Page 1 of 1

PG: GPU bandwidth

Posted: Mon Apr 13, 2009 12:29 am
by alpha754293
How much of the 4-8 GB/s bandwidth (depend on PCIe version) does the GPU client actually use? 100%? 50%? ??

Re: PG: GPU bandwidth

Posted: Mon Apr 13, 2009 12:52 am
by P5-133XL
People can fold on 2x PCI-e without much PPD penalty. So the answer is very little.

Re: PG: GPU bandwidth

Posted: Mon Apr 13, 2009 12:53 am
by alpha754293
P5-133XL wrote:People can fold on 2x PCI-e without much PPD penalty. So the answer is very little.
Excellent. Good to know. Thanks.

Re: PG: GPU bandwidth

Posted: Mon Apr 13, 2009 1:38 am
by Slash_2CPU
2x version 1.1 PCIe incurs a small decrease.

2x PCIe 2.0 is twice as fast, and has no loss at all. Same for 4x PCIe 1.1.

Re: PG: GPU bandwidth

Posted: Mon Apr 13, 2009 1:44 am
by alpha754293
Slash_2CPU wrote:2x version 1.1 PCIe incurs a small decrease.

2x PCIe 2.0 is twice as fast, and has no loss at all. Same for 4x PCIe 1.1.
That doesn't make sense. The PCIe 1.0 Spec calls for 250 MB/s (2 Gb/s) per lane, while PCIe 2.0 calls for 500 MB/s (4 Gb/s) per lane. hmmm...

Re: PG: GPU bandwidth

Posted: Mon Apr 13, 2009 5:45 am
by 7im
Total bandwidth is obviously not the only gating factor in the older spec.

Re: PG: GPU bandwidth

Posted: Mon Apr 13, 2009 9:19 am
by alpha754293
7im wrote:Total bandwidth is obviously not the only gating factor in the older spec.
Depends on which metric you're talking about.

Re: PG: GPU bandwidth

Posted: Thu Apr 16, 2009 5:27 am
by 7im
All of them except bandwidth, what else? ;)

Re: PG: GPU bandwidth

Posted: Thu Apr 16, 2009 5:41 am
by alpha754293
7im wrote:All of them except bandwidth, what else? ;)
But I think either all or almost all of the other gating factors affects the bandwidth, so...