Archives Discussions

[apollo_maverick]

With the newest catalyst 10.7b driver GPU->CPU PCIe bandwidth's as low as ~800MB/s, which should be close to CPU->GPU bandwith that's ~4.9GB/s.

tested by PCIeSpeedTest_v0.2.

HD5870 with ASUS P6T7 WS X58, OS is Windows 7 x64, 12GB (6x2GB config) DDR3.

seems appeared only on Windows x64 systems...

2010.8.18

i've just found there is a new benchmark program PCIeBandwidth.exe in %atistreamsdksamplesroot%\samples\opencl\bin\x86_64 dir.

interestingly, the result is:

Host to device : 2.61799 GB/s

Device to host : 1.97843 GB/s

could u guys run the benchmark and send the results for a comparison?

[rotor]

I think this is an well-known problem and AMD promised to fix it a while ago but I have not seen yet 😉

[apollo_maverick]

any guy from amd to clarify this?

[zeland]

Originally posted by: rotor I think this is an well-known problem and AMD promised to fix it a while ago but I have not seen yet 😉

No it is not an well-known problem. PCIeSpeedTest v0.2 works fine now , but OpenCL PCIeBandwidth shows low results.

Update And I realy want to see source code of  PCIeBandwidth. Can anyone give it?

[nou]

source code of PCIeBandwith is in OpenCL SDK samples.

and i think this is becuase in current OpenCL is no DMA transfers or somewhat limited.

[zeland]

10.8

/PCIeBandwidth
Host to device : 1.4857 GB/s
Device to host : 1.55345 GB/s

i72CIeSpeedTest_v0.2 # ./PCIeSpeedTest
Devices found: 2

===> Testing device 0 <===
Device type: Unknown
Max resource 2D width/height: 16384/16384
Total GPU memory size: 2048 MB
Total CPU cached space size: 508 MB
Total CPU uncached space size: 1788 MB
GPU engine clock: 0 MHz
GPU memory clock: 0 MHz
Number of timing loops: 100
[ 16 bytes] CPU->GPU= 320.000 KB/sec, GPU->CPU= 200.000 KB/sec
[ 32 bytes] CPU->GPU= 1.067 MB/sec, GPU->CPU= 800.000 KB/sec
[ 64 bytes] CPU->GPU= 2.133 MB/sec, GPU->CPU= 2.133 MB/sec
[ 128 bytes] CPU->GPU= 4.267 MB/sec, GPU->CPU= 1.829 MB/sec
[ 256 bytes] CPU->GPU= 8.533 MB/sec, GPU->CPU= 6.400 MB/sec
[ 512 bytes] CPU->GPU= 17.067 MB/sec, GPU->CPU= 17.067 MB/sec
[ 1024 bytes] CPU->GPU= 34.133 MB/sec, GPU->CPU= 34.133 MB/sec
[ 2048 bytes] CPU->GPU= 68.267 MB/sec, GPU->CPU= 68.267 MB/sec
[ 4096 bytes] CPU->GPU= 102.400 MB/sec, GPU->CPU= 136.533 MB/sec
[ 8192 bytes] CPU->GPU= 273.067 MB/sec, GPU->CPU= 273.067 MB/sec
[ 16384 bytes] CPU->GPU= 546.133 MB/sec, GPU->CPU= 546.133 MB/sec
[ 32768 bytes] CPU->GPU= 1.092 GB/sec, GPU->CPU= 819.200 MB/sec
[ 65536 bytes] CPU->GPU= 2.185 GB/sec, GPU->CPU= 819.200 MB/sec
[ 131072 bytes] CPU->GPU= 3.277 GB/sec, GPU->CPU= 504.123 MB/sec
[ 262144 bytes] CPU->GPU= 3.277 GB/sec, GPU->CPU= 468.114 MB/sec
[ 524288 bytes] CPU->GPU= 4.033 GB/sec, GPU->CPU= 472.332 MB/sec
[ 1048576 bytes] CPU->GPU= 4.194 GB/sec, GPU->CPU= 461.928 MB/sec
[ 2097152 bytes] CPU->GPU= 4.280 GB/sec, GPU->CPU= 448.109 MB/sec
[ 4194304 bytes] CPU->GPU= 4.280 GB/sec, GPU->CPU= 443.842 MB/sec
[ 8388608 bytes] CPU->GPU= 4.215 GB/sec, GPU->CPU= 440.578 MB/sec
[ 16777216 bytes] CPU->GPU= 4.258 GB/sec, GPU->CPU= 441.041 MB/sec
[ 33554432 bytes] CPU->GPU= 4.067 GB/sec, GPU->CPU= 442.671 MB/sec
[ 67108864 bytes] CPU->GPU= 4.087 GB/sec, GPU->CPU= 446.084 MB/sec
[ 134217728 bytes] CPU->GPU= 4.145 GB/sec, GPU->CPU= 450.849 MB/sec
[ 268435456 bytes] CPU->GPU= 4.148 GB/sec, GPU->CPU= 458.966 MB/sec
[ 536870912 bytes] CPU->GPU= 4.218 GB/sec, GPU->CPU= 472.419 MB/sec

and halt on it

[d_a_a_]

Originally posted by: apollo_maverick  

i've just found there is a new benchmark program PCIeBandwidth.exe in %atistreamsdksamplesroot%\samples\opencl\bin\x86_64 dir.

interestingly, the result is:

Host to device : 2.61799 GB/s

Device to host : 1.97843 GB/s

could u guys run the benchmark and send the results for a comparison?

./PCIeBandwidth
Host to device : 1.42255 GB/s
Device to host : 1.78671 GB/s

HD5750 / Asus P6T7 WS / Xeon 3680 / 12GB DDR3 / Debian GNU/Linux 64-bit / ATI SDK 2.2 / Driver fglrx 10.7

[ryta1203]

Seems the CPU->GPU went down and the GPU->CPU went up?

[HarryH]

Mainboard Gigabyte GA-MA785GMT-UD2H

PCIeBandWidth (OpenCL):

Host to device : 2.72026 GB/s
Device to host : 1.04368 GB/s

PCIeSpeedTest 0.2 (CAL):

Peak CPU->GPU Bandwidth =   5.518 GB/sec [data size = 536870912 bytes]
Peak GPU->CPU Bandwidth =   4.706 GB/sec [data size = 33554432 bytes]

[zeland]

OpenSUSE 11.2 x64,  10.7 driver,SDK 2.2, ASUS P6T7 WS SuperComputer (X58), i7,12Gb ddr3 1066 ram ,ATI 5870.

./PCIeSpeedTest -tdf pcietest1

Peak CPU->GPU Bandwidth = 4.510 GB/sec [data size = 8388608 bytes]
Peak GPU->CPU Bandwidth = 6.554 GB/sec [data size = 262144 bytes]

./PCIeBandwidth
Host to device : 1.48225 GB/s
Device to host : 1.55237 GB/s