Question to the NVIDIA wizards (TESLA CLUSTER)

I am willing to deploy a computational finance platform on a tesla cluster sometime in the future and am willing to know if NVIDIA already has the intention to release an updated version of the Tesla Computing server with double precision arithmetics and 1.1 compute capabilities. If the answer is yes, when are we to expect such product (end of this year ? first quarter of next year ?).

I would appreciate your answer.

PS : thank you for this amazing platform :devil: O:)

All future chips will have 1.1 (or a superset of 1.1) Compute capability. Double precision will be available on professional solutions at some point.

Various news sources reported that double precision would be available late 2007 alongside the 65nm die shrink. Apparently, plans have changed. They might as well, the software stack has some maturing to do.

EDIT: The “double precision on professional solutions” line saddens me a bit. I can already see nvidia ripping CAD users off with the “high-precision Quadro” marketing. As it is, poor fools don’t realize that even 16bit floats are complete overkill that’s only useful for HDR games. Yet the “128-bit preciision” stickers on the Quadro boxes get the cards sold. Now nvidia is betting that it’ll find suckers for double-precision too. How can such intelligent, technical people get so duped? Maybe they won’t be, and that’s why the plans got scrapped.

While single precision is enough for graphics in 99.5%, double precision is crucial for significant amount of existing general-purpose applications and algorithms, so the point is not very clear.

i mean that by “professional”, nvidia means the regular workstation 3d market. the card apparently won’t be marketed exclusively or even mainly as hpc or stream processing. after all, it would indeed sell much more in the large pro market (as opposed to enterprise or hpc) than in the infant general purpose. but for snake oil-type reasons.

besides. it also means the die retains all the graphical baggage, hurting price and power consumption.

Double-precision will be available on the Tesla line and on the high-end Quadro boards.
Tesla is a solution for HPC.

.< OK now, Sometime in the future can be next month or one year from now. we certainly do not want to buy a platform knowing that the next generation will come out a month later. You will tell me, that is the reason why you are not telling (because you want to sell your current platform). I will tell you that we need the 1.1 comp cap and the double precision.

I think that everybody would appreciates a release date ± 3 months. Come on … you can do that. :ph34r:

Well, are the current Teslas even available right now? The web page is delightfully vague, and I’ve not had anything back since filling in the web form.

I very strongly doubt that you will see a Tesla version of the G9x GPU until after the consumer version is released. From what I have heard, the earliest they will even announce the new GPU will be November and I wouldn’t expect to be able to actually go to and buy one at anything near MSRP until after the holidays.

any news about when double precision will be out?