Folding@Home

[P5-133XL]

I was not trying to get you to uninstall the SMP client, but rather the GUI GPU client (if you want to run the console GPU client). I think they are conflicting with each other.

The other alternative is not to use the GPU console client and go back to using the GUI GPU client and simply do a date change to before 2/1/08 before running it and then change the date back after it is started.

 

[Fushigi]

I know, but as I said the SMP client wasn't being used so there wasn't any reason to keep it around anyway. The GUI GPU client has no uninstaller. It runs like the 5.04 console version - just run the .exe and it loads up.

And right now I am doing the date change cheat. That's a minor annoyance, but I can live with it for the moment as I rarely reboot.

 

[Fushigi]

Bozo, the SMP client uses .NET: http://folding.stanford.edu/English/FAQ-SMP

 

[Bozo]

Bozo, the SMP client uses .NET: http://folding.stanford.edu/English/FAQ-SMP

Thanks,

Bozo :joker:

 

[Fushigi]

Forgot to say it a couple of days ago, but congrats to Pradeep for hitting the 500K mark. Will is approaching that milestone as well & should hit it in the next few weeks. About that same time Mark should surpass the 4 million point mark.

I use the ExtremeOverclocking stats page and noticed that the person who runs it is soliciting donations for a beefier server to process the ever-increasing stats database. Looks like they should get sufficient funds without any problem, but I thought I'd pass it along in case anyone else here would like to donate to the cause.

 

[Pradeep]

The PS3 is helping a lot with my point production. Far outweighs my old p4 2.4

The kids need a couputer or two so it may be time to build a new box, and trade the p4 down to them.

Edited to add: The 40GB PS3 costs $399 and consumes 135W when folding, with an output of 900 points per day.

 

[Will Rickards]

So my SMP client keeps dying with I/O errors in windows XP. I can't take it anymore. I think what I want to do is boot linux and use that for folding and then boot into windows when I need to do other stuff. I could surf the internet on the linux install and that is what I use it for most of the time.

Maybe I should consider linux full time...

 

[P5-133XL]

File I/O errors should not be common, unless there is something wrong with the config or the file system.

If you'd like, try running Linux in a Virtual Machine and then run the folding SMP client in the virtual machine. That way you can have Linux and Windows running simultaneously and can switch back and forth whenever you want.

 

[ddrueding]

What is the performance hit of running it in a linux VM?

 

[P5-133XL]

It's kinda hard to tell because Linux running in a VM has a clock issue (the Linux system clock in a VM runs faster than the Windows system clock) that invalidates the frame time comparisons. My guess is that there is aprox. a 10%-15% decrease in performance running in a VM.

Note, some people have reported a significant increase in total points per day (50%+) by running two instances of the SMP client. Now there are issues with running multiple SMP client instances within Windows, but not if you put them in separate VM's or a Windows and a VM configuration. You also have to have a machine that can easily finish the WU's in at least 1/2 the deadline because performance goes down quickly when you start getting 0 points because the WU's are returned late...

 

[Fushigi]

Four Million Points for Mark!

 

[P5-133XL]

Thanks

 

[Fushigi]

You're welcome.

BTW, after a recent reboot the old GPU beta wouldn't even work with the date change trick. The new system tray one still doesn't run for some reason - it starts but never processes anything - so I switched to the beta console GPU. That's running fine.

FWIW, my current Folding farm consists of 16 cores + 1 GPU. CPU clients vary from the SMP to the V6 beta to the 5.04 traditional. I'm considering taking my last two single-core PCs (server & wife's workstation) to quad-cores, but it depends on the deals I find.

 

[Fushigi]

At the moment, what I'm trying to do is replace 9 machines with 2 as an summer energy saving and a summer temperature comfort goal. So, I'm going with two quads that will run multiple VM's rather than having multiple machines dual booting to different OS's.

Current farm:
My main PC: Q6600 on Gigabyte (see building new PC thread from May '07). 4GB, Vista, 3870. Runs 3xConsole + 1 GPU2.
Server: Q6600 on ECS 671T. 2GB, Vista. Runs 3xConsole + 1 System Tray. I use the term Server loosely; it really is a repository for sharing files & backing up our local documents. That's why it's running a desktop OS & not something more robust.
Dedicated Folder 1: Q6600 on ECS 900T-M. 2GB, Vista. Runs 3xConsole + 1 GPU2.
Dedicated Folder 2: Q6600 on ECS something. 2GB, Linux. Runs SMP.
Wife PC: Athlon XP 2700 on something old. 1GB, XP. Runs 1 5.x Service.
Laptop 1: Dell D830 w/C2D T7800. 4GB, Vista. Runs 2xConsole.
Laptop 2: Dell D800 w/PentiumM 2GHz. 1GB, XP. Runs 1 Console.

Plan:
Server: Replace System Tray with GPU2 on a 4xxx Radeon once they're released.
Wife PC: Replace with Dedicated Folder 2's hardware by doing a wipe/reload with XP. Will run 3xConsole + 1 GPU2.
Dedicated Folder 2: Rebuild with Q9450. Will run XP or Vista, probably XP. Remaining system specs TBD but leaning towards MSI X48C for the motherboard. It looks the most reasonable (along with the Intel DX48BT2) for a DDR3 platform with dual PCI-E x16 slots. Will eventually run 2xConsole + 2xGPU2 on 4xxx.

Originally I was going to shut down my old server (an Athlon XP), replace it by rebuilding the Linux box, and still replace the Wife's PC with a new C2Q leading to one less system but things didn't work out so instead of reducing my box count it's staying the same. However my core count is going up by 6 (2 single cores becoming 2 quads) and I'm deploying GPU wherever I can; eventually in all non-laptop systems.

The machine I'm currently setting up is:
MB: Ga-X48-DS4
CPU: Q9450 - OC'ed to 3.2GHz (It seems to be very stable at that speed getting aprox 3200PPD with SMP folding)
Video: Sapphire HD3870
RAM: 4x2GB - Geil PC2-1066
5 small HD's - One for each VM (Vmware Server Beta 2.0)
Power Supply: PC power and cooling Silencer 750
Case Antec P80

Base OS: Microsoft 2008 Server (X64).

As I'm deploying GPUs and looking at potential power draw that PCP&C 750 does look good. If you haven't bought yet there's currently a $30MIR at NewEgg.

I'm surprised you're going with discrete HDs. From a power and cost perspective I'd think mirrored 500s or 750s would be better.

My expectation is that my Power bill will go down significatly, my folding PPD (with two SMP + 2 GPU2 clients) should stay the same or go up slightly. I have no idea yet, what I will do with the surplus machines

If my power draw for the summer become overwhelming I'll shut down the Dedicated Folders. Those boxes and my Server are in the basement, though, so at least there's virtually no impact to my AC bill.

As to my displaced Athlons, I've no idea what to do with them either. If I can find a good home for them, I'll likely just give them away. There's really nothing on them I want anymore. Not even the HDs & opticals as I will
have moved entirely to SATA. One thought was to keep one of the systems as a Linux playground, but while playing with Kubuntu was OK I really didn't care for it all that much.

 

[Fushigi]

Congrats to sf-[SPLASH]-gh for passing the million point mark!

 

[P5-133XL]

This really is a better place for this conversation than the for-sale forum.

Interesting, and you are definately creating a folding farm. If and when Stanford improves the SMP client to better mix with the GPU2 client may make it worth your while to SMP fold again.

You brought up a question about my machine. Specificly why lots of little drives rather than a few much larger drives. I already have accumulated them over time so I don't need to spend money unnecessarily. While drives do generate heat and fewer is better than more on the scale of power/heat savings that I'm striving for, the HD's are small potatoes. I could easily see replacing them with a larger drive in the future. Mirroring is not a an issue, everything will be backed up and easily replacable. There rarely is a situation I need 24x7 reliability for this lab.

As to the future, I could easily add more video cards because The MB's have two PCI-E X16 slots so I could easily see folding two GPU's per machine and doing less SMP folding.

In adition to the dual PCI-e, I also specificly bought those MB's because I intended to OC: I was just plain annoyed at getting 1600-1800 PPD on my Q6600 where others were getting 3000-3100.

Yes, I have already purchased all the equipment for the entire setup. I will see if the PC Power & Cooling power supplies MIR's dates will include when I purchased it. Since the purchase is so recent it may be possible: Thanks for the heads up.

In the process of fiddling I did discover that the XP X64 OS has the same problem with mixing SMP and GPU2 clients as Vista. You will need VM's to SMP fold in addition to GPU2 folding.


Also, ATI/AMD is no longer supplying new video drivers for Windows 2000 or Windows 2003 and specificly for the x1900/x2900/x3900 cards -- No GPU2 folding for them :sad. I contacted ATI and they sent me a conversion utility so I could install the newer drivers but GPU2 folding didn't like it (constant client-core communication errors). Now that Microsoft has EOL'ed XP I am now starting to wonder how long before it gets the same treatment.

 

[P5-133XL]

P.S. The MIR didn't qualify by two days.

 

[Fushigi]

Re: ATI drivers. While I'm a little surprised about W2003 losing support, W2K doesn't surprise me at all. I imagine XP drivers will be available for some time to come. Realistically they'd be dumb to stop XP driver offerings as long as XP had a 15% or higher market share. IIRC XP came out in 2002 so it's reasonably impressive that 2K support lasted for about 6 additional years.

Re: SMP v. multiple consoles. I'd prefer the SMP client but until Stanford improves the client I'm pretty stuck. Ideally I'd like to see a single client in which you could configure the number of cores. One better would be a unified CPU & GPU client but that's probably not worth the effort.

I would like to OC more than I do (which is very little indeed) but really haven't been bothered to take the time to figure out what these cheap board & my sole more expensive one can reliably handle. I sometimes go days at a time between checking on some of the machines so stability is paramount.

I'll be monitoring the sales to start buying the parts for my Q9450 build. While I could use it in an existing box in the mean time I don't think it'll take that long to get the parts so I won't bother. It'll probably be done in the next 3-4 weeks.

 

[P5-133XL]

My ECS board wouldn't OC at all. Even going from Stock 2.4GHz to 2.5GHz, made Windows intermittantly unstable. However, at stock speeds, I really couldn't complain about its stability.

One thing I did consider, with all these X2's that I'm de-commisioning is to buy a GPU2 capable video card for each and just GPU2 fold and I may choose to do that if I haven't gotten rid of the machines by the time Winter arrives and I need more heat. However, summer is not the time to add alot more heat to my place. I did one summer with all these heaters running and when the outside temps get 80+ it gets really very uncomfortable inside.

 

[LiamC]

Mark, how much better/faster/ppd is the Q9450 than the Q6600 in SMP?

 

[P5-133XL]

I actually haven't personally compared a stock Q6600 (@2.4 GHz) to a Stock 9450 (@2.66GHz). My Q6600 won't OC in the MB it is in and I have only run SMP on the Q9450 OC'ed. I was able to safely OC to 3.56GHz, but for stability purposes I deliberately dropped it significatly. I may fiddle some more however. What I'm getting for the Q9450 @3.2GHz is 3200-3300 PPD My Stock Q6600 was getting between 1600-1800 PPD. Most people that OC the Q6600 get around 3000-3100 at somewhare around 3.6GHz. If you'd ask, I could run the Q9450 stock for a while and see what I get and you can compare better. The alternative is that you can wait till I start OC'ing the Q6600 and you can compare the OC'ed PPD's as opposed to the stock PPD's -- Your choice.

 

[LiamC]

No, the figures you've given are enough, it's obviously significantly faster clock-for-clock. Thanks. But if you'd put up the o'clocked Q6600 scores when you get them, that would be informative and appreciated.

 

[Fushigi]

So my Q9450 is built & installing Kubuntu as I type this.

Hardware:
Core2Quad Q9450
MSI X48C motherboard
OCZ DDR3 10666 Low Latency RAM (2x1GB)
Corsair TX650W 650 Watt PSU
Samsung 500GB HD
Sapphire Radeon X1950 Pro
Philips SPD2513BD DVD-R
Cooler Master Elite 330 case

Issue with the X48C: It did not assign RAM or CPU voltage correctly with the AUTO setting. CPU was overvolted at 1.25 (should be 1.15V). RAM was way undervolted at 1.5V (should be 1.8). Funny thing was that it would still run memtest86+ from the Kubuntu install disk just fine but would lock up installing Kubuntu itself.

The install & setup of the Folding SMP client will run at stock speeds to begin with. I'll toy with overclocking later.

 

[Fushigi]

Running the newest version of memtest86+ reveals errors. I'm working with the OCZ forum folks to find stable settings.

 

[LiamC]

TGDAily is reporting that the NVIDIA client is going to be some serious whoop-ass.

..."The unveiling of the client is set for the next week as part of the launch of Nvidia’s GT200 GPU series. Owning such a card will have its benefits in Folding@Home and will outrun Radeon 3870 cards. The new GeForce cards are expected to hit more than 650 nanoseconds of protein simulation in a single day, while the Radeon HD 3870 is stuck at about 170 ns. The Playstation 3 is able to produce "only" 100 ns of simulation, while a quad-core CPU creates an output of just four nanoseconds. For those who are keeping count: The GeForce GPU will be about 163 times faster than a quad-core processor in this specific application."...

http://www.tgdaily.com/content/view/37931/113/

PPD

 

[Fushigi]

One stick of the memory is bad but the machine is up & Folding onthe other stick. An RMA has been submitted. I will say the OCZ forum is a good place to go if you have one of their products. It's monitored by OCZ staff and when we determined the RMA was needed they gave me the info to bypass have tech support also deal with the issue.

Liam - Interesting. I will be in the market for GPUs for the Q9450 setup eventually and may give this consideration.

 

[P5-133XL]

I have now switched to my summer folding setup. I specificly changed it because during the summer months it gets outright very uncomfortable inside with my previous setup of nine machines. So I have only two machines running, though I tried my best to keep my folding performance at or above my previous levels.

Machine #1:

Intel Q9450 (OC to 430x8=3.44GHz); 2x2=4GB Ram; Gigabyte GA-X48-DS4 Motherboard; PC Power and Cooling Q750 PS; Sapphire 3870 video card; Windows Server 2008 X64 (SP1).

I am folding using two instances of Ubuntu 8.04 under Hyper-V beta (M$ virtual machine) each one uses two proccessors and is running the SMP client producing roughly 3000PPD (p2662) each for a total of 6000+ PPD for that machine. The machine uses 232W while folding.

Machine #2:

Intel Q6600 (OC to 400x8=3.20GHz); 2x2=4GB Ram; Gigabyte GA-X48-DS4 Motherboard; PC Power and Cooling Q750 PS; Sapphire 3870 video card; Windows XP Media Center Edition (SP3).

Here I am folding two clients the GPU2 and the SMP. I am getting 2200PPD for the GPU2 client (p4731) and 1900PPD for the SMP client (p2665) for a total of 4100PPD. The machine is using 307W while folding.

So in theory, I should be generating around 10,000 PPD with the two machines which is significantly higher than I was getting with the 9 machines.

 

[LiamC]

Mark,

I've been reading up on the F@H forums about the Q9450/Q9300/Q6600 and the consensus seems to be that the Q9450 is only about 5% faster clock-for-clock. Yours seems to be 60% faster (comparing the SMP ppd--1900 v 3000 for a single client). What gives? Your Q9450 is clocked a little higher, but not much. OS?

 

[P5-133XL]

There are several factors. The first is OC'ing. Next is Windows vs Linux. Third is processor scaling.

I find that for a Q9450 I'm getting a very good OC. Most Q9450 chips are not stable beyond 3.2GHz in a non-folding enviornment and folding is very stressfull on the chip. My Q6600 was not stable beyond 3.2GHz, while folding, and it is supposed to be a very much better OC'er. I have, for compareson purposes OC'ed to 3.2 (8x400) vs 3.44 (8x430) the values I got were 2700PPD vs 3000PPD, per instance, so there is a significant OC factor.

Next is Windows vs Linux. The Windows SMP client is just plain less efficient. Also, the different OS's get different WU projects and Linux just plain consistantly gets better performing projects.

Third, has to do with running two instances, as opposed to one instance of the SMP client. The current SMP clients do not scale well at all. When you go from two processors, to four processors the PPD does not go up much. So running two instances, each using two cores, can almost double PPD which more than compensates for the loss in productivity that comes from running inside a virtual machine. From what I can tell, from the Hyper-v performance monitors, I'm losing around 6% efficiency by using a virtual machine.

Do note, I did alot of effort to optimize my PPD efficiency on that machine. I tried a lot of different combinations till I got where I am. I have yet to test one major senaero -- running 4 processors in linux but two SMP client instances within that one virtual machine. I'm not sure if it is even worth the effort considering the productivity I am currently getting.

One thing of interest to me, and I don't know why yet is that FAHMON is consistantly telling me that I should be getting 10,100 PPD between all the clients on two machines but EOC (Extreme OverClocking) is only reporting around 6500PPD. When I actually do the calulations manually from frame times FahMon appears to be correctly reporting. Now part of it may be that while I was experimenting I was killing a lot of WU's which kills my EOC results. I'm going to give it a week or so, to stablize, and then if EOC does not come much closer to matching FahMon, I'll record all the WU's completed from the various logs then contact Stanford and have them crosscheck to see what was turned in for what values. I really don't want to do this, because there are so many WU's and I don't want to cause any bother but I think it is important to pin this discrepency down. If it is affecting me, then it is likely to be affecting lots of others too. This observation is not new. I was seeing the same when I had all nine machines working. Back then EOC was reporting 4500PPD while FahMon was reporting 6500PPD). It is just a pain to go through all the logs and all the WU's so I was not willing to do it.

 

[P5-133XL]

After reading the above, I wanted to append the following, but the editing time limit had expired.

When I was running Linux with Vista dual-core virtual machines, the Linux was still getting 3000PPD while Vista was only getting 2400PPD.

If you PM'me with an Email address, I'll send you a screen shot, rather than hosting it on the web. So you can see what FahMon is reporting and know that I'm not pulling your ear with the numbers I'm quoting.

 

[P5-133XL]

It took quite a while, but my EOC stats are starting to approach my FAHMon stats.

All those running Linux SMP: It is worthwile to manually replace the FAHCore_A2.exe by deleting it (the new 2.0 version will auto-download when folding restarts). It is also good to add the -advmethods parameter because then you too will start getting WU's that use the new A2 core. For Quads running a single SMP folding client, it will almost double your productivity.

Windows fan's, will have to wait a while longer because the Windows A2 Core has not yet been released.

 

[P5-133XL]

Almost all the beta SMP/GPU2 clients expire 8/2/08. Please download new clients as needed. As long as you don't shut off the client, it will continue working even though not updated.

 

[LiamC]

Congratulations Mark on hitting 5 000 000 points

 

[P5-133XL]

Thanks

 

[Splash]

Congrats to sf-[SPLASH]-gh for passing the million point mark!

And, congrats to CougTek for joining the SF millionaire club.


>>> 1,549,264 <<<

 

[CougTek]

Thanks.

 

[P5-133XL]

And we have Fushigi who today broke the 4,000,000 mark. Congrats.

 

[LiamC]

Well done Gentlemen

 

[P5-133XL]

LiamC: Cudos for reaching 3,000,000

Congrats Liam for getting to 3M!

 

[CougTek]

Milestones celebrations will be rarer. Both Fushi and LiamC have considerably slowed down their productions. I suspect LiamC's reason is because it's the middle of Summer down there and he tries to limit the heat production made by his computers. Fushi, I haven no idea. I hope it's not something bad like a dead computer.

_________________________________________________

Server 2008 is a lot less optimized for multiple multimedia rendering like playing several videos at the same time or doing anything graphical while the GPU Folding client is running, than Vista 64bit can be. The unit output is equal though.

Interesting note : there's no GeForce 8800GT driver from Nvidia for Server 2008. The one for Vista works perfectly.