11th Annual Folding@Home Holiday Season 3-way Race

[Markfw]

If you guys need some motivation, here is what one of their higher end users said on the 17th on the Toms hardware forum:
"
UnolocogringoNov 17, 2016, 5:47 PM

Is MOONING still allowed ????????

On the recruiting part.

All of that equipment in the labs is "Fair Game" According to their "RULES".

And maybe contact Asus, MSI Gigabyte etc... and see about a few giveaways for new folders.Mostly Graphics Cards, (Better PPD folding) but anything to draw attention to folding is very much appreciated. If any bite make sure parts are donated to both teams to give away.
Thanks AGAIN guys!!
Rick
"

I hope they don't get the lab or the freebees, we don;t get that, here !!!!

 

[alpha754293]

Does anybody know if there is any real performance differences between Windows and Linux F@H clients anymore?

I've also recently read that the bigadv has ended as of Jan 31 2015, so would there be any real use for me to continue using either the --client-type=bigbeta flag in the console client anymore?

My systems are currently running CFD right now, but I might actually interrupt that and start firing up the systems to get them primed and configured for folding.

(I have three Core i7-3930K systems that I can use. One of them is only running on 4 cores because I burnt out a core OCing it. The other two has all 6-cores operational, but because of my experience OCing it literally to death, I no longer OC any of them. The two systems that still has all 6 cores operational has 64 GB of RAM each, and the one with only 4 cores operational only has 32 GB of RAM. VIdeo cards are Quadro 5000, GTX 660, and I think a GTX Titan Superclocked (original, not X, Black, or Z).)

I also have my 6700K daily driver (4-core, 4 GHz, HTT enabled, but I want to disable it again), 32 GB of RAM as well.

Waiting for feedback re: OS.

and my F@H username is alpha754293 (pending...).

Thanks.

 

[Markfw]

The linux vs windows thing Ithink is pretty mute, except GPU is the way to go by far, and I think its way harder to set them up on linux.

Yes, the BIGADV ended, so don't bother with that flag. But leave 2 cores open for every video card to use. On my 3930k, I have one card and 10 threads allotted to CPU.

HT disabled helps on slower systems.

 

[alpha754293]

The linux vs windows thing Ithink is pretty mute, except GPU is the way to go by far, and I think its way harder to set them up on linux.

Yes, the BIGADV ended, so don't bother with that flag. But leave 2 cores open for every video card to use. On my 3930k, I have one card and 10 threads allotted to CPU.

HT disabled helps on slower systems.

Is there any real performance advantage by leaving HTT enabled?

It's been a while since I've folded.

And what's the reasoning for leaving a physical CPU core open for a GPU slot? Can you clarify please? Thank you.

 

[Ken g6]

The linux vs windows thing Ithink is pretty mute, except GPU is the way to go by far, and I think its way harder to set them up on linux.

I'm not sure I'd agree with that. For Nvidia on Ubuntu Linux, all you have to do is install the card, download the latest Linux drivers, disable Nouveau, and install the new drivers from the command line. To do that, press ctrl-alt-F2, log in, stop your window manager service (usually a file in /etc/init.d that ends in "dm"), run the Linux driver file with sudo, and restart your window manager service.

Well, OK, maybe it's a little harder.

And what's the reasoning for leaving a physical CPU core open for a GPU slot? Can you clarify please? Thank you.

Many CUDA and OpenCL programs loop on one CPU constantly waiting to send or receive signals or data to or from the GPU. Even for those that don't, having a CPU core handy when such a signal comes in is useful.

 

[Markfw]

Yes, the CPU for the GPU is easy. Only install the GPU slot, do nothing with the computer, and watch the utilization, it at LEAST on complete core. My E8400 with both cores trying to push the card was maxed at 100% ! So at least 2 threads with HT, or one without for every video card.

On linux, I had a nightmare time, but I defer to Ken, or anybody who is good at it.

As far as disabling HT, the client can't really use more than one CPU, or 2 if HT is enabled. The CPU client produces WAY less points than the GPU. For example, by 3930k with 10 threads is currently getting 20k ppd. The 980ti on that system is getting 584k ppd.

 

[TennesseeTony]

.....And what's the reasoning for leaving a physical CPU core open for a GPU slot? Can you clarify please? Thank you.

You will find that Folding CONTINUOUSLY uses 100% of one core per GPU. The faster the core the better for fast cards, therefore the recommendation to disable HT.

Welcome to the race!!!

 

[Ken g6]

I just noticed something over on Tom's board:

We really need to stress to everyone how important it is to get a pass key and 10 work units in before the race officially starts.

That's 10 WUs per person (or per passkey, but you only need one per person), so it doesn't matter to us old-timers. But the new people joining will probably want to get started on that early.

Click here to get a passkey if you don't already have one.

(or they can send a lost passkey to your original sign up email address if you still use it)

 

[alpha754293]

As far as disabling HT, the client can't really use more than one CPU, or 2 if HT is enabled. The CPU client produces WAY less points than the GPU. For example, by 3930k with 10 threads is currently getting 20k ppd. The 980ti on that system is getting 584k ppd.

I realise this, but you're running it with HTT enabled?

(All of my 3930K systems have HTT disabled, and the only one that I've got it enabled on is my 6700K, but there's no discrete GPU in that one, so I'm not sure how well it will or won't handle running it sans discrete GPU.)

Is there any speed/performance difference between the command-line interface vs. FAHControl?

Yeah, I miss the days of bigbeta. You got a LOT of points (due to the bonus structure), which in turn incentivised people to work on those bigger, more difficult WUs. Pity that they don't have any more of those.

P.S. One of my systems is having difficulties getting WUs assigned to it, but two of the other systems are already fully up and running. So we'll see what kind of results it gets...

*edit*
Not really sure what happened, but when I uninstalled and re-installed the client, it was able to pick up CPU work without any issues. Weird...

*shrug* Go figure...

 

[Markfw]

I left HT on for my 3930k, as I only had one video card, and the PPD was in range. For all other systems I turned it OFF. The 6700 has 2 monster 1080's to feed, and I want to take no chances on loosing ppd on that system. The rest need it off, as they have one or 2 cards, and their ppd on 2 cores would be next to nothing.

 

[alpha754293]

I left HT on for my 3930k, as I only had one video card, and the PPD was in range. For all other systems I turned it OFF. The 6700 has 2 monster 1080's to feed, and I want to take no chances on loosing ppd on that system. The rest need it off, as they have one or 2 cards, and their ppd on 2 cores would be next to nothing.

What kind of PPD do you get with it HTT on on your 3930K?

And if you were to turn HTT off, what kind of PPD would you be getting?

(I'm only asking because it might be worthwhile for me to it turn it back on if it yields some measureable benefit.)

Hahaha....my 3930K with the Quadro 5000 - the GPU was expected to take almost the time limit to finish a WU, so yeah....nixed that.

lolll....

(It's now running CPU only)

 

[Markfw]

Since CPU is such a small part of my PPD, I have not really looked or tried. Its all about feeding the video cards. Right now its at 40k, but earlier it was 20k. But the video card in the system ranges from 400k to 650k, so its inconsequential. Right now the video card is on a bad unit at 515k.

 

[alpha754293]

Since CPU is such a small part of my PPD, I have not really looked or tried. Its all about feeding the video cards. Right now its at 40k, but earlier it was 20k. But the video card in the system ranges from 400k to 650k, so its inconsequential. Right now the video card is on a bad unit at 515k.

Gotcha. Thanks.

I've taken the system that has the Quadro off completely now, for a similar reason - the CPU only PPD is too low.

Sadly, my 4-core 3930K+GTX 660 and one 6-core 3930K+GTX Titan Superclocked will likely yield around 300k PPD, so that's what my contribution to this is going to look like.

The system with the Quadro is back on to running CFD for me now... (it would have only added around 20k PPD or so...).

That's in a way, kind of sad, because you can sort of get by with like a "mediocre" 4-core CPU and still, like you said, be generating over 500k PPD with just GPUs, but the CPUs don't get as much credit. Pity, being that the GPU WUs are fundamentally different than the CPU WUs. Oh well...guess if that's how Dr. Pande wants to set up the point/reward system such that it incentivises "investment" into GPUs rather than CPUs, it is what it is.

For GPUs though, which would be better? GTX 1080 or GTX Titan X? (Or would the top of the line AMD be better for PPD?)

Thanks.

 

[Markfw]

I have no real idea on GTX titan X. Are you looking at power ? or ppd/$ ? The 1080 is king, except no real data on the Titan X from personal experience.

BUT if you have 3 boxes, and can put 2 cards in each, the 1070 at $389 (or thereabouts) at about 550-600 ppd each would be the best way to kick hiney ! Over 3 million ppd for $2400

Pretty sure AMD is not a good option, I have one, and its the worst in my farm. Newer ones may be a little better.

See sig. This farm is doing 6.5 million ppd

 

[alpha754293]

I have no real idea on GTX titan X. Are you looking at power ? or ppd/$ ? The 1080 is king, except no real data on the Titan X from personal experience.

BUT if you have 3 boxes, and can put 2 cards in each, the 1070 at $389 (or thereabouts) at about 550-600 ppd each would be the best way to kick hiney ! Over 3 million ppd for $2400

Pretty sure AMD is not a good option, I have one, and its the worst in my farm. Newer ones may be a little better.

lol....thanks for your input.

I'm pretty sure that wife would kill me right now if I dropped $2400 on video cards for F@H for this ca. month's worth of competition.

That's rather surprising.

The GTX Titan (regular, not X, Z, or Black) Superclocked is currently averaging an estimated 240k PPD, and I bought it off eBay used for like $500. Hmmm....damn. (On the other hand though, I am using it for other stuff besides F@H, and really wanted/needed its double precision compute capability without going to the Teslas.)

I actually technically have four systems if you include my daily driver into that mix (the 6700K). And two servers. And that's assuming that I don't fire up my old workstation as well.

Unfortuantely, the "arms race" for F@H PPD gets rather expensive after a while, and between what I normally do/want the systems to do vs. F@H - it used to be that my servers were quote "king" like almost a decade ago because it was largely CPU driven. Nowadays, the apps that I normally run don't run on consumer-grade GPUs (which means Teslas) which means $$$$.

If the GTX Titan has 2688 SPs and the GTX Titan X has 3072, assuming linear scalability (which I know it isn't perfectly linear), the estimated performance (also taking the differences in clock speed into account) can be guestimated to be around 327k PPD. Hmmm....interesting...

Once again, thank you for your input.

 

[Markfw]

Not sure what Titan you have, but the X is way better for some reason. (180k vs 600k). See here: http://www.overclock.net/t/475163/gpu-projects-ppd-database

But again, if you want to increase ppd, just buy as many of these as you can afford: ~550-600 k ppd http://www.newegg.com/Product/Product.aspx?Item=N82E16814487279

There may be others a little less ($10-20) but I can only recommend ones I have.

 

[alpha754293]

Not sure what Titan you have, but the X is way better for some reason.. See here: http://www.overclock.net/t/475163/gpu-projects-ppd-database

But again, if you want to increase ppd, just buy as many of these as you can afford: ~550-600 k ppd http://www.newegg.com/Product/Product.aspx?Item=N82E16814487279

There may be others a little less ($10-20) but I can only recommend ones I have.

Thanks.

 

[HutchinsonJC]

You can put me down for the race this year.

 

[alpha754293]

Playing a bit with the settings and optimising it.

It's weird because apparently, if the number of threads is a prime number (or multiples of a prime number), it has problems getting WUs, but if it isn't, then no issues.

So with 6 cores on one and 4 cores on another, and a GPU each, that puts the systems in a bit of a pickle.

 

[Ken g6]

2 is a prime number too.

Also I've never had a problem getting three core work.

 

[MrCommunistGen]

Anyone else having issues getting CPU WUs assigned? The GPUs I fired up today seem to be folding fine, as are my low-thread machines. But my 2 big thread CPU clients are failing to pull down WUs with this error:

05:50:50:WARNING:WU00:FS00:Failed to get assignment from '171.67.108.45:8080': Empty work server assignment
05:50:50:WU00:FS00:Connecting to 171.64.65.35:80
05:50:51:WARNING:WU00:FS00:Failed to get assignment from '171.64.65.35:80': Empty work server assignment
05:50:51:ERROR:WU00:FS00:Exception: Could not get an assignment

I launched the FAH client for the first time on my old home rig (an i7-3770K) about 24hrs ago and it has been looping that error the whole time. I've rebooted Windows and restarted the client several times. I pulled up the NaCl client but the PPD is pretty pitiful - I'm assuming at least partly due to the lack of bonuses on NaCl.

I'd gotten my new workstation (i7-6900K) folding over the weekend and it was doing fine, but when I added my HD 7870 and adjusted the number of threads going to the CPU it killed the WU I was working on, and now that client won't pull down any more CPU WUs either (same error as above). I went ahead and launched NaCl there as well, but I'm at ~15K PPD instead of ~75k PPD.

Interestingly enough, I've got an old i5-650 that I threw my HD 7950 into - I set it up from the get-go to have 2 CPU threads folding and 1 GPU slot. That's been folding to both the CPU and GPU slots perfectly - although I don't think either slot has turned in a WU yet.

Environment Variables:
The i7-6900K and i5-650 are on the same network at my office. The i7-3770K is at home. All are running Win10 x64 and they should be pretty well up to date.

I've considered reinstalling the FAH client on the misbehaving machines, but at least on the 6900K I'll wait until the GPU WU has finished.

Suggestions? Thanks

-mcg

 

[TennesseeTony]

Try going to your drive letter of choice/users/(user name)/AppData (hidden file)/Roaming/FAHclient, then find and open your config.xml file (right click and choose edit).

Copy and paste your passkey as this can be a pain in the butt to type in, save that somewhere else. (Or just make a duplicate file I guess would be easier and name it .old).

Finish all work, shutdown/exit the client, delete the config.xml, restart and I THINK this will give you a clean install, more or less. Start over with your username/team 198/passkey.

Choose FINISH, let that automatically downloaded CPU task get out of there, then make your changes to your CPU slot.

 

[Pokey]

I assume, as usual, Petrusbroder will be the scorekeeper for our Anandtech sub-team race, but who will be the official score keeper for the Tom's & Canucks races?

 

[alpha754293]

Anyone else having issues getting CPU WUs assigned? The GPUs I fired up today seem to be folding fine, as are my low-thread machines. But my 2 big thread CPU clients are failing to pull down WUs with this error:

05:50:50:WARNING:WU00:FS00:Failed to get assignment from '171.67.108.45:8080': Empty work server assignment
05:50:50:WU00:FS00:Connecting to 171.64.65.35:80
05:50:51:WARNING:WU00:FS00:Failed to get assignment from '171.64.65.35:80': Empty work server assignment
05:50:51:ERROR:WU00:FS00:Exception: Could not get an assignment

I launched the FAH client for the first time on my old home rig (an i7-3770K) about 24hrs ago and it has been looping that error the whole time. I've rebooted Windows and restarted the client several times. I pulled up the NaCl client but the PPD is pretty pitiful - I'm assuming at least partly due to the lack of bonuses on NaCl.

I'd gotten my new workstation (i7-6900K) folding over the weekend and it was doing fine, but when I added my HD 7870 and adjusted the number of threads going to the CPU it killed the WU I was working on, and now that client won't pull down any more CPU WUs either (same error as above). I went ahead and launched NaCl there as well, but I'm at ~15K PPD instead of ~75k PPD.

Interestingly enough, I've got an old i5-650 that I threw my HD 7950 into - I set it up from the get-go to have 2 CPU threads folding and 1 GPU slot. That's been folding to both the CPU and GPU slots perfectly - although I don't think either slot has turned in a WU yet.

Environment Variables:
The i7-6900K and i5-650 are on the same network at my office. The i7-3770K is at home. All are running Win10 x64 and they should be pretty well up to date.

I've considered reinstalling the FAH client on the misbehaving machines, but at least on the 6900K I'll wait until the GPU WU has finished.

Suggestions? Thanks

-mcg

Try this - so that I can try and help you debug it (because I have a feeling that you're encountering the same or a very similiar problem that I ran into when I fired up my systems).

On all of the systems, can you list your CPU, your GPU (if/where appliable), how many cores your CPU has, and how many threads your CPUs are using for Folding?

(for the last one, you can click on the Folding@Home tray icon, and then click on Advanced Control" and under cpu, there should be a number (like "cpu: 5") and that will tell you the number of threads that it is running.)

I ran into the same problem yesterday and after doing a little bit of research on the Folding@Home forums, apparently, there is an issue with WU assignment if the number of threads is a prime number like 7, 11, 13, etc. (and also 5, as I experienced/encountered yesterday), which presents an interesting dilemma/pickle.

If you have a CPU and GPU, GPU takes up a thread so CPU is now (numcores-1) threads (i.e. if you have a 6 physical core processor, it will run with 5 threads and one GPU), which has problems with WU assignment.

So your choices are to either go down to 4 threads, which means you have an idle core not doing anything, or for it to go up to 6 threads, which now you might have conflict between the CPU FAHCore and the GPU FAHCore (conflict for resources to feed the GPU).

(That's also if you have HyperThreading Technology disabled in the BIOS.)

If you have a 6-core processor with HTT enabled, then you might be able to force it to run with 10 threads so that all 5 physical cores will be fully loaded.

Give that a shot. See that helps you out in terms of you getting more WUs assigned to your systems.

 

[petrusbroder]

I assume, as usual, Petrusbroder will be the scorekeeper for our Anandtech sub-team race, but who will be the official score keeper for the Tom's & Canucks races?

I am quite sure there will be several.
But I will be doing all the stats for this site and I will use folding.extremeoverclocking.com as my stats site.