9th Annual Holiday Season Folding@Home race

[petrusbroder]

I've added the poll to this thread. You have 3 days to vote - and then we start!
I'll do the teams tonight and post the proposed teams tomorrow morning (UTC).

 

[Plimogz]

Sign me up, please!

<---Name's the same

 

[petrusbroder]

Participants' list updated.
Welcome, Plimogz!

 

[Markfw]

With 2 days to go, teams ???

 

[petrusbroder]

In a few hours ... teams!
Will be done by tomorrow morning.
Start your engines, hoard WUs, and upload just after the starty ...

 

[GLeeM]

How many ppd (estimate) can I expect from GTX 970?

From a 4770K at 4.1 using 7 cores?

 

[VirtualLarry]

Hmm, I loaded F@H 7.44-something (newest) Windows client on both of my Q9300 @ 3.0 rigs w/GT630 384SP cards.

My estimated PPD is just north of 5K. For both CPU + GPU. I expected the CPU PPD to be a bit on the low side, but since my 9600GSO cards cranked out 4K PPD back in the day, and these have 384SP instead of 96SP, I thought I would see a marked improvement.

Especially, since my 860-something SP 7790 1GB GDDR5 card would crank out 30K per card, during last year's F@H race.

Oh well. I guess, don't expect much output from me this year, sorry.

Edit: It appears that the web client is misleading. The "PPD" it displays, depends on whether you click the CPU or GPU WU. The CPU WU alone was 5K PPD. I opened the "Advanced client", and it gives an overall PPD, which is about 12-13K for each machine. (Q9300 @ 3.0, w/GT630 384SP). That's not horrible. Not great, but I'm still on older hardware.

 

[Plimogz]

How many ppd (estimate) can I expect from GTX 970?

From a 4770K at 4.1 using 7 cores?

I'd say somewhere between 200K and 300K PPD. Mainly depending on which project it's working on.
There's a good database over at overclock.net featuring most higher-end cards. Link

 

[Markfw]

I have something in the mail....

 

[Plimogz]

Hey Mark, did you finally set up that R9 290, or what? It's not listed in your sig...

I was wondering if you could chime in on PPD.

 

[GLeeM]

I'd say somewhere between 200K and 300K PPD. Mainly depending on which project it's working on.
There's a good database over at overclock.net featuring most higher-end cards. Link

Thanks Plimogz

Anyone have an idea about CPU ppd? Any CPU?
I'm wondering if it is worth folding on CPU?

 

[Markfw]

Hey Mark, did you finally set up that R9 290, or what? It's not listed in your sig...

I was wondering if you could chime in on PPD.

Yup, running win 10, whenever they don't keep rebooting the box...

And I have like 10 boxes that don't even ft in my sig.....

 

[Pokey]

Thanks Plimogz

Anyone have an idea about CPU ppd? Any CPU?
I'm wondering if it is worth folding on CPU?

I think it might be. But there are so many variables it is hard to predict exactly. I typically keep one core per GPU free. The remaining cores can get anywhere from 1800 to 2300 PPD each core. So six cores out of an eight core CPU would be 10,800 to 13,800. all stock, no OC.

An older quad core using 2 cores might get 1,000 each = 2,000.

This is just me of course, your mileage...............

 

[Pokey]

I'd say somewhere between 200K and 300K PPD. Mainly depending on which project it's working on.
There's a good database over at overclock.net featuring most higher-end cards. Link

I would add a caveat to this; when you look at the data pay attention to the work unit. If it is one that draws a x17 core you will get good production. If you get assigned a wu that needs core x18 or x15, (they don't seem to have quick return bonuses) your production will drop dramatically. It is based on where the assignment server puts you..........based on the work they need to have done.
Again, this is just me............

 

[GLeeM]

I think it might be. But there are so many variables it is hard to predict exactly. I typically keep one core per GPU free. The remaining cores can get anywhere from 1800 to 2300 PPD each core. So six cores out of an eight core CPU would be 10,800 to 13,800. all stock, no OC.

Thanks for the answers Pokey

 

[GLeeM]

I don't remember anything about how to optimize for best points.

Are there any settings or flags to help get better WUs for best ppd?

 

[VirtualLarry]

I typically keep one core per GPU free. The remaining cores can get anywhere from 1800 to 2300 PPD each core. So six cores out of an eight core CPU would be 10,800 to 13,800. all stock, no OC.

An older quad core using 2 cores might get 1,000 each = 2,000.

This is just me of course, your mileage...............

It seems to do that automatically in the newest F@H client for Windows. For example, on my Q9300 rigs, it uses 3 cores for CPU WUs, and reserves one CPU core for the GPU.

For my Thuban (hex-core), it saved two CPU cores, and assigned one CPU core for each of my four GPUs. Personally, I thought CUDA allowed one to drive a GPU with hardly a percent of a CPU core, so this seems kind of extreme. I would have rather assigned two CPU cores for the four GPUs, and saved four CPU cores for CPU WUs.

My Q9300 @ 3.0 w/GT630, shows 14K PPD. My 1045T with (four) GT630s, shows 30K PPD. Doesn't seem quite right.

Edit: Correct me if I'm wrong, but with these new installs of F@H on my three machines, I needed to input the passkey (done), and then each machine needs to complete 10 WUs successfully, before getting bonuses, right?

 

[Pokey]

VirtualLarry
The passkey is to identify you as a personal donor. Not the computer. So any ten work units with that key will do the trick if I understand correctly. They explain it in their faq.
As far as cpu assignments I was just saying what I do. The reason is that I noticed when I was running certain gpu's alone the cpu usage was more than 25% on a quad core. So I started allocating more cpu's, sacificing them to make sure the gpu was at full capacity. It is not scientific. And I do it as a general rule whether I need to or not. Again, that is just my way. Call me a cynic but I don't trust any of these cores or drivers to automatically do the right thing...................

 

[Pokey]

GLeeM
There really isn't much you can do. The way the client-type's are set up now, beta means just that, and can cause a good bit of heart burn. Advanced does not give you any more points, but can give you more problems. So the default settings appear to be the best, at least at this point in time.
I personally prefer the smoother route anyway, and have dropped all the client-type settings.

 

[Plimogz]

[...]I thought CUDA allowed one to drive a GPU with hardly a percent of a CPU core[...]

From what I gleaned looking over some F@H threads these last few days, it does seem like the rule of thumb for nVidia folding is to give each GPU slot it's own core; it's the AMD cards which don't need (deserve?) quite so much cpu.

But then again, most information floating around dates back to before it was feasible to get any decent return on red cards, so this may have changed somewhat.

 

[VirtualLarry]

I have F@H "Advanced Control" open on one of my Q9300 rigs, and the second WU slot, at the bottom of the details page on the right-hand side, shows "Collection Server: 0.0.0.0". Is that a problem? The other WU has what appears to be a valid IPv4 address listed.

 

[Plimogz]

Mine's the same most of the time, but I think I saw it showing a plausible address earlier. Maybe it only connects when actually transferring work back to stanford. I'm sure you'll be fine. I've certainly never had to enter or configure anything with regards to IP addresses and my clients have always worked fine.

 

[biodoc]

Nvidia Foundation awards two teams pioneering cancer research. Link

Vijay Pande (Folding@Home) was one of the recipients.

 

[Pokey]

I have F@H "Advanced Control" open on one of my Q9300 rigs, and the second WU slot, at the bottom of the details page on the right-hand side, shows "Collection Server: 0.0.0.0". Is that a problem? The other WU has what appears to be a valid IPv4 address listed.

I think they have a problem at their end. Lots of ports showing red. Most likely an outage or some such at the campus. When stuff happens it always happens on the week end................
Collection server usually shows a valid number.

 

[Rattledagger]

Collection server usually shows a valid number.

AFAIK collection server for Amd-GPU's never shows a valid number, since AFAIK FAH has never had a collection-server for these cards.


In any case, since I love a race, count me in with folding-name Baldurs_Gate.

Since the old i7-920 crapped-out I've for the moment only replaced it with a bottom-end Celeron G1820, meaning my FAH-production will be fairly low this year.