CUDA 2.0b & C870

Does anybody have encountered any problems in using the 2.0beta together with a C870?
I just installed the complete package (toolkit, SDK & driver) on Linux and get a strange behavior running the bandwidthTest example.
bandwidthTest runs without any problems on a second card in the same system (a Quadro FX3700). But if I select the C870 (using the newly introduced --device switch in CUT_DEVICE_INIT, thanks for that btw.!) bandwidthTest produces ridiciulously small bandwidths for pageable memory (~100MB/s) and segfaults for pinned memory. The other examples are also kind of weird: they work fine for the FX3700 but the system load goes up if I use the Tesla card.

I’m running RHEL4.6 on an HPxw8600. Driver/SDK/TK is 2.0b(174.55).

Anybody made similar experiences on a C870 (or multiGPU) setup?


Make sure that you have the latest SBIOS for the xw8600.

I’ve encountered a very similar problem with the C870. I posted about it here: