[Info-vax] The Future of Server Hardware?

[JohnF]

Stephen Hoffman <seaohveh at hoffmanlabs.invalid> wrote:
> JohnF said:
>> Stephen Hoffman <seaohveh at hoffmanlabs.invalid> wrote:
>>> One size does not fit all.
>>>> JohnF wrote:
>>>> Is there a pretty standard whitebox configuration people typically
>>>> put together as a gpu platform -- MB make and model, power supply
>>>> and cooling options, memory, etc? And is nvidia/cuda the current
>>>> favorite among the gpu/"language" options to put in that box?
>>> 
>>> There are probably almost as many options as there are opinions.
>> 
>> There seem to be several out-of-the-box vendor-assembled solutions,
>> e.g., http://www.nvidia.com/object/personal-supercomputing.html
>> as well as build-your-own recommendations, e.g.,
>> http://www.nvidia.com/object/tesla_build_your_own.html
>> (and plenty of non-nvidia pages about all that, too)...
> 
> Intel Xeon Phi, as well.
>   <http://www.intel.com/content/www/us/en/high-performance-computing/
>    high-performance-xeon-phi-coprocessor-brief.html> 
> I'd expect others.
Thanks for the additional link. Yes, google searches on cuda, and/or
similar terms and phrases, turn up lots of links, some of which
seem pretty good (others, as usual, not so much).

> If you're thinking of basing this on Linux as I might suspect,
Yeah, why not? Seems like the most straightforward approach.

> see which GPU vendor is currently making and keeping the Linux folks 
> happier; who has the better or more open drivers has varied over the 
> last couple of years.
A not-so-veiled plug for nvidia?

> I haven't paid all that much attention to that, 
> but which bunch is binary-blob and which is open-source and 
> particularly which is stable will be as interesting here as the 
> hardware itself.
Yes, I see it's clearly an nvidia plug (whether you know it
or not:). And, yes, the software "framework" in which developers
would have to work is way more important (to me) than the
specific hardware configuration, which seems to be approaching
some kind of "concensus", if not "standard". Have to be able
to migrate the single-threaded app in some straightforward
way, or it's a no-go. Vectorizing some of the matrix-like
calculations seems like the only reasonable approach.

> There's almost certainly a Linux HPC group or forum or resource around.
Yeah, sorry for yet another off-topic, non-vms discussion.
Happy 35th (though for os's, I'm afraid that's about the age
when social security benefits kick in).

> Current OS X includes OpenCL and some clever GCD threading syntax baked 
> into the clang compiler, if you have one of those boxes around.
Thanks, again. Those are great search terms for an alternative
framework. (Though no matter how open OpenCL is, OS X is way too
closed for my liking. Linux gets the job done just fine, without
leading me around like a bull with a ring through its nose.)
-- 
John Forkosh  ( mailto:  j at f.com  where j=john and f=forkosh )