AMD Awarded $32 Million for 'Extreme Scale' High-Performance Computing Research

[btarunr]

AMD (NYSE: AMD) today announced that for the third straight year it was awarded research grants for development of critical technologies needed for extreme-scale computing in conjunction with projects associated with the U.S. Department of Energy (DOE) Extreme-Scale Computing Research and Development Program, known as "FastForward 2."

The two DOE awards, totaling more than $32 million, will fund research focused on exascale applications for AMD Accelerated Processing Units (APUs) based on the open-standard Heterogeneous System Architecture (HSA), as well as future memory systems to power a generation of exascale supercomputers capable of delivering 30-60 times more performance than today's fastest supercomputers.

FastForward 2 is a jointly-funded collaboration between DOE Office of Science and the National Nuclear Security Administration (NNSA) focused on initiating partnerships with multiple companies to accelerate the research and development of critical technologies needed to enable exascale computing. Specifically, the FastForward 2 research grants awarded to AMD are for:

• Node Architecture: AMD will conduct research for an integrated exascale node architecture based on its HSA-enabled APUs. Particular areas of research include power efficiency, reliability, programmability, component and network interface integration, APU microarchitecture, advanced memory architectures and efficient data movement.
• Memory Technology: AMD will collaborate with the DOE and others to help define a new standard for memory interfaces that meets the needs of future-generation memory devices, including non-volatile memory and processing-in-memory (PIM) architectures.

"We're honored to be selected once again by DOE to aid research efforts for exascale-computing," said Alan Lee, AMD's corporate vice president for Research and Advanced Development. "Our customers have long looked to AMD to develop innovative compute, graphics, and memory technologies and products that solve real-world problems. This research is focused on energy-efficient node architectures and memory systems to improve the capability of the world's fastest supercomputers."

The DOE strategic plan seeks to address the nation's most pressing scientific challenges by advancing simulation-based scientific discovery made possible by the world's highest performing exascale supercomputers. The technologies developed as part of this initiative could positively impact low-power embedded, cloud/datacenter and mid-range high-performance computing applications for diverse fields ranging from medical science to astrophysics and climate modeling. FastForward 2 investments will result in an increase in DOE's ability to leverage commercial developments for future systems.

AMD Opteron compute processors and AMD FirePro graphics processors are used today in many of the world's leading supercomputers.

View at TechPowerUp Main Site

 

[natr0n]

I bet they blow the money.

Lets hope it goes to something worthwhile.

 

[FordGT90Concept]

$32 million really isn't much. AMD would have to find other means to fund the project as well (e.g. IBM).


I think what DoE is looking for is Mangycores and similar technologies where they pack as much data execution into one package as possible. I can't see APUs+GPUs being the future of supercomputing.

 

[HumanSmoke]

$32 million really isn't much.

What do you mean? $32m is at least a dozen milspec ratchet screwdrivers and a toaster oven in DoD/DoE currency!

AMD would have to find other means to fund the project as well (e.g. IBM).
I think what DoE is looking for is Mangycores and similar technologies where they pack as much data execution into one package as possible. I can't see APUs+GPUs being the future of supercomputing.

AMD = HSA, IBM = OpenPOWER. Can't see them buddying up somehow, especially when IBM and Nvidia (another OpenPOWER signatory ) just got a $325m contract for the DoE's Sierra and Summit supers

 

[FordGT90Concept]

•Memory Technology: AMD will collaborate with the DOE and others to help define a new standard for memory interfaces that meets the needs of future-generation memory devices, including non-volatile memory and processing-in-memory (PIM) architectures.

This is right up IBM's alley and is pretty much completely foreign to AMD.

The "node architecture" sounds stupid. That is, HSA is already done and it is already mature. No research grants should be awarded to AMD for this and I'd argue AMD is in no position to address "memory technology" unless there's some prototype they revealed that I missed.

 

[BUCK NASTY]

This is right up IBM's alley and is pretty much completely foreign to AMD.

The "node architecture" sounds stupid. That is, HSA is already done and it is already mature. No research grants should be awarded to AMD for this and I'd argue AMD is in no position to address "memory technology" unless there's some prototype they revealed that I missed.

32 Mil will buy a lot of Hookers and Blow. After they are done celebrating, maybe they will have a little left to apply towards research.

 

[GhostRyder]

This is right up IBM's alley and is pretty much completely foreign to AMD.

The "node architecture" sounds stupid. That is, HSA is already done and it is already mature. No research grants should be awarded to AMD for this and I'd argue AMD is in no position to address "memory technology" unless there's some prototype they revealed that I missed.

There is plenty more research needed to be done before it can really be called complete in my book because while its there its still a new technology that has bugs to work out so I do not see this being a bad thing.

Alas, 32million isn't much to be sure in the grand scheme of things but it will help.

 

[TRWOV]

This is right up IBM's alley and is pretty much completely foreign to AMD.

The "node architecture" sounds stupid. That is, HSA is already done and it is already mature. No research grants should be awarded to AMD for this and I'd argue AMD is in no position to address "memory technology" unless there's some prototype they revealed that I missed.

Wasn't AMD developing HBM along with Hynx?

 

[FordGT90Concept]

Wasn't AMD developing HBM along with Hynx?

Indeed: AMD Working With Hynix For Development of High-Bandwidth 3D Stacked Memory

 

[Steevo]

What do you mean? $32m is at least a dozen milspec ratchet screwdrivers and a toaster oven in DoD/DoE currency!

AMD = HSA, IBM = OpenPOWER. Can't see them buddying up somehow, especially when IBM and Nvidia (another OpenPOWER signatory ) just got a $325m contract for the DoE's Sierra and Summit supers

IBM and AMD have been and continue to be good friends with sharing tech, its unfortunate they got 32M for this though as I am fairly certain they will find a way to fuck it up by adding another layer on the already too deep shit pile they have.

AMD needs to go back to the drawing board and fix their latency issues in the CPU, Its pitiful when separate components outperform a singular solution due to the inherent issues, for example a i3 and equal AMD GPU outperforms APU's, even though the APU is all on die in CPU bound cases, and in many GPU bound.

 

[xenocide]

IBM and AMD have been and continue to be good friends with sharing tech, its unfortunate they got 32M for this though as I am fairly certain they will find a way to fuck it up by adding another layer on the already too deep shit pile they have.

AMD needs to go back to the drawing board and fix their latency issues in the CPU, Its pitiful when separate components outperform a singular solution due to the inherent issues, for example a i3 and equal AMD GPU outperforms APU's, even though the APU is all on die in CPU bound cases, and in many GPU bound.

If you use Mantle the APU will pull ahead, but lolwhousesthatshit...

AMD really needs to do something about how god awful their CPU's really are.

 

[Thefumigator]

If you use Mantle the APU will pull ahead, but lolwhousesthatshit...

AMD really needs to do something about how god awful their CPU's really are.

I don't see my FX 8320 a god awful cpu
BTW excavator is on the way... but who knows...

 

[HumanSmoke]

I don't see my FX 8320 a god awful cpu
BTW excavator is on the way... but who knows...

Excavator is the Carrizo APU on 28nm. Hardly likely to trouble Intel's current 22nm offerings let alone 14nm when it arrives.

 

[TRWOV]

I don't see my FX 8320 a god awful cpu
BTW excavator is on the way... but who knows...

Last year AMD's roadmap stated that Vishera would be the only core for AM3+ through 2014 and 2015. I suppose that they won't launch another full fledged CPU until they switch architectures (the rumored Zen Core) or at least change the socket and chipset.

http://www.techpowerup.com/195355/vishera-end-of-the-line-for-amd-fx-cpus-roadmap.html

And the FX are not awful. They're pretty good for what they offer. There's the argument about power consumption but if you don't keep your PC on 24/7 I think you could use it for years before having a noticeable impact on your power bill.