6800NU's 3DMark2K3 score is leaked by Galaxy! Updated!

All three of you

But you had them - according to FM both passed their internal testing so as far as they are concerned the performance increase isn't through any "funny business". According to the convervsation I had with worm the other night the beta is no longer accepted because there is a WHQL version about now.

WHQL certification can take up to about a month, and in the early sets of drivers you can find a fair amount of performance, so its not really surprising that they supplied a newer DX driver for the boards they are reviewing, whilst the WQHL release can at least enable vendors to sell vertified boards (meaning no surcharge for MS).

 

So would you recommend benchmark using the official 4.5 or the 4.5 beta?

Reminder: Using official cat4.5 will give you about 1000 points less.
________
O309

 

I don't need reminding thanks - I do remember, seeing as I replied.

However, I'm not going to make any recommendations as to whatever driver should be used - thats entirely up to the policy of the reviewer and the publication they work for. For this site, we'll use the official WHQL drivers if they are available for the board we are trying to review, however we, at the moment, aren't doing any competetive comparisons. You might like to question how fair it would be for a site to compare WHQL drivers for one manufacturer against brand new, non-certified drivers for another if such a situation arose, given the difference in development that goes into them. In this situation my preference would be to seek the latest official WHQL's for both manufacturers, but if certified drivers are not available for one then the latest releases for all would probably be the fairest comparison (seeing as you would hope that the performances you see here would be representative of performances in later WHQL drivers).

 

I suppose Ati didn't provide the reviewers with the WHQL drivers because they would be beaten by Nv's 6800? ~+20% in GT4 with the beta drivers is not bad.
*edit* didn't see the last message.

 

Personally I'd assume they supplied non-WHQL drivers for the same reasons as NVIDIA.

 

I'm a little mystified as to how a 6800 can outperform a X800P in 3DM03, considering ATi has the lead in theoretical specs (fillrate [and thus I'd imagine shader power], bandwidth) and should have more mature drivers.

Heck, 9,200 in 3DM03 with a 12P@325MHz 6800 seems awfully high considering a 16P@400MHz 6800U scores 12,000 on an A64 3400+. TR's review shows the (12.5%/9% core/mem) slower 6800GT scores ~8% less than the 6800U, so I'd acccept the 6800 losing slightly less performance than its theoretical speed differences would indicate. But the 6800 has ~40% (core & mem) lower theoretical numbers than the 6800U, and yet its 3DM03 score is only 23% slower. I might explain the small difference b/w the 6800GT and 6800U as being memory-limited, but that doesn't work with the 6800. Granted, I'm comparing benchmarks across different CPUs, but that's a pretty huge disparity. Surely a P4E isn't that much faster than an A64? And doubly surely PCPop can't have a (16P@475MHz) X800P score only 8,700 on the same system that a 6800 scores 9,200?

To put it bluntly, is nV cheating in 3DM03, or is the NV40 really that much better than the R420 at 3DM03's tests?

 

3DMark03 prefers (scores better on) Intel CPUs over AMD cpus. 3DMark2001 prefers AMD cpus over Intel CPUs. At least this is the case for the P4s vs A64s/FX51s.

 

I had noted that previously, but surely this preference can't manifest itself in a win for the cheaper and ostensibly slower 6800? I haven't really focused on P4 vs. A64 benches, so I don't know how much two CPUs of relatively equal speed can affect the normally GPU-limited 3DM03 score, but I doubt the CPU alone can explain such a discrepancy. I'm thinking a faster CPU can result in significantly different GT1 and CPU scores, but I don't think it'd affect GT2-4 scores that much (AFAIR from Ace's article). Plus, the 6800 and X800P should be fast enough to make themselves the bottleneck in such a GPU-limited benchmark.

I'm try to find other 3DM03 scores, though, to see how wrong my assumptions and hypotheses are.

 

Tweaking a system can make a pretty big difference in 3dmark scores.

Macci is listed #1 over 13k with a barely overclocked XT (stock cooling) and not really much more on the processor either, yet he is way ahead of the score DB had.

 

At 325/350, P4 2.4B, the 6800 scores 8242.
At 350/400, P4 2.4B, the 6800 scores 8647
At 350/400, P4 3.4E, the 6800 scores 9232
________
Tube 8

 

I didn't think the 6800 was run OCed on the 3.4E, but even so, its score still looks too high compared to the X800P.

Who knows, maybe the NV40 is that much more efficient than the R420, but I don't see this with the handful of other DX9 games. I guess I'll have to wait for further PCPop numbers.

 

Isn't the WHQL 4.5 the reason for so little differencies?

 

That is interesting... I wonder why it is higher?

 

4.5 beta are later drivers than 4.5WHQL.

I suppose it's like that because Ati didn't release the 4.5 WHQL before hand, and didn't wanted to name them 4.6 beta...

 

I thought they were released before the WHQL 4.5, lol. Remembered wrong. Edit: By earlier I meant an earlier build.

 

So the 4.5 beta may include a more advanced shader compiler/other valid tweaks? If it is indeed a later build than the WHQL 4.5, why didn't ATI communicate this to reviewers to ensure that they would benchmark beta against beta, instead of beta against WHQL? :?

 

for the 4.5 betas...is that the work of the DEC engineers (wasnt it ati that acquired some of them recently) ?

 

Or you could wait a bit longer like me and buy an x800xt and be very happy. Well, at least I know I will.

 

It is too high. The drivers used in that test were the 61.11 set, which is broken and inflates scores by not rendering certain objects. The X800 Pro typically gets around 9200, I would expect the 6800nu to get around 7500-8000 under normal circumstances.