Mothballing the 8800GTX Worried about the supposed GTX 300

Accelerated Computing CUDA CUDA Programming and Performance
1

As more and more of my process moves out of Java and into OpenCL, hopefully resulting in a Cloud implementation, I see the use of atomics as key. My memory requirements do not actually require a C1060 for development. I can use anything with 100mb. The 8800 needs to go though.

I am using Java concurrency to implement an OpenCL context/GPU pool. This can schedule 1000’s of iterations of kernel sets with large worksizes on multiple GPUs, so a multiple GPU test environment is what I am looking for. The GTX 295 looks like it, but rumor of a 40 nm version called the GTX 300 has me worried that my decision could look dumb a month from now. Is this just wishful thinking by some gamers?

I should also mention that I already have a multi-GPU MacbookPro, so I can development there for a while. Everything I have heard here is about the future with “ferni” . Spreading this rumor to here is not my intension. Can anyone shoot this thing down?

2

Fermi’s not really a rumor:

[url=“Inside Fermi: Nvidia's HPC Push”]Real World Tech
[url=“GF100 graphics architecture unveiled - The Tech Report”]http://techreport.com/articles.x/18332[/url]
[url=“Inside Fermi's graphics architecture - The Tech Report”]http://techreport.com/articles.x/17815[/url]

3

Thanks, This is plenty of real info. I’ll look around for dates & come to some decision.

FYI, I always thought Ferni was for real, but way in the future. I can see how you could interpret that I was saying it was a rumor. I meant the GTX 300 was the rumor. Ferni was only in quotes, because it is an internal code name / architecture & not an actual product. GTX 300 turns out to be just guesses by the peanut gallery of what coming out might be called. Seems similar to iSlate or iTablet.

4

According to this, the first Fermi based Tesla card should be available in early Q2 this year. If you can’t wait that long, you may consider a lower end card (such as a GT 240, which is a CUDA 1.2 device) for now.

5

Thanks for the link. I might have waited that long, using the Mac till then, if it was a PC/graphics version coming out in early Q2. Since that amount of memory is overkill, I thought of only using the Tesla line for production. That link also does not show Windows 7 on the downloads tab (that’s got to be a typo).

My gut is telling just do it on the Mac regardless, and re-evaluate when this becomes critical path.

Thanks, again to both of you. I do not have forever to do blind keyword searches. You saved me a lot of time.