Folding@Home

[Splash]

Well, I should march past "i" in the next few hours, then begin my conquest of SteveC and Mercutio over the next couple days.

Since I've started effin' @ home (cough), the F@H donater count has gone from around 576000 up to the current 585300!

 

[P5-133XL]

F@H is attracting new donators because of the recent marketing of their GPU client: No one else has one.

 

[Splash]

F@H is attracting new donators because of the recent marketing of their GPU client: No one else has one.

I think the vast majority of new donators are just donators like me with a conventional X86 CPU. They seem to be high schoolers and whatnot -- just people coming out of the proverbial woodwork. Donator count is at 586415 at this hour, which is up over 1100 since my last post.

Nonetheless, the ATI GPU donators are making a huge mark, such as ATI Technologies. For 2 or 3 weeks solid ATI Technologies were top dog team at climbing the ladder (i.e. -- F@H team rank). They started off at around #51000 and are now at #500 with almost a million points collected.

http://folding.extremeoverclocking.com/team_list.php?s=&srt=9



But, they are making their huge advances up the team list using somewhere over 400 GPUs or CPUs (very likely the former at 99:1).

http://vspx27.stanford.edu/cgi-bin/main.py?qtype=teampage&teamnum=51394

 

[P5-133XL]

The ATI team had the benefit of being in the "official instructions" in how to configure the GPU client with the ATI beta drivers. Part of the instructions was to put in the ATI team number.

But the number of people that are actually using the GPU client is still very small (less than 500 active GPU clients). However, the amount of publicity that GPU folding garnered was very large. This marketing produced as you say "people comming out of the woodwork".

The impressive part is that those 500 active GPU clients are now producing aprox 15% of all the terraflops of all 186,000 computers currently folding. When the number of GPU folders get to aprox 3000 then they will be out producing all of the conventional folding computers. See: Client stats by OS.

 

[Splash]

In the past few days, I've managed to breeze by SteveC, Mercutio, miksmi, and "i" up to #29 on the StorageForum_net member point ranking list. I'm now pursuing F**KBAG. I also noticed that "Gregory_Santilli" (#25) has once again come alive after a l-o-n-g period of hibernation.


http://folding.extremeoverclocking.com/user_list.php?s=&t=10047


I'm I wrong in assuming that sometime in the past it was significantly easier to make points than it is now?

 

[P5-133XL]

I'm I wrong in assuming that sometime in the past it was significantly easier to make points than it is now?

While occasionally, there are a series of good WU's that can produce very good results your assumption is generally wrong. Newer machines produce far more points than older machines ever did. The WU points are assigned via a reference machine (2.8 GHz P4 w/o SSE == 100PPD). As better, more capable CPU's have been created the points have scaled upwards appropiately.

So CPU's with faster floating point processors or faster SSE implementations can get far more PPD -- i.e C2D's when matched with WU's that use those functions. A high end C2D with an appropiate WU (1495-1498) can exceed 2000PPD while a P4 would never get beyond 350PPD no matter what WU or configuration.

 

[Handruin]

I decided to turn my machine off this week because I was away.

 

[CougTek]

Total points during the last week :

• 
  F. Bostrom : 6623
  Bill C. : 6608
  CougTek : 6537

We three are really toe to toe. Unfortunately, I'm the last one of the trio. For February, Bostrom is ahead by a little more than a thousand points and I'm (currently) less than two hundred points in front of BillC. I used to race against Groltz, but since he deserted the team, I had to find new opponenents. We miss your ~650ppd, Pullayop (or whatever it is spelled) guy.

I lost a Core 2 Duo E6400 two weeks ago (sold the motherboard) and I should lose my X2 3800+ this week or the other because I sold the entire system. Because of this, I'm pretty sure I won't walk on the second or even third step of the podium for February. Hopefully, I'll have enough money to buy a motherboard to bring back the Core 2 Duo in (and kick your sorry arses in March!). The rest of my pathetic fleet is composed of three P4, one feeding an X1600Pro with it's second hyperthreaded phantom core, and one Athlon XP 1800+. As I said, pathetic.

 

[LiamC]

Coug,

what GPU are you using for folding? Only the X1600Pro?

 

[CougTek]

Yes, only this low-end mickey mouse 3D wannabee card. Still, it adds ~220 points to my daily production, which is more than any other processor I have.

 

[P5-133XL]

This weekend, I'm doing some revamping and setting up the 64 bit SMP Linux client on two more machines, for a total of three. According to FAHMON, on each machine I'm getting 1,000PPD for a SMP Linux client as opposed to aprox 300PPD for two Windows clients or even 150+500 = 650PPD for a Windows + GPU client.

 

[LiamC]

Decided to follow up on suggestions and purchased an X1650XT AGP (HIS) for the P!!!. This has 24 pixel pipelines vs 12 in an X1650 Pro, and 36 in the X1950 Pro.

The board is AGP 4X. Good news, the board works fine in the P3V4X.

The bad news is that I get an "mdrun_gpu returned 99" error right on start up. I've posted on the folding forum:

http://forum.folding-community.org/ftopic18075.html

but so far haven't received in meaningful feedback/assistance.

CATALYST version: 6.10
2D driver version: 6.14.10.6441
Direct 3D version: 6.14.10.0440

This is a clean install of XP, chipset drivers, DX9.0c & Cat drivers . Sigh.

Might just wait for the 7.2 drivers to be released.

 

[P5-133XL]

You are using a valid driver set: the 6.10 or the 6.11 should do fine. Ask CougTek, He has a x1650 too and thus can undoubtably hand hold you through a driver installation.

 

[P5-133XL]

I moved one of my GPU's to another machine and by golly, I can't get it to run either: Same problem as you. It won't initialize the GPU and then kicks out with an error mdrun_gpu returned 99.

Very annoying because I haven't been able to figure out whats wrong..

 

[LiamC]

I moved one of my GPU's to another machine and by golly, I can't get it to run either: Same problem as you. It won't initialize the GPU and then kicks out with an error mdrun_gpu returned 99.

Very annoying because I haven't been able to figure out whats wrong..

Ha! I don't feel so silly now.

 

[CougTek]

...I'm getting 1,000PPD for a SMP Linux client...

What's the specifications of those 1000ppd systems?

Ask CougTek, He has a x1650 too...

I wish. As stated 5 posts above yours, I only have a X1600 Pro.

1. Download the console GPU client.
2. Create a folder named FAH in your user profile (c:\Documents and Settings\[LIAMC_AKA_BILL]\FAH for instance).
3. Put the GPU console client (named FAH5.91beta3-conle.exe by default) in your newly created FAH folder.
4. Rename the FAH5.91b3Zzzz... executable to FAHGPU.exe for simplycity purpose.
5. Start -> Run -> CMD.exe
6. type cd FAH, press enter
7. type FAHGPU.exe -config, press enter
8. type LiamC (or if you're a really nice guy, you can type CougTek), press enter
9. type 10047, press enter
10. type no, press enter (there are problems with the GPU client when launched as a service)
11. press enter again
12. press enter again
13. press enter again
14. type yes, press enter
15. type yes, press enter
16. press enter again
17. press enter again
18. press enter again
19. press enter again
20. press enter again
21. press enter again
22. press enter again
23. type yes, press enter
24. press enter again
25. press enter again, OR, if you have a dual core or Hyper-threading-capable CPU AND you are already folding with the regular FAH client on your CPU, then type "2" and then press enter.

After each reboot, repeat steps 5 and 6, then type : "FAHGPU.exe" and then press enter.

If the above method doesn't work, address your complain to the Folding@home-on-your-GPU whining department.

* We make no warranty of the success of the above method. Use the above guide at your own risk. You agree not to hold CougTek or any of his associates responsible for any damage or lost caused by following the above method. You agree to give your full cooperation in case of any legal dispute to CougTek and his associates regarding your use of the above guide so that we can eventually sue your ass, suck all the money you (may) have and deprive your descendants of any heritage they might have otherwise gotten from you.

 

[P5-133XL]

What's the specifications of those 1000ppd systems?

I am using AMD X2 4600+ (939); 512MB-PC3200 DDR RAM; Gigabyte K8N51PVM9-RH MB's; Onboard Video; Ubuntu Linux (ubuntu-6.10-desktop-amd64.iso); and F@H's Linux SMP 64 bit folding client.

I installed the FAH client start out by DL'ing finstall by typing "wget http://ra.vendomar.ee/%7Eivo/finstall" from a terminal window and type "chmod +x finstall" to make it execuatable. Once installed, I then install the 32 bit libraries before executing FAH for the first time by typing the following into a terminal window "sudo apt-get install ia32-libs". Then I run Finstall by typing "./finstall smp". The next step is to run ./installService and then ./folding start.

You get about 700PPD running under VMWare but 1000PPD if you actually install the OS to a 2nd HD (I then dual boot, if Windows is needed). PPD measurements come from FAHmon 2.1.5b.3 using a SAMBA share.

 

[P5-133XL]

As a side note Stanford is about to redefine the point values for the SMP and GPU clients. So in the future, your points may vary ...

 

[LiamC]

Are they going to drop the points?

BTW, have you solved your error yet?

 

[CougTek]

I thought SMP folding was preferably for systems with more than two cores, hence why I didn't use it on my X2 3800+ and E6400 C2D. You don't have any issue using the SMP client on such a relatively low-end dual core box? Great, I'll install a 64bit Linux (Kubuntu) right away on my X2 3800+.

Congratulations for passing the two million bar.

 

[P5-133XL]

Are they going to drop the points?

BTW, have you solved your error yet?

I have no idea if they are going to drop or raise. I just know that they are going to adjust, specificly to encourage the clients that have more current scientific value. My guess is the GPU clients will go up and the SMP clients will go down because the amount of flops that a GPU can do is far greater than any of the other clients and the SMP client really can't do more than the aggregate of the individual processors.

No I haven't solved my GPU problem. I have no idea what the problem is.

 

[P5-133XL]

I thought SMP folding was preferably for systems with more than two cores, hence why I didn't use it on my X2 3800+ and E6400 C2D. You don't have any issue using the SMP client on such a relatively low-end dual core box? Great, I'll install a 64bit Linux (Kubuntu) right away on my X2 3800+.

Congratulations for passing the two million bar.

The SMP core was designed for a dual processor with dual cores (a total of 4 cores run using four threads), but works fine with a single processor with two cores. The only real issue is whether the two cores can get the WU's done in time because of the short expiration on the SMP WU's: I'm running 1.7 days for a 4 day WU worth 1,700 points using an X2 4600+ processor.

Stanford is working on a Windows SMP client and that should be out shortly.

Thanks, its been a 4? year effort.

 

[Corvair]

P5-133XL,

Congratulations on surpassing the 2 000 000 mark!

F@H also recently broke through a statistical barrier when they surpassed 600 000 registered users.

As for me, I'm about to creep past the 40 000 mark.


PS: If any of you recently experienced a sudden a gust of wind and a flash of light as you folded along the F@H raceway, that was probably a F@H user named "Burn_In" blowing your doors off. Burn_In started off only 3 months ago (December 2006) and is now ranked at 248th overall with over 2.1 million points! Burn_In is currently averaging 35000 points per day.

http://folding.extremeoverclocking.com/individual_overtake.php?s=&u=204353


http://vspx27.stanford.edu/cgi-bin/main.py?qtype=userpage&username=Burn_In

 

[LiamC]

Burn_In is currently averaging 35000 points per day.

Holy sh!t Batman!

BTW, congratulations Mark.

Also, the 7.2 Catalysts get me past the 99 error, but now I get a 114 error in its place...

 

[P5-133XL]

Holy sh!t Batman!

BTW, congratulations Mark.

Also, the 7.2 Catalysts get me past the 99 error, but now I get a 114 error in its place...

Thanks,

I tried 7.2 on the machine that refuses to GPU fold and it did absolutely squat. It just plain refuses to initialize the GPU. I suspect that it is a registry issue, since I have an almost identical machine (Same OS, MB, Processor, RAM, video card, etc.) and it GPU folds. The drivers are probably detecting the reminants of some old video card. The machine that refuses to fold has been upgraded from actual Windows for workgroups 3.11 ->98 -> me -> NT -> Windows 2000 -> Windows 2003 with four MB changes along the way and lots of video cards too. What it really needs is a format and reinstall but I really don't like doing that to working domain controllers because my history has been wrought with unpleasentries when doing that in the past.

What I am going to do is do some RAM adjustments and see if I can get up to 2GB and then use VMWare to SMP fold on that machine. Currently, SMP folding is producing better than GPU folding ... The only other machine I can place the card into and GPU fold requires a molex->PCI-E adapter and I don't currently have one (or it's lost somewhere in the multiple boxes of spare cables).

For GPU folding, what I'd like to do is add it to my P4 Dells, but the card is not suitable because they need the video card to exhaust heat out the back. This card doesn't and I can't SMP fold in the P4's either. So the P4's are stuck with low performance folding unless I want to buy a bunch of expensive video cards for them.

 

[LiamC]

My experience with the mdrun 99 error is on a machine that was freshly formatted and installed—so there was no registry holdovers. Mind you it was a P!!! so I wasn't expecting too much. Still, it would have been nice. I'm going to try it out on a friends Athlon and see if its an AGP 4X/8X thing. Maybe the card is tolerating the 1.5V but doesn't really like it.

 

[CougTek]

PS: If any of you recently experienced a sudden a gust of wind and a flash of light as you folded along the F@H raceway, that was probably a F@H user named "Burn_In" blowing your doors off.

Someone you know? Lucky bastard he is, 231 CPUs at his disposal. He's a network admin for a big department or what?

Is he the guy who does the job you did some 17 years ago, before you became a paper shuffler?

 

[P5-133XL]

PS: If any of you recently experienced a sudden a gust of wind and a flash of light as you folded along the F@H raceway, that was probably a F@H user named "Burn_In" blowing your doors off. Burn_In started off only 3 months ago (December 2006) and is now ranked at 248th overall with over 2.1 million points! Burn_In is currently averaging 35000 points per day.

http://folding.extremeoverclocking.com/individual_overtake.php?s=&u=204353


http://vspx27.stanford.edu/cgi-bin/main.py?qtype=userpage&username=Burn_In

How about looking at Inpharm or Inpharm with his 152 processors. He just started 10/06 or just 4 months ago to get 12 million points. He's generating 144,000 points per day or over 1 million a week. At my current rate, it will take me a little less than a year to produce 1 million points. To consistantly average 1000 PPD per processor he must be running most of those processors as Linux 64 bit SMP clients.

 

[ddrueding]

He has been getting 1400+ points per WU lately....

 

[P5-133XL]

Well, I've finished with my SMP conversion: I added two more machines today.

I have four 4600+ X2's running Linux SMP natively; two 4600+ X2's running SMP Linux under VMware; two 3.2 P4's running Windows clients and one 3.2 P4 running a uni-linux client. No GPU clients at all.

This is as point-optimized as I can do, with the current hardware. My prediction is that i've boosted my average PPD from 3,200 to about 5,700. Of course, not all the machines are going to be running Linux all the time: As I need to, I will dual boot into Windows.

This has been my first signifigent voyage into the Linux world. What an interesting and time-consuming pain it has been. The biggest time consumer was trying to get VMWare to bind to my Microsoft Proxy Server's ethernet cards without the proxy server blocking everything (I totally failed, and ended up replacing the Microsoft proxy server with Smoothwall, just so I could SMP fold). One just has to deal with the fact that whenever, one does something new it is going to be a time-consuming pain.

Coug, did you actually convert a machine to SMP linux, like you said you were going to? I tried Kubuntu, and came to the conclusion I prefered Ubuntu.

 

[ddrueding]

I prefer Ubuntu as well, though I don't prefer either enough to actually use it for anything. What type of performance hit do you have running in a VM session vs. running it normally? Microsoft ISA blows major chunks; smoothwall is far easier to deal with.

 

[Clocker]

Mark-What version of Linux are you using? Ubuntu all the way?

 

[P5-133XL]

I prefer Ubuntu as well, though I don't prefer either enough to actually use it for anything. What type of performance hit do you have running in a VM session vs. running it normally? Microsoft ISA blows major chunks; smoothwall is far easier to deal with.

It's really hard to judge because the clocks are all messed up running linux under VMWare: You can't trust the time stamps. My best estimate is that I'm running about 30% slower.

The only bug fixes that I saw, when I reasearched it, for the VMWare clock problem don't seem to apply to multiple cores or require me to recomple Linux.

Which version of Linux do you prefer enough to actually do something? I actually fiddled around with Gentoo, but not enough because I never got it to actually work on my HW. I also fiddled with Novell but that was over-kill for my use.

I'm used to the proxy server and didn't normally find it particularly hard to deal with. I guess, I'll see for myself, over time, if Smoothwall is actually better or not. To me, it seems rather limited in what it does. But then I'm using their free product. I'm actually running their new Express 3.0 alpha-test because their release product doesn't work with SATA drives.

 

[P5-133XL]

Mark-What version of Linux are you using? Ubuntu all the way?

Currently all are Ubuntu except for the Smoothwall box and it is only running the uni-processor Linux client.

 

[ddrueding]

Ubuntu is my favorite Linux flavor. Let's leave it at that

Smoothwall 2.0 has a modified version called "superkernel" that includes support for SATA drives and a number of other enhancements. It's more stable than the last 3.0 beta I tried. I'm running it on a Tyan 1U S939 Server with SATA drives

 

[LiamC]

Mark, any word on by how much the Linux SMP points will be reduced?

 

[P5-133XL]

Mark, any word on by how much the Linux SMP points will be reduced?

Not yet: I don't get any inside information that isn't posted in their forums...

 

[CougTek]

Coug, did you actually convert a machine to SMP linux, like you said you were going to?

I tried to, but failed so far. I tried both with a Xubuntu and a Kubuntu (6.10 both) to install Linux on my X2 3800+, but for an unknown reason, both 'buntu versions I tried were incapable to mount the installation media during the installation process. I can boot with the CD/DVD. I can start the installation (I tried both the graphical and the text-only), but for both Xubuntu and Kubuntu, the installation failed the step where it has to mount the media. Both iso images where ok according to the shsum. I burned them using Nero 7.5.

This has caused me a major point hit during the last week. I had two units more than halfway thru under Windows on an external hard drive. I then destroyed the partition on the main hard drive in the system (not the external one where the FAH folders are) in order to install Linux. But since I failed twice to install Linux (add the time to download a full DVD iso to try again after the first failed attempt) and I then add to re-install Windows because it seemed like the only way I could fold on this shitbox, I lost something like 10 hours of processing, for both cores. But it's not only that. I don't know why, but after I unplugged the external hard drive because it wasn't needed under Linux, it no longer works. It no longer powers up. I had two units taking around 40 or 50 hours each to complete ~80% done on it. And it looks like they are both lost because the hard drive kind of like, you know, died. I calculated that I lost around 600 points worth last week because of this.

Criss de calice de tabarnack de calvert! I tried that Thursday evening and I had to work early Friday morning. At the end of the Windows installation and after I realized the external drive had failed, the only thing that folded during the evening/night were the metal arm rests of my computer chair. I bent them to make them in line with the seat. It was much harder to re-bend them in place afterward and the plastic covers are now loose.

I downloaded the iso of Fedora Core today and I'll try again in a few minutes to install Linux on the dually. Since I don't have to work before noon tomorrow, I should have the time to fixe whatever I'll break in the process (OS or furniture). I don't know why, but it looks like Linux doesn't support the SATA controller of my GeForce 6150-based M2NPV-VM. That's the most likely cause why it can't mount the media during the installation.

 

[CougTek]

CPU F@h potential : http://www.techreport.com/reviews/2007q1/quad-core/index.x?pg=10

 

[Mercutio]

Since I rarely make a contribution to this thread...

Those looking for fast and cheap GPU folders would probably be interested in knowing that x1950 Pros are starting to show up at under $150. I hadn't looked at them in a while and thought they were all $250+ cards but new ones are considerably cheaper.