Hi, some silly questions about mem limits:
as far as I benchmarked memory transfers it was not possible to get more than 1.5 GB/s to the device and 1.1 GB/s back. I thought 16x PCIe (PEG) should get 4 GB/s in each direction?
On my 8800GTS (which also runs the GUI) I can allocate 582 MB linear space for CUDA. Is GDI eating 38 MB VRAM? Can I use all VRAM for CUDA if the GUI runs on another board (like my ancient quadro) so the G80 is for GPGPU only…?
Is there some QuadroFX board with 1GB (or even more) planned?
I read something of virtual memory for graphics with Vista/D3D10. Is something like that possible with CUDA? (something like automatic PCIe DMA paging) ?