufon)   8           208   1130    488     443%   135%
5950U (B3D)     8           268    783    533     192%    99%

3DM03 scores compared with TR / B3D:

Code:

3DM03:    6800b2 | 6800b1 | 5950U (TR: 52.16 & A64 3400+ / B3D: 60.72 & P4 3GHz)
Overall   8978     7565     5595
GT1        251      236      220
GT2         68       56       39
GT3         58       47       33
GT4         48       41       26
Fillrate:
Single    2392     2244     1516
Multi     3083     2670     3166
VS          25       23       14 / 18
PS2.0      108       83       51 / 52
RagTroll    40       36

Edit: Added BIOS v2 (12-pipe) numbers. GL_REME pixel & texel fillrates (3657 & 3735) are in line with theoretical numbers (3900).

 

The 5950 Ultra has 256MB of ram while the 6800 only has 128MB which could be another reason for the performance hit, in addition to bandwidth, assuming I'm guessing right.

 

the gt and the ultra sould have 16 pipelines and any thing below that sould have 12
check out www.gainward.com for pdf file

 

You think you could try 3DMark03 running at 335/800?

If it's not bandwidth limited in the fill rate tests, running it at 335/800 (same core, faster ram) shouldn't change the fill rate results very much, correct?

 

this the specs for Leadtek A400 TDH 6800 Ultra
Vertex Units 6
Shader Units 16
Texture Units 16
SM 3.0 Yes
64-bit Floating Point Frame Buffer Blending Yes
64-bit Floating Point Texture Filtering Yes
On-Chip VideoProcessor Yes
UDA Yes

 

It's probably bandwidth-limited in the fillrate tests, but how do you explain the remarkably similar game benchmarks? Is that a fillrate or CPU limitation?

 

Perhaps the 6800NU need to flash with the correct bios.
I remember someone at gzeasy mentioned poor performance of some 6800NU cards were due to wrong bios.
________
Vaporizer affiliate programs

 

You can enable/disable pipes on a 6800 with a bios flash?!?

 

Ufon need a new 6800 bios , then the 3dmark03 result can reach 87xx~92xx .