Now thats more like it

[GhosT15]

A little late on this beast finally coming online fully, but its finally pulling some nice numbers

Project ID: 2686
Core: GRO-A3
Credit: 8955
Frames: 100

Name: MainBox
Number of Frames Observed: 30

Min. Time / Frame : 00:23:24 - 54,408.0 PPD
Avg. Time / Frame : 00:25:23 - 48,157.5 PPD
Cur. Time / Frame : 00:24:28 - 50,830.4 PPD
R3F. Time / Frame : 00:24:03 - 51,711.0 PPD
All Time / Frame : 00:25:23 - 48,994.7 PPD
Eff. Time / Frame : 00:25:37 - 48,548.5 PPD


This is my updated Hex i-7 running at 4.13GHz


Lets retake first

 

[capreppy]

Nice!!!

Looking forward myself to going dual hex.

 

[musky]

A little late on this beast finally coming online fully, but its finally pulling some nice numbers

Project ID: 2686
Core: GRO-A3
Credit: 8955
Frames: 100

Name: MainBox
Number of Frames Observed: 30

Min. Time / Frame : 00:23:24 - 54,408.0 PPD
Avg. Time / Frame : 00:25:23 - 48,157.5 PPD
Cur. Time / Frame : 00:24:28 - 50,830.4 PPD
R3F. Time / Frame : 00:24:03 - 51,711.0 PPD
All Time / Frame : 00:25:23 - 48,994.7 PPD
Eff. Time / Frame : 00:25:37 - 48,548.5 PPD


This is my updated Hex i-7 running at 4.13GHz


Lets retake first

Are you pushing video cards with this machine? The reason I ask is, as good as those numbers are, they are actually low for that processor at that speed on that unit. You should be right around 60K - 65K, so unless your video cards are putting up much more than than 10K, you may want to shut the GPU client down and save some power.

 

[zero2dash]

you may want to shut the GPU client down and save some power.

I hate my 260's just sitting there doing nothing, but this is totally right. (even moreso in my case being an SR-2)
The PPD drop from running gpus on hexas (let alone 2) is not worth running the gpus anymore.

 

[GhosT15]

I hate my 260's just sitting there doing nothing, but this is totally right. (even moreso in my case being an SR-2)
The PPD drop from running gpus on hexas (let alone 2) is not worth running the gpus anymore.

Are you pushing video cards with this machine? The reason I ask is, as good as those numbers are, they are actually low for that processor at that speed on that unit. You should be right around 60K - 65K, so unless your video cards are putting up much more than than 10K, you may want to shut the GPU client down and save some power.

It's a single 280 OC, puts out about 8k-9k , Odd that it effects the CPU that much, only ever shows 1%. Guess I'll try killing it remotely and see how much lower the TPF goes. Seems counter productive, but you guys know your F@H stuff!

 

[capreppy]

Are you using any affinity programs? The GPU shouldn't be taking much? Are you running anything else on this box? This will also affect PpD

With 12 cores you can try assigning the GPU to one core and the SMP to 11 cores.

 

[GhosT15]

Are you using any affinity programs? The GPU shouldn't be taking much? Are you running anything else on this box? This will also affect PpD

With 12 cores you can try assigning the GPU to one core and the SMP to 11 cores.

I noticed overall better TPF letting windows assign out the 1% the GPU is using, it seems to roate the cores so no one core is always assigned to the GPU. My Sample size, however is small so take that into consideration.

It is my main PC, so I do use it, however it picked up this unit last night and completed those sample frames with me doing nothing more than quick email check on it this morning (probobly why the "max frame" was about 2 minutes longer then average)

When I remoted in, it was averaging 24:01 TPF , I turned off the GPU and will see what happens to the TPF, at the least, its worth an experment.

 

[musky]

Are you using any affinity programs? The GPU shouldn't be taking much? Are you running anything else on this box? This will also affect PpD

With 12 cores you can try assigning the GPU to one core and the SMP to 11 cores.

Do not switch to -smp 11 (ask jebo and his 120 EUE'ed standard smp units about why.)

As far as whether the GPU should or shouldn't take so much CPU time, that is probably debatable. However, it does. I think that it gets magnified in a big way with these faster machines. I have a thread on this subject somewhere. The summary of it is, after beating my head againast a wall for a weekend trying to get an I7 970 and a GTX 460 to play well together, I finally gave up and shipped the 460 to someone who could use it to fold. My 970 does just under 70K ppd on a good unit, which is at least as good and probably actually better than any combination I could come up with using the 460.

 

[GhosT15]

Looks like turning off the GPU folding netted me about 22:06 TPF average now, about 2-2.5 Minutes faster, and brought up the PPD to 60k even with the "old slower" TPFs effecting the bonus.

Not that I didn't belive you guys, but wow, I did not expect it to be that great, as you said Musky it seems the effect is magnified on the hex core vs the quad core. (the quad core only lost about 1 minute TPF, so the GPU out-performed the loss from the SMP, as the bonus was not as high either)

I could probobly squeze a little more out of it, if I turned off the GPU/CPU Gadgets on the desktop, they are the next highest 'cpu hog' Xfire/GTalk use next to nothing idle.

 

[musky]

This is a relatively new thing. With the significant increase in CPU folding perfromance of the single hex Intel machines, GPU folding is just not worth it. This is unfortunate since there are so many good folding GPUs out there for a reasonable price.

 

[APOLLO]

This is a relatively new thing. With the significant increase in CPU folding perfromance of the single hex Intel machines, GPU folding is just not worth it..

There's something going on architecturally that is impacting all the hex-core machines. Haven't put my finger on it yet but it sure seems like it's more than the bonus factor of the lower TPFs that is magnifying the detriment of running GPUs alongside hex-cores. Something else must be involved is my hunch.

 

[sirmonkey1985]

This is a relatively new thing. With the significant increase in CPU folding perfromance of the single hex Intel machines, GPU folding is just not worth it. This is unfortunate since there are so many good folding GPUs out there for a reasonable price.

i wouldnt say its not worth it.. its just that we have become point whores instead of caring about the science because of PG's point system..

 

[SpeedyVV]

A little late on this beast finally coming online fully, but its finally pulling some nice numbers

You kidding? The race has only started. There is quite the way to 5Billion

 

[APOLLO]

i wouldnt say its not worth it.. its just that we have become point whores instead of caring about the science because of PG's point system..

Exactly. You got it right on the bulls eye.

For those who lament not being able to fold with your shiny new GPUs, just put together a low end single socket system with used parts if you have to, and stuff the cards in there. That way you won't need to fuss about the blessed bonus and you can make an extra 5-15k PPD per card while you're at it with more science to boot. Besides, we will need all the production we can muster to retake the #1 spot. Believe me when I say it's going to take much more than the oft lauded hex-core to accomplish that.

 

[zero2dash]

I could probobly squeze a little more out of it, if I turned off the GPU/CPU Gadgets on the desktop, they are the next highest 'cpu hog'

Nah, shouldn't make any difference.

i wouldnt say its not worth it.. its just that we have become point whores instead of caring about the science because of PG's point system..

It helps save on the electric bill and puts less stress on my psu. If the points were equal running SR-2+gpus compared to running SR-2 only, I still wouldn't run the cards. As it is, I get more PPD and lower TPF's without the cards than I would with the cards. The points don't mean everything; there's a + there, but the savings on power and going cpu only is why I built an SR-2 to begin with.

 

[jebo_4jc]

Here's my theory. Something about the way the Smp client uses threads gets bogged down by one thread being used for the gpu. So the rest of the cores have to wait for the core that is also pushing the gpu client.

Hey I just thought of something. Did anybody "disable CPU affinity lock" on their gpu clients?

Anyway, Smp 11 WOULD actually work great. It salvaged most of my CPU ppd and let's me run my gpus. However, it is only stable on bigadv units. If you try a Smp unit, it will crash. A lot. a whole lot. So, I run Smp 11 -oneunit when I have a bigadv Wu. Then switch to Smp 12 once the bigadv unit finishes. Requires baby sitting, but it works.
Posted via [H] Mobile Device

 

[APOLLO]

If the points were equal running SR-2+gpus compared to running SR-2 only, I still wouldn't run the cards. As it is, I get more PPD and lower TPF's without the cards than I would with the cards.

Not trying to start a debate nor am I going to pursue another subthread on the evils of the points system, but I think you essentially confirmed what sirmonkey was saying at least in essence. If the bonus system did not exist and -bigadv was still in beta, I harbor a lot of doubt there be would this proliferation of SR-2 rigs we have witnessed here since last summer. With the exception of myself, Tiger and a few other members, hardly anyone was considering dual-socket systems and very high core counts before last fall. Recall how the state of folding was in '08 and most of '09, and it would be that way to this day...

The points don't mean everything; there's a + there, but the savings on power and going cpu only is why I built an SR-2 to begin with.

Can't speak about everyone on this team, but like I mentioned above, folding wouldn't be very much different than it was last year at this time if the new points system was not implemented. Sure, points don't dictate everything. They definitely weigh heavily upon participants' purchase decision, I can state this with absolute certainty.

 

[jebo_4jc]

update: the "Do NOT lock cores to specific CPU" setting does nothing. In fact, I think it hurts PPD more than helps.

 

[GhosT15]

Nah, shouldn't make any difference.

To my suprise, it effected the TPF by 30 Seconds, nothing crazy sure, but 30 seconds x 100 frames = 50 minutes faster

 

[musky]

To my suprise, it effected the TPF by 30 Seconds, nothing crazy sure, but 30 seconds x 100 frames = 50 minutes faster

That is significant, and may be a very good clue as to why GPU folding is killing these high-end hex core machines.

 

[GhosT15]

That is significant, and may be a very good clue as to why GPU folding is killing these high-end hex core machines.

the form doesnt do quote inside quote, but incase it was missed. the 30 seconds faster TPF average was by killing gadget.exe , which had a CPU Monitor and GPU Monitor widget/gadget, and Rivatuner (which feeds the GPU monitor).

 

[jebo_4jc]

I stick to my theory that some CPU processes hold individual FAH threads hostage.

 

[Nathan_P]

I stick to my theory that some CPU processes hold individual FAH threads hostage.

There some that hog a load of time.

GPU-z, in fact any hardware monitoring program
Nvidia system tools
AV software

I'd say they were the biggest 3, if i get some time in the next few dats i'll see if i can get some numbers.