AMD Opteron 4P -- SETI@Home Benchmarking
SETI@Home Gets its Chance
This content was originally featured on Amdmb.com and has been converted to PC Perspective's website. Some color changes and flaws may appear.Introduction
Part of AMDMB’s ongoing look at a 4-way AMD Opteron system, we have taken a look at how well it performs using the SETI@Home Distributed Computing program. While, at first glance, the numbers do not look good, there are several reasons why they are not very “solid” either.
Using the benchmark work unit setup by the AMDMB SETI@Home team (http://seti.amdmbpond.com), two sets of tests were run, giving us a total of 5 work units processed. The work unit was selected last year and has an angle range of .417 which is the average of all work units sent out to computers by the SETI@Home project. Here are the vitals of the work unit used:
The benchmarks were completed with little or no other tasks being performed on the computer. Two sets of tests were done, the first to test the single processor, single work unit performance of the system, the second to test the 4-processor, 4-work unit performance of the system.
Be sure to check out Ryan’s First Look at the 4-way system for its vitals.
Single Processor, Single Work Unit
The SETI@Home client is not designed to run on multiple processors, hence the reason for two sets of tests. In the single processor, single work unit test, the computer completed the work unit in approximately 2.217 hours (2:13) as noted with SetiSpy by Roelof J. Engelbrecht shown in use below. This puts the 1.8GHz Opteron somewhere in the range of a Thoroughbred based AthlonXP running at 2.3GHz as see in the AMDMB SETI@Home Benchmark Database.
4 Processor, 4 Work Unit
Running four separate instances of the client at one time with each assigned to it’s own processor gives us the chance to really stretch the system as a whole. Each process will eat up as much RAM bandwidth as you can throw at it, but the numbers were actually surprising when compared to the Single Processor, Single Work Unit test. Most would expect that performance would suffer quite a bit because while each SETI@Home client has it’s own processor, they still have to share other system resources, with the Opteron based system, the performance hit was almost nil. Here are the times for each of the processors:
CPU0 – 2.2110 hours (2:13)
CPU1 – 2.2654 hours (2:16)
CPU2 – 2.2145 hours (2:13)
CPU3 – 2.2121 hours (2:13)
As you can see, the performance hit was less than a minute on all but one processor. At this point, we are thinking that CPU1 has a head cold and decided to take it’s time. Keep in mind that all of the clients were using the exact same work unit file and were therefore, doing the same exact work.
So how does all this stack up against the Itanium 2 from Intel? Well, at first look, it seems the Opteron has a long way to go. Of course, if you have been reading the news lately, you would know that’s not really true. The Opteron is fairing quite well in many tests (and even really well in some). If you looked at the Benchmark Database earlier, you may have noticed two things: 1) the Opteron times aren’t listed – not yet anyway, and 2) the fastest processor listed is a 900MHz Itanium 2.
There are two major reasons to note the roughly 40 minute difference in processing times – all of them bode well for the Opteron.
- The Opteron test system is running Microsoft Windows 2003 Server – 32bit edition, meaning it’s not taking advantage of the full potential of the processor(s). On the Itanium 2 system, HP-UX 11.22 was being used, a 64bit iteration.
- The SETI@Home client used was a 32bit client, again, not taking advantage of the full potential of the processor(s). Unfortunately, there is no SETI@Home client available for the AMD-64 architecture yet (but, try running the 32bit client on the Itanium and see what hairball it coughs up).
If we are fortunate enough to receive a 64-bit client from the folks at SETI@HOME in the future, we will be sure to run through these tests again as I am sure it will make a huge difference.
So what does all of this tell us? Mostly, it tells us that the 32bit performance of the Opteron is on par with a much faster version (about 500MHz) of it’s cousin, the Thoroughbred core AthlonXP. It also tells us that the architecture AMD has built can sustain high loads on multiple processors without dragging down the overall performance.
Stay tuned for more of our look at the 4-way Opteron.