Image: Cray

One of the USDOD’s top computing providers is due for a massive upgrade in computational performance.

The Navy Department of Defense Supercomputing Resource Center (DSRC) has announced that it is getting a multi-million dollar Cray Shasta supercomputer capable of a healthy 12.8 petaFLOPS. That’s 12.8 quadrillion floating point operations per second.

Making that possible is “290,304 AMD EPYC (Rome) processor cores and 112 NVIDIA Volta V100 General-Purpose Graphics Processing Units (GPGPUs), interconnected by a 200 gigabit per second Cray Slingshot network.” It also features “590 total terabytes (TB) of memory and 14 petabytes (PB) of usable storage, including 1 PB of NVMe-based solid state storage.”

What this means is that the machine can run Crysis, but the Navy will be flexing its power for more important applications such as military technology and informational systems.

Image: Navy DSRC

“The investment and increase in supercomputing power at the Navy DSRC at Stennis Space Center is absolutely critical to Naval Oceanography delivering future capability upgrades to global and regional ocean and atmospheric prediction systems, to include later this year the Navy’s first Earth Systems Prediction Capability,” said Commander, Navy Meteorology and Oceanography Command (NMOC) Rear Adm. John Okon.

“Naval Oceanography’s ability to be the Department of Defense’s authoritative source for characterizing and applying data of the physical battlespace into a decisive advantage for naval, joint and allied forces hinges on the continual upgrade and advancements in high-performance computing from the High Performance Computing Modernization Program.”

The Navy DSRC’s system should be up and running by early fiscal year 2021. It’s expected to rank in the top 25 of the most powerful supercomputers on the planet.

According to Top 500, IBM’s Summit and Sierra machines hold the top spots. These are capable of 148.6 and 94.6 petaflops of performance, respectively.

Join the Conversation

7 Comments

  1. Keep in mind that a human brain is theoretically capable of 100 teraflops of calculations.. This system is on the order of 190 people… Or as smart as two of me. And they are putting it in charge of battle field predictions a and potentially command. Skynet?
  2. Keep in mind that a human brain is theoretically capable of 100 teraflops of calculations.. This system is on the order of 190 people… Or as smart as two of me. And they are putting it in charge of battle field predictions a and potentially command. Skynet?

    It can probably play Crysis.

  3. And I thought 64 cores was a lot…

    Good god Imagine how expensive it would be to license SQL on that scale…

    Lets see.. lets be REALLY nice and presume an INSANE contract.. for 10k per 2 cores..

    HAHA… HAHAHAHAHAHAHHAA,,,, Weeeze… One BILLION, four hundred and fifty one MILLION, five hundred and twenty thousand dollars..

    HAHAHAHAHAHHAA!!

    Wheeeze…

    Sorry that’s just TOO damn funny.

    Of course if they used ALL 64 core epyc CPU’s… they would need 4536 of them. At 7522.00 each that purchase was 34,119,792 dollars. Presuming a 60% discount on price.. we’re looking at 13,647,916.80.

    For memory they are running about 130GB per 64 core socket so memory they can save some money on. Lets say they want that 8 channel goodness.. and lets call it 128gb of ram so that would be 8 16gb sticks per socket. And ECC ram. Middle of the road we’re looking at 100 bucks a stick or 800 per socket. So 3,628,800 but after that 60 percent discount… 1,451,520 bucks.

    Now lets look at those video cards…

    There are a lot of models so lets say 10k per card. So easy math on this one 1,120,000 then with that 60 percent discount 448,000 dollars.

    Before Chassis, power supply and board not to mention custom interconnects.. I think it’s fair to say you’re looking at a 500 million dollar system all told.

    Even then you could buy THREE of these buildings with hardware for the base cost of the SQL license. LOL.

Leave a comment