X2 vs. Opteron hardware questions

[BHawthorne]

I'm in the process of designing a boinc farm that will be based on cheap diskless AM2 based machines. My biggest question atm is L2 cache. The X2 3600+ Brisbane has 2 x 512KB L2, while the Opteron 1210 Santa Ana has 2 x 1MB L2. Taking into account I can get the X2 for $63.99 and the Opteron for $95.99 just now much better is the larger L2 comparing these 2 chips on a properly overclock/tweaked system? Going the X2 would allow me to keep each cruncher to around ~$100-110 USD, while the Opteron would push that price higher. Is the cost increase justified to get the Opteron?

Anyone care to share thier boinc CPU benchmarks for these 2 chips? Thanks

[LeBo]

I am running a 3000+ X2, a FX-60 and an Opteron 180. And from my point of view the thing that matters most is CPU Ghz.
My Opteron is 2.4 Ghz, the FX-60 is 2.6 Ghz and the X2 is 3.0 Ghz. The Opteron is the slowest. IMHO, go with the most Ghz rating you can get, provided it's duel core.
In fact I had a 3800+ X2 OCed to 2.52 Ghz and it would out crunch the Opteron.

[Bender10]

I have an X2 4800 (2 x 512 L2) and a X2 5200 (2 x 1mb L2). Both running at ~2.8 Ghz. Without doing a 'real' benchmark, here are some BOINC benchmarks. Both machines have the same motherboard and 1 gig of the same ram. Identical systems execpt for the cpu.

4800 ~2.8 Ghz:

2685.32 float pt.
4785.53 int speed


5200 ~2.8 Ghz

2685.31 float pt.
4753.48 int speed

The BOINC benchmarks don't seem too stable. But these are both from the same project site.

[BlackAdder]

I would have thought L2 size would matter most...since Opterons have 1 meg each core don't they spank X2's with a bit more speed but only 512 of L2 ??

[NeoGen]

It depends on the software being run on it. L2 cache size is only as good as the number of times the software can make use of it.
If a given program running on it makes many different or random operations in a row, there are few to no cache hits and you get almost no performance benefit from having either 512kb, 1Mb or any other value. (In which case, higher GHz equals higher performance)

[NeoGen]

One example that just crossed my mind is table based math operations. If you have in your application a precalculated table of sine and cosine values, and you access those values every time you need instead of asking the processor to calculate, you get alot of benefit from the cache, because those static values will be stored there.
Pre-calculated values may have less precision than cpu made calculations, but that's a trade-off that software developers always have to be aware of. If the extra precision is not really necessary why losing time with deep calculations?

[drezha]

Neogen got to it first.

It depends on the project. IIRC, Folding at home has various work units that benefit massively from the added cache.

http://fahinfo.org/ can help you out there.

Possibly it's not worth it on a BOINC based project as what people have said above doesn't show a large difference.

[Steve Lux]

Another difference between the x2's and the Opterons: I exchanged my 3800+ X2 for a 185 Opty and quadrupled the number of crawlers I could open for Majestic-12. Increased from 20 crawlers to 80. The Opty seems to be better designed for communications. There wasn't any improvement in the crunching output though - at least in the projects I run. Actually I'm crunching a bit slower, but that may be because Majestic-12 is now taking up so much more processor resources than before.