Extremely poor bandwidth performance, any hints ?

Accelerated Computing CUDA CUDA Programming and Performance
1

Dear Cuda Users,

We recently bought a TESLA S2050 and plugged it to a Supermicro server.

We’re getting extremely poor bandwidth performance and I don’t know why. Here are the results :

bandwidthTest Starting...

Running on...

Device 0: Tesla S2050

 Quick Mode

Host to Device Bandwidth, 1 Device(s), Paged memory

   Transfer Size (Bytes)	Bandwidth(MB/s)

   33554432			2314.1

Device to Host Bandwidth, 1 Device(s), Paged memory

   Transfer Size (Bytes)	Bandwidth(MB/s)

   33554432			2114.8

Device to Device Bandwidth, 1 Device(s)

   Transfer Size (Bytes)	Bandwidth(MB/s)

   33554432			85397.5

The installed NVIDIA packages are :

- CUDA Toolkit 3.2.16

- CUDA SDK 3.2.16

- NVIDIA Driver : NVIDIA-Linux-x86_64-260.19.44

The system where the Tesla S2050 is plugged is :

- H8QG6F Supermicro Motherboard 

- AMD Opteron processor

- Ubuntu 10.04.2 amd64 server, kernel 2.6.35-020635rc1-generic

Motherboard specs

Do you know what might be causing the problem ?

Also, what should be the “right” bandwidth for a system like that ? Just in case I have to complain to the vendor.

Please any help will be appreciated.

Best Regards,

David McGiven

2

The answers to similar questions that I supplied in the following threads may be of help: