Nvidia announces Fermi-based Tesla GPUs

Don’t we all love a good paper launch? Fermi-based graphics processors may not be ready for stores just yet, but that hasn’t stopped Nvidia from announcing its first Tesla products based on the new architecture. These Tesla 20-series offerings include two workstation cards and two quad-GPU workstation systems, all of which Nvidia expects to make available in the second quarter of next year.

We already went over what Fermi brings to the high-performance computing stage in our detailed look at the architecture in September. Some of the highlights include: compliance with the IEEE 754-2008 standard, ECC memory capabilities, a multi-level cache hierarchy, and of course, a sizeable increase in processing resources. Nvidia claims single Tesla 20-series GPUs can achieve 520-630 gigaFLOPS of double-precision, floating-point number-crunching power.

The two workstation cards are the Tesla C2050, which will cost $2,499 and support up to 3GB of GDDR5 memory, and the Tesla C2070, which will set you back $3,999 and have room for 6GB of onboard GDDR5 RAM. Nvidia quotes a 9.75" board length and 190W "typical" power consumption for both products.

For datacenters, Nvidia will also offer the $12,995 Tesla S2050 and $18,995 Tesla S2070. Both products will cram four Fermi-based GPUs within a 1U system with total double-precision performance of 2.1-2.5 teraFLOPS. Per-GPU memory capacities will be the same as for the workstation cards, so the S2050 will offer up to 12GB in total, and the S2070 will carry up to twice that.

What about Fermi-based GeForces? Nvidia hasn’t announced any products just yet, but it does say in the press release, "As previously announced, the first Fermi-based consumer (GeForce®) products are expected to be available first quarter 2010."

Comments closed
    • lycium
    • 11 years ago

    i’ve been rendering the recently-popular 3d mandelbrot (or “mandelbulb”) in double precision on the geforce gtx 260 i bought today and it’s… very slow unfortunately 🙁

    basically fermi will give me the same performance using doubles that i get with single precision fp now, together with added precision for all modes (thanks to denormals, common in tricky numerical code like surface normal estimation), which is going to be really awesome. (not to mention full c++ and debugging support, which means a physically-based renderer can theoretically be ported to the platform, complete with programmable materials.)

    • wira020
    • 11 years ago

    Is the card shown just for illustration or is it the one Nvidia working on?

    p/s: There wasnt any demo right?.. which would means a mock up…

    • ClickClick5
    • 11 years ago

    More vaporcards. At least AMD had the small chance of releasing a few 5xxx before they ran out of chips…nvidia just cant get their chips together.

    But if Nvidia keeps up their code rigging, the Geforce 8800 will be able to play for a very long time. Why release this?

      • Shining Arcanine
      • 11 years ago

      The IEEE 754-2008 standard compliance is a fairly important feature. After all, no one cares how fast you can do computations if you produce the wrong answers.

      It also will make things easier for programmers who presently need to hack around non-conformance in current GPU floating point implementations, at the expense of performance.

    • Spotpuff
    • 11 years ago

    I don’t know what any of the words Tesla, Fermi, or Nvidia mean but I would like to buy one of these future cards.

    • maxxcool
    • 11 years ago

    I would like to announce that in 5-10 years I should be able to release a operating system that is 100% compatible with windows XP.

    • Mystic-G
    • 11 years ago

    It can run Crysis?

    • flip-mode
    • 11 years ago

    q[

      • l33t-g4m3r
      • 11 years ago

      [prime1]this isn’t a paper launch, the 5870 is a paper launch.[/prime1]

        • MadManOriginal
        • 11 years ago

        This isn’t the paper launch you’re looking for *waves hands*

      • Meadows
      • 11 years ago

      Prime1 hasn’t appeared so far, he’s lost.

    • BoBzeBuilder
    • 11 years ago

    I announce flying cheese. To be released sometime in the future. Stay tuned, don’t buy any cheese and wait.

    BTW, it can fold too. Take that AMD.

      • Buzzard44
      • 11 years ago

      I was gonna get a 5850, but now I’m saving up for this folding flying cheese.

      Now I’ve just got to decide the order in which I will fold/fly/eat it (when it is released, of course).

        • SomeOtherGeek
        • 11 years ago

        And the best part, it grows! So, buy the specially designed case to keep the smell in and let you spread your cheese on crackers and such.

    • FuturePastNow
    • 11 years ago

    9.75″ is a smaller card than I’d expect, assuming the graphics version uses the same PCB. Maybe they’re getting pressure from OEMs to make the cards shorter.

    • Hattig
    • 11 years ago

    Wow, a 5 month advance paper launch? Is this a record, there aren’t even reviewable devices!

      • Deanjo
      • 11 years ago
        • CampinCarl
        • 11 years ago

        That’s the 10 series. They just launched the 20 series.

        • Hattig
        • 11 years ago

        That’s the existing line of C10xx devices, not the C20xx devices this story is talking about.

        • Deanjo
        • 11 years ago

        My mistake wrong card.

      • Jigar
      • 11 years ago

      What can you expect ? They are caught with their pants down…

      • dpaus
      • 11 years ago

      Not even close to a record – ARM just announced new processors to be available in… wait for it…. 2013!

      • stmok
      • 11 years ago

      l[

    • ssidbroadcast
    • 11 years ago

    So… I guess this means that the NDA on Fermi can be lifted?

      • bLaNG
      • 11 years ago

      I guess non-disclosure agreement gets a whole new meaning here: non-disclosure of any real product to the market agreement (or NDOARPTTMA)

Pin It on Pinterest

Share This