Visit additional Tabor Communication Publications
December 11, 2009
As we reported on Monday, the life story of Intel's manycore GPU processor architecture, codenamed "Larrabee," has taken an interesting twist. Late last week, the company canceled the rollout of the first Larrabee product, slated to debut next year, while leaving the door open to future graphics and HPC offerings. IT analysts reacted in force, offering a range of speculation on Intel's decision and what the company's next move might be.
Like many graphics industry watchers, analyst Jon Peddie figured Intel looked at the flattening growth numbers for discrete GPUs and decided Larrabee didn't have what it takes to compete in a tightening market. But he's less sure if Intel will regroup and make a second stab at a discrete GPU platform or switch the focus of the technology to HPC.
In his blog this week, Peddie implied -- as Intel also seemed to do in their brief announcement -- that the next step for the technology would, in fact, be in high performance computing:
Intel has made a hard decision and we think a correct one. Larrabee silicon was pretty much proven, and the demonstration at SC09 of measured performance hitting 1 TFLOPS (albeit with a little clock tweaking) on a SGEMM Performance test (4K by 4K Matrix Multiply) was impressive. Interestingly it was a computer measurement, not a graphics measurement. Maybe the die had been cast then (no pun intended) with regard to Larrabee’s future.
Of course, 1 teraflop of single precision (for an overclocked chip, by the way) is no great shakes for GPUs these days. Although SGEMM benchmark results are unavailable for NVIDIA and AMD GPU silicon, in general, graphics chips are already well into multi-teraflop territory, with even more performance on the way. And as NVIDIA quickly found out, most of the action in HPC is in the double precision arena anyway.
GigaOM's Stacey Higginbotham opines that Intel maybe tried to push the x86 beyond its natural abilities by forcing it into the GPU mold. She writes this week:
For Intel, the question becomes, how far can the x86 architecture stretch? Its Larrabee delay suggests that using x86 to develop a decent graphics processor may work, but it can’t compete against specialty GPUs.
The implication here is that Intel learned the wrong lesson from Itanium, that is, x86 compatibility trumps everything else. In the case of discrete GPUs, that's less likely to be true. The software ecosystem is focused at the API level (e.g., OpenGL and DirectX) rather than the ISA. But for general-purpose GPU computing, x86 compatibility still may make some sense. Like Peddie, Higginbotham thinks the second coming of Larrabee is likely to appear in HPC.
Nathan Brookwood, principal analyst at Insight 64, also expects Intel to deliver a new and improved Larrabee. He writes:
Eventually, although not before 2011, a next generation Larrabee will emerge. Intel always learns from its mistakes, so this new design will avoid the problems that crippled the first generation design. They can replace the archaic Pentium-era core used in the Gen 1 Larrabee with their new, power-efficient Atom core (that was not yet complete when they started Gen 1 in 2007) and thus tame the chip’s power appetite.
However Brookwood goes on to say that the delay blunts Intel's natural advantage in x86 software, since OpenCL, DirectCompute and CUDA will be that much more mature two years down the road. He points to the recent SC09 conference as evidence of how much momentum has already beem built around GPU computing.
The wildest reaction to the Larrabee news comes from industry reporter Bob Cringely, whose commentary headline says it all: Intel Will Buy nVIDIA. Cringely speculates that Intel wants NVIDIA mainly for its integrated Tegra chips for mobile platforms, and had to get rid of Larrabee to remove the appearance of creating an uncompetitive GPU market. He writes:
.Intel had to do something the minute AMD bought ATi. Now with Larrabee gone Intel has no real choice but to buy another company to remain in contention. The only such company available is nVIDIA.
He also speculated that the resolution of the Intel-AMD lawsuit was a setup to prevent AMD from objecting to a future Intel-NVIDIA merger. If you're a fan of corporate intrigue, you'll love Cringely's take on the situation. But this merger idea has a whole host of problems, not the least of which is that these two companies are fundamentally incompatible. As Jon Peddie writes in a follow-on blog post: "The cultural differences, acrimony, and belligerences between Intel and Nvidia run so deep it would be impossible to blend the organizations without a few homicides." Having seen NVIDIA CEO Jen-Hsun Huang espouse this business philosophy at the recent GPU Technology Conference, I tend to agree.
My take is that it's still too early to count Intel out of the GPU computing game. Although this latest delay isn't going to help its cause, the overall idea of using a manycore, vector-accelerated x86 design for general-purpose data parallel apps is a reasonable strategy. The chipmaker's biggest mistake is that it tried to shoot at two fast-moving targets -- traditional graphics and GPGPU -- without the software expertise to pull it together. I don't think they'll make the same mistake again.
Posted by Michael Feldman - December 11, 2009 @ 7:35 AM, Pacific Standard Time
Michael Feldman is the editor of HPCwire.
No Recent Blog Comments
In quieter times, sounding the bell of funding big science with big systems tends to resonate further than when ears are already burning with sour economic and national security news. For exascale's future, however, the time could be ripe to instill some sense of urgency....
In a recent solicitation, the NSF laid out needs for furthering its scientific and engineering infrastructure with new tools to go beyond top performance, Having already delivered systems like Stampede and Blue Waters, they're turning an eye to solving data-intensive challenges. We spoke with the agency's Irene Qualters and Barry Schneider about..
Large-scale, worldwide scientific initiatives rely on some cloud-based system to both coordinate efforts and manage computational efforts at peak times that cannot be contained within the combined in-house HPC resources. Last week at Google I/O, Brookhaven National Lab’s Sergey Panitkin discussed the role of the Google Compute Engine in providing computational support to ATLAS, a detector of high-energy particles at the Large Hadron Collider (LHC).
May 23, 2013 |
The study of climate change is one of those scientific problems where it is almost essential to model the entire Earth to attain accurate results and make worthwhile predictions. In an attempt to make climate science more accessible to smaller research facilities, NASA introduced what they call ‘Climate in a Box,’ a system they note acts as a desktop supercomputer.
May 22, 2013 |
At some point in the not-too-distant future, building powerful, miniature computing systems will be considered a hobby for high schoolers, just as robotics or even Lego-building are today. That could be made possible through recent advancements made with the Raspberry Pi computers.
May 16, 2013 |
When it comes to cloud, long distances mean unacceptably high latencies. Researchers from the University of Bonn in Germany examined those latency issues of doing CFD modeling in the cloud by utilizing a common CFD and its utilization in HPC instance types including both CPU and GPU cores of Amazon EC2.
May 15, 2013 |
Supercomputers at the Department of Energy’s National Energy Research Scientific Computing Center (NERSC) have worked on important computational problems such as collapse of the atomic state, the optimization of chemical catalysts, and now modeling popping bubbles.
05/10/2013 | Cleversafe, Cray, DDN, NetApp, & Panasas | From Wall Street to Hollywood, drug discovery to homeland security, companies and organizations of all sizes and stripes are coming face to face with the challenges – and opportunities – afforded by Big Data. Before anyone can utilize these extraordinary data repositories, however, they must first harness and manage their data stores, and do so utilizing technologies that underscore affordability, security, and scalability.
04/15/2013 | Bull | “50% of HPC users say their largest jobs scale to 120 cores or less.” How about yours? Are your codes ready to take advantage of today’s and tomorrow’s ultra-parallel HPC systems? Download this White Paper by Analysts Intersect360 Research to see what Bull and Intel’s Center for Excellence in Parallel Programming can do for your codes.
In this demonstration of SGI DMF ZeroWatt disk solution, Dr. Eng Lim Goh, SGI CTO, discusses a function of SGI DMF software to reduce costs and power consumption in an exascale (Big Data) storage datacenter.
The Cray CS300-AC cluster supercomputer offers energy efficient, air-cooled design based on modular, industry-standard platforms featuring the latest processor and network technologies and a wide range of datacenter cooling requirements.