GTX 480 / 470 Double Precision Reduced?

[Seducer]

Hi everyone,

It should be my mistake that the same question may be asked before but I am with my own issue and it doesn't matter for me but I hope if we ask enough to someone official from NVIDIA . Is the double precision performance of the consumer fermi cards decreased (by 75%) compared to that of the Tesla Line?

[kiLLer]

The thing about that you are talking can be from different sources. It also depends on your card performance that explanation that you are giving here about your issues or confusion is seems that gaming cards such as GTX480 do not have increased dp performance. One thing that need to be sure that nvidia advertised this type of feature in GTX480?

[Seducer]

Actually, I didn't got any official news about the gaming cards which shows the decreased dp performance - they use the fermi chip, which is similar as the Tesla and the Quadro Cards will use. And the architecture of the fermi chip provide the double precision performance at half acceleration of single precision. So it is never a query of whether the tesla cards have increased DP performance but if the consumer cards have a decreased one. If they are going with the minimum DP Performance than the Tesla cards,it will be because its either disabled via drivers or some types of hardware jumper. So the query is still remains if NVIDIA decided to "criple" the DP performance on the consumer cards in order to have several advantages for the Tesla cards besides a lot memory, higher flexibility (I think here that NVIDIA "hand picks" the chips for their professional cards and check them more thoroughly than the chips for consumer cards] and ecc compatibility.

[Birdie]

The same question is also occurred when I was one of the member of a hardware forums, he asked me the question that is related to the NVIDIA:

He asked me about the same. In the GeForce family, double-precision throughput has been decreased to 25% of the complete design. Was this decision made to discourage the use of these components for professional use (where Quadro and Tesla are getting detected in the operation ?) Thinking about the fused compatibility of single- and double-precision manipulation in the CUDA cores, how was this configured even applied?

[David23]

I have the solution for the above question that is asked to you before, full-speed double precision performance is a property that we reserve for our professional customers. Consumer applications have very less use for double precision, so it does not actually affect GeForce users. Running with the differentiated properties and pricing is actually fairer for entire. Given the option of enabling all professional properties on GeForce and having gamers pay for them, or disabling them on GeForce and providing a much compelling price, we assume that the latter is the good option.

[Archibald]

It looking much bad. At least currently there is a remarkable feature to drive particular Tesla sales aside from memory size (and ECC). I didn't got any compelling causes to have a C1060 into a developer workstation unless you required 4 GB of space.

My code is going on to ignore double precision (probably because development began on compute 1.0 devices), and it assumes like it will be effective to continued that trend when possible, if only to achieved GeForce cards.