[Beowulf] Re: Opteron 275 performance

[Joe Landman]

Robert G. Brown wrote:
> Steve Cousins writes:
> 

[...]

>> Hi Joe,
>>
>> Thanks a lot.  I just took a look and it seems to make a good case for
>> getting the Dual Dual Core machine. 
>> I'm fairly certain that the memory latency issue that Vincent was warning
>> about won't be an issue, although I'm a bit clueless about how to know 
>> for
>> sure.  How would I go about finding out if our model is TLB trashing main
>> memory?  I feel like I just bit the hook... I don't want to start a huge
>> discussion on this but if there are some quick tell-tale signs of it I'd
>> be interested to find out.

There are some good analysis tools out there you can use for setting up 
and watching various processor counters.  Have a look at 
http://user.it.uu.se/~mikpe/linux/perfctr/ and the announcements 
http://user.it.uu.se/~mikpe/linux/perfctr/current/ANNOUNCE-2.6.15 . 
There are numerous others such as http://icl.cs.utk.edu/papi/ . 
Oprofile is focused upon a slightly different performance measurement.

That said, the only thing that really will give you convincing data will 
be running your code, and seeing if it has performance anomolies.  Its 
when you have those anomolies that you need to start looking at why you 
are having the anomolies.  Thats when it makes sense to start looking at 
Perfctr etc.

> Why bother with tell-tale signs?  Like I said, your previous post was
> dead on the money.  Get a loaner (which can physically be far far away
> and should be "free"), install YOUR application and run the only
> benchmark or test that matters.

Agreed.

> On paper, the memory access schemes used by the Opterons should largely
> ameliorate the kind of difficulty encountered with the dual PIII's --
> they ought to do better than just divide single processor bandwidth
> between two processors at any rate.  You can visit the hypertransport
> site and look at white papers, e.g. --

I am of the opinion that the proof is always in the pudding as it were. 
  That is, regardless of the white papers say (or benchmarks say), your 
code is going to beat on the processor in its preferred manner, which 
may or may not mesh with what is in the benchmark, whitepaper, etc.

-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 734 786 8452
cell : +1 734 612 4615