Quote Originally Posted by informal View Post
I can tell you one thing folks. This level we are seeing from vrzone,coolaler was not the projected one 9 months ago. Something changed in the meantime and we now get "this". What changed is anyone's guess. The 2 benchmarks from vrzone ,3dmar06 cpu test and c11.5 are clear signs of this "change". 9 months ago,Thuban couldn't touch zambezi in both. Today Zambezi can't touch Thuban. Situation looks grim.
I found a few posts by JF which you might find interesting, I think he was commenting the sandra benchmarks:

I have no idea if they are even real, but I have good reason to believe that they are not representative. That is all I can say.
First, how does anyone know if the results are real? How do we know if they even had an actual processor? And if they did have a processor, how do we know what level of silicon they had?

Secondly, there are some reasons why even if they did have a chip, I can feel pretty confident that those numbers don't represent actual shipping silicon. Can't get into that right now, but given a little time you'll see why.

The real problem is I can't say the results are fake because nobody on this board can say that. Only the person that posted them can say that. But, just stop for a second and think what if they are real. Do you know if the person who ran them optimized the BIOS settings? Do you know if they specifically had the latest BIOS, the latest drivers, etc.?

There is better probability that there is something wrong than not.
I said that the benchmarks are "not representative", not inaccurate.

We just hit production on server processors. So are those benchmarks on final silicon? Highly unlikely. Are they using a final BIOS? Final drivers? OS optimizations? There are a ton of things that all need to be in place in order to get the best performance.

If I posted a benchmark showing the my product against an intel product but refused to give you the detail on the config, would you trust me? Probably not. But you would trust me if I gave you my score and compared to the best intel score that they ran, right?

There have been so many fake benchmarks, including those that were published before the first silicon hit our test labs.

Up until we hit production, none of the benchmarks could be considered representative. And, the only thing that we have publicly said is production on server (I cannot confirm if zambezi is in production or not - that is not my job.)

Once they are in production, even then, there are lots of variables that still have to be considered. So there are plenty that are not representative out there. Showing both lower and higher performance. Both sides of the coin seem to want to push an agenda. I recommend not trusting anything until launch.

If AMD does anything prior to launch, I would gladly confirm it and most likely would have a blog post as well.

These are from [H] and were posted in august, but there are quite a few hints there.

I think everyone should relax, there are so many fake benchmarks flying around no one can tell which one (if any) is true.

We do however know that IPC increases and clock is increased shouldn't that result in better performance?