[Beowulf] General thoughts on Xeon 56xx versus E5 series?

[Vincent Diepeveen]

AVX roughly 25% faster for the fastest prime number transforms as  
used by GIMPS.
That is boost by just AVX over the previous SSE 4.2 and similar codes.

Code is still in beta.

See for more details postings there www.mersenneforum.org

However it is unclear how much of that can be attributed to improved  
transform optimizations overall and
faster multiplication instruction on the newer i7's.

These codes are so optimal that you feel the throughput constraints  
of the multiplication instruction.

Note that this library is floating point. The other library used a  
lot is integer code. It's overall slower than the floating point,
yet it's advantage is that it is lossless, so objectively it's a lot  
better.

For the biggest transforms that use big clusters, it is easier to use  
integer transforms.

The AVX floating point runs on what can be run at a single machine.




On Sep 14, 2012, at 1:00 PM, Igor Kozin wrote:

> if memory bandwidth is your concern then there are models which boost
> it quite significantly. e.g.
> http://ark.intel.com/products/64584/Intel-Xeon-Processor- 
> E5-2660-20M-Cache-2_20-GHz-8_00-GTs-Intel-QPI
> probably very few codes are going to benefit from AVX without extra
> efforts but BW is a clear win.
> i'm seeing a good speed up on some applications which can be
> attributed to higher BW.
>
> On 13 September 2012 15:48, Orion Poplawski <orion at cora.nwra.com>  
> wrote:
>> On 09/13/2012 08:09 AM, holway at th.physik.uni-frankfurt.de wrote:
>>>> Any general thought out there on Xeon 56xx versus E5 performance  
>>>> wise?
>>>
>>> We were comparing the performance of an older AMD with a newer  
>>> AMD a while
>>> ago and found that the newer AMDs floating point performance was
>>> significantly reduced.
>>>
>>> It turned out that the kernel did not yet support AVX or FMA4 or  
>>> some
>>> other new fancy bit of register blah blah so it reverted to some
>>> compatibility support for floating point operations. (I am not a  
>>> kernel
>>> dev :)
>>>
>>> If your going to play with newer processors you have to make sure  
>>> your
>>> kernel supports all the functions of that CPU. Kernel development  
>>> can be a
>>> little way behind on mainstream releases.
>>>
>>> Ta,
>>>
>>> Andrew
>>
>> Thanks for the reminder of kernel support.  RedHat states that  
>> RHEL5.7 should
>> support the E5 processors, so hopefully that means supporting the new
>> instructions.
>>
>> It seems the question before us is core scaling versus raw MHz.   
>> I've seen
>> many codes saturate the memory bandwidth before filling all of the  
>> cores and
>> it seems you have a trade off between MHz and cores these days at  
>> a given
>> price point.  But perhaps that doesn't even matter if you're  
>> memory bandwidth
>> limited anyways.
>>
>>
>> --
>> Orion Poplawski
>> Technical Manager                     303-415-9701 x222
>> NWRA, Boulder Office                  FAX: 303-415-9702
>> 3380 Mitchell Lane                       orion at nwra.com
>> Boulder, CO 80301                   http://www.nwra.com
>> _______________________________________________
>> 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