4GB system memory required for each installed TESLA C0160? Is this really true?

Accelerated Computing CUDA CUDA Programming and Performance
1

I have found in some page of the official nvidia web site “Guide to Building Your Own Tesla Personal Supercomputer System:”
nvidia.com/object/tesla_build_your_own.html
the following statement:
“Since each Tesla C1060 has 4GB of GPU memory, we recommend 4GB of system memory per Tesla C1060. So, for a 3x Tesla C1060 system, include at least 12 GB of system memory and for 4x Tesla C1060 system, configure with 16GB of system memory.”
Is this a requirement or just a recommendation? Does this have something to do with NVIDIA graphic card drivers internals?
Is this true also for consumer cards? E.g. for a 3xgtx295 configuration at least 3x1792=5376MB of installed system memory is required?

2

The reasoning probably is:
If you want to feed 4GB of input data to GPU, you must atleast have that much RAM…

3

I think it is just a recommendation, but it may be a requirement for some mainboards. I know in the first box that tmurray built with 4x Tesla cards, the BIOS was just simply unable to cope with more graphics memory in the system than main memory. So you should at least plan your system to be capable of such if it turns out to be a requirement…

4

Thank you Sarnath and MrAnderson for your answers

5

I have a box with the C1060, and 2 gigs of ram without any problem