The Week in Review
Here’s a collection of highlights, selected totally subjectively, from this week’s HPC news stream as reported at insideHPC.com and HPCwire.
>>10 words and a link
High-level .NET libraries for data-parallel computation;
NEC and Sun join forces on HPC services;
IDF: Harpertown due November 12;
IDF: Nehalem – 16 threads and QPI;
HPC companies working hard to win Wall Street’s continued love;
World Community Grid runs world’s largest anti-virus program;
SDSC Pegs Astrophysicist Norman to Bring in NSF Track 2c;
PSC announces Executive Director;
>>PNNL to install Cray XMT
Cray announced this week (http://investors.cray.com/phoenix.zhtml?c=98390&p=irol-newsArticle&ID=1053229&highlight=) that the Department of Energy’s (DOE) Pacific Northwest National Laboratory (PNNL) has taken delivery of the first Cray XMT supercomputer (or at least the first publicly admitted one).
The XMT processors can execute 128 threads in hardware, which makes the system ideal for many data intensive computing applications. The XMT is a follow on to Cray’s previous MTA systems, of which only a few were ever sold.
>>Intel Gets Cozy With FPGAs
Not to be left out of the FPGA acceleration party, Intel has teamed up with Altera and XtremeData to deliver a Xeon-friendly version of FPGA coprocessing. The XD2000i module consists of three Stratix III FGPAs hooked onto Intel’s front side bus (FSB). From the Altera press release (http://www.altera.com/corporate/news_room/releases/products/nr-altxdatafsb.html?f=hp&k=wn3):
“With today’s release of the XD2000i, it’s an exciting time for the industry,” said Ravi Chandran, CEO of XtremeData. “Using Altera’s Stratix III FPGA, the XD2000i delivers up to five times the logic capacity of previous generations of ISAs. Additionally, it offers unparalleled double-precision floating performance per watt, the key benchmark requested by our customers in high-performance computing and financial analytics.”
Nallatech and Mitrionics have also announced an FPGA-FSB solution, in this case, with Xilinx hardware. The press release is at http://www.mitrionics.com/default.asp?pid=6&iid=54.
These are the first instances of FPGA coprocessing for an Intel platform. XtremeData as well as DRC Computer Corp. and Celoxica have similar offerings for AMD’s HyperTransport technology.
>>AMD’s tri-core processor
The Inquirer has a story on AMD’s reported tri-core processor (http://www.theinquirer.net/default.aspx?article=42369). Seems like this is NOT just a salvage strategy:
AMD is probably doing this for two reasons; the lesser being salvage, the more important one being that Intel can’t do it. Intel would have a far harder time making a tri-core part until Nehalem next September — it is easy to fuse off a core, far harder to MCM disparate cores.
The Inquirer goes on to say that Hector (Ruiz) and the rest of the AMD boys are learning a thing or two from the ATI crowd:
…it seems AMD is learning from ATI. Most non-top SKU GPUs are simply top SKU die with features turned off, and if you look at the success of people unlocking that, you will see that it is far more than salvage.
>>New Opterons not very powerful. Enough for some?
Last week, TechReport.com released the first thorough review of AMD’s Barcelona two-way quad-core CPUs (http://techreport.com/articles.x/13176), including a comparison with Intel’s Clovertown chips. In terms of absolute performance, they conclude:
Barcelona’s gains in performance per clock aren’t quite what we expected, especially in floating-point-intensive applications like 3D rendering, where it looks for all the world like a quad-core K8. As a result, Barcelona is sometimes faster, sometimes slower, and oftentimes the equal of Intel’s Core microarchitecture, MHz for MHz.
That’s bad news for AMD, considering 1) the current top clock-speed option is 2/3 of Intel’s, and 2) they’re more than a year behind Intel in manufacturing technology (Intel’s 45nm Penryn quad-core chips are just around the corner). However:
…power-efficient performance is the new key to processors, especially in the server space. By that standard, AMD now has the lead. By any measure — and we have several, including idle power, peak power, and a couple of energy use metrics — the quad-core Opterons trump Intel’s quad-core Xeons.
The Xeon platform’s FB-DIMM memory consumption seems to be the major factor holding it back compared to Barcelona, as well as Xeon’s comparatively high power drain at idle. Any way you look at the findings, it should be nice as a consumer to know that the CPU market remains highly competitive.