Tuesday, February 18th 2020

AMD Gets Design Win in Cray Shasta Supercomputer for US Navy DSRC With 290,304 EPYC Cores

AMD has scored yet another design win for usage of its high-performance EPYC processors in the Cray Shasta supercomputer. The Cray Shasta will be deployed in the US Navy's Department of Defense Supercomputing Resource Center (DSRC) as part of the High Performance Computing Modernization Program. The peak theoretical computing capability of 12.8 PetaFLOPS, or 12.8 quadrillion floating point operations per second supercomputer will be built with 290,304 AMD EPYC (Rome) processor cores and 112 NVIDIA Volta V100 General-Purpose Graphics Processing Units (GPGPUs). The system will also feature 590 total terabytes (TB) of memory and 14 petabytes (PB) of usable storage, including 1 PB of NVMe-based solid state storage. Cray's Slingshot network will make sure all those components talk to each other at a rate of 200 Gigabits per second.

Navy DSRC supercomputers support climate, weather, and ocean modeling by NMOC, which assists U.S. Navy meteorologists and oceanographers in predicting environmental conditions that may affect the Navy fleet. Among other scientific endeavors, the new supercomputer will be used to enhance weather forecasting models; ultimately, this improves the accuracy of hurricane intensity and track forecasts. The system is expected to be online by early fiscal year 2021.
Source: HPC Wire
Add your own comment

30 Comments on AMD Gets Design Win in Cray Shasta Supercomputer for US Navy DSRC With 290,304 EPYC Cores

#26
DeathtoGnomes
R0H1TThe thing with the 56 core Xeon is though that they do not come anywhere close to the AMD chips in terms of efficiency. Besides Intel's probably making these parts only for demonstration because I see no price listed & of course the yields would be horrific, even after accounting for the fact that they're 2 Xeon platinums glued together.
On a chip vs chip basis efficiency might be in Intels favor, but not in this instance. I'd say the side that uses 621 less CPU chips is more efficient over all.

Be careful throwing that glue comment around, your Fanboism is showing. :p :rolleyes:
Posted on Reply
#27
Cheeseball
Not a Potato
renz496if nvidia is one step ahead with hardware with software nvidia is 10 step ahead of AMD. and it seems in professional world the big player are moving away from openCL and try to push their own version of "open standard". apple with metal. intel with one API.
Technically those are all open. You can use any GPGPU with the Metal, oneAPI, ROCm, CUDA (Radeons work through transpiling) and pure OpenCL frameworks.
Posted on Reply
#28
ratirt
renz496if nvidia is one step ahead with hardware with software nvidia is 10 step ahead of AMD. and it seems in professional world the big player are moving away from openCL and try to push their own version of "open standard". apple with metal. intel with one API.
10 steps ahead? I wouldn't say so. Besides Intel was ahead too and look now. There is nothing certain and I wouldn't be so sure if I were You.
Posted on Reply
#29
INSTG8R
Vanguard Beta Tester
Epyc!
Posted on Reply
#30
Vikutorusama
12.8 Peta FLOPS Is not much compared with the Atlas 900 from Huawei with 256 to 1024 Peta FLOPS
Posted on Reply
Add your own comment
Dec 18th, 2024 07:48 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts