nvidia chipset versus onboard memory

Accelerated Computing CUDA CUDA Programming and Performance
#1

I am brand new to cuda still. I have a geforce 9500 gt card to learn on for the time being. The specs on that chip say 512 mb of memory, but the card (evga) says 1 gb. I assume the discrepancy is that evga has simply added 512 mb of memory directly onto the card, is this correct? If so then the nvidia chipset memory should be dramatically faster than the on-the-card memory I would assume, in which case is there anyway to specify that you want to allocate memory only on the nvidia chipset, and ignore the space memory on the card? Or am I all wrong together (very possible, as I said, I am very new to this). Oh, querying the card for how much memory it has, it reports a whole 1 gb.

-John

#2

The GPU has no memory fabbed on the die apart from a few kb of instruction and data cache - just SDRAM chips on the board. The only difference between any of the 512Mb and 1Gb 9500GT models is the size and type of the SDRAM chips used.

#3

Very interesting. Thanks. I wrongly assumed the memory was internal to the gpu.

-John Robertson