AMD Gains HPC Market Share, Interlagos On The Way

AMD Gains HPC Market Share, Interlagos On The Way

AMD announced its share of the TOP500 supercomputer list has grown 15 percent in the past six months. The company credits industry trends, upgrade paths, and competitive pricing for the increase. Of the 68 Opteron-based systems on the list, more than half of them use the Opteron 6100 series processors. We covered the launch of Magny-Cours more than a year ago; AMD's current lineup includes the dodeca-core Opteron 6180 SE at 2.5GHz at one end and two low-power parts at the other.

Magny-Cours adoption is important. Organizations typically don't upgrade HPC clusters with new CPUs, but AMD is billing Interlagos as a drop-in option for MC. As such, it'll offer up to 2x the cores (16x as compared to 8x) and equal-to-faster clockspeeds. AMD's corporate marketing VP, Leslie Solon, said the following:
The introduction of the AMD Opteron processor coincided with the HPC industry’s shift to x86 and cluster-based architecture, away from expensive proprietary or RISC-based systems,” said Leslie Sobon, corporate vice president, product marketing, AMD. “AMD provided the server technology that helped drive the democratization of supercomputing and has shown steadfast leadership in helping HPC break performance boundaries. Our new 16-core processor codenamed “Interlagos” will add features specifically for HPC and offer the world’s highest core count for x86. In addition to CPUs, we are also now seeing adoption of our GPU technologies in HPC and believe this will be a significant area for further development.


If current trends continue, AMD's share of the supercomputer market will slowly rise. The chart is interesting in its own right. Intel's share of the industry exploded virtually overnight, from a bare footnote in 2001 to a huge slice of the market in 2005. AMD's Opteron debut went extremely well, but topped out in 2006. It fell steadily for several years thereafter, thanks to increased pressure from Intel's Core architecture. Thus far, 2011 looks like a winning year, though we suspect Interlagos (currently scheduled for a Q3 debut) will make the difference.

According to the firm, Interlagos will feature multiple enhancements for HPC workloads, including a "Flex FP" technology. Presumably this refers to the fact that Bulldozer modules share an FPU. AMD is attempting to spin this as a positive and refers to Flex FP as "an innovation that allows for significant overall processor power savings when the Flex FP is not fully utilized."

That's not necessarily untrue. Bulldozer is quite different from any of AMD's previous architectures. Previous K8/K10 processors were built for high performance first and foremost. Bulldozer, in contrast, is designed to maximize instruction execution efficiency and die size. AMD's decision to build a single, powerful, shared FPU unit rather than building two completely separate cores may well reduce power consumption. The big question now is whether or not Bulldozer can deliver performance good enough to challenge Intel's dominance.
0
+ -

That is quite the graph there. Really shows the dynamic of business I think. In the beginning a decent number of competitors until the numbers dwindle to just a few top selections. Thanks for the info and illustration :)

0
+ -

Wow that is really impressive. I am amazed at Intel's current dominance of the HPC market and wish AMD luck in capturing more of the market share.

0
+ -

I would like to see the Intel share in this graph show the split between X86 and Itanium.

0
+ -

Um question? how do you read that graph... lol Like a normal line graph?

Keep up the work AMD :)

0
+ -

Yes, business choices - It shows the usual trend of buying hardware that runs the popular software. Specs only compared after this point.

Specs are based on comparisons of how well a particular language/API compiles to the target hardware. Ie: RISC architectures get a raw deal out of stacking mentalities. Even the old 80386 would do better if there was less stack based software. This is particularly true now that AMD has finally given x86-64 a decent number of general registers to play with, and, of course, along with x86 long having been a RISC design itself but with a little CISC support still attached.

Better late than never as they say.

0
+ -

Actually, the slowness of the PC architecture advancements could almost be compared to evolution! It's been quite the waiting game.

0
+ -

Interesting perspective, Evan. And I half agree with you in spots here. :-)

0
+ -

I hope AMD can become competitive again as we all benefit.

I'm no electrical engineer by any means but I think the days of high clocks do not matter as much as the pipeline length and how many instructions per cycle can be processed and this is where AMD is heading with multi core processors.

Will it be enough, only time will tell, they do have their work cut out for them though as Intel is on a roll.

0
+ -

great graphics

0
+ -

JWhite1: Other historical development graphs can be generated at the Top500.org site which hosts this data.

These two separate the Intel x86 (IA-32), Intel Itanium (IA-64), and Intel 64 (EM64T).

http://top500.org/overtime/list/37/procfam

The Number of Systems graph shows the number of systems in the Top500 list, not number of processors they contain. Notice SPARC barely shows up in the number of systems even though the #1 system is a SPARC system and provides almost 14% of the total performance of the Top500, 8.2 of 58.8 petaflops.

The Performance Share Graph shows the performance contributed to the total Top500 performance by systems that USE that CPU processor architecture, NOT necessarily performance contributed BY that processor. Notice that co-processors and GPUs are not listed: NVidia does not appear on the graph. For example, the performance share of systems [like #2 Tiahne-1A, #4 Nebulae, or #5 Tsubame-2] that use Intel64 CPUs with NVidia GPUs may inflate the apparent performance share of the Intel CPUs.

0
+ -

fantastic graph!

Login or Register to Comment
Post a Comment
Username:   Password: