Since the GeForce, Quadro, and Tesla products are all based on the same graphics chip from Nvidia, one of the things that people have done is to buy the lower cost product (GeForce) and use it like a higher cost product (Quadro and Tesla). Sure you may have to give up on a few features, but not everyone needs them. Sure you may have to give up some of the extra memory, but it may be worth the price. This is a trade-off that purchasers have to make. Hexus has found yet another trade-off, and one that I had not heard of yet. The GeForce GTX 480 will have 1/8th of the speed as it does in single precision. That may not be a deal breaker to some people. After all, the GeForce GTX 285 had a similar problem. But for those expecting the double precision speed to have increased to ½ that of single precision, it does make a difference.
As much of a general-purpose computer as a GPU, the parallel architecture is also designed for the high-performance computing segment in mind. The enhanced cache structure, detailed above, helps with general computations, and GF100’s adherence to the IEEE 754-2008 floating-point standard means that it can run high-accuracy tests (double-precision support) at an increased rate when compared to anything NVIDIA has designed before.
Delve a little deeper, handily not mentioned in any briefing, and NVIDIA is limiting the double-precision speed of the desktop GF100 part to one-eighth of single-precision throughput, rather than the one-fifth speed of the Radeon HD 5000-series. We’ll have to wait for the Tesla parts before that’s restored to the one-half speed the GF100 is capable of.
via HEXUS.net – Review :: NVIDIA’s GeForce GTX 480 finally unleashed. Reviewed and rated. : Page – 3/17.
@Jed G. It is my understanding that yes, it will. Although you’ll only be using a quarter of the card anyway, double precision arithmetic will run 8x slower than single-precision arithmetic, no matter how much of the card you use.
Now, if you can offset that penalty with other operations (remember, the Fermi architecture support multiple simultaneous kernels), then you may not notice.
Are there details as to how the double precision performance limited? If, my application, for example, were limited primarily by memory bandwidth, such that I was only able to use 25% of the compute capabilities anyway, would the double precision limitation still affect my application performance?
Cheers