GPU Folding Facts and Fiction - Configuration and Settings

[Deadpato]

Hi!

 

I have a question, i have a Nvidia GT 630 4gb ddr3 laying around and I was thinking if it was worth installing in my system.

I'm currently runing an i5 9400f and an AMD RX 560 4gb ddr5. And if I can, how do i make them work.

 

Thanks!

[ykdave]

21 minutes ago, Deadpato said:

Hi!

 

I have a question, i have a Nvidia GT 630 4gb ddr3 laying around and I was thinking if it was worth installing in my system.

I'm currently runing an i5 9400f and an AMD RX 560 4gb ddr5. And if I can, how do i make them work.

 

Thanks!

whats your definition of "worth it"

Ive got a couple 650s i can throw together on an old board and fold with it, but IMO its a waste of power as it would take at least a week to pump out the WU's the 2080ti in my rig will do in just one day

[SansVarnic]

On 3/27/2020 at 11:49 AM, plasticbomb1986 said:

I have my own question: does it possible to prioritize between gpus, whichone gets WUs first? Have two VEGA64 in my system, and fah always load up the first gpu, what time to time id wanna use for gaming, and the second gpu is there sittin doing nothing, plus no game using two gpu on linux land anyway. (When i built this rig, i was still on windows, shortly after win upset me too much, and just switched over to linuxland !)

No. The F@H software automatically assigns the packets.

You can still game, the F@H software can prioritize pc usage of the main gpu for display use, it will just slow down the proces a bit depending on the load you place on the gpu.

14 hours ago, Randarr said:

I'm having an issue getting my Vega 64 folding, without it I'm not gonna get much work done, maybe someone already answered a question like this here and I'm just too lazy to read through the whole thread, but are there any recommendations to get it going? I did see something about the pro drivers is that going to get it running?

 

System info: ryzen 5 1600, 16gb DDR4 3200, x370 motherboard, vega 64 strix software version 20.2.2, Windows 10 version 1909

My game server with an r9 290x and an e5 2640v2 seems to be chugging along just fine and has the GPU active more often, it's also using the regular game drivers not the pro ones, but different GPU architecture and so forth making the difference?

This is better posted in it own topic but ill do what I can help here.

Try reinstalling your drivers, in most cases this will fix the problem.

24 minutes ago, Deadpato said:

Hi!

 

I have a question, i have a Nvidia GT 630 4gb ddr3 laying around and I was thinking if it was worth installing in my system.

I'm currently runing an i5 9400f and an AMD RX 560 4gb ddr5. And if I can, how do i make them work.

 

Thanks!

The F@H software will recognize gpus separately and utilize accordingly. You can install it and it should work fine. A GT630 is a pretty low end card so you may not see much in the way of production... but every bit helps.

[Deadpato]

6 minutes ago, SansVarnic said:

The F@H software will recognize gpus separately and utilize accordingly. You can install it and it should work fine. A GT630 is a pretty low end card so you may not see much in the way of production... but every bit helps.

Well, as it is my main pc, i don't mind it beeing there, as the majority of the day, the work I have to do it's pretty light, and wennever I need more resources, I could easily stop folding.

I know the GT630 is low end, it was my previous GPU and it would be used as a secondary GPU to fold when I'm gamming, my current one is not new too but it gets me going on a budget.

 

About the drivers of the nvidia, will there be a conflict with the amd drivers? I don't know if it's just as easy as plug and play or I have to tinker with something.

 

Thanks!

[SansVarnic]

1 minute ago, Deadpato said:

Well, as it is my main pc, i don't mind it beeing there, as the majority of the day, the work I have to do it's pretty light, and wennever I need more resources, I could easily stop folding.

I know the GT630 is low end, it was my previous GPU and it would be used as a secondary GPU to fold when I'm gamming, my current one is not new too but it gets me going on a budget.

 

About the drivers of the nvidia, will there be a conflict with the amd drivers? I don't know if it's just as easy as plug and play or I have to tinker with something.

 

Thanks!

Unfortunetly you cannot use two gpus of different generations or brands together in your daily pc for display purposes.

The F@H software will use both but separately, F@H doesnt use gpus for simultaneous wu assignments.

[Casparj]

ok i dont understand anything now
f@h is reporting that the 290x is working now the % is going up and its getting points per day but acording to CAM and HWmonitor theres no activity from the card the exposed heatpipes is not even getting hot
anyone got ANY idea what is going on ?

 

 

[lexxR]

is there any way to configure the config file to have a max time limit on retry time (like 60 seconds) as the Folding client to have a more sane max retry count (it wouden't to bad if it would que up work units like rosetta@home, but i set to 0 que on rosetta@home as it can reliably get work units)

 

as i keep command back to it and sometimes its on a 2-3 hour cool down when i pause and resume it 2-3 times both CPU and GPU get a working server and starts doing work (the client only trys 2 servers per go not all of them and typically keeps trying the same server that is not giving work units or the server times out)

 

apart from using a program to just restart the F@H control every 60 seconds when it does not detect 22 and A7 is not running (it wouldn't really be a problem if there was not any work units but there are and the client keeps trying the same servers that don't have any with increasing wait retry times times)

[SansVarnic]

3 hours ago, lexxR said:

is there any way to configure the config file to have a max time limit on retry time (like 60 seconds) as the Folding client to have a more sane max retry count (it wouden't to bad if it would que up work units like rosetta@home, but i set to 0 que on rosetta@home as it can reliably get work units)

 

as i keep command back to it and sometimes its on a 2-3 hour cool down when i pause and resume it 2-3 times both CPU and GPU get a working server and starts doing work (the client only trys 2 servers per go not all of them and typically keeps trying the same server that is not giving work units or the server times out)

 

apart from using a program to just restart the F@H control every 60 seconds when it does not detect 22 and A7 is not running (it wouldn't really be a problem if there was not any work units but there are and the client keeps trying the same servers that don't have any with increasing wait retry times times)

Due to the massive influux of new folders the F@H servers are overwhelmed for packet requests, that said wait times are current 30min-12hrs.

But to answer your question: No

[lexxR]

9 hours ago, SansVarnic said:

Due to the massive influux of new folders the F@H servers are overwhelmed for packet requests, that said wait times are current 30min-12hrs.

But to answer your question: No

guess i just stick to rosetta@home i have lowered it to 90% and left the default for quing work (0.1 days work minimal and 0.4 days works additional i think) as that keeps about 12-13 work units ready (i am completing them in about 7-8 hours)

 

the problem i have with folding at home is it is not geared for more then 2 servers and does not try every server when it retries resulting in it not downloading work that is available and lack of automatic pre queing of work units (if it's only going to take 40 minutes to 2 hours to complete a work unit it should be able to que up 1 work unit right away, the client should work that out automatically due to how fast i am completing them)

 

if i pause and resume the CPU que (i am not doing GPU any more as its crashed my system 2 times doing GPU) about 5-6 times it actually trys a different work server and gets a work unit

 

i long time ago (about 7 years ago) i did have my F@H client setup to buffer up 5 work units (wish i remembered how i did it) so i always had work units as this has always been a problem with F@H with its inability to successfully download a new work unit when currant work is at 99% and resulting in not doing any work for hours at time because it does not want to try a different server

 

it also help if the project makers would make the work more Harder/Bigger as i am finishing most of them in under 1 Hour as with the amount of new people who have joined F@H they need about 10x more work units (i say about 5 million work units vs the 500k they have right now, assuming they don't make them bigger) as a lot of people are joining with quite high end hardware and finishing them very quickly (and reconfigure the client to try all servers after a server fails to respond or says no work unit available it should try the other 10+ servers one after another once all have failed to provide a work unit then the fall back to the default retry counter should multiply like it does now, there server have throttling so it shouldn't be a problem if they are busy)

[SansVarnic]

1 hour ago, lexxR said:

<snip>

I believe you are over thinking this a bit, though you have a point on how packets are que'd up, this you need to contact and give that feed back to those that run F@H.

Unfortunately I dont have much else i can offer other than my own experiences of using the software. Albeit the software has improved considerably over the last several years on how it utilizes and prioritizes system resources.

[lexxR]

just seems the same as it was 8-7 years ago but with a new manager, no que and only trying 2 servers not all servers

 

at a minimum trying all servers before giving up needs to be done, as over 12-15 hours is unacceptable when there are work units available

 

they are because if i pause and resume the client enough times (about 4-5) it then trys a new work server instead of trying the same server that is not going to give you work, i can see in the logs it's trying the same server over and over instead of trying a different one

[SansVarnic]

1 hour ago, lexxR said:

just seems the same as it was 8-7 years ago but with a new manager, no que and only trying 2 servers not all servers

 

at a minimum trying all servers before giving up needs to be done, as over 12-15 hours is unacceptable when there are work units available

 

they are because if i pause and resume the client enough times (about 4-5) it then trys a new work server instead of trying the same server that is not going to give you work, i can see in the logs it's trying the same server over and over instead of trying a different one

That is a conversation to have with the F@H people to be frank with you.

[RCD3T]

On 3/29/2020 at 6:59 PM, lexxR said:

is there any way to configure the config file to have a max time limit on retry time (like 60 seconds) as the Folding client to have a more sane max retry count (it wouden't to bad if it would que up work units like rosetta@home, but i set to 0 que on rosetta@home as it can reliably get work units)

 

as i keep command back to it and sometimes its on a 2-3 hour cool down when i pause and resume it 2-3 times both CPU and GPU get a working server and starts doing work (the client only trys 2 servers per go not all of them and typically keeps trying the same server that is not giving work units or the server times out)

 

apart from using a program to just restart the F@H control every 60 seconds when it does not detect 22 and A7 is not running (it wouldn't really be a problem if there was not any work units but there are and the client keeps trying the same servers that don't have any with increasing wait retry times times)

I'm having the same issue that you are. F@H keeps cycling through the same 2 servers over and over. 

[Macaw2000]

4 hours ago, RCD3T said:

I'm having the same issue that you are. F@H keeps cycling through the same 2 servers over and over. 

I've just recently added a cron job to my clients that stops and starts the F@H software twice a day. The issue is that F@H uses exponential backoff and can leave a client sitting there idle indefinitely. With the app you can sit there and hit pause and resume but for headless machines, I had to use cron.

[Macaw2000]

0 11,23 * * * sudo /etc/init.d/FAHClient stop && sudo /etc/init.d/FAHClient start

There's the cron job if useful to anyone.

[DonGuano]

Hi!

 

FAHCore 0xa7 is using most of my CPUs power even after  lowered thread count to 6 (R3600)??? It's herder to just type this message cos of the lag...

 

Is there a delay lowering threads?

 

-Don

[PAPO1990]

On 3/25/2020 at 11:51 PM, Casparj said:

do i need to tweak any settings if i have an AMD card and a nvidia card in the same computer i cant seem to get my r9 290x to work along side my 1060

I don't think the R9 290X is properly supported anymore, before I upgraded my GPU recently I was trying to fold on a 390X and on the rare occasion it gave me a WU for it it would hardly make it past 1% before saying it was done and sending it back. Eventually I just disabled GPU folding until my 5700XT arrived.

[chaozbandit]

7 hours ago, PAPO1990 said:

I don't think the R9 290X is properly supported anymore, before I upgraded my GPU recently I was trying to fold on a 390X and on the rare occasion it gave me a WU for it it would hardly make it past 1% before saying it was done and sending it back. Eventually I just disabled GPU folding until my 5700XT arrived.

 

As per the website

Quote

GPU3 – OpenCL – FAHCore_21)
OpenCL compatible GPU, 5xxx series or newer*, see full list
 

 

Anything GCN is almost certain to still be compatible, and I've had my 290s completing WUs without issues.

[PAPO1990]

9 hours ago, chaozbandit said:

 

As per the website

 

Anything GCN is almost certain to still be compatible, and I've had my 290s completing WUs without issues.

when's the last time the website was updated though?

[Bigtruck747]

GTX 1080ti vs. NVIDA Tesla K80 for Folding.

 

This question comes after reading through SansVarnic's post about his folding server; (Personal Folding Server Rig - Project Steamroller)

 

Looking at the K80 on eBay right now it looks like I get them used for $250 USD ... & the 1080ti would be at least double that.  So then the question would be, would the 1080ti Fold so much more, or use so much less electricity than the K80 that it would be better to get the the 1080ti???

 

In this article by SQream Technologies they talk about comparing a 1080 vs a K40 while doing "heavy database operations".  They mention that the 1080 used less energy(???)

 

I was thinking of putting two 1080ti's in my person machine to Fold with ... but now I'm wondering about two K80's!?!?

 

Thank you in advance for all your help!!

[SansVarnic]

7 hours ago, Bigtruck747 said:

GTX 1080ti vs. NVIDA Tesla K80 for Folding.

<snip> So then the question would be, would the 1080ti Fold so much more, or use so much less electricity than the K80 that it would be better to get the the 1080ti???

 

In this article by SQream Technologies they talk about comparing a 1080 vs a K40 while doing "heavy database operations".  They mention that the 1080 used less energy(???)

 

I was thinking of putting two 1080ti's in my person machine to Fold with ... but now I'm wondering about two K80's!?!?

The difference is the architecture between the two, K40's and K80's are of the older Kepler architecture and the 1080TI are the newer Pascal architecture. Pascal is better all around and much more energy efficient.

[DonGuano]

In your experience, what is the optimal slot configuration for Ryzen 5 3600 when folding both CPU and GPU. My GPU is RTX 2070 Super. Atm I'm using 10 threads and my FahCore_a7 CPU usage is 81.3 to 83% while GPU FahCore_22 is 9 to 9.2%. Overall CPU usage is 95 to 96%.

[DonGuano]

okey, now overall CPU usage hit 100%, CPU FahCore_a7 rise to 87. Lowering to 9 threads...

[BourbonOrBust]

I had a TIL moment the other day while playing around to maximize PPD and wanted to share
 

Quote

CPU Impact

This is primarily a NVidia issue, mostly attributed to NVidia drivers. So, the rule of thumb when folding with a NVidia GPU and CPU folding, is to leave one core free per GPU to maximize your GPU Folding performance. In the case of AMD, it is simply recommended to leave a single core free for one or more cards as the impact isn't as great.

This was always the advice I had heard and followed. I have a 4790k and was folding on 6 threads, leaving two free. After playing around with my OCs and power settings, I wound up removing my CPU task entirely just to see what would happen. Not only has my PPD gone UP by an appreciable amount, but my thermals (and therefore noise) are better now as well since the CPU isn't dumping heat into my loop anymore.

 

Obviously YMMV, but it may be worth trying

[Bigtruck747]

I think this thread answers one of my questions, but I will appreciate any advice given.  If a person is planning on building a computer, and optimize it for Folding at Home, would it be better to use an Ryzen CPU since GPU folding w/ AMD isn't as core intensive (per GPU)?

 

Thank you!