FireStream 9370, 9350 bring Cypress computing to servers

In conjunction with today’s announcement of the Opteron 4100 series of CPUs, AMD has also raised the curtains on a pair of FireStream cards. The FireStream 9370 and 9350 are both based on the Cypress GPU first deployed in the Radeon HD 5870 graphics card, but they’re very geared for use in servers as high-performance GPU computing engines.

The FireStream 9370, for instance, fits into a PCI Express x16 slot and has a total of 4GB of GDDR5 memory onboard. It has only a single DisplayPort output, and that may go unused in most cases. New in this generation of FireStreams is a wholesale move to passive coolers, to allow for tighter integration into servers. The cards will rely on chassis-based airflow to keep them cool. Like a high-end graphics card, the 9370’s cooler occupies the width of two expansion slots.

The adoption of a Cypress GPU gives the 9370 more than double the peak throughput of the 9270 it replaces, with a single-precision floating-point peak of 2.64 teraflops and a double-precision peak of 528 gigaflops. At least as importantly, Cypress has a handful of new compute-specific features built in, including higher precision computation, better support of atomic operations, and improved data sharing and thread synchronization. We’ve covered these features in more detail here. The upshot is a GPU that’s well suited to the requirements of OpenCL and other emerging tools for GPU computing.

The 9370 is very much a premium product, with a power requirement of “under 225W,” according to AMD, and a suggested price just one dollar shy of two grand.

The FireStream 9350 may prove to be more popular; it’s a scaled-down variant with 2GB of GDDR5 memory, a single-slot passive cooler, and a power requirement of “under 150W.” Lower clock speeds yield a peak SP arithmetic rate of 2.0 teraflops and a peak DP rate of 400 gigaflops. The price is nicer, too, at $799.

AMD expects both products to be available in the third quarter of this year, “a few weeks to a couple of months from now,” according to Patti Harrell, AMD’s Director of Stream Computing.

The system pictured above will give you a sense of the sort of compute density upcoming integrated FireStream-based solutions could provide. This 1U system from SuperMicro packs in two FireStream 9370 compute accelerators, dual Opteron 6100-series (12-core) CPUs, 16 DDR3 DIMM slots with (if our math is right) eight memory channels, and three hot-swappable SATA/SAS drive bays. The power supply is rated for a staggering 1.4kW. In the back is room for a low-profile PCIe x8 add-on card, possibly an Infiniband interconnect. The GPUs alone should reach over 5.2 teraflops.

Other vendors are working on larger systems with four FireStream 9370s or eight 9350s; an eight-way 9350 system would peak at over 16 teraflops.

The most obvious competitors for the FireStream 9300 series are Nvidia Tesla cards based on the Fermi GPU architecture. AMD cites the Tesla C/M2000-series as a point of comparison, noting that the Tesla tops out at one teraflop for single-precision math and 514 gigflops for double-precision. Both figures are below the FireStream 9370’s peak rates. The Teslas require power from an 8-pin connector and a 6-pin one, while the 9370 needs only dual 6-pin power inputs. This comparison won’t be unfamiliar to those who know the desktop graphics cards based on these same GPUs. Nvidia has a much larger chip but no real edge in performance.

In the GPU computing space, though, Nvidia has some noteworthy advantages, including ECC protection throughout its memory hierarchy, a true L2 cache, and what is probably a more robust set of development tools than those produced so far by AMD’s Stream initiative. ECC support will likely open doors for Nvidia in large supercomputing clusters that AMD can’t yet open, and the Fermi architecture’s additional computing features could allow it to achieve higher performance and superior efficiency when running certain algorithms.

When we asked Harrell how she would address the tricky question of ECC support with potential customers, she said that AMD tests both software and hardware reliability, down to neutron beam testing in its labs. Those tests have revealed that the memory interface is the most vulnerable point in the system, and GDDR5’s error correction adds a measure of protection at that point.

On the topic of software development tools for Stream-capable GPUs, Harrell sounded more confident than in our past forays into this area. She claimed AMD has a substantial investment in tools with third parties now, and she expects it to bear fruit in the form of product announcements later this year.

Comments closed
    • Silus
    • 11 years ago

    Is AMD really trying to compete with Tesla by citing theoretical numbers ? LOL

    How about some comparison numbers and I mean real ones, done by sites such as TR, and not some AMD provided PR.

    And I wonder why the news of NVIDIA’s Fermi based Teslas was labeled as a paper launch by Cyril (which was deserving of course, since it was a paper launch), but this one isn’t ?

      • flip-mode
      • 11 years ago

      LOL, funny boy Silus nv lover angst about AMD product much cover for Nvidia sweetheart so sad

        • Silus
        • 11 years ago

        “AMD product much cover for Nvidia sweetheart so sad”

        Has “Tarzan speak” struck you or something ?

          • flip-mode
          • 11 years ago

          he he, defense of nvidia and insult of the ati fan is so important to you, sad man, poor man.

            • Silus
            • 11 years ago

            pot meet kettle

            • flip-mode
            • 11 years ago

            kettle meet stew meet spoon meet stomach meet arse

        • Deanjo
        • 11 years ago

        He does have a point about the theoretical peak rates. When it comes to HPC on GPUs without proper ECC those rates drop to about 2/3’s of actual usable data with the possibility of dropping even more. It’s a weak point in their current solution and ATI is aware and knows it has to improve it’s product to make a real dent in nVidia’s offerings.

      • SubSeven
      • 11 years ago

      How can you conclude that this a paper launch? This product is said to be available in “a few weeks to a couple of months”. Wouldn’t it be prudent to give AMD the benefit of a doubt and at least wait until after the time has past to call this a paper launch? If they do deliver this product in said time do you realize that you will look like an Equus africanus asinus? Your remarks make you sound like a fool and a heavily biased one at that.

        • Silus
        • 11 years ago

        Ok…then what do you call the announcement/launch of a product, without availability ? It’s always been “paper launch” until now.

        And by your reasoning of benefit of the doubt, the news I mentioned before where Cyril labeled this – §[<https://techreport.com/discussions.x/17980<]§ - a paper launch, should no longer be considered a paper launch, since NVIDIA said it would deliver these products in the first quarter of 2010 and they did... But no, in the real world, it was really a paper launch. Just like this. Terms don't change as you go along...and they won't change because you want to use this to call me a fool either. But then again, there are no arguments anymore, just insults...

          • SubSeven
          • 11 years ago

          I stand corrected. Using what you’ve said above to define what paper launch means, then this is by all means a paper launch. But note that your remarks still make you sound like a fool (and a childish one at that) because you get worked up and feel the need to defend Nvidia every single time! You realize that to a “byreader” your “you did this to x, how come you aren’t doing this to y too?” remarks are pedantic and childish. Though I see your point and agree that to be perfectly fair, Cyril should have slammed AMD as well. But what are you accomplishing by being the one to consistently point this and other “injustices” to Nvidia out? Are you trying to suggest that Cyril is pro AMD biased? If so, that is to Cyril’s detriment; if this is indeed true, then Cyril’s credibility and reputation as a reporter stand to suffer. Let the readers worry about Cyril and focus on yourself instead. You don’t need to defend Nvidia every time, they are a big & great company and I assure you, they can handle themselves just fine. I hope you get where I’m going with this. Despite what you may believe, I’m not writing this to insult you. My goal is to reflect your image back to you so that you can see what you look/sound like to most of the readers in here. Unlike Pr1me, who in my opinion is hopeless, I sense there is still hope in you.

    • dpaus
    • 11 years ago

    I believe you’re thinking of the inverse tachyon /[

    • Flying Fox
    • 11 years ago

    How much Folding ppd?

      • SomeOtherGeek
      • 11 years ago

      Ahhh! Probably not as good as the Fermi POS as long as the software side sux on it.

    • lilbuddhaman
    • 11 years ago

    But does it run Crysis ?

      • Jigar
      • 11 years ago

      No it can walk Crysis…

    • iatacs19
    • 11 years ago

    How is that passively cooled? LOL

      • Farting Bob
      • 11 years ago

      Really, really high airflow running across the server(s), as you can see from the 16 extra thick fans in the photo. Creates a big ass current across the whole board.

      • OneArmedScissor
      • 11 years ago

      There were passive 4850s. The card isn’t picky about where the air comes from so long as the heatsink is drawing it away. :p

      The way the heat is drawn out is a lot more important than how close the source of air current is. My passive 9600GT idles about 20C lower than my 6600GT idles with the fan on 100%.

      If copper wasn’t so pricey, most cards could be passive.

      • Duck
      • 11 years ago

      You don’t want to know how loud this thing will be. I bet those fans pull 1A each or something :O

        • SNM
        • 11 years ago

        We have 3 servers of that size in a closet at work for some of our luckier devs to work on (besides our big datacenter) and when you’re in there it sounds like a vacuum cleaner.

    • Dposcorp
    • 11 years ago

    “………..AMD tests both software and hardware reliability, down to neutron beam testing in its labs…………”

    Yawn.
    Wake me up when they start using a inverse tachyon beam.

    §[<http://img2.moonbuggy.org/imgstore/id-hit-it-with-an-inverse-tachyon-beam.jpg<]§

      • iatacs19
      • 11 years ago

      Or when they start writing stable drivers.

Pin It on Pinterest

Share This