NVidia quotes the peak bandwidth of the GTX 280 at about 140 GiB/s. I’d be very interested to know the “real world” bandwidth as measured by the “bandwidthTest” sample program included in the CUDA SDK. Has anyone with a 200 series card run this test program?
Thanks,
Kip
On a Zotac GTX280 (stock clock 602Mhz, no overclock)
I have :
For -mode=quick -memory=pinned
Using device 0: GeForce GTX 280
Quick Mode
Host to Device Bandwidth for Pinned memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 4748.1
Quick Mode
Device to Host Bandwidth for Pinned memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 5739.7
Quick Mode
Device to Device Bandwidth
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 114139.5
And for: -mode=quick
Using device 0: GeForce GTX 280
Quick Mode
Host to Device Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 2530.0
Quick Mode
Device to Host Bandwidth for Pageable memory
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 2331.9
Quick Mode
Device to Device Bandwidth
.
Transfer Size (Bytes) Bandwidth(MB/s)
33554432 113922.8
So not 140GB/s but real 113GB/s.
thats better than the G80 with its 40GB/sec
I’m looking forward to test my raytracer on the GTX280. Programs that are memory intensive would be very happy with this new GPU.
My raytracer is working very nicely on GT200. The register-space helps a LOT.
?? G80 (8800 GTX) gets ~70 GiB/s
really?
Hmm… I think I need to run my bandwidth test again :blush:
I got 40 GB/sec on an 8800 GTX only when there was a driver bug in CUDA 1.1. For some reason I was required to load X at least once after booting to get full Device-to-Device bandwidth.