[A51] NVIDIA Tesla C1060 GPGPU

[Cal Leeming [Simplicity Media Ltd]]

http://www.siliconmechanics.com/i20995

Anyone seen this? Looks quite beefy..


-- 

Cal Leeming

Operational Security  Support Team

*Out of Hours: *+44 (07534) 971120 | *Support Tickets: *
supp...@simplicitymedialtd.co.uk
*Fax: *+44 (02476) 578987 | *Email: *cal.leem...@simplicitymedialtd.co.uk
Simplicity Media Ltd. All rights reserved.
Registered company number 7143564
___
A51 mailing list
A51@lists.reflextor.com
http://lists.lists.reflextor.com/cgi-bin/mailman/listinfo/a51

Re: [A51] NVIDIA Tesla C1060 GPGPU

[Frank A. Stevenson]

On Mon, 2010-07-26 at 18:00 +0100, Cal Leeming [Simplicity Media Ltd]
wrote:
 This might be a silly question but.. 
 
 
 If there was a miscalculation due to hardware, would you be able to
 detect such an occurrence, and re-process that calc again?
 
 

One of the big advantages with Tesla is the use of registered memory for
the GPU. This significantly reduces the risk of a bad memory bit
screwing up your calculations, and the hardware can let you know if such
an error exists. ATI doesn't offer this feature on any of their boards,
and that is pretty much the reason for them being absent from the HPC
space, despite having better performance in relation to both price and
power.

f


___
A51 mailing list
A51@lists.reflextor.com
http://lists.lists.reflextor.com/cgi-bin/mailman/listinfo/a51

Re: [A51] NVIDIA Tesla C1060 GPGPU

[Doktor Jeep]

For those of us who have more old boards and CPUs laying about, what are the 
chances of a distributed computing application?

 


 
 Date: Mon, 26 Jul 2010 19:11:54 +0200
 From: sascha.kriss...@web.de
 To: a51@lists.reflextor.com
 Subject: Re: [A51] NVIDIA Tesla C1060 GPGPU
 
 On Mon, Jul 26, 2010 at 06:00:54PM +0100, Cal Leeming [Simplicity Media Ltd] 
 wrote:
  This might be a silly question but..
  
  If there was a miscalculation due to hardware, would you be able to detect
  such an occurrence, and re-process that calc again?
  
 
 There is no obvious way to add some form of checksumming to the algorithm.
 The next best thing would be to compute every chain twice. Which is what
 i do during testing (and also with 2 different algos, one on the GPU and
 a reference on the CPU) and would be feasible if you cared for correctness.
 But then again even with a 1% error rate you would not do this for the
 rainbow tables, because the incorrect chains cannot lead to incorrect
 results and would only waste some space and some time during lookup when
 the lookup code works with them until they turn out not to produce
 a result. Then you would produce 1.01% more chains and live happily ever
 after.
 ___
 A51 mailing list
 A51@lists.reflextor.com
 http://lists.lists.reflextor.com/cgi-bin/mailman/listinfo/a51
  
_
The New Busy is not the too busy. Combine all your e-mail accounts with Hotmail.
http://www.windowslive.com/campaign/thenewbusy?tile=multiaccountocid=PID28326::T:WLMTAGL:ON:WL:en-US:WM_HMP:042010_4___
A51 mailing list
A51@lists.reflextor.com
http://lists.lists.reflextor.com/cgi-bin/mailman/listinfo/a51