Cambridge-1 Supercomputer is nearly ready for service

[HEXUS]

Nvidia DGX SuperPOD architecture system will deliver 400 petaflops of AI performance.

Read more.

[fend_oblivion]

Hope they are able to find answers and bring about awesome developments soon!

[ik9000]

and cue the "will it run crysis" posts. But what about its mining capability. could they stick it on that for a month or two to recoup the build costs?

[lumireleon]

Calling it most powerful supercomputer yet its GPU only is pretty MISLEADING.

[DanceswithUnix]

could they stick it on that for a month or two to recoup the build costs?

Having such a powerful supercomputer enabling people to play cryptokitties would say a lot about the modern age I think

[kalniel]

Calling it most powerful supercomputer yet its GPU only is pretty MISLEADING.

Please do give us the official definition of a (super) computer then and why this doesn't meet it?

[lumireleon]

WIKI....overall applicability of GPGPUs in general-purpose high-performance computing applications has been the subject of debate, in that while a GPGPU may be tuned to score well on specific benchmarks, its overall applicability to everyday algorithms may be limited unless significant effort is spent to tune the application to it.

[ik9000]

WIKI....overall applicability of GPGPUs in general-purpose high-performance computing applications has been the subject of debate, in that while a GPGPU may be tuned to score well on specific benchmarks, its overall applicability to everyday algorithms may be limited unless significant effort is spent to tune the application to it.

That's nice. Now in your own words please?

[kalniel]

I don't mind someone quoting wiki (though one should probably link/attribute it correctly), but it doesn't answer my question about what the definition of a (super) computer is and why this doesn't meet it.

Since wiki is a favourite source, here's the wiki article about super computers:
https://en.wikipedia.org/wiki/Supercomputer

Which defines it as

A supercomputer is a computer with a high level of performance as compared to a general-purpose computer.

In which case, this appears to meet the criteria.

[philehidiot]

A dragster isn't a car because I can't use it on normal roads.

Supercomputers are built, like any other performance machine, to optimise certain areas at the expense of others. Some workloads will absolutely run terribly on a GPU only machine. But then the fault is yours for running it on the wrong hardware. Every workload will need optimising to run on a specific supercomputer as they are built differently.

Some workloads will run, on a single GPU, several thousands times faster than on a single CPU. Some will run badly. The onus is on the people using the data to select an appropriate supercomputer for their needs and optimise their workloads accordingly.

Optimising CUDA, etc is a very specialist and lucrative game for precisely this reason.

If you build a supercomputer with the aim of it being awesome at everything, you'll build an expensive dog that no one wants to rent. Engineering is compromise at every level.

[tygrus]

It's so good at AI it can play Crysis for you while you watch.