AMD: R9xx Speculation

http://h-5.abload.de/img/69704mxh.jpg

3dMark feature test scores 1680x1050

Interestingly, the 6970 loses significantly in Perlin Noise but beats it / edges it at everything else

What can we extrapolate from that?

Also, is that device ID correct?

 

It may not mean anything once the reviews are out but for now I would wait to see what Cat 10.12 offers. We already have an understanding of what a 6970 does with cat 10.11 and it's derivatives. Once Cat 10.12 is used then we will see if Cat 10.11 was actually using VLIW4 architecture correctly or not.

 

Please .. it can't read SPs...

 

So how do you explain it showing 1120 SP 6850s for example?

 

no idea if this is what happens or not, but that number could just be input data, it reads the processor type and outputs what is known about it from the file.

 

Cayman has 1536 SP´s or 6970 has 1536 SP´s?

 

I think it may be able to read the SIMD count on Evergreen but apparently not on Northern Islands, otherwise it would have been able to get the right texturing throughput for Cayman, since the number of TMUs per SIMD has not changed.

 

So I'm beginning to think that the 1920 figure from a while ago is refering to Cayman, but 6970 has got 1536. 1920 ALUs would be equivalent to 480 Nvidia style SPs with VLIW4, and the transistor count is 2.65bn according to the latest GPU-Z build.

480/512 = 0.9375
2.65bn/3bn = 0.8833

The numbers are too similar to be a coincidence. What I wonder is how ATi have managed to pack their chip so densely for Cayman while Nvidia have 15% more transistors but nearly a 33% larger die. That's some good design and engineering from ATi!

 

One thing I can guarantee gpu-z isn't doing is counting transistors. Is that an official source number or someones guess?

 

Probably not fake this time .. , it will all come down to pricing now ..

 

Well it's the updated build which people are assuming is based on the official slides. I can only go by the data we have. If the number of transistors is closer to 2bn then it makes more sense for 1536, but also a lot less interesting...

 

Thsese are AMD marketing slides , they are chosen specifically to represent best outcome possible , I suspect they used 2500x1600 with 8XAA just to run the competition into FB limitation (pretty obvious in Crysis and Metro).

Using 8X AA is probably a good thing to do though , I think AMD just improved 8XAA performance again (enhanced ROPs) .

 

Agreed, but even a 5-10% improvement over GTX480 will be 20-30% above 5870 which is a decent improvement for a 20% increase in die size.

 

You want to say 1.5GB isn't enough for Crysis or Metro? Besides, isn't high resolutions and AA settings something an enthusiast would run?

 

It's 19x12.

 

against the 5870 it's benchmarked at 25x16, against the 480 it's benchmarked at 19x12.
driver rev for the 480 is 260.99, all cayman and cypress benchmarks are done with 10.11.

 

If so , then that is probably good for them .

 

Ill call it fake because:

1 - Would a marketing slide say "on average"?
2 - On HD6950 heading is too close to graphic top.

They dont look very professional to me.

 

This is getting to be really silly... all of the fake rumors/slides won't come to a stop, not even 2-3 days before release. I don't think I've ever seen this much fake rumor/slides for any new upcoming release, ever.. not even close! All of the people who waste their time creating fake slides, well, whatever.. I'll just wait until launch day before making a decision anyways--not as if the fake information is gonna affect me.