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

[VirtualLarry]

So, anyways, I forgot whether I already said so or not, but I'm in, as folding handle "VirtualLarry".

I've got two 7950 3GB cards, running on G4400 CPUs, slightly overclocked. (Cards are stock 800/1250 thought).

Along with my "new toy", another Skylake 1151 board, with an i5-6400 BLCK OCed to 150 (50% OC), @4.05Ghz, with a Sapphire Nitro RX 460 4GB card. (Folding@Home on less than 75W, baby!)

Ok, so probably outclassed entirely by a single GTX 980 cards, but w/e, I've spent my money for Dec. already on a 4K screen, no more budget for GPUs.

I might fire up a fourth box, and throw in my two BNIB R9 270X 2GB cards, haven't decided yet.

 

[TennesseeTony]

Well I am glad to apparently be wrong regarding the CPU usage, but I'll keep providing my GPU's with spare horsepower. I'd rather have 17 + 1 in the chamber and not need it than to only have 5 in the snubby and run out.

EDIT: Larry: 50% OC. Freaking awesome!

 

[iwajabitw]

Well I am glad to apparently be wrong regarding the CPU usage, but I'll keep providing my GPU's with spare horsepower. I'd rather have 17 + 1 in the chamber and not need it than to only have 5 in the snubby and run out.

EDIT: Larry: 50% OC. Freaking awesome!

Or 33+1 for the Glock

 

[TennesseeTony]

Yeah but makes me look like I have bon.....err, nevermind.

You don't think it looks, I dunno, weird?

 

[iwajabitw]

LMAO! I know what you mean.

 

[TennesseeTony]

Regarding CPU utilization, is everyone more or less using the same yardstick? I am confident everyone here knows a thing or two about computers, so forgive me if this is in any way condescending, not my intention at all.

The pic below is from Mjolnir. HT is off so it has 6 physical cores to use. 100% divided by six cores equals 16.6667% each core. So anything using 16.67% or more is 100% using that core. (Turbo Boost/overclocking has been said to cause task manager to report elevated percentages, thus the slightly more than 16.67%)

BumbleBee has 14 physical cores, HT off, TurboBoost on. 100% / 14 = 7.14%, so 7.14% on any task equals 100% core usage on Bee.

Again, TaskManager is over-reporting a bit, I don't like seeing 100% total here, but MSI afterburner and CPUID HardWare Monitor both assure me I'm only at 94% total (which is basically an entire core in standby, to run the background processes, virus scans, TeamViewer, etc).

 

[iwajabitw]

My 2 instances in task manager are topping at 9% each.

 

[iwajabitw]

Sorry If this is a big pic!

 

[TennesseeTony]

So you too are at 100%. HT is on, so 100 / 12 threads = 8.33%, and you're overclocking like a mad man, so task manager has over-reported your usage a touch also. Cool, thanks for the back up, with all 33+1.

 

[Markfw]

I'd consider AMD, except I've never tried getting an AMD card to work with Linux.

The CPU utilization is for Nvidia cards, Yes, AMD does not seem to suffer that. However, their ppd is vastly inferior on F@H

Edit, another thing. My quads cores *even the 6700k) seem to need a core or two to get max ppd, but my 3930k is at 12 cpu's for F@H uses 7-8% and ppd is normal/ I think that the 2011 quad channel ram helps a lot. Both the other 2 are quads.

 

[iwajabitw]

Gotcha Tony, I didn't do the math.

 

[TennesseeTony]

Hey, it's ok man. We're from the South, and countin's hard. Once I run outta fingers and toes, I start countin' teeth. And if you do the same, you're from 'Bama, so, you know, you run outta stuff to count faster'n the rest of us.

edit: If anyone should happen to have any dental issues, I sincerely apologize, just being silly!

 

[iwajabitw]

LMAO!

 

[VirtualLarry]

I find that Task Manager over-reports CPU usage percentages, by the actual percentage of my OC, up to a point... once CPU usage hits 100%, it doesn't seem to, the totals for each process seem to get limited.

For example, my i5-6400 running WCG on several cores. It's a quad-core, so any process that is using one thread at full load, should show 25% CPU usage. Yet, when I have only two of them running, they show 37.5% each approx. With three running, they hit 31%, and CPU shows 100% load due to other processes.

I had a thread in the OS forum about this, that I had just discovered the over-reporting of CPU usage, but it got lost when the forum got rolled back.

Also, OCCT has a CPU usage graph, and it shows around 60% CPU usage, when Task Manager reports 87-90% CPU usage with two WCG tasks taking up 100% of a CPU core each, plus misc threads.

 

[Ionstream]

My Xeons can contribute 60k - 80k PPD (which honestly pales in comparison to what GPUs can pull off), but the more the merrier I guess?

 

[alpha754293]

What is FEA? Pure curiosity. Also, pure curiosity, at what price are you expecting to purchase these six K40 cards (each)? Sounds a wee tiny bit expensive.

They're about $3200 a piece from B&H Video.

The difference is that with this, I can probably justify it as a business expense since I can run the finite element analysis on it, but with the GTX 1080s, I can't do the same.

It'd be housed in a Supermicro 2U server if I go that route. (Not specifically looking to build a folding system only - just something that's folding-capable, much like the current systems that I've got enlisted to fold.)

FEA = Finite Element Analysis. A very double-precision intensive workflow, which the GeForce cards gut.

The program that I am using for FEA is double precision, but that isn't to say that all of them are.

As always, YMMV, but if their marketing material is to be believed, using the Teslas can double or triple the throughput so the FEA runs can complete in half to a third of the time it would normally takes CPUs-only.

But it does come with a price and being that I run both computational fluid dynamics (CFD) (which can be either single or double precision, which though the Tesla single and double precision throughput is through the roof, the CFD codes don't fully support the Teslas yet), so I need something that will work well.

Like I said, ideally, I would be getting these cards to do something that pays for itself and provides income and folding would just be a benefit on top of that.

For me, it's an optimisation between cost, FEA, CFD, and folding.

 

[TennesseeTony]

Welcome Ionstream! Is your folding user name Ionstream as well?

 

[alpha754293]

Regarding CPU utilization, is everyone more or less using the same yardstick? I am confident everyone here knows a thing or two about computers, so forgive me if this is in any way condescending, not my intention at all.

The pic below is from Mjolnir. HT is off so it has 6 physical cores to use. 100% divided by six cores equals 16.6667% each core. So anything using 16.67% or more is 100% using that core. (Turbo Boost/overclocking has been said to cause task manager to report elevated percentages, thus the slightly more than 16.67%)

BumbleBee has 14 physical cores, HT off, TurboBoost on. 100% / 14 = 7.14%, so 7.14% on any task equals 100% core usage on Bee.

Again, TaskManager is over-reporting a bit, I don't like seeing 100% total here, but MSI afterburner and CPUID HardWare Monitor both assure me I'm only at 94% total (which is basically an entire core in standby, to run the background processes, virus scans, TeamViewer, etc).

I'm not so sure if it matters or whatnot.

I'm using just the default Windows Task Manager.

TB has no impact on the results, but rounding to the units precision does and very well do/might.

e.g. my 3930K with all 6-cores running/active, HTT enabled, I've got a cpu slot with 10 threads going and 1 gpu slot, so it's showing me 91-92% CPU usage. Stock it runs at 3.2 GHz, but with TB enabled, it is currently running at 3.5 GHz.

If you want to measure it more accurately, you'll likely need something along the lines of ProcessExplorer, but I'm not entirely convinced that this is even an issue or that it matters.

 

[Ken g6]

My 2 instances in task manager are topping at 9% each.

I guess you didn't like me passing you. Not that I expected to pass you permanently anyway.

 

[Pokey]

Hey, it's ok man. We're from the South, and countin's hard. Once I run outta fingers and toes, I start countin' teeth.

We have to take our clothes off to get past 20..............you are too polite Tn............Sorry...............

 

[TennesseeTony]

It's okay Ken, at least your still kicking Assimilator_1

 

[TennesseeTony]

I've been running with the Alzheimer's preference for awhile. It seems like they have no Alzheimer's GPU WUs.

I have run through all the different preferences, and all I get are either cancer or Cystic Fibrosis tasks, so I'd have to agree, those other projects must be intermittent. There is no known way to avoid the lower paying 92xx tasks it seems.

I've been all in for at least 12 hours now, rearranged cards again on one box, so a bit of downtime there, but looks like I may top out at about 6.6M on a good day. Disabling hyper-threading only seemed to assist the 1070 cards, they seem to consistently outperform the previously reported figures from Markfw's GPU result tracking thread.

The 980Ti I moved into the 980 box (thermal reasons) is now under-performing by 10-15%, but the 980 that took it's place in the 980Ti box is OVER-performing. lol Still a bit of a overall ppd loss, but at least I'm no longer worried about frying the Ti.

 

[Markfw]

I have run through all the different preferences, and all I get are either cancer or Cystic Fibrosis tasks, so I'd have to agree, those other projects must be intermittent. There is no known way to avoid the lower paying 92xx tasks it seems.

I've been all in for at least 12 hours now, rearranged cards again on one box, so a bit of downtime there, but looks like I may top out at about 6.6M on a good day. Disabling hyper-threading only seemed to assist the 1070 cards, they seem to consistently outperform the previously reported figures from Markfw's GPU result tracking thread.

The 980Ti I moved into the 980 box (thermal reasons) is now under-performing by 10-15%, but the 980 that took it's place in the 980Ti box is OVER-performing. lol Still a bit of a overall ppd loss, but at least I'm no longer worried about frying the Ti.

I think you need more boxes. It looks like F@H is adjusting things so all cards run at the same as the slowest card.

 

[alpha754293]

So here's a serious and practical question - based on the best projections, I'm likely only to generate about 6 million points for the entire month what you guys will do/be able to generate in a day.

Is it really still worth it for me to participate?

 

[crashtech]

So here's a serious and practical question - based on the best projections, I'm likely only to generate about 6 million points for the entire month what you guys will do/be able to generate in a day.

Is it really still worth it for me to participate?

Everyone matters, alpha. All are welcome! Bring what you can, I'll go ahead and speak for the team and say that no matter the size of the contribution, it's appreciated just the same!