[Pharmadryg]

Could Nvidia go into CPU production instead? They aren't getting along with Intel right now.

JM

[Si8jy8HN]

They would have to go into a non-x86 CPU design (not production, they don't make anything...they're fabless). ARM is a big market right now. But I'm not sure they'd enter the market well, it's already hyper-competitive.

Intel won't license x86 to them.

And I'm not going to say anything about the stock. I don't think it'll perform too well, but I'd have sold a year ago. I'm not sure if it's worth selling now.

[indocrew]

GPU cluster computing is awesome.

Our currently code, with just a few parts set to use GPUs would prefer to have 10 GPUs per CPU.

It is possible that we could design it to use GPUs even more, only one person has tried and with only one module of code.

I am not sure I would build a cluster without GPUs right now.

JM

[BashBeissedat]

The basic code to recognize gestures from a stream of points is something on the order of 1000x faster with OpenCL on a Radeon 5770 (800 stream processors). It's absurdly responsive.

[Frodogzzz]

I'm very interested to play with AVX in the new Sandy Bridge processors also. It doubles the SIMD throughput on the CPU, upgrading the SSE engines from 128-bit to 256-bit.

[AntonayPina]

http://en.wikipedia.org/wiki/Advanced_Vector_Extensions

New features

* The size of the SIMD vector registers is increased from 128-bits XMM registers to 256-bits registers called YMM0 - YMM15. Existing 128-bit instructions use the lower half of the YMM registers. Further extensions to 512 or 1024 bits are expected in the future.
* Non-destructive instructions. The AVX instruction set allows all two-operand XMM instructions to be modified into non-destructive three-operand forms where the destination register is different from both source registers. For example a:=a+b is replaced by c:=a+b so that register a is unchanged after the instruction. AVX does not support non-destructive forms of instructions using general purpose registers (e.g. EAX), but such support may be added in future instruction sets.
* The alignment requirement of SIMD memory operands is relaxed.

Applications

Suitable for floating point-intensive calculations in multimedia, scientific and financial applications. Increases parallelism and throughput in floating point SIMD calculations. Reduces register load due to the non-destructive instructions.

[Psymoussy]

Still wouldn't want to use integrated GPUs to play games today.

[Laqswrnm]

Not very.

[blankostaroe]

I'm very interested to play with AVX in the new Sandy Bridge processors also. It doubles the SIMD throughput on the CPU, upgrading the SSE engines from 128-bit to 256-bit.

Are AMD's bulldozer specs the same?

[newspetty]

Are AMD's bulldozer specs the same?

No. They will support AVX, but at half speed. It's still 128-bit.

[23tommy]

The only way they could take over completely is if they become good enough for games.

It also means good things for the PC gaming industry if that happens. Right now the percentage of PCs capable of playing modern games is very small...