[Beowulf] Large amounts of data to store and process
pbisbal at pppl.gov
Mon Mar 18 08:49:57 PDT 2019
On 3/15/19 9:23 PM, Gerald Henriksen wrote:
> On Fri, 15 Mar 2019 05:28:42 +0000, you wrote:
>> I think what I was getting at is why not include the current HPC practices to every day desktops in the sense since we are reaching certain limits and have to write code to take advantage of more and more cores. Why not use MPI and the like to help distribute the software side of things to the cores?
> I suspect the simple answer is a combination of most of the software
> running on desktops doesn't really need all those extra cores we are
> now getting, and some of the more common desktop applications don't
> really lend themselves to parallel processing anyway.
> It's great that you can go out and get 32 core Threadripper and sit it
> on your desk for an affordable price, but for 90% or more of the
> market at least 28 of those cores would spend most of their life idle.
Not really, the OS can still assign those cores to different
applications, so that each application has it's own core, or each core
is shared amongst a smaller amount of applications. That would still
yield a noticeable speed up to the user, just not single application
performance, which is typically what we concern ourselves with in HPC.
More information about the Beowulf