Unboxing Canada's BIGGEST Supercomputer

[SuicideRabbit]

4 hours ago, AnonymousGuy said:

Is it UV sterilized and particulate filtered or is there some sort of biocide in it?  Is it straight glycol+RO water or did you add a scale inhibitor as well?

 

I ask for details because I have a 25 gallon loop in my house with a lot of similarities to how datacenters do cooling, and I want to know the "proper" implementation.  

No, it isn't UV sterilized, and I'd be lying if I knew all of the exact components of the mixture; we left those details to the plumbing company which installed the cooling system, but afaik, yes, there are both a biocide and a scale inhibitor in it. Unfortunately, I can't help with any specific implementation in the context of your setup. Also, to help with both weight and rust, all of the piping is a special ABS pipe imported from Germany - very esoteric stuff. After the plumbing company installed the system, they basically said they would never work with that piping again, it was so difficult. 

[Orical]

I use to build similar systems at EMC in a sub division called VCE that built VBlocks as custom builds and yes they get insanely expensive and a pain to configure. Sadly Dell bought them out and I had to move on, enjoy the pics and being around these beasts is amazing and the performance of even a 4 cab system is just incredible. 

[danwat1234]

On 7/3/2017 at 9:19 AM, ggbaker said:

The facility is available to any Canadian researcher who needs high-performance computing. In general, it wouldn't be used as single monolithic supercomputer: many jobs would be running scattered across the cluster. The job that needs 3TB of RAM on one box can be running there, while the highly-parallel job is running on 1000 cores on the next rack.

 

Apparently, they expect the facility to be utilized at 100%. If there are any unallocated cycles, the particle physicists will snap them up: some part of the LHC project uses the Compute Canada facilities and they always need more compute power.

LHC as in similar work to the LHC Distributed Computing project run thru the BOINC manager? https://lhcathome.cern.ch/lhcathome/

Suppose it would be large data set work only and your server would not be seen as a user in the top participants list.

 

 

11:00 GPUs being used for running AI in 2017 back before the GPU crypto mining craze, and now AI is front and center in GPU farms!

 

What happens to your old hardware when you need to make room for new compute?