I'm thinking of upgrading my graphics cards, but first I need a piece of information I haven't been able to find online. How well do cards meant for PCIE3 slots work if placed in PCIE2 slots on the motherboard? Do they even fit in such slots, and so on? Going to a computer store and looking at the boxes such cards come in seems likely to provide that information, but is seldom practical for me.

Don't worry, PCIe is backwards-compatible. The only concern is bandwidth, which is fine if you're using a 16x PCIe 2 slot (less than a few % difference, depending on the card). At 8x or 4x PCIe 2 the lack of bandwidth starts to hurt GPU-Grid.

MrS

---

Scanning for our furry friends since Jan 2002

I've got a GTX 750 running on a PCIe 2.0 motherboard. The current cuda 6.0 GPUgrid work running the GPU at 98% in 64-bit Linux puts bandwidth usage at ~40%.

From here, it looks like 2.0's bandwidth would be an issue for top performing cards. If bandwidth demands and benchmarks are proportional, a GTX 780 would be at about 100%.

I'd assume my numbers are typical, but I've read others talking about bandwidth demands being low.

Robert,

Yes, no problem putting a gen3 card in a gen2 slot.

Matt

I just got an older mb, older cpu, same os and drivers(331) going on a GTS 450. If it was proportional to my GTX 750's by performance on GPUgrid, it'd have the bandwidth somewhere around 10%, but it's more like 1%.

The GTX 750 seems to be completing units in good time looking at other people's results, but without cutting at least 2 of the 8 cores from boinc or tweaking core affinities, gpu utilization is generally well below 90%. GPU and bandwidth fluctuate together. Without any considerations, it's around 70% gpu, 25% pcie. Bandwidth seldom exceeds 2% on other projects.

Has anyone else seen high bandwidth usage or have an idea where this might be coming from?