The US government is auctioning the Cheyenne supercomputer: 145,152 CPU cores and 313 TB of RAM


TL;DR: It is not each day that one will get the risk to buy a world-class supercomputer, however that is precisely the chance afforded to bidders. GSA Auctions, a central authority surplus public sale operator, is accepting bids for the Cheyenne supercomputer. The gadget used to be built by means of Silicon Graphics (SGI). It’s an SGI ICE XA gadget consisting of four,032 dual-socket gadgets configured as quad-node blades.

Cheyenne is pushed by means of 8,064 Intel Xeon E5-2697v4 processors. Every Broadwell-class chip is constructed on a 14nm procedure and contours 18 cores / 36 threads with a base frequency of two.3 GHz that may Turbo Spice up as much as 3.6 GHz. In general, that works out to a staggering 145,152 processing cores.

The gadget makes use of DDR4-2400 ECC single-rank reminiscence, and numerous it. In general, the brand new proprietor gets a gadget with 313,344 GB of reminiscence.

The auction listing mentions 14 E-Mobile gadgets, every weighing 1,500 kilos. Additionally referenced are 28 E-Rack gadgets, all of that are watercooled. Significantly, fiber optic and CAT5/6 cabling is excluded from the public sale, so you can wish to provide your personal.

Bidding began on April 26 with the public sale scheduled to finish on Would possibly 3 in a while after 6 p.m. Central Time. As of this writing (with lower than 3 days closing), 13 bids were positioned, and the present top bid sits at $28,085. Of observe is the truth that the reserve has no longer but been met.

Cheyenne went to paintings in January 2017 and used to be taken offline on the finish of final yr. At one time, it used to be ranked as the twentieth quickest supercomputer on the earth. As of final November, the Top500 listing ranked it at quantity 160.

events will have to learn throughout the complete public sale checklist prior to filing a bid. The gadget is obtainable as-is in its present situation, and is experiencing “upkeep boundaries” because of some misguided fast disconnects. What is extra, about one % of nodes skilled failure all over the supercomputer’s final six months of operation, that are most commonly attributed to DIMMs with ECC mistakes.

Be the first to comment

Leave a Reply

Your email address will not be published.


*