Is anyone here using an FX CPU for distributed computing (e.g. Rosetta)

I am considering an upgrade from a Thuban (x6) to FX8120 (x8) but there are a lot of posts claiming a lesser performance at the same clock speed for the new platform. Therefore I would be interested in other people's experience running SETI or Rosetta or some other distributed project.

I am not running distributed computing. However, for example on a multithreaded benchmark like wp1024m - I can run a 1090T at around 5.7GHz and finish the benchmark in about 134s. In order to do roughly the same time on an fx-8120 or fx-8150, I need to run over 7GHz.

Most people claim its a bad idea to use these for DC, because under full load they suck a lot of power as well. You are more efficient on the Thuban than you would be on a BD, though depending on your cooling setup - if you can get your clocks up higher than on Thuban, you might get better performance.

And a free bump, I think your post got lost in the holiday.

[rant]
I have mixed feelings here, at certain settings a 1090T will stomp a BD in terms of effiency (default is a good example). At stock voltage (1.30v) my BD (8120) will overclock to 4.0Ghz+, where my 1090T requires 1.4625v...a definite loss in terms of points/watt for Thuban when it is overclocked. BD then looses when pushed beyond that trying to catch the Thuban because the BD requires higher clock speed for the same amount of work (clock for clock X6>BD). The reason I believe BD fails so bad is because the default voltage is far too high, both the Core and the IMC on my BD can be massively undervolted and remain stable where my Thuban can be undervolted at stock speed but only slightly (one bump down for both the Core and IMC voltage). Also look at turbo of BD, on mine it jumps to 1.400v for 4.0Ghz, it can easily do that on 1.30v (default) and pass a battery of Intelburntest. Perhaps it's the poor yields that resulted in such bad voltage selections I don't know but it is bad on mine and makes BD look a lot worse than it is.

To sum it up my BD does more work (edit: in folding at home, rosetta is most likely the same) then my Thuban does at ~225W, rest of the system held constant.[/rant]

Yes I'm an AMD fanboy, and no I'm not saying this for that reason alone, there are plenty of BD's being overclocked by~1.0Ghz on stock voltage and stability tested.

If you want to help AMD out a bit buy BD or Thuban. If you're after points/watt and/or just raw points buy an i7 2500K or 2600K, they overclock like crazy and use less power and generally do more work clock for clock.

Clear opinion, and pretty well said for what you were trying to get across. Some actual DC experience helps.

If I have some time I'll post some efficiency results of various distrubuted computing programs (SETI, Rosetta and FAH) to compare points/watt at varying frequencies. Neither my 8120 or my 1090T are "golden chips" from results I've seen so it should be a good comparison of an average BD to Thuban.

Well initial results for Folding at Home using Win 7 x64 don't look good. Project 8004 looks amazing @ ~24,000ppd @ 4.0Ghz.
Project 7611 on the other hand is amazingly bad @ ~6,500ppd. What is even more dumbfounding is the tpf/core count. At X8 tpf is ~8:40, at X6 tpf is ~9:05, in other words 2 "cores" (1 Compute Unit) yields a ~4% increase in speed on that project (7611)...I think it maybe the scheduler problem that is to be addressed in the coming months. I will continue to look into this, 6,500ppd @ ~225W .

The scheduler problem already has MS patches out, I haven't seen good independent tests ran for before and after the patch to summarize its impact, but it's not making waves.

EDIT: Nevermind, they are saying Q1 2012 now I see, and that the previous patch was only a partial fix.

I posted a new thread in the Rosetta section.

I.M.O.G andPsykoikonov - thank you for your very insightful responses. I was looking to replace my T1100 x6 with an FX 8120X8 for Rosetta, but the feedback I read here and in other places makes me think that the effort may be futile - too many posts that report per core performances at the same clock speed to be 10-30% lower - depending on benchmark, so that even having the extra cores, and even if overclocking on water (to say an optimistic 4.7GHZ) would not provide a significantly better overall output. OTOH I like Psykoikonov's observation regarding the lowering of voltages. This may come in handy for a "high performance quiet system", which is another project I am looking into

Yomama said:

This may come in handy for a "high performance quiet system", which is another project I am looking into

Click to expand...

Lol, that was my goal for many years. I just found it too hard to keep a system stable whilst being quiet without spending a bucketload of money.

In the end I have different computers for different purposes. An old Mac Pro for my HTPC, an old Core2Duo notebook running Linux Mint for my web surfing/download PC and my gaming rig is the one in my signiture.