Hi, I am running both tests on HD6970, WinXP x64, CCC 11.2, SDK2.3. I have searched the forum and found there are discussions on this issue too.
With OpenCL PCIeBandwidth test, the Host to device: 0.286868 GB/s, Device to host: 1.65559 GB/s, both are very slow!
With PCIeSpeedTest_v0.2, CPU->GPU Bandwidth is 5.304 GB/s, GPU->CPU Bandwidth is 5.913 GB/s, both for 536MB data.
What is the reason for such big different btw the two tests? Becasue of data size? Can any one give a hint? Thank you!
i have no such problem.Radeon 5850, Ubuntu 10.04 Catalyst 10.1, SDK 2.3
CAL - 5.825 GB/sec/6.671 GB/sec
OpenCL - 4.43402 GB/s 5.52109 GB/s
but in rpevious version of SDK i had similiar speed in OpenCL. so in next release it will be better.
Hi nou, thanks for your info. Is this true because I am running Windows? Has anyone verified the performance with catalyst 11.2, SDK 2.3 on linux x64? Thanks!
My results: 11.2, sdk2.3, x86_64, 5870
PCIeBandwidth:
Platform Vendor : Advanced Micro Devices, Inc.
Device 0 : Cypress
Host to device : 4.63619 GB/s
Device to host : 5.61816 GB/s
PCIeSpeedTest:
Peak CPU->GPU Bandwidth = 5.825 GB/sec [data size = 524288 bytes]
Peak GPU->CPU Bandwidth = 6.673 GB/sec [data size = 268435456 bytes]
Hi Harry,
Thanks for the information, I shall test again on ubuntu 10.04 while I get my spare harddrive.