Monday, September 14th 2020

AMD COVID-19 HPC Fund Adds 18 Institutions and Five Petaflops of Supercomputer Processing Power to Assist Researchers

AMD today announced a second round of high-performance technology contributions to assist in the global fight against the COVID-19 pandemic. AMD is now contributing high-end computing systems or access to Penguin-On-Demand (POD) cloud-based clusters powered by 2nd Gen AMD EPYC and AMD Radeon Instinct processors to 21 institutions and research facilities conducting COVID-19 research. With 12 petaflops of total supercomputing capacity now awarded, the combined compute capacity donated through the AMD COVID-19 HPC Fund would rank among the fastest supercomputers in the world according to the most recent Top500 list.

"AMD is proud to be working with leading global research institutions to bring the power of high performance computing technology to the fight against the coronavirus pandemic," said Mark Papermaster, executive vice president and chief technology officer, AMD. "These donations of AMD EPYC and Radeon Instinct processors will help researchers not only deepen their understanding of COVID-19, but also help improve our ability to respond to future potential threats to global health."

The AMD COVID-19 HPC fund was established to provide research institutions with computing resources to accelerate medical research on COVID-19 and other diseases. In addition to the donations of $15 million of high-performance computing systems, AMD has contributed technology and technical resources to nearly double the peak system of the "Corona" system at Lawrence Livermore National Laboratory which is being used to provide additional computing power for molecular modeling in support of COVID-19 research.

The second round of AMD-donated compute capacity is expected to be operational starting in Q4 of this year and will be used for a number of pandemic-related workloads including genomics, vaccine development, transmission science, and modeling. Research projects range from evolutionary modeling of the virus, to understanding the virus spike protein activation that occurs prior to first interaction between the coronavirus and human cell, and large scale fluid dynamics simulations of COVID-19 droplets as they travel through the air. To maximize the impact of the research, AMD is also initiating a working group for COVID-19 HPC Fund recipients and AMD engineers to jointly discuss research areas and findings as well as hardware and software optimizations that can accelerate their collective work.

To date, the AMD COVID-19 HPC fund has donated computing systems or cloud-based computing capacity to: Cambridge University, Carnegie Mellon, GENCI / French National High-Performance Computing Agency, Harvard Children's Hospital, High Performance Computing Center (HLRS) / the University of Stuttgart, MIT, NYU, CSIR Fourth Paradigm Institute in India, Leibniz Supercomputing Centre (LRZ), Rice, Stanford School of Medicine, Texas State University, The University of British Columbia, The University of Texas at Austin, UCLA, University of Arkansas, University of Toronto, University of Trento, University of Vermont, Virginia Commonwealth University, and Washington University.

AMD COVID-19 HPC Fund Recipient Support
Stanford School of Medicine
"We are very grateful to receive AMD's donation of high performance servers. The diverse evolution of COVID-19 epidemics in various parts of the world raises numerous, time-critical, location-specific, policy-relevant questions," said Dr. Jeremy Goldhaber-Fiebert, associate professor of Medicine and core faculty member of Stanford Health Policy. "Computational power like that provided by AMD allows our team to attempt to answer such question more quickly, more deeply, and for more places."

UCLA
"UCLA is grateful for the generous gift from AMD, which will make an important impact on our research at UCLA," said Todd Yeates, professor of biochemistry and director of the UCLA-DOE Institute for Genomics and Proteomics. "We have numerous research groups working on diverse aspects of the COVID19 problem, from epidemiology to molecular studies aimed at developing therapeutics and vaccines. Many of those studies are limited by computing needs. This new influx of power will be an opportunity for UCLA and AMD to work together in exploiting their processor technologies in new areas of computational research."

UT Austin
"AMD CPUs and GPUs will allow our researchers to expand the scope of their COVID-19 work while also helping them accelerate their timeliness significantly," said UT's Interim Vice President for Research Alison R. Preston. "This includes everything from identifying viral spike protein mutations that could help improve vaccines to using machine learning technology to study COVID disease severity in patients of all ages, from newborns to the elderly."

University of Toronto
"This new partnership is critical in helping the University of Toronto accelerate research on COVID-19 and develop innovations that will mitigate the impact of the pandemic on communities," said Alex Mihailidis, associate vice-President of International Partnerships. "The donation of cutting-edge computing resources will allow the University of Toronto and our affiliated hospitals expand our current infrastructure and launch SciNet4Health, a platform that can securely process big data sets of private health information to advance the discovery of therapies, vaccines and disease mutations. This donation brings us one step closer to the healthcare of the future."

AMD Ecosystem Partners
AMD continues to work with well-known HPC and AI solutions firm Penguin Computing, a division of SMART Global Holdings, Inc. (NASDAQ: SGH), to define, build, and deliver on-premises systems and Penguin Computing's POD clusters, powered by AMD. Penguin Computing's POD support is collocated in data center space donated by DataBank. Contributions from Gigabyte, Nvidia Networking, and others are also supporting the AMD HPC Fund.

"Penguin Computing is proud to announce that the first petaflop of compute capacity in our datacenter is now online and available for researchers to begin advancing the science of pandemic research to help find new ways to fight COVID-19. We look forward to continuing to support AMD and the latest round of universities and labs to participate in this global effort with Penguin Computing's applications and technology expertise," said Sid Mair, President, Penguin Computing Inc.

Gigabyte is supplying its G291-Z20 compute nodes for the Penguin Computing clusters, built around a single, 48-core AMD EPYC 7642 processor paired with eight Radeon Instinct MI50 GPU accelerators. The system R182-Z91 management nodes, also from Gigabyte, each utilize two 16-core, AMD EPYC 7302 processors.
Add your own comment

14 Comments on AMD COVID-19 HPC Fund Adds 18 Institutions and Five Petaflops of Supercomputer Processing Power to Assist Researchers

#2
Steevo
TheButcherNL
Lol, for a normal flu virus....
But but but.. . Be skared, really skared, cause it's gonna kill us all unless you are a protestor with the correct ideas!!
Posted on Reply
#3
Mark Little
TheButcherNL
Lol, for a normal flu virus....
Yeah! AMD should really be putting money into more police to help Q fight Hollywood liberal pedophiles!
Posted on Reply
#4
prtskg
Considering this virus has already killed more than 925,000 people globally, I don't think it's a 'normal flu' virus.
Posted on Reply
#5
dragontamer5788
Okay, as much as I love a COVID19 discussion (and yall know I'm willing to verbally fight on the issue)...
AMD is now contributing high-end computing systems or access to Penguin-On-Demand (POD) cloud-based clusters powered by 2nd Gen AMD EPYC and AMD Radeon Instinct processors to 21 institutions and research facilities conducting COVID-19 research
Gigabyte is supplying its G291-Z20 compute nodes for the Penguin Computing clusters, built around a single, 48-core AMD EPYC 7642 processor paired with eight Radeon Instinct MI50 GPU accelerators. The system R182-Z91 management nodes, also from Gigabyte, each utilize two 16-core, AMD EPYC 7302 processors.
Wait, what? AMD has a cloud-based cluster with EPYC + Radeon Instinct GPUs available for rent? This is news to me.

I mean, I guess I assumed that AMD would have an EPYC + GPU combo available somewhere. But this "Penguin-on-demand" cluster is the first actual name I've seen for the Radeon "Instinct" brand. There's GPUEater, which uses Rx 580 and Rx Vega GPUs. But the Radeon Instinct (aka: MI50) type GPUs have faster communication between the GPUs, which could be useful in some supercomputing instances.

I did check out Penguin HPC briefly: but they only seem to have documentation on Intel clusters: pod.penguincomputing.com/documentation/account-creation.html#about-our-clusters-mt1-mt2.
Gigabyte is supplying its G291-Z20 compute nodes for the Penguin Computing clusters, built around a single, 48-core AMD EPYC 7642 processor paired with eight Radeon Instinct MI50 GPU accelerators. The system R182-Z91 management nodes, also from Gigabyte, each utilize two 16-core, AMD EPYC 7302 processors.
This does sound like a great system. MI50 is Vega 7nm, almost equivalent to Radeon VII, which was proven to be good at compute (though weak compared to NVidia with regards to deep-learning... its still a very wide SIMD compute system with 1TBps VRAM bandwidth). IIRC, those MI50 GPUs can transfer data between each other at ~180+ GBps speeds, meaning each of those nodes has a significant amount of "traditional" GPU-compute power. (Protein folding won't use Tensor Cores: IIRC, chemical simulations like that are large 64-bit double-precision matrix multiplications. So MI50 is a good choice in this particular use case).
Posted on Reply
#6
theoneandonlymrk
dragontamer5788
Okay, as much as I love a COVID19 discussion (and yall know I'm willing to verbally fight on the issue)...





Wait, what? AMD has a cloud-based cluster with EPYC + Radeon Instinct GPUs available for rent? This is news to me.

I mean, I guess I assumed that AMD would have an EPYC + GPU combo available somewhere. But this "Penguin-on-demand" cluster is the first actual name I've seen for the Radeon "Instinct" brand. There's GPUEater, which uses Rx 580 and Rx Vega GPUs. But the Radeon Instinct (aka: MI50) type GPUs have faster communication between the GPUs, which could be useful in some supercomputing instances.

I did check out Penguin HPC briefly: but they only seem to have documentation on Intel clusters: pod.penguincomputing.com/documentation/account-creation.html#about-our-clusters-mt1-mt2.



This does sound like a great system. MI50 is Vega 7nm, almost equivalent to Radeon VII, which was proven to be good at compute (though weak compared to NVidia with regards to deep-learning... its still a very wide SIMD compute system with 1TBps VRAM bandwidth). IIRC, those MI50 GPUs can transfer data between each other at ~180+ GBps speeds, meaning each of those nodes has a significant amount of "traditional" GPU-compute power.
I have disagreed with you a few times but not today sir.
The only post not trolling bs.

And on point, shame others can't be civil about tech.

AMD needs to up it's Pr game, this isn't enough, adds in a tech forum.
Posted on Reply
#7
AnarchoPrimitiv
dragontamer5788
Okay, as much as I love a COVID19 discussion (and yall know I'm willing to verbally fight on the issue)...





Wait, what? AMD has a cloud-based cluster with EPYC + Radeon Instinct GPUs available for rent? This is news to me.

I mean, I guess I assumed that AMD would have an EPYC + GPU combo available somewhere. But this "Penguin-on-demand" cluster is the first actual name I've seen for the Radeon "Instinct" brand. There's GPUEater, which uses Rx 580 and Rx Vega GPUs. But the Radeon Instinct (aka: MI50) type GPUs have faster communication between the GPUs, which could be useful in some supercomputing instances.

I did check out Penguin HPC briefly: but they only seem to have documentation on Intel clusters: pod.penguincomputing.com/documentation/account-creation.html#about-our-clusters-mt1-mt2.



This does sound like a great system. MI50 is Vega 7nm, almost equivalent to Radeon VII, which was proven to be good at compute (though weak compared to NVidia with regards to deep-learning... its still a very wide SIMD compute system with 1TBps VRAM bandwidth). IIRC, those MI50 GPUs can transfer data between each other at ~180+ GBps speeds, meaning each of those nodes has a significant amount of "traditional" GPU-compute power. (Protein folding won't use Tensor Cores: IIRC, chemical simulations like that are large 64-bit double-precision matrix multiplications. So MI50 is a good choice in this particular use case).
Well, if I remember correctly, isn't the El Capitan supercomputer contract AMD won going to be Epyc and Instinct? I know AMD is working overtime on their own heterogeneous interconnect akin to NVLink, though obviously involving CPUs, so there must be some promise to it if it convinced El Capitan and then another supercomputer project just last week if I recall.
Posted on Reply
#8
dragontamer5788
AnarchoPrimitiv
Well, if I remember correctly, isn't the El Capitan supercomputer contract AMD won going to be Epyc and Instinct?
I think El Capitan is pure EPYC (CPU-only cluster). You're probably thinking of Frontier, which is EPYC + Radeon Instinct of some kind.

EDIT: I'm confused. El Capitan will be EPYC + Radeon like you said. I know there was a EPYC-only CPU cluster somewhere, but its hard to remember all these supercomputer names.

I tried to get details on Frontier months ago by emailing the Frontier PR Reps. The details (at least, a few months ago) were still under NDA and they wouldn't give me much info at all.
Posted on Reply
#9
Redwoodz
dragontamer5788
I think El Capitan is pure EPYC (CPU-only cluster). You're probably thinking of Frontier, which is EPYC + Radeon Instinct of some kind.

EDIT: I'm confused. El Capitan will be EPYC + Radeon like you said. I know there was a EPYC-only CPU cluster somewhere, but its hard to remember all these supercomputer names.

I tried to get details on Frontier months ago by emailing the Frontier PR Reps. The details (at least, a few months ago) were still under NDA and they wouldn't give me much info at all.
www.techpowerup.com/271961/gigabyte-northern-data-ag-and-amd-join-forces-to-drive-hpc-mega-project
www.techpowerup.com/271963/covid-19-drives-rise-in-global-fab-equipment-spending-semi-reports
Posted on Reply
#10
mtcn77
I can safely state, more hardware accelerated vaccine development is definitely welcome.
In case you missed, targetted therapies are really strict with their therapeutic requirements. Vaccines are not. I've heard vaccinologists state the response is non-uniform.
It opens up a great opportunity for computer simulated molecular rearrangement studies.
We'd be set for good just by solving the yellow bit where it does mostly good, no harm done.
theoneandonlymrk
AMD needs to up it's Pr game, this isn't enough, adds in a tech forum.
Isn't this, though?
Posted on Reply
#11
theoneandonlymrk
mtcn77
I can safely state, more hardware accelerated vaccine development is definitely welcome.
In case you missed, targetted therapies are really strict with their therapeutic requirements. Vaccines are not. I've heard vaccinologists state the response is non-uniform.
It opens up a great opportunity for computer simulated molecular rearrangement studies.
We'd be set for good just by solving the yellow bit where it does mostly good, no harm done.



Isn't this, though?
They need general population Pr not this.
Posted on Reply
#12
dragontamer5788
theoneandonlymrk
AMD needs to up it's Pr game, this isn't enough, adds in a tech forum.
I dunno if "PR" is what AMD needs.

What AMD needs is cloud GPU-servers available for people to play with. Or another cheap platform (IE: NVidia Jetson Nano) which can provide AMD-Supercomputer "demos". I can right now, rent a server from Amazon Web Services and get a CUDA-app loaded with just a few minutes of playing around.

GPUEater is close, but AMD needs to provide more services like that, and to more popular cloud providers. Especially if they want someone trying out their $2000+ GPUs like MI50.
Posted on Reply
#13
theoneandonlymrk
dragontamer5788
I dunno if "PR" is what AMD needs.

What AMD needs is cloud GPU-servers available for people to play with. Or another cheap platform (IE: NVidia Jetson Nano) which can provide AMD-Supercomputer "demos". I can right now, rent a server from Amazon Web Services and get a CUDA-app loaded with just a few minutes of playing around.

GPUEater is close, but AMD needs to provide more services like that, and to more popular cloud providers. Especially if they want someone trying out their $2000+ GPUs like MI50.
You realise how niche you are though , right?!.
AwS rent epyc servers but never will ,one that runs cuda, if all you want is to run cuda on VMS your niche in a niche is covered.
How many. Millions should they invest?.

Yeh servers for random people to mine on in reality.
Posted on Reply
#14
Alduin
I wonder what Researches are doing with that kind of processing power.
nothing in almost a year
Posted on Reply
Add your own comment