Tom (noemail@yahoo.com) on 4/11/12 wrote:
---------------------------
>Oscar Eddington wrote:
>>>> If Nvidia's graphics and compute products diverge
>>>> as explained in this article, doesn't that make
>>>> it difficult to use both a graphics card and a
>>>> tesla card in a workstation for compute?
>
>Gary M. wrote:
>>> It will be nearly impossible to keep both a graphics card and a Tesla card running
>>> at full speed if these cards have different capabilities. It is too difficult to
>>> do load balancing across chips with different capabilities. Nvidia's strategy to
>>> diverge their graphics and compute products seems similar to the transition from
>>> heterogeneous pixel shaders and vertex shaders to unified shaders except it's going in the opposite direction.
>
>David Kanter wrote:
>> It's not really opposite at all. Conceptually, you can think of hardware vertex
>> or pixel shaders as heterogeneous cores within a single GPU. That's not at all what's happening.
>>
>> What you see is different cores for different GPU (e.g. different double precision)
>> and also different memory controllers (e.g. ECC support).
>>
>> In the CPU world, we are used to seeing the same core, but different 'uncore'.
>
>Having a workstation with both a current Tesla card and a Kepler GK104 graphics
>card would be like taking a dual socket motherboard and putting a Xeon with 20MB
>of L3 cache in one socket and a Xeon with 5MB of L3 cache and no ECC in the other
>socket, assuming that is even possible.

>In other words, it makes no sense to have
>two different versions of GPUs in the same system just like it makes no sense to
>have two different versions of CPUs in the same system. I bet no one will use the
>integrated graphics in Sandy Bridge for GPU computing at the same time that they
>use a real graphics card for GPU computing.

How is this relevant to what I said?

>It wouldn't surprise me if the Kepler GK104 chip actually has ECC but Nvidia just
>disables it for their consumer graphics cards.

It does not have ECC for DRAM or for SRAM. Also, from the standpoint of soft errors...the ECC on SRAM is actually more important.

>Once they fix any hardware bugs,
>they will relabel this chip as a Quadro/Tesla chip.

I doubt it. I think you'll see that the GPGPU versions are quite different.

David