[Beowulf] El Reg: AMD reveals potent parallel processing breakthrough

[Arnold, Galen Wesley]

1.  The K20X isn't necessarily as hard to program as one might imagine.  OpenACC codes take to it well , and both Cray and PGI are making nice progress with OpenACC performance compared to CUDA.  Programming it to get 85% of peak is hard, but that can be said for every architecture--GPU or not.


Galen Arnold
NCSA - systems engineer

________________________________________
From: beowulf-bounces at beowulf.org [beowulf-bounces at beowulf.org] on behalf of Vincent Diepeveen [diep at xs4all.nl]
Sent: Saturday, May 11, 2013 3:56 AM
To: Lux, Jim (337C)
Cc: beowulf at beowulf.org
Subject: Re: [Beowulf] El Reg: AMD reveals potent parallel processing   breakthrough

On May 10, 2013, at 6:04 AM, Lux, Jim (337C) wrote:

>
> On 5/8/13 6:41 PM, "Prentice Bisbal" <prentice.bisbal at rutgers.edu>
> wrote:
>
>> On 05/08/2013 09:41 AM, Lux, Jim (337C) wrote:
>>> The game console business is a strange one, and I don't know that
>>> it has
>>> much to bring to the HPC world (whoa, that will provoke some
>>> comment).
>>
>> Roadrunner's body isn't even cold yet, and everyone's already
>> forgotten
>> about it. :(
>>
>> http://en.wikipedia.org/wiki/IBM_Roadrunner
>> http://en.wikipedia.org/wiki/Cell_microprocessor
>>
>>
>
> I think roadrunner is an example of a one-off stunt..
> In the long run, "easy programming" is probably a bigger cost driver.
>

The top500.org of today completely refutes your statement there.

november 2012 list http://www.top500.org/list/2012/11/

number 1: cray with gemini interconnect and K20x.

That's not *easy* to program. Interconnect nor CUDA.

number 2:

BlueGene/Q

also not 'easy' to program for with those dead slow latencies it has.

number 3:

sparc64 VIIIfx 2.0Ghz with 8 cores

Yet again it's not standard x64 cpu's. It's Fujitsu cpu's.

Also heavily focussing upon vectorization. ILP type vectorization
simply is *not* easy.

number 6:

possibly the only relative 'easy' to program in the entire top10. As
it has seemingly Mellanox FDR with Xeon E5 cpu's.

Because of their interconnects, supercomputers by definition are not
*easy* to program.
In the long run always the struggle is there to get the maximum out
of the cheapest processing power.

That simply is not 'easy programming'.

This is both true in HPC as well as the embedded world.

>
>>
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin
> Computing
> To change your subscription (digest mode or unsubscribe) visit
> http://www.beowulf.org/mailman/listinfo/beowulf

_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf