Couple of nOOby questions

[Big.Wayne]

Greetings fellow folders,

I am slowly getting back to grips with Folding@home and have been playing around with the various different bits of software and options, I have a few questions . . .

Windows, SMP vs Multiple Console Clients

I have no problems getting the single console client set-up, I have done this for both cores on my Wolfdale HTPC. I had a go at the SMP/Deino software but found it a bit complex, I don't have a password on my account as I prefere my PC to boot straight to desktop with faffing around, anyway the question is what advantage is there to running the SMP instead on multiple single console clients?

At the moment two separate consoles installed as a service seem to be working ok?

-advmethods

This is the same as using the -advmethods flag. This will allow you to grab workunits with bigger requirements (often with more points value) than the normal units. Keep in mind this will often have a higher risk of failure so use this flag only if you are ready to accept the risks.

Do you guys use this option? . . . I thought I'd give it a spin to see what is does, it's download some MAHOOSIVE WU called Translocon_GLX2 which seems to be taking forever to crunch . . .15 hours and I'm nearly halfway lol!

Not sure if its worth it point wise or not but I'll leave this -advmethods thing on for a week just to see how it acts etc. I was interested what your viewpoint on this advanced thing was?, I've also used it for the GPU client below, is that ok?


Windows, GPU Console Client

I was using the nOObling system tray GPU Client for a week and thought I'd give the console client a spin, all seems fairly straightforward but I was wondering, if I was to install a second graphics card how then do I set-up another GPU console client? . . . I don't see a way to specify which card uses which client etc

Overclocking your Folding GPU

The first week I was folding on the GPU using the system tray client I left the card at stock and all was fine, this week I thought I would apply a known working GPU overclock to see how that affected PPD . . . seems like the overclock isn't as stable as I thought lol!

[00:58:59] Starting GUI Server
[01:05:05] Completed 1%
[01:11:13] Completed 2%
[01:13:03] mdrun_gpu returned
[01:13:03] NANs detected on GPU
[01:13:03] Folding@home Core Shutdown: UNSTABLE_MACHINE
[01:13:07] CoreStatus = 7A (122)


[01:13:21] Starting GUI Server
[01:19:33] Completed 1%
[01:25:36] Completed 2%
[01:28:00] mdrun_gpu returned
[01:28:00] NANs detected on GPU
[01:28:00] Folding@home Core Shutdown: UNSTABLE_MACHINE
[01:28:04] CoreStatus = 7A (122)


[01:28:19] Starting GUI Server
[01:34:26] Completed 1%
[01:40:29] Completed 2%
[01:46:34] Completed 3%
[01:47:47] mdrun_gpu returned
[01:47:47] NANs detected on GPU
[01:47:47] Folding@home Core Shutdown: UNSTABLE_MACHINE
[01:47:50] CoreStatus = 7A (122)


[01:48:04] Working on Protein
[01:48:04] Starting GUI Server
[01:54:15] Completed 1%
[02:00:21] Completed 2%
[02:06:28] Completed 3%
[02:12:34] Completed 4%
[02:18:39] Completed 5%
[02:24:45] Completed 6%
[02:30:51] Completed 7%
[02:36:56] Completed 8%
[02:43:02] Completed 9%
[02:49:33] Completed 10%
[02:55:39] Completed 11%
[03:01:45] Completed 12%
[03:07:51] Completed 13%
[03:13:57] Completed 14%
[03:20:04] Completed 15%
[03:26:11] Completed 16%
[03:32:18] Completed 17%
[03:38:25] Completed 18%
[03:44:32] Completed 19%
[03:50:39] Completed 20%
[03:54:56] mdrun_gpu returned
[03:54:56] NANs detected on GPU
[03:54:56] Folding@home Core Shutdown: UNSTABLE_MACHINE
[03:54:59] CoreStatus = 7A (122)

Inbetween each failure above I kept downclocking the card to improve stability but it kept finding errors, in the end I have just put the card back to stock to get a few units crunched error free but I am suprised how sensitive folding is to overclocking. I was curious to know what your experiences have been with overclocking your folding GPUs.

 

[ChrissyT88]

Hi Big.Wayne

1. Smp Client

The main advantage of the SMP client is that a single client will use multiple cores, which means less fiddling with multiple clients. On the other hand, it is a piece of BETA software, so expect some problems occasionally. In general, the SMP client also generates a lot more points than multiple single core clients (although this is not always the case). Not sure what sort of ppd you currently get from multiple single clients, but on a dual core with a reasonable clock i would expect around 1500 ppd from the SMP client, although im not too familiar with the windows client. If you are lucky with the units you get from the standard client, you can get more ppd than SMP clients, but i think its pretty rare! So, in short, less clients to handle (potentially less hassle), generally more ppd.

EDIT: Just seen you have a wolfdale with some pretty good clocks - will probably get more than 1500 ppd from the SMP client. Also worth mentioning that the SMP client deadlines are a LOT shorter - a few days to a week compared with months for some of the standard client WUs. If your PC is not on that much, you may struggle to meet the dealines of the SMP client, although 8hrs+ per day and i doubt you will have a problem.

2. Advanced Methods Flag

Personally, i do use this flag. Its meant to give new experimental units, but i can't say i have ever had a problem with any unit i have recieved that is flagged under advanced methods. I dont know how many of the units are flagged, but the new 59xx series of nvidia units are, and these give pretty good ppd. I would say go for it, unless you are really worried about an experimental WU going wrong.

3. Multiple GPU Clients

Not an ATI man, but with nVidia cards its easy. For the console client, just install as normal and make a copy of the installation folder and rename it (GPU2 for example). When you make the shortcuts to run the .exe, add the -gpu x flag to instruct the client to target a particular gpu. You need to replace 'x' with '0' for the first card, and with '1' for the second card etc (but make sure you start with 0!). As i say, not too familiar with ATI folding, but i presume the client set up is the same. Other than that, you dont need to do anything special - run the client as normal and add the -gpu x flags to target the right gpu, and you're away. You will need to make sure that crossfire/SLI is disabled (also PhysX on nvidia hardware) otherwise only one card will fold - SLI/crossfire folding is currently a no go.

4. GPU OC

Not dabbled tool much in this area - only ever used mild overclocks, but some units are definately more sensitive to overclocking. On the other hand, some people have monsterous clocks on the GPU and the units fold just fine! The shaders are the only thing that really make a difference to F@H perfromance, so maybe try reducing your core/memory clocks and just push the shaders up until the client gives errors? Be prepared for some to error if they introduce a new unit that is sensitive to overclocks, but i think with some modest clocking (if you could bare such a thing!) will be fine! Of course, folding on a team other than Team 10 severly reduces your chances of being successful with the OC!

 

[Big.Wayne]

Excellent ChrissyT88, many thanks!

Yup your spot on with your PPD estimation, a Wolfdale 3.6GHz/6MB is worth about 1500 PPD using two separate Console/Service clients and it's a real doddle to install or uninstall. In comparison the Windows SMP client is well fiddly with having to customise the services option and edit the Windows registry etc

I think I may give it another go after I made a real pigs ear of the first attempt (I ignored the Must Have Windows Password bit). It's not so much the extra PPD but the fact that it's a technical achievement just to get the damn thing installed, configured and running heh! . . . I want my Windows SMP stripes!

The Multiple Single Consoles works great, I don't have to do anything, the HTPC is on 24/7 and even with both cores folding away the PC is still able to deal with HTPC duties, plays DVD's/iPlayer etc perfectly.

Thanks for the GPU tips, I think I got it but won't know until I try, so your saying add that flag to the -configonly shortcut and the rest will be taken care off

-configonly -gpu 0

-configonly -gpu 1

Sorry I'm quite new to the flags thing and have used -configonly only.

Re the overclocking on Radeons as far as I know you can only adjust either the GPU speed or Memory speed, no separate shader clocking. I guess I will have to slowly bump the clocks and see if it was Memory or maybe the GPU?

I intend to try all the versions (Linux, Virtual Machines etc) for the crack, I'm finding the whole experience interesting and I hope Stanford Uni gets some new software sorted in the future, a single client that can auto configure itself to use multi CPU's and whatever GPU's are installed, that would be handy!

I have one last question heh but it's not that important . . .

Windows 7, GeForce GPU and Radeon GPU folding on same machine

One of the guys told me that its possible to have a Radeon (with Cats) and a Geforce (with Dets) installed and running under Windows 7, I just wondered if any of the crew here are GPU folding like this?

 

[ChrissyT88]

With the flags, im sorry i should have made that a bit clearer. You just need to add '-gpu x'. Only add the config flag to enter the config/setup mode. I did word that very badly and i'll edit it out in case anyone else reads it! To make things more confusing, you can also add any flag via the config mode (!) - somewhere under the advanced options it asks for any additional client parameters, and here you can enter '-gpu x'. I guess this cleans up the shortcut a bit, but they both achieve the same things.

I also forgot to mention that (on Vista at least) you need to extend the desktop onto a second monitor, either by using a second physical monitor, or a 'dummy plug'/separate input on your normal monitor/tv. Im not sure if you have to do this on windows 7 - they do seem the have cleared things up a bit in 7, but i wouldnt be surprised if this quirk still remained.

Its quite complicated getting everything set up, and its hard to put everything into words. It really has taught me a lot! I certainly would not have spent as much time in Linux/used VMware ever if it were not for f@h. Running Linux virtual machines is definately the best points wise, and i would say that generally the Linux SMP client is a little more stable + a heck of a lot easier to set up! Plus its what all the cool kids do . If you go down this route you need to use VMware (either server or a paid for version) or Microsofts Hyper-V if you have access to a copy of Server 08, since virtualbox etc wont let you have more than one virtual processor (and the SMP client will not be happy about that! ).

With regards to the combined NV + ATi folding - i have no idea! Be sure to let us know what you find out though!

 

[Big.Wayne]

you can also add any flag via the config mode (!) - somewhere under the advanced options it asks for any additional client parameters, and here you can enter '-gpu x'.

Yup I know the exact stage, its right near the end of advanced options, so just enter -gpu 0 or -gpu1 got it!

I will have to find out what the system thinks GPU0 or GPU1 is, I guess the primary PCI-E slot would be GPU0 but will find out soon enough!

I've got some basic experience of VMware, think the whole Virtualization arena is very interesting, I was hoping to use it to test out drivers and stuff but it doesn't seem to allow direct access to hardware yet so no joy, great for software testing though.

I have no idea how Folding@home runs better on Virtual Linux . . . I'm guessing that either its been compiled by the kings of geekdom and is uBer well coded or perhaps there is just something bad about Windows? It can't be the Windows overhead because thats still there when running a VM, I dunno heh!

Thanks again for taking the time to clear up those points, cool beans!

 

[ChrissyT88]

Basically it comes down to the cores available for each client. The SMP client was developed first for linux, and then ported to windows (i think its something to do with the implementation of MPI that makes this easier in Linux). As such, the Linux SMP client is not a BETA, whereas the windows client still is and is developing much more slowly. The Linux SMP client is, however, much more advanced and can use the much more efficient A2 core (as opposed to the A1 core for windows SMP). Its much better at using processor resources and so folding on linux is better than on windows.

You can set up a virtual machine for folding on linux really easily - just create a VM, install Ubuntu (or any flavour of Linux), install the client (MUCH easier than the windows install) and then you're set. The overheads of VMware server are really low. I did some comparisons between native folding on Linux (4 cores of a Q6600) and on Linux VMs within windows (each running on half of the quad core) and the points difference is really small. Gfx card folding is much more hassle free on windows to boot, and since you can set the VMs to a low priority, they have almost no impact on normal windows use (unless you want to do something really RAM/processor heavy). Id recommend giving it a go - as i said, its taught me loads about virtualisation, got me more involved in Linux, and gets more ppd on top of all that!

Must get on with some work now - lousy exams! I like talking about F@H much more (i actually know something about it... )!

 

[shadowscotland]

Hi wayne - stealing our secrets again are you...

anyway check out SB's smp folding guide linky has lots on vm/ubuntu and stuff on smp clients in general (not 100% sure it totally up to date)

Nvidia gpu folding is the current king, but all the newish cards for both companies can be used. Part of the standard driver set for ATI now too