CougTek
Hairy Aussie
Congratulations on passing me. Was it your last unit or do you have another one coming before you sell the system? I want to know if it'll take me two days or a week and a half to retake my second place.
Folding@Home
- Thread starter The JoJo
- Start date
ddrueding
Fixture
I haven't had time to pull the system yet. It may be mid-next week.
ddrueding
Fixture
Done. After being the 122nd most productive user (with only one machine) for a while, this machine is finally going down. Coug only needs to make up a bit over a million points to get his spot back.
CougTek
Hairy Aussie
I was about to ask when it was going to be done. Did you get a good price for it? I hope you didn't lose too much money on it.
In other news, I spotted a Cisco UCS C210M2 with two (aging/old generation/borderline obsolete) Xeon E5620 2.4GHz (4c/8t each) CPU and 24GB of DDR3 memory into it for a scant 1000$. I don't know if it'd be able to complete the fat units on time to get the bonus (doubtful), but I'm tempted. It's only one tenth of the price of your mad box. Almost reasonable.
ddrueding
Fixture
It actually is still running, the client wants 64GB of RAM instead of the measly 32GB that is in the box now. Hopefully that will arrive tomorrow. I am not getting all the money I spent for the parts originally, but I am getting current market value for the parts. Lost about $3k I'd imagine; I don't want to actually do the math.
CougTek
Hairy Aussie
I should take back what's mine by the end of the day. It will have taken me 35 days.
CougTek
Hairy Aussie
Done.
ddrueding
Fixture
Next time I build a monster rig I might hold on to it for a little bit just to give you a scare 
With a little luck, you should see my PPD go sky high with no change in HW: http://foldingforum.org/viewtopic.php?f=66&t=22808&start=15#p227210
A preliminary estimate indicates a change from ~70K PPD to ~450K PPD with the new WU's and new GPU benchmarking system.
A preliminary estimate indicates a change from ~70K PPD to ~450K PPD with the new WU's and new GPU benchmarking system.
CougTek
Hairy Aussie
Ok, so now I have to fill my empty PCI-E16X slots with Kepler-based cards. I browsed the FAH forums, but i cannot find a comparison between the FAH performances of a GTX 670 versus the one of a GTX 660Ti. I'll probably go with the former. I can't see you fly away that fast without answering. I don't think I'll be able to afford more than 2 GTX 670, so you'll probably beat me anyway with your 5 GTX 460.
ddrueding
Fixture
Tempting to throw my GTX 690 back in there...
CougTek
Hairy Aussie
If you do that, I'll have to spend ~600$ just to conserve my second spot on the team's ranking.
Kelper's currently have issues. They require a new core that still has problems and is still in closed-beta. Currently, Fermi's are much more reliable. I'm sure that will eventually change but when I cannot predict. That is not to say you can't fold on them but I wouldn't invest large amounts till the problems have diminished. But if you have one, you can try it.
That being said, with the new benchmarking very high-end Nvidia GPU's may end up being the best way to go. So much more so that a MB with lots of PCI-e slots filled with high-end dual-GPU's (SLI status is irrelevant to folding) may supplant multi-processor servers running bigadv's with more PPD, cheaper upfront costs, and perhaps even cheaper PPD/W. I could see that perhaps very high end dual-GPU cards while outrageously expensive for gaming could easily end up being optimum configuration for folding by maximizing the usage of PCI-E slots on a single machine. There needs to be more time for that type of analysis to be done.
You do not have to fill your machines with very high-end GPU's. It is just better if you do. I'm using 5x GTX 460's (cost a little over $100 each) and theoretically getting 90,000+ PPD each. 570's are reporting 150,000 PPD; 580's are supposed to be getting 220,000 PPD each. That will at least get you started with some points on the graph. These are all very early results and PG could easily end up making modifications to the system.
The whole point is to give you guy's a heads up that people can drastically improve their folding PPD inexpensively by just GPU folding.
There is one big issue with GPU folding and that is LAG. The video subsystem has no way to resolve resource conflicts and that means that you can end up with a machine that stutters while GPU folding (Folding and Windows both trying to use the GPU). The amount of stuttering varies greatly from perfectly fine unless playing a game or watching videos to a permanently jerky mouse with a 1/2 second delay between pressing a mouse button and the computer registering it. If you have a dedicated folding machine that typically doesn't matter much. If it is your main usage machine that LAG may be a show stopper for GPU folding. The LAG problem is not likely to go away until the GPU manufacturers and Microsoft get together and create a priority system applicable to GPGPU usage. There are some techniques that do seem to help: Aero theme, the right driver (some driver versions do better with Lag than others), and multiple video cards as well as just plain pausing the primary video card.
That being said, with the new benchmarking very high-end Nvidia GPU's may end up being the best way to go. So much more so that a MB with lots of PCI-e slots filled with high-end dual-GPU's (SLI status is irrelevant to folding) may supplant multi-processor servers running bigadv's with more PPD, cheaper upfront costs, and perhaps even cheaper PPD/W. I could see that perhaps very high end dual-GPU cards while outrageously expensive for gaming could easily end up being optimum configuration for folding by maximizing the usage of PCI-E slots on a single machine. There needs to be more time for that type of analysis to be done.
You do not have to fill your machines with very high-end GPU's. It is just better if you do. I'm using 5x GTX 460's (cost a little over $100 each) and theoretically getting 90,000+ PPD each. 570's are reporting 150,000 PPD; 580's are supposed to be getting 220,000 PPD each. That will at least get you started with some points on the graph. These are all very early results and PG could easily end up making modifications to the system.
The whole point is to give you guy's a heads up that people can drastically improve their folding PPD inexpensively by just GPU folding.
There is one big issue with GPU folding and that is LAG. The video subsystem has no way to resolve resource conflicts and that means that you can end up with a machine that stutters while GPU folding (Folding and Windows both trying to use the GPU). The amount of stuttering varies greatly from perfectly fine unless playing a game or watching videos to a permanently jerky mouse with a 1/2 second delay between pressing a mouse button and the computer registering it. If you have a dedicated folding machine that typically doesn't matter much. If it is your main usage machine that LAG may be a show stopper for GPU folding. The LAG problem is not likely to go away until the GPU manufacturers and Microsoft get together and create a priority system applicable to GPGPU usage. There are some techniques that do seem to help: Aero theme, the right driver (some driver versions do better with Lag than others), and multiple video cards as well as just plain pausing the primary video card.
This is the very first day these WU's and the new benchmarking system have been released. Everything is still in closed beta. However, the forums are read-only to non beta-team members, so if you do some research and fiddling you can possibly figure out how do make it work.
You are not likely to find public benchmarks for a variety of different cards yet because it is so new. You'll need to come up with your own 670 vs. 660Ti data or wait till a bunch of people have tried it and someone out there has published it.
You are not likely to find public benchmarks for a variety of different cards yet because it is so new. You'll need to come up with your own 670 vs. 660Ti data or wait till a bunch of people have tried it and someone out there has published it.
ddrueding
Fixture
At work I currently have a single GTX690, probably moving to two when I get another Dell 30" screen there. It is my main machine, but I only log 4-6 hours in the office a day, so if I could set up some kind of idle-time configuration it would likely still put some points on the board.
At home I have the two ATI cards (7970s), but it doesn't look like they are worth bothering with?
At home I have the two ATI cards (7970s), but it doesn't look like they are worth bothering with?
Currently, the new system is Nvidia only and even under the old system ATI video cards suck for folding (personal opinion).
You can configure an idle schedule script. I have not done that but others have. It tends not to be as simple as you'd think. Perhaps this will help: http://foldingforum.org/viewtopic.php?f=67&t=22292.
I just keep the client up on the screen and just pause and unpause the client/slot before and after anything that is problematic.
Just note that regardless of what you do, the big value here is in the bonus points for quick return just like bigadv's. Big delays produce big point drops so turning off the machine at night could kill your PPD.
CougTek
Hairy Aussie
From what I can read, AMD still has issues with OpenCL and that is the primary cause why their client cannot be optimised as much as Nvidia's. The point reward for AMD's graphic cards is miserable compared to what you can get on Nvidia's stuff.
The GTX 580 still is quite expensive, but I've found a GTX 570 for 200$. For 150000ppd, that's tempting.
ddrueding
Fixture
I don't think that will be a problem, operating even 18 hours a day will likely meet the time constraints for maximum bonus, no?
CougTek
Hairy Aussie
Well...I should add 150K ppd to daily output sometimes next week. Meanwhile, we should help Dave to setup his monster card even if the client isn't quite ready for Kepler. If we can make it work, the point reward will be tremendous.
Yes and no. Assuming a continuous 6 hour non-folding block then For any WU's that start and end inside that 18 hour block, you will get the maximum bonus. Those outside, will get less than the maximum. The bonus point system is exponential with no limit so any time not folding will decrease the bonus points.
If you have an idle routine then rather than removing a big chunk of time then you'll be just taking some time, rather than all, from each WU during that 6 hour it will produce more PPD than the above choice.
A pause unpause scenario where you pause before something will cause a problem and un-pause after doing will work similarly as the idle routine. Which is more effective depends upon how you configure the idle routine as opposed to personal judgement and habits.
Once you get it working, you can fiddle and see what works best for you.
If you have an idle routine then rather than removing a big chunk of time then you'll be just taking some time, rather than all, from each WU during that 6 hour it will produce more PPD than the above choice.
A pause unpause scenario where you pause before something will cause a problem and un-pause after doing will work similarly as the idle routine. Which is more effective depends upon how you configure the idle routine as opposed to personal judgement and habits.
Once you get it working, you can fiddle and see what works best for you.
CougTek
Hairy Aussie
Compared to his currently daily contribution (nothing), I think adding his GTX 690 for 18 hours per day will be pretty substantial. The first thing to do would be to join the beta team.
It can be done without joining the Beta team. The beta team forums are readable by anyone so all the information is there to access.
Download the latest client and install.
adjust FAHControl to expert mode. Configure the client with your username, team, and passkey. Go to the expert tab click on "add expert client options" insert into box name: client-type value:http://foldingforum.org/viewtopic.php?f=67&t=21567&start=15 (not allowed to say). Save it all and let it run.
That should be enough. You are likely to get two GPU slots and an SMP slot. The SMP slot is optional. The first GPU WU's will not be ones you want because they were likely gotten before the modifications above. But when they finish, with a little luck, you should get good ones (p8057's).
There are several potential problems that can interfere but I'll only go into those if we run into one.
You can pause an individual slot by simply right clicking the slot and picking pause. The big pause button on the top pauses everything. Also, If you need to pause, try to pause only the primary GPU (the one attached to the display) and allow the secondary to continue running. Lag is primarily based on the primary GPU. Because of how the client enumerates the GPU's It won't necessarily be the first one in the list. You'll need to try each and find the one that is the primary.
Download the latest client and install.
adjust FAHControl to expert mode. Configure the client with your username, team, and passkey. Go to the expert tab click on "add expert client options" insert into box name: client-type value:http://foldingforum.org/viewtopic.php?f=67&t=21567&start=15 (not allowed to say). Save it all and let it run.
That should be enough. You are likely to get two GPU slots and an SMP slot. The SMP slot is optional. The first GPU WU's will not be ones you want because they were likely gotten before the modifications above. But when they finish, with a little luck, you should get good ones (p8057's).
There are several potential problems that can interfere but I'll only go into those if we run into one.
You can pause an individual slot by simply right clicking the slot and picking pause. The big pause button on the top pauses everything. Also, If you need to pause, try to pause only the primary GPU (the one attached to the display) and allow the secondary to continue running. Lag is primarily based on the primary GPU. Because of how the client enumerates the GPU's It won't necessarily be the first one in the list. You'll need to try each and find the one that is the primary.
Last edited:
By the way, I'm not even sure that Kepler's are getting them at all. So far, all the reports in the forums that I've seen have been for Fermi's and even more specificly FAHCore_15 v2.22. The v2.25 of the core is required for Keplers...
CougTek
Hairy Aussie
I'll soon be in the market for a Xeon W3680 to pair with my spare GigaByte X58-USB3 motherboard. 3 PCI-E16X slots (one being only 4X electrical though), all just waiting for upcoming GTX 570.
CougTek
Hairy Aussie
BTW, there were talks on Stanford's FAH forum that a Linux GPU client was in the making. Is there a beta version out yet and can you get the p8057s with it?
I can try adding in my GTX 560 Ti 448 core if you think it'll help? I'm guessing it should come close to the GTX570 with respect to speed. I saw you edited your post and removed the last step. Do I need to dig through that URL to figure out how to setup my system to get this work units?
Since you have not been previously GPU folding you will probably need to revert to an old core v2.22 too. A new install will auto DL the new core version and as far as I can tell the only people getting these WU's are using an old core v2.22. That means you will need this link to get the old core: http://fahwiki.net/index.php/Downloading_FAH_Core_files_manually and then you will need to follow this process to convert the source file: http://fahwiki.net/index.php/Downloading_FAH_Core_files_manually into a usable core that you will use to replace all copies of the current v2.25 core. Yes, needing to use a hex editor is archaic and the process is inconvienant at best.
The P5-133XL secret sauce for getting these WU's: http://foldingforum.org/viewtopic.php?f=66&t=22808&start=45#p227355
I added it but it didn't seem to make a difference. I deleted the projects that were downloaded originally and let it retrieve new ones but I didn't see a p8057. Do I need to complete a non p8057 or any number of them prior to getting the good ones? I see what you mean about the primary display chugging.
Looks like I cross posted with this new info. I'll take a look tomorrow and see if I can get it working. Thanks for putting the info together.
Check the core version in the log. It will look something like this:
If it is v2.25 then you will need to go through the get old core process given above. I do not know if you have a passkey but it is needed to get the bonus points.
Code:
05:59:50:WU01:FS00:0x15:Folding@Home GPU Core
05:59:50:WU01:FS00:0x15:Version 2.22 (Thu Dec 8 17:08:05 PST 2011)
05:59:50:WU01:FS00:0x15:Build host SimbiosNvdWin7
05:59:50:WU01:FS00:0x15:Board Type NVIDIA/CUDA
05:59:50:WU01:FS00:0x15:Core 15If it is v2.25 then you will need to go through the get old core process given above. I do not know if you have a passkey but it is needed to get the bonus points.
CougTek
Hairy Aussie
I'm almost certain Handruin has a passkey, otherwise he wouldn't have climbed to ~5 million points so far. If the big GPU units are like the CPU big units, you have to complete a certain amount of "ordinary" units before getting the good ones. So I wouldn't systematically delete the other units you receive in the meanwhile. Anyway, it is unethical to do so.
Found it. Looks like i have the wrong version like you mentioned:
Code:
06:18:13:WU00:FS00:0x15:Folding@Home GPU Core
06:18:13:WU00:FS00:0x15:Version 2.25 (Wed May 9 17:03:01 EDT 2012)
06:18:13:WU00:FS00:0x15:Build host AmoebaRemote
06:18:13:WU00:FS00:0x15:Board Type NVIDIA/CUDA
06:18:13:WU00:FS00:0x15:Core 15I do have a passkey and I plugged it into the client install like you documented. I've used it many times in the past like CougTek speculated.
OK, I won't delete any more. I only did it the once to see if it would download the work unit described above, but I understand that it isn't guaranteed and it's obvious I'm missing other pieces. I'll work on fixing them and then letting it plow through a bunch of normal work units so that i can start getting the good ones.
My video card makes all sorts of high pitch whining noises while working on these things. I remember others including P5-133XL mentioning this observation also.
CougTek
Hairy Aussie
The noise the card makes is crazy loud. Had I known, I'm not sure I'd have gotten one. Also, maybe because I've only slept two and a half hours during the past 48 hours, but I haven't figured out how to get the beta units come in. All the older core files from the site Mark linked are 1.5MB and the FahCore_15.exe installed by default in AppData/Roaming/[...]/ is ~7.5MB large. I'll see to that tomorrow when I wake up or maybe even just after the job.
Off to bed.
Off to bed.
ddrueding
Fixture
I've currently worked 65 of the last 80 hours, and I'm not going home yet.
There are 2x GTX590 currently in my work machine, and there is room for one more. Does SLI support more than quad graphics cards?
There are 2x GTX590 currently in my work machine, and there is room for one more. Does SLI support more than quad graphics cards?