Compare Announced ATI Crossfire/Brook+ and CUDA

Accelerated Computing CUDA CUDA Programming and Performance
1

As several of you must have seen (e.g. at
[url=“http://ati.amd.com/technology/streamcomputing/stream-computing.pdf”]http://ati.amd.com/technology/streamcomput...m-computing.pdf[/url]
or http://ati.amd.com/products/streamprocessor/specs.html) ATI is coming to the stream computing party as well with the crossfire boards

Is anybody using these in trial versions? Is there a beta of Brook+ out?

I see a peak speed similar to G80 (500 vs. 330)

Also, there are 320 stream processors vs 128

Both architectures now only appear to have generic processors … no vertex and fragment

the local memory seems to be arranged a bit differently on ATI

Please post any info. you have or thoughts here …

2

I don’t know, but I’m certainly interested in this as well. Especially feature-wise. We all know that ATI and NVidia try to beat each other in speed every generation. Though it’s nice to know who is on top.

Can this CTM (with Brook+) do the same as CUDA? Why does ATI have more stream processors but comparable performance? Etc.

There must be someone that worked with both that can answer these, right :)

3

Processor frequency is much lower, something around 800 MHz. And, clearly, some other microarchitectural issues. Not much information about ATI cards/API.

4

It will be interesting how Brook+ is different than Brook (or at least the Stanford implementation of Brook for GPUs). I worked with Brook briefly before switching to CUDA, and that language also had a C-like syntax. It was far more constrained than CUDA, given that it was targeted at the previous generation of graphics cards.

5

What about writing directly to the framebuffer? Can the Crossfire boards do it? And what about the memory bandwidths?