For the F@H brain trust

[Insidious]

Originally posted by: Markfw900

I really need a life.

I can relate! It's been pretty ....directionless here since my son left. I really don't want to do the bimbo thing, but....

ah well, summer is comming. The toys can come out of hiding. That'll help :sun:

-Sid

 

[Duvie]

Do you have the shortcut tabs set up correctly? With the gpu0 and gpu1 callouts as well as the forcegpu command. MIne did that before I had it set up right.....

Also make sure you add the affinity system variable....

 

[Markfw]

Originally posted by: Duvie
Do you have the shortcut tabs set up correctly? With the gpu0 and gpu1 callouts as well as the forcegpu command. MIne did that before I had it set up right.....

Also make sure you add the affinity system variable....

Exactly ! I use the below:
C:\folding_gpu1\Folding@home-Win32-GPU.exe -gpu 0 -forcegpu nvidia_g80 -advmethods
C:\folding_gpu2\Folding@home-Win32-GPU.exe -gpu 1 -forcegpu nvidia_g80 -advmethods

 

[Drsignguy]

Well, I did some bios checking and added voltage to my ICH (from 1.5 v to 1.6v ) and my second card now has jumped back up to its normal PPD. I seem to think PCTC2 is correct on this.

All my settings in the bios with my current OC are stable and never gave me any indication of instability, but by adding a bit more voltage to my ICH, it helped my second card and now both cards PPD are over 6000+.

 

[Insidious]

Great news Drsignguy!

And nice catch PCTC2 :thumbsup:

-Sid

 

[Drsignguy]

Yes this is good news, but I would like to add some more information.

As you all know, with the latest nvidia drivers (182.06) it allows you to possibly run 2 Smp's and a Gpu relatively without any problems. So, by adding a bit more voltage to your ICH, it helps with a second card, BUT both cards still have troubles with PPD production even with the adjustments. So, I did a bit of testing this weekend and this is what I found:

Tested with ( Main Rig )Q9450@ 3.4, 2 GTX 260/216 55nm @ 576/1512/999 (sync) on a EP45-DS3R. Vista 64.
Q6600@ 3.3, 2 GTX 260/216 65nm @ 576/1512/999 (sync) on a EP45-UD3P. XP 64.
Q6600@ 3.1, 1 GTX 260/216 55nm, GTX 260/192 @ 576/1512/999 (sync) on a (Intel) XBX2. XP.


Main Rig seems to have the most difficulties (vista): Secondary, with only the basics installed, seem to be producing approx the same (XP). 2 Other rigs using Vista.



I looked at the Affinity in Vista. At first it sets the Gpu FAH_core to all 4 Cpu cores. Once it started working, it sets the Gpu FAH_core to only 1 Cpu core while the Smp is set to each Cpu core ( if running 1 smp ). If running 2 Smp and Affinity changer is installed, the affinity will set the Smp to 2 Cpu cores each and Vista will still set the Gpu FAH_core to only 1 Cpu core. In XP, it sets the Gpu FAH_core to all Cpu cores. I removed NV_FAH_CPU_AFFINITY ( value 0 ) to see what happens and there is no change due to the latest nvidia drivers. I set the Affinity to all Cpu cores and the PPD has improved a lot. But once the Wu has completed, it sets it back to 1 Cpu core, but only in Vista.

F@H affinity changer is installed on all my rigs but if I am not mistaken, this only pertains to the Smp's ( if 2 are running ) and not Gpu. So vista has something to do with the affinity changing on the Gpu's. I uninstalled FAH Affinity changer and Vista still does the same thing. Since then, I reinstalled Affinity changer.

I am still trying to find out why Vista ( checked all rigs with Vista installed and found that they do the same) places the Gpu FAH_core to only one Cpu core as opposed to XP placing the Gpu FAH_core to all Cpu cores. Now since the latest drivers don't require much Cpu, 1 Cpu core should be sufficient. But when a Smp is running, it's better to have all Cpu cores for the Gpu FAH_core. It does affect the Cpu, but not much in PPD with 1 and or 2 Smp's running.


Take a look for yourselves and see what you can come up with. Any and all feedback is appreciated.



Edit: I was looking around and found set affinity changer for Gpu and other things. I just installed it and testing it out. Will post back later with some news.

 

[Insidious]

It sounds like the system environment variable doesn't do anything in Vista. :Q I'm wondering why Pande group wouldn't forward this functinality when it had such a large effect in WXP.

I re-checked all my WXP affinities to verify they are still managed (to see if Core_14 had anything to do with this) They're fine. SMP's are putting one instance on each of the 4 cores and GPUs are remaining set to use all 4 cores. (NV_FAH_CPU_AFFINITY = 0)

I'm going to put a thread at F@H and see if this has been discussed/corrected for Vista

/me puts on asbestos undies :evil:

-Sid

When you configured your GPUs, you did check the box for 'Do NOT lock the core to a specific CPU' right?

 

[Drsignguy]

Originally posted by: Insidious

When you configured your GPUs, you did check the box for 'Do NOT lock the core to a specific CPU' right?[/q]






Correct Sid, I do have the box checked and the same thing happens even if unchecked; Start the Wu, Affinity to all cores then once it starts working, Affinity goes to 1 core.

Thanks for checking this out Sid.

 

[GLeeM]

Hey Sid (or somebody) once you are over there check out a thread called Double your PPD or something like that about drivers for NV/Vista to make sure the TeAm already knows about.

 

[Insidious]

Originally posted by: GLeeM
Hey Sid (or somebody) once you are over there check out a thread called Double your PPD or something like that about drivers for NV/Vista to make sure the TeAm already knows about.

I was following that one. It's got a threadcrapper busy, but as far as I can tell, it is only for Vista users and there seems to be some doubt about the ability of those leaked drivers to do correct calculations.

Ihaque has been testing and promises some sort of recommendation in the near future.

-Sid

 

[Insidious]

So far, one guy has suggested using 'Configurable Affinity Changer for Windows' (no link, but maybe a Google will find it.)

Another guy is saying he doesn't loose any PPD since the CPU is only used 2-4% on his Vista-64 setup. He suggested making sure 'slightly higher priority' is selected on the GPU installation.

-Sid

 

[Drsignguy]

I also removed the variable and rebooted. After reboot, Made sure that the box was checked (again) in the configuration section for the Gpu and replaced the variable, then rebooted. So far, after first initial startup on WU, all cores are taken up and after works starts, All cores are taken up.

I am wondering what the hell is going on here. I am going to let the Wu complete and then see if it continues. Something is a strange in my machine(s):roll:

 

[Drsignguy]

So far so good. Something must have happened to my system when I cleaned out all the crap last week that was not needed. Ran ccleaner and just made my system a bit more organized. So, something was very wrong with the clients and the variables and with the operator.
Hopefully this trend continues with the Wu's and not the operator errors.....

 

[Insidious]

If I had a dollar for everytime I have loved my PCs to death...

 

[PCTC2]

Originally posted by: Insidious
Great news Drsignguy!

And nice catch PCTC2 :thumbsup:

-Sid

It's my contribution to the TeAm until I can afford the electricity on the computers. I have a problem (in addition to my health. ). I tend to buy more hardware with my extra cash, rather than allow my computers to run 24/7. It seems I will have a new AMD system (because I haven't had one in FOREVER). Well, eventually I will run out of room for hardware and I will pay for electricity.

I noticed this on a few of my older systems. With the strain of running lots of RAM and multiple graphics cards, some chipsets "throttle" the PCIe lanes to the PCIe_1 slot to 1x if their voltage is not high enough. Just bump it +0.1v and it should return to whatever lane width it should be at.

Well, my new addition I doubt will run F@H very well, but it's going to be my media center.
AMD Phenom II X3 720BE
Gigabyte GA-MA790X-UD4P
4GB DDR3-1333
XFX HD 4850 1GB

 

[Drsignguy]

This is interesting to know. I will be doing some more checking on this. :thumbsup: So far, everything seems to be running much better.

So, this is what I have done to increase PPD and consistency:

Made sure my variables are working correctly ( NV_FAH_CPU_AFFINITY, value 0 )
Made sure that my "Do NOT lock cores to specific CPU" is checked
I am using the "slightly higher core priority"
Adjusted my ICH voltage to a bump of +1.

Make sure when cleaning your machine that all these work. If not, go back and reset everything after the cleaning and rebooting. Sometimes, even though it's there, doesn't mean its working.

Everything seems up to par. Depending on the WU, some are reaching as high as 8000+ PPD with the 216/55, over 7500+ PPD with 216/65 and as low as 4800+ PPD with 216/192 card. This card does produce more but this is the lowest I have seen so far. Even my 8800's do much better. So with this being said, my daily production has been averaging over 60,000+/- give or take a few.

Thanks Sid, PCTC2, Pokey and MarkFW. Nice to know that we have wise folks in here that truly give..........

 

[Pokey]

Just bump it +0.1v and it should return to whatever lane width it should be at.

Pokey is not an overclocker so bear with me.....................

I have a couple of DG35EC MBs and looking in the bios setup and they don't seem to have the ability to change anything voltage wise.
Also, there is a EP45-UD3P and a DX48BT2. So which setting am I looking to change +0.1v?

 

[Insidious]

Originally posted by: Insidious

Originally posted by: GLeeM
Hey Sid (or somebody) once you are over there check out a thread called Double your PPD or something like that about drivers for NV/Vista to make sure the TeAm already knows about.

I was following that one. It's got a threadcrapper busy, but as far as I can tell, it is only for Vista users and there seems to be some doubt about the ability of those leaked drivers to do correct calculations.

Ihaque has been testing and promises some sort of recommendation in the near future.

-Sid

The "double your PPD" question has been answered now. You may have noticed that yesterday, your client (automatically) downloaded a new version of the FahCore_14.exe for the 59xx series work units. This is because there was indeed an issue with the 185 (Leaked) NVidia driver. It was yielding higher PPD in Vista because it was not doing all of the calculations. This led to full credit for partially crunched work units. Thus, a higher PPD.

Unfortunately, you will see some PPD decrease in your 59xx work because the fix affects all driver versions. It's not terribly big... I am loosing ~5% as compared to before the new core.

Anyway, it just goes to show there really is no free lunch. Life is good and we can be comfortable using the latest NV drivers and there is no need to go googling for a special version or feeling like you're missing out with what you presently use.

-Sid

PS: I hit 9 Million yesterday!

 

[Drsignguy]

Ah, Now I see what was happening... Thanks for the info Sid. :thumbsup:

BTW, Congrats on hitting 9 mil. I noticed that yesterday too.



Pokey, I am trying to figure out if PCTC2 means the PCI-E (mhz lock) or the actual (clock mv) settings. You should have both in each of the Ep45 and the DX48. But as for the DG35, that would be unknown as I am not familiar with that board....Maybe he or someone else could make that clear..

 

[Pokey]

Thanks Drsignguy. I can wait, I am getting a little paranoid about changing things anyway.............I moved one of my boxes to another desk the other day and when I started it back up it came up with funny colors. I got that fixed, and then it started freezing up after a couple of hours and require a reboot. It took me two days of fiddling to get it back to some semblance of normalcy. Sometimes I feel like I'm spinning plates.

Congratulations on the 4mil MM by the way.

Also Sid congrats on the 9 mil.

But you better watch out I'm gaining on you................ I'll have you in say a year and a half.................

 

[PCTC2]

Sorry. I mainly used nVIDIA boards for LGA775 so I was trying to remember the voltage. MCH Core +0.1v (on your EP45-DS3P) and you can also try modifying your PCIe Frequency if that doesn't work by 1 or 2 MHz.

I'll be picking up an EP45T-DS3P soon to replace an aging Abit AW9D 975X board. I'll let you know if I have similar problems. (I'll be running an 8800GTS and an 8800GT on it). I'm trying to consolidate computers and do a few minor upgrades.

 

[Drsignguy]

Thanks Pokey, congratulations to you with reaching 4 million as of today.
Nice work, now go get Sid!


Sid, nice BIG fat number of 9 MillioN you just reached......Congrats to you sir...


PCTC2, thanks for the info. I shall put it to good use. :thumbsup:

 

[PCTC2]

Well, I just bought an EP45-DS3P. Let's see how this puppy handles.

 

[PCTC2]

Also, I was wrong about the PCIe slots... both the x16 and x8 slots run from the P45 ICH NorthBridge... hmmm. So when 2 cards are inserted, they drop to x8/x8. Forgot about that in the P45 chipset. The voltage can help with stability with your ppd due to NB stability, but I don't know why your ppd are so slow. I'm perplexed now.

 

[Drsignguy]

Ah, thats right. I forgot about that. Was wondering why my slots were at 8x. I had to bump most of my voltages up a notch or 2 in my main rig as it has 8Gb Ram, and the dual vid cards. It helped the stability of it all.

As for why the slowness, here is what I noted earlier and posted it somewhere...As per Sid, if you set your client configuration to "slightly higher" CPU core, the results are so much better. The Smp's do not take a very noticeable hit at all. Also make sure the "do NOT lock cores" is checked. Mine was, but I had some strange software issues and I repaired all the problems. And of course, make sure all the variable settings are correct.
I hope that all this information has helped Pokey with his production...Haven't heard much if he did or not.