So what about linux on Cuda then?

Seems to me that with the arrival of the MIMD fermi architecture it should be possible to run linux on the GPU and leave the CPU to handle disc, network and other I/O.

Prolly wouldn’t run too fast yet, but …

Anyone else been thinking about this?

No because it wouldn’t be possible. There is no evidence that Fermi (nor any of the GPUs that preceded it) have a “proper” hardware MMU. Just about all memory management seems to be done in the host driver. No MMU=no Linux, irrespective of all of the other reasons why it wouldn’t work.

Ha! Of course, so the limit would be something akin to CP/M really, shame really.