Reference Build: Z8NA-D6 "Computing" Rig

[Jmtyra]

Used this calculator. Using the TPF and Project, it estimates 16,426.72 PPD w/3m11s TPF. Not sure if it's accurate, because this calculator shows 846,634.95 PPD w/the same TPF. Neither seem right, IMHO. Guess I'll have to wait for it to complete a WU to really know.

 

[ihrsetrdr]

The first calculator you link'd to looks correct, it included the K factor. The second calculator is for bigadv units.

 

[Jmtyra]

Hmmm...how can I figure out if the WU is a bigadv one or not?

 

[harlam357]

It's not a bigadv WU... if that's all this platform has to offer then I may be out of the picture. Man, I hate to put it so bluntly. I can't believe that many more threads aren't producing better. I guess this proves that clock speed is king.

Here's a benchmark from my i7 920 main rig (@ 3.8GHz)... same TPF w/a single CPU.

 Project ID: 6077
 Core: GRO-A3
 Credit: 481
 Frames: 100

 Number of Frames Observed: 200

 Min. Time / Frame : 00:03:04 - 17,373.0 PPD
 Avg. Time / Frame : 00:03:11 - 16,426.7 PPD

 

[Jmtyra]

Wow, that's pretty sad indeed. Perhaps I'll have to look at the VMware route instead.
What's the wattage draw for your main rig?

Oh -- and how does one identify a bigadv WU vs a 'normal' one?

 

[HayesK]

psummary currently lists four bigadv wu, p2684, p2685, p2686, p2692, all with 1MM+ atoms, base credit=8955, k=26.4.
http://fah-web.stanford.edu/psummary.html

 

[pik4chu]

wow that is seriously disappointing Looks like Ill be going the sfu route for my builds

Guess its good news in a way, that way I dont have to spend nearly as much on my server. Still would have liked to see some good numbers come from this.

 

[ihrsetrdr]

Hmmm...how can I figure out if the WU is a bigadv one or not?

In order to get a bigadv WU you have to use the "-bigadv" flag; otherwise you'll just get the 'normal' A3 type WUs.

 

[GIXXERGUY6]

You could run 2 clients with the -bigadv -smp 8 - verbosity 9

I noticed your flags didn't have -smp 16 on your single client.

I take it you can't overclock them? Put them on water and clock them to somewhere near 3.5-4.0ish or something.

sfu has the setup though. I can't wait for my time!

p.s. that classified board allows for overclocking!

 

[harlam357]

Good call Gix... you are running -smp 16 are you not Jmtyra?

 

[GIXXERGUY6]

Good call Gix... you are running -smp 16 are you not Jmtyra?

Just grabbed the kill-a-watt and the latest SMP client. -bigadv -smp -verbosity 9 just got started. Pulling 220w AC (external), and that's with 2x H50s + 2x Velociraptors + 3x 1.5TB HDs. So, the ppd/kWh looks very promising.

No he is not According to the above quote he's not even running -smp 8

 

[harlam357]

But... his screen shot of Task Manager also showed all 16 threads pegged out.

I'm still going to go with a more conventional desktop setup for my next box... got the parts mobo/cpu/ram spec'd out on the egg for ~$620. I'm forcing my finger not to hit the trigger.

 

[spitter3]

Yeah. I don't think you need the core count flag in windows. Windows client detects the number of cores you have. only need a flag if you want to folding on less then you have.

 

[GIXXERGUY6]

But... his screen shot of Task Manager also showed all 16 threads pegged out.

I'm still going to go with a more conventional desktop setup for my next box... got the parts mobo/cpu/ram spec'd out on the egg for ~$620. I'm forcing my finger not to hit the trigger.

If you are going with a low cost addition to or a start of a farm we can get that price down some

 

[GIXXERGUY6]

Yeah. I don't think you need the core count flag in windows. Windows client detects the number of cores you have. only need a flag if you want to folding on less then you have.

My understanding was that you need the -smp 8 flag due to the cpu being hyperthreaded instead of physical 8 cores.

I may be wrong as it's happened in the past

 

[GIXXERGUY6]

But... his screen shot of Task Manager also showed all 16 threads pegged out.

Yes but doesn't folding take more ram resources?

I know when I have my folding up and running it takes almost 2.5-4gb out of my 6 for folding.

I see all 16 pegged but he could have taken the SS while running a stress test?


As previously stated many times I could be wrong as it's happened in the past

 

[pik4chu]

easy on the post spam there Gixx lol, just multiqoute next time

Also for the client in windows it is correct as mentioned above, windows properly detects the cores available to it so you dont need a # after -smp unless you want to use less cores. As for memory it can vary. only using a few gigs for 8 logical procs, it doesn't increase too much after the inital group so adding more cores really shouldnt boost the RAM requirement by too much.

 

[harlam357]

Also for the client in windows it is correct as mentioned above, windows properly detects the cores available to it so you dont need a # after -smp unless you want to use less cores.

That must be something new that came with the A3 core... I was totally not aware. The A1/A2 cores were hard coded to 4 core processes (not true threads) and I think you HAD to use -smp 8 (or otherwise) to get more than 4 cores working. I just hadn't changed my practices nor did I ever read where the A3 core was auto-running the maximum available. I know there's a core count check, that data is stored in the queue.dat file, I just didn't know the client was automatically respecting that count now.

OK, all that said to say this... I'm definitely out on this dual socket build. I think I'll be ordering my next i7 within the week.

 

[pik4chu]

That must be something new that came with the A3 core... I was totally not aware. The A1/A2 cores were hard coded to 4 core processes (not true threads) and I think you HAD to use -smp 8 (or otherwise) to get more than 4 cores working. I just hadn't changed my practices nor did I ever read where the A3 core was auto-running the maximum available. I know there's a core count check, that data is stored in the queue.dat file, I just didn't know the client was automatically respecting that count now.

OK, all that said to say this... I'm definitely out on this dual socket build. I think I'll be ordering my next i7 within the week.

I still use the -smp ## flags as well regardless of where or how I setup the client just out of habit as well but I noticed with the recent client when I was reinstalling that it picked up all cores automatically when I left out the flag on accident so I dunno.

And I agree, its sad because I really wanted a nice duallie setup for crunching but looks like im going the I7 farm route after all.

 

[Jmtyra]

Morning folks -- quick update, sorry work/business trip so very little time for anything fun, forums included.

psummary currently lists four bigadv wu, p2684, p2685, p2686, p2692, all with 1MM+ atoms, base credit=8955, k=26.4.
http://fah-web.stanford.edu/psummary.html

Ah, cool, thank you. Had no idea.

wow that is seriously disappointing Looks like Ill be going the sfu route for my builds

Guess its good news in a way, that way I dont have to spend nearly as much on my server. Still would have liked to see some good numbers come from this.

100% agreed. I'm considering RMA'ing the whole damn thing if I can't get the throughput when I get home.

You could run 2 clients with the -bigadv -smp 8 - verbosity 9

I noticed your flags didn't have -smp 16 on your single client.

I take it you can't overclock them? Put them on water and clock them to somewhere near 3.5-4.0ish or something.

sfu has the setup though. I can't wait for my time!

p.s. that classified board allows for overclocking!

Good idea -- I'll have to try that. I did update to -smp 16 remotely (LogMeIn free FTW ) but didn't get time to reply.

This board doesn't allow overclocking. Fail.

Classified board does, true, but omg pricy. Dunno though. Hmmm.

Good call Gix... you are running -smp 16 are you not Jmtyra?

Aye, as of mid sunday-ish (maybe 4pm EST?). PPD is still fail IMHO.