anands nv38 not really an nv38?

[mczak]

Who in their right mind would publish some results without an actual product.

Actually, if the 5950 Ultra is indeed just an overclocked 5900 Ultra why would it be so wrong? I mean it even uses the exact same PCB as the current boards etc. So if you know for a fact that an overclocked 5900 Ultra would produce identical results then why not publish it? (Of course while mentioning that the benchmarked board is indeed "just" a overclocked 5900 Ultra.)

Yes if you mention it it's not that bad. However, there are some issues with that. First, with the NV38 you just can't know if it doesn't have any changes which could change performance. And even if it really is the same, some things could still be wrong. Remember that the GFFX5900 overclocked to the GFFX5900Ultra level didn't perform the same - I haven't followed that, but it could have been because of different memory timings. This won't change things dramatically, but since you're talking about an only 5% improvement for the FX5950U over the FX5900U a change of say 2% (up or down) could easily make the numbers worthless. Same goes for the 8500LE vs. 8500 - depending on the ram/bios there was some difference (never exceeded 4-5% I believe, but still) when they were clocked the same.

 

[Ante P]

Who in their right mind would publish some results without an actual product.

Actually, if the 5950 Ultra is indeed just an overclocked 5900 Ultra why would it be so wrong? I mean it even uses the exact same PCB as the current boards etc. So if you know for a fact that an overclocked 5900 Ultra would produce identical results then why not publish it? (Of course while mentioning that the benchmarked board is indeed "just" a overclocked 5900 Ultra.)

Yes if you mention it it's not that bad. However, there are some issues with that. First, with the NV38 you just can't know if it doesn't have any changes which could change performance. And even if it really is the same, some things could still be wrong. Remember that the GFFX5900 overclocked to the GFFX5900Ultra level didn't perform the same - I haven't followed that, but it could have been because of different memory timings. This won't change things dramatically, but since you're talking about an only 5% improvement for the FX5950U over the FX5900U a change of say 2% (up or down) could easily make the numbers worthless. Same goes for the 8500LE vs. 8500 - depending on the ram/bios there was some difference (never exceeded 4-5% I believe, but still) when they were clocked the same.

As for knowing if it's the same: you could just ask nVidia or their board partners.

5900 vs 5900 Ultra that's due to the extra amount of memory on the 5900 Ultra. The 5900 non-Ultras with 256 MBs (Gainward makes one) performs just the same as the 5900 ultra at the same clockspeed).

I'm just saying that it could be a viable method of doing a fairly accurate forecast on upcoming hardware. Of course such an article should be utterly clear on the point that it is indeed not the "real" thing.

 

[Arun]

That reminds me...
I heard, but I'm not sure at all how reliable it is, really no idea, nada, that the NV38 was already used to make some 5900Us. Basically, that'd mean if they can't achieve 475Mhz safely, they put it at as a 400 or 450Mhz 5900.

So for his peace of mind, he could have tried to find one of these rare specimens, and then overclock it to 475Mhz.

Of course, that's pretty far fetched - and I'm really not sure of reliable the original info is. But eh!


Uttar

 

[Joe DeFuria]

Well, we already know that Lars at Tom's hardware was given the offer to go ahead and benchmark the NV38 with Det 52 drivers, so I think it's safe to assume that Anand got the same offer.