SMP2 issues

Moderators: Site Moderators, PandeGroup

Re: SMP2 issues

Postby bruce » Fri Mar 05, 2010 9:15 pm

zygrene wrote:The way I see it, even if someone else is going to end up with the a1 I dump, chances are they don't have the same issue I'm having. I wouldn't mind running the a1's if it didn't require me to exit the GPU client. Also, how would it be bad for science if all the work units are going to get assigned one way or another?


That's pure rationalization.

FahCore_a1 does not require you to exit the GPU client. They work quite well together. It's FahCore_a3 core that suffers significantly when it's interrupted. You were already running FahCore_a1 plus a GPU client plus one or more classic clients. How has any of that changed (except maybe for the env vars)?

It's bad for science in that it delays the results. The scientific value of the results is related to both the successful completion of WUs and to the promptness with which they're returned. The new bonus plan explicitly makes that clear by granting a bigger bonus for more rapid returns and by explicitly disqualifying you if you can't maintain the 80% completion rate.

The flaw in the old points system was that it didn't encourage prompt returns in accordance with the scientific needs. Nobody can stop you from exploiting that flaw, but it's strongly discouraged as others have already said.
bruce
 
Posts: 21270
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: SMP2 issues

Postby zygrene » Fri Mar 05, 2010 9:28 pm

Alright, I won't argue with you. I wasn't aware of the 80% completion rate thing nor the fact that dumping a WU to someone else would significantly reduce responsiveness. I'm going to set up Environment Variables and change -smp 3 back to -smp and see what happens.
zygrene
 
Posts: 11
Joined: Fri Mar 05, 2010 4:19 pm

Re: SMP2 issues

Postby zygrene » Fri Mar 05, 2010 9:44 pm

Ok so I set up the variables with these values:

BROOK_YIELD ................2
CAL_NO_FLUSH ..............1
CAL_PRE_FLUSH ............1
FLUSH_INTERVAL ...........128

Now the GPU client is using 6-8% CPU while SMP retains the rest. Are these values optimal?

EDIT: Also, I just found a major discrepancy in reported PPD between Fahmon and HFM-NET. Fahmon reports 850ppd for the 484-point a3 unit while HFM reports over 3800ppd. It takes 825 minutes to complete this unit (8.25 min per 1%). I did a calculation by hand and found that Fahmon is correct: (24 hrs/day) * (60 min/hr) * (484 pts/unit) / (825 min/unit) = 845 ppd.

Does HFM report PPD differently than Fahmon?
zygrene
 
Posts: 11
Joined: Fri Mar 05, 2010 4:19 pm

Re: SMP2 issues

Postby ChelseaOilman » Fri Mar 05, 2010 10:19 pm

zygrene wrote:Does HFM report PPD differently than Fahmon?

FahMon doesn't have the capability to calculate bonus points. It's showing you your PPD based on the base points of A3 WUs. HFM.NET can calculate the bonus points of A3 WUs. That's an option than can be turned on or off in the preferences of HFM.NET. Just because HFM.NET can calculate bonus points for the A3 WUs doesn't mean it knows if your qualified to receive them yet.
User avatar
ChelseaOilman
 
Posts: 1742
Joined: Sun Dec 02, 2007 3:47 pm
Location: Colorado @ 10,000 feet

Re: SMP2 issues

Postby ChasR » Sat Mar 06, 2010 12:41 am

Increase FLUSH_INTERVAL to 192. CPU usage should be below 5%. It might work at a slightly higher value and lower cpu utilization a bit more, but if you go too high you'll encounter vpu recovers.
Image
User avatar
ChasR
 
Posts: 759
Joined: Sun Dec 02, 2007 5:36 am
Location: Atlanta, GA

Re: SMP2 issues

Postby zygrene » Sun Mar 07, 2010 4:42 am

Ok, so right now I'm getting just over 1000ppd with a Project 6012 A3 unit, not including bonus ppd. I'm running the GPU client at the same time but it's only using around 6% of my CPU. Does this sound right? I'm averaging 6.5-7 minutes per 1%.
zygrene
 
Posts: 11
Joined: Fri Mar 05, 2010 4:19 pm

Re: SMP2 issues

Postby P5-133XL » Sun Mar 07, 2010 5:37 am

I came to the conclusion that with my two 3870's and an OC'ed Q6600@3.42 that I actually gained more total PPD by just SMP2 folding with A3's. I lost more PPD from the delay in the SMP2 client than I was getting for my two video cards. I make the assumption that the PPD value closely correlates to the PPD so I retired both my 3870's from folding. Not only do I get more points, the scientific value increases, The lag from GPU2 folding disappears, and I get to decrease my power costs and the wear and tear on the video cards.

You really need to include the bonus when making these types of judgments. The bonus may be 5-6x the base point value. HFM.net can be easily configured to produce PPD including the bonus.

Different CPU's at different clock rates; different Video cards; different OS's will all change the PPD's that you get. So my decision may not apply to all. Each person needs to test their setup independantly and come to the decision that works bets for them.
Last edited by P5-133XL on Sun Mar 07, 2010 5:58 am, edited 1 time in total.
Image
P5-133XL
 
Posts: 4345
Joined: Sun Dec 02, 2007 4:36 am
Location: Salem. OR USA

Re: SMP2 issues

Postby bruce » Sun Mar 07, 2010 5:44 am

HFM is fine, but these two calculators will also allow you to get a fairly accurate idea of what bonus you might receive.
-bigadv bonus point calculator and SMP2 bonus point calculator
bruce
 
Posts: 21270
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: SMP2 issues

Postby zygrene » Sun Mar 07, 2010 6:28 am

I understand that there are many variables to account for when looking at PPD, but right now I'm just trying to make sure I'm not doing something wrong.

If i incorporate the bonus points then ppd from SMP2 goes up from 1000 to ~5000. I'm just a little confused since other people are claiming around 10k ppd with quads.
zygrene
 
Posts: 11
Joined: Fri Mar 05, 2010 4:19 pm

Re: SMP2 issues

Postby Grandpa_01 » Sun Mar 07, 2010 6:55 am

zygrene wrote:I understand that there are many variables to account for when looking at PPD, but right now I'm just trying to make sure I'm not doing something wrong.

If i incorporate the bonus points then ppd from SMP2 goes up from 1000 to ~5000. I'm just a little confused since other people are claiming around 10k ppd with quads.


If they are claiming 10K at 3.4Ghz they are stretching it a little I have a AMD Phenom II 940@3.6Ghz and it gets around 6500 to 7500 PPD depending on Wu. I have a Q9650 @ 4aghz and it gets between 9500 and 11500 PPD depending on Wu. 5000 PPD is about right for your OC of 3.4Ghz
Image
2 - SM H8QGi-F AMD 6xxx=112 cores @ 3.2 & 3.9Ghz
5 - SM X9QRI-f+ Intel 4650 = 320 cores @ 3.15Ghz
2 - I7 980X 4.4Ghz 2-GTX680
1 - 2700k 4.4Ghz GTX680
Total = 464 cores folding
User avatar
Grandpa_01
 
Posts: 1873
Joined: Wed Mar 04, 2009 7:36 am

Re: SMP2 issues

Postby P5-133XL » Sun Mar 07, 2010 8:00 am

I get ~7500 with a Q6600@3.42 with a p6021; ~7300 with a Q6600@3.42 on a 6012
P5-133XL
 
Posts: 4345
Joined: Sun Dec 02, 2007 4:36 am
Location: Salem. OR USA

Re: SMP2 issues

Postby Wrish » Sun Mar 07, 2010 12:54 pm

I claimed 10,000 PPD for a 3.8 GHz Q9550. To downscale that to expected yield at 3.4 GHz, raise the relative performance fraction to the power of 1.5.

10,000 * [ (3.4 / 3.8) ^ 1.5 ] = 8463

Remember, that's dedicated (a GPU takes more resources than Task Manager says), with a slightly newer chip that may be faster per clock, and that RAM and controller timings also play a factor (DDR2-1066-5-5-5-15-1T). HFM benches:
Code: Select all
"Q9550 SMP @ 3804 MHz" Win7 x64, affinity lock enabled
Project   #frames    avg PPD
6012      300        9863
6013      100        7394
6014      300        10529
6015      200        10230
6020      300        9364
6023      300        10040
6024      200        10432
6025      300        10267
Wrish
 
Posts: 390
Joined: Thu Jan 28, 2010 5:09 am

Re: SMP2 issues

Postby zygrene » Sun Mar 07, 2010 9:09 pm

I didn't realize that giving 6% of my CPU to the GPU client would reduce my ppd by around 30-40%. I'm running SMP2 exclusively right now and I'm getting 7600ppd on a P6012. I think I'll just forget about the GPU client for now since the extra power consumption isn't worth getting around the same net ppd.

I just need to meet the bonus requirements now so all that ppd actually gets taken into account... Is there any way to check my WU completion rate? I know I've dumped a lot of WU's in the past due to technical issues.
zygrene
 
Posts: 11
Joined: Fri Mar 05, 2010 4:19 pm

Re: SMP2 issues

Postby 7im » Sun Mar 07, 2010 9:38 pm

No, sorry, there is no tool to check completion rate, other than submitting an 11th A3 WU and see if you got a bonus.
How to provide enough information to get helpful support
Tell me and I forget. Teach me and I remember. Involve me and I learn.
User avatar
7im
 
Posts: 15237
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: SMP2 issues

Postby P5-133XL » Sun Mar 07, 2010 9:43 pm

zygrene wrote:I didn't realize that giving 6% of my CPU to the GPU client would reduce my ppd by around 30-40%. I'm running SMP2 exclusively right now and I'm getting 7600ppd on a P6012. I think I'll just forget about the GPU client for now since the extra power consumption isn't worth getting around the same net ppd.

I just need to meet the bonus requirements now so all that ppd actually gets taken into account... Is there any way to check my WU completion rate? I know I've dumped a lot of WU's in the past due to technical issues.


I believe that that was the conclusion that I came to ...

P5-133XL wrote:I came to the conclusion that with my two 3870's and an OC'ed Q6600@3.42 that I actually gained more total PPD by just SMP2 folding with A3's. I lost more PPD from the delay in the SMP2 client than I was getting for my two video cards. I make the assumption that the PPD value closely correlates to the PPD so I retired both my 3870's from folding. Not only do I get more points, the scientific value increases, The lag from GPU2 folding disappears, and I get to decrease my power costs and the wear and tear on the video cards.

You really need to include the bonus when making these types of judgments. The bonus may be 5-6x the base point value. HFM.net can be easily configured to produce PPD including the bonus.

Different CPU's at different clock rates; different Video cards; different OS's will all change the PPD's that you get. So my decision may not apply to all. Each person needs to test their setup independantly and come to the decision that works bets for them.
P5-133XL
 
Posts: 4345
Joined: Sun Dec 02, 2007 4:36 am
Location: Salem. OR USA

Previous

Return to V6.34Beta SMP2 with passkey [Not Bigadv]

Who is online

Users browsing this forum: No registered users and 1 guest

cron