HD2900XTX Doesn't Exist!

[ShadowOfMyself]

Why is everyone ignoring the fact that it has 1 GB of ram instead of 512 mb?

Take notice of the performance drop from the 640 mb GTS to the 320 GTS...

 

[TecHNooB]

Originally posted by: ShadowOfMyself
Why is everyone ignoring the fact that it has 1 GB of ram instead of 512 mb?

Take notice of the performance drop from the 640 mb GTS to the 320 GTS...

It was mentioned in a post on DT. Someone said that the ram wasn't bottlenecking the card so there were no gains going from 512MB to 1GB.

 

[Gstanfor]

Originally posted by: munky

Originally posted by: Gstanfor

Originally posted by: Sable

Originally posted by: Gstanfor
Well, I think that all that the most zealous of the fanatics will now agree that R600 is ATi's nv30.

not even close. NV30 got raped by ATI in DX9, at least R600 has some performance about it.


It IS a complete disaster for ATI though. (unless they pull the refresh out of there arse at the same time NV do. And I don;t mean that Ultra crap from NV)

Whatever makes you feel better Sable.

To me, the most amazing part about all is ATi appears to have swallowed David Kirk's "not going unified" comments hook, line and sinker, slacked off on R600 because they didn't see a real threat and have now paid a big price for it. Its all the more amazing given that if you read Kirks comments carefully its obvious he was talking about the current gen (at the time) & unified shaders, not about next gen.

I think this also demonstrates quite nicely the the "common wisdom" that both sides have a very good idea of what the other is doing is nothing more than an internet myth.

Last point, I don't think even a refresh will save r6xx (see nv3x for proof) unless the "refresh" goes for full scalar ALU's intsead of vector ALU's (thats a fundamental architectural change bTW - not something you are going to see in a refresh), don't forget nvidia will have a g80 refresh also (and I'm not talking about the Ultra either)

Scalar archtecture only helps in worst case scenarios. Most shader ops are performed on vector data, and I dont believe going scalar will solve the performance issues of the r600, especially when it already "supposedly" has more theoretical shader power than the 8800gtx.

Vector architectures mean the shader compiler has to try and pair instrctions up to get full usage from the ALU's, not always an easy thing to accomplish and adds much compiler complexity. Scalar avoids the issue allowing the GPU to be more fully utilized at all times. It also simplifies "non traditional" tasks such as physics or GPGPU computing.