Folding Community Board

[Zberg]

55 minutes ago, ForgottenNavigator said:

I've decided to refocus the TR build into something a bit more reasonable. This case isn't going to be 3D printed, beyond some mounts and aesthetic doodads. But I think I've refined the design into something workable. Still have a lot of work to do but I should be able to at least finish the case before the contract here is finished.

Plan is now for a Ryzen 9 3950X, two water cooled 2080 Supers, and a pretty crazy custom loop including 2 240mm rads. I worked for a plumbing contractor for a summer so I'm pretty confident I can pull off the frame and reservoir. The final outer dimensions will be 420mm x 450mm x 200mm. I'm still messing around with the loop layout and some final structural work. I want to keep it easy to fill/drain and keep it well ventilated.

I'll be doing a build log for somewhere else on the forum if anyone's interested.

 

Any chance you can make case a little bigger with 2x 280 at least?  or 2x360?  3950x + 2x 2080S might ask for some more rad surface area

[Zberg]

5 hours ago, rodarkone said:

I'll be back in sometime ... F@H impact due to holidays ahahaha

You going to request that mythic badge before going ?!  

[justpoet]

59 minutes ago, ForgottenNavigator said:

I've decided to refocus the TR build into something a bit more reasonable. This case isn't going to be 3D printed, beyond some mounts and aesthetic doodads. But I think I've refined the design into something workable. Still have a lot of work to do but I should be able to at least finish the case before the contract here is finished.

Plan is now for a Ryzen 9 3950X, two water cooled 2080 Supers, and a pretty crazy custom loop including 2 240mm rads. I worked for a plumbing contractor for a summer so I'm pretty confident I can pull off the frame and reservoir. The final outer dimensions will be 420mm x 450mm x 200mm. I'm still messing around with the loop layout and some final structural work. I want to keep it easy to fill/drain and keep it well ventilated.

Spoiler

 

I'll be doing a build log for somewhere else on the forum if anyone's interested.

 

 

 

Are the tubes structure only, or intended to be part of the loop for more coolant capacity?

 

Also, which app are you using to model, and are any of the generic PC parts in a library for it?  One of these days I'm going to fab my own case too, whenever I need to build a replacement for my daily driver system.

[rodarkone]

7 hours ago, Zberg said:

You going to request that mythic badge before going ?!  

No  will let it happen naturally  ...

[ForgottenNavigator]

8 hours ago, Zberg said:

Any chance you can make case a little bigger with 2x 280 at least?  or 2x360?  3950x + 2x 2080S might ask for some more rad surface area

Thanks for the advice, I think you may be right but this will be my first water cooled PC (I built a mineral oil rig in college using a high viscosity pump, a 10 gallon fish tank, and a heater core out of a car).

I'm trying to figure out how to cram a third 240mm rad in there. I might be able to get one in on the right side of the motherboard tray. Theoretically I could but I won't really know until I start assembling the case. I have a few days before parts start arriving so I may take the time to design an alternative layout. Barring that I could probably do a 240 on top and a 360 in front if I sacrifice my planned IO layout.

If you have any experience mind weighing in on my planned loop? I want to go CPU -> Rad 1 -> GPU 1 -> Rad 2 (If I can get 3 in there) -> GPU 2 -> Rad 3 -> Reservoir -> Pump -> CPU. From my research loop order is either critical or irrelevant (save reservoir feeding pump).
 

 

8 hours ago, justpoet said:

Are the tubes structure only, or intended to be part of the loop for more coolant capacity?

 

Also, which app are you using to model, and are any of the generic PC parts in a library for it?  One of these days I'm going to fab my own case too, whenever I need to build a replacement for my daily driver system.

Structural only unfortunately. I had been trying to figure out how to make an effective water loop with them. Or at least a portion as a reservoir. I was left with three options. 1) Braze the entire outer pipe assemblies to the inner assemblies in order to use inner assembly as a resevoir. 2) Internally cap the portions I intended to drill through for attaching the side panels or 3) Braze the bolts to the inner assembly and introduce the risk of galvanic corrosion to my loop.There's probably a solution but the cooling benefit seems minimal to me and greatly increases the chance of leaks. I believe there was actually a LTT video on it?

I do most of my work in Fusion360. I've used sketchup as well but found it to be severely lacking. I use grabcad to find most of my reference models. If I cannot find a ready made model I usually just build the model myself from references. I've compiled quite a list over the years. If you're looking at building your own case I'd recommend starting with one of the Lian-Li removable motherboard trays. IO geometry can be a bit tricky to the uninitiated. I've been case building for years (My first computer was used parts I scavenged and mounted inside a milk crate with zipties).

[OIIOOIOI]

On 6/24/2020 at 11:16 AM, dogwitch said:

so am taking a break for a bit on folding.

1 machine is a old fx and a gpu. doing its thing.

atm am working on a mining mobo and doing multi gpu set up.

testing etc will take a month or so. also need to figure out power draw. per month,

I have an old FX too and 2 graphics cards. While no event i turn down FX in advanced settings->configure->CPU Slot [edit] to 2 or just 0 on warm days cause the cpu has a horrible TDP to PPD ratio anyway consuming 135 Watt when running on all Kernels. Lowering this it ends up at 65~85 Watts and CPU cooler aint getting noisy and produced heat is 20° below full power. The loss of WU´s per day is only 1 to 2 at 8-core FX. When made enough money in my 2nd job i will buy me a new mobo with ryzen cpu and ddr4 ram to upgrade my computer again ... think that will take place in 3 months if no unexpected coasts aproach.

[Zberg]

8 hours ago, ForgottenNavigator said:

If you have any experience mind weighing in on my planned loop?

I have mine set up alternating currently (rad - pump-GPU-rad-CPU), and have previously had it rad-pump-CPU-rad GPU, and before I got second radiator, rad-pump-CPU-GPU.

 

Because I am usually running continuous workloads (folding with both CPU and GPU, or gaming in which my GPU is pegged at 100% for hours at a time with the CPU pinging up and down) none of the changes in my loop did anything significant to temps.  Only notable difference was when I had CPU after the biggest rad (after pump).  I like where my pump sits now and the loop order / aesthetics, so I have it GPU first now.  But when it was CPU first, I noticed temp spikes seemed the tiniest bit lower, not significantly, and only if I was mostly doing low level usage and not folding or giving it a constant workload.  

 

With constant workloads and gaming, the system comes to equilibrium at some temp, the GPU tends to sit around that temp (because it is producing the majority of heat to get to said equilibrium) and the CPU has more wide variance but no different in any loop set up I had.  Id be interested to line up all my current rads in series, then to CPU-GPU and back to rads and see if its any different but for folding and constant tasks I would bet money that would be the same equilibrium too.

[ForgottenNavigator]

23 minutes ago, Zberg said:

I have mine set up alternating currently (rad - pump-GPU-rad-CPU), and have previously had it rad-pump-CPU-rad GPU, and before I got second radiator, rad-pump-CPU-GPU.

 

Because I am usually running continuous workloads (folding with both CPU and GPU, or gaming in which my GPU is pegged at 100% for hours at a time with the CPU pinging up and down) none of the changes in my loop did anything significant to temps.  Only notable difference was when I had CPU after the biggest rad (after pump).  I like where my pump sits now and the loop order / aesthetics, so I have it GPU first now.  But when it was CPU first, I noticed temp spikes seemed the tiniest bit lower, not significantly, and only if I was mostly doing low level usage and not folding or giving it a constant workload.  

 

With constant workloads and gaming, the system comes to equilibrium at some temp, the GPU tends to sit around that temp (because it is producing the majority of heat to get to said equilibrium) and the CPU has more wide variance but no different in any loop set up I had.  Id be interested to line up all my current rads in series, then to CPU-GPU and back to rads and see if its any different but for folding and constant tasks I would bet money that would be the same equilibrium too.

Yeah, it makes sense that under load you'd end up reaching a point where the heat generation and dissipation would equalize. Fluctuations in temperature would decrease and you'd reach a steady state. I'd of liked to have integrated the structural frame for cooling as well but short of brazing thin copper sheets along the top I'm skeptical that it would provide any real cooling beyond having extra volume to absorb heat short term.

[RollinLower]

Another 2080 is about to join the fold!

 

Edit:

It's gonna be a tight squeeze

[TVwazhere]

14 hours ago, rodarkone said:

No  will let it happen naturally  ...

[TVwazhere]

13 hours ago, ForgottenNavigator said:

From my research loop order is either critical or irrelevant (save reservoir feeding pump).

I think the only thing that truly matters is that if you have a reservoir, it needs to feed directly into the pump so that it doesn't run dry. (I've never built a custom loop myself, just what I have gathered from my own research)

[dogwitch]

 i turn off most pc and hell even the mode today.  we where hitting 110 outside.

[TVwazhere]

We will surpass Curecoin in less than a year (11.9 months) at this rate.  

 

Meanwhile the Default team will surpass them in two days

[GOTSpectrum]

3 minutes ago, TVwazhere said:

We will surpass Curecoin in less than a year (11.9 months) at this rate.  

 

Meanwhile the Default team will surpass them in two days

Default team is FAKE NEWS!!!!

[rkv_2401]

How do you guys check if your GPU O/C is Folding stable? Curious to see how high I can get my PPD to go. 

 

Also, @yaboistar my R5 3600 can do 3.7Ghz with 1V(!) - but needs 1.125V to stably run at 3.9V. Seems like that's the most efficient zone. My thermals aren't good enough to go any higher on voltage.

[GOTSpectrum]

24 minutes ago, yaboistar said:

-snip-

Personally I actually do the opposite, that being undervolt and clock the card to reduce power and the like while maintaining 90+% of the production 

[viming_aint_easy]

Whelp, much like my promise in my initial post, I am beginning to upgrade my desktop to 20.04, to see if I can get Navi GPU folding working.  The drivers were last updated on 2020-06-12 (not sure how recently other people have tested it here), but even if they don't work, it really isn't a downgrade from my current position, folding-wise.

 

Wish me luck!

 

[viming_aint_easy]

Well, I did successfully upgrade to 20.04 on my RX5700 system.  Thanks to the guide at https://foldingforum.org/viewtopic.php?f=81&t=35141, I was able to install the drivers to the point that F@H is able to recognize the graphics card and starts "working on" a WU:

 

Quote

02:13:56:WU00:FS00:0x22:Project: 13414 (Run 2777, Clone 52, Gen 0)
...
02:13:56:WU00:FS00:0x22:Folding@home GPU Core22 Folding@home Core
02:13:56:WU00:FS00:0x22:Version 0.0.10
02:13:56:WU00:FS00:0x22:  Checkpoint write interval: 50000 steps (5%) [20 total]
02:13:56:WU00:FS00:0x22:  JSON viewer frame write interval: 10000 steps (1%) [100 total]
02:13:56:WU00:FS00:0x22:  XTC frame write interval: 250000 steps (25%) [4 total]
02:13:56:WU00:FS00:0x22:  Global context and integrator variables write interval: 250 steps (0.025%) [4000 total]

 

 

But it just stops there.  I confirmed with radeontop (which I had to compile from source to get it working with a navi card), that the GPU is doing nothing.

 

At least, I'm not (yet) seeing any errors related to bad WU's ( last post of https://linustechtips.com/main/topic/1171449-rx5700xt-on-linux-works/page/2/ )

 

Is anyone aware of anything that I may be missing?

 

[justpoet]

29 minutes ago, viming_aint_easy said:

Well, I did successfully upgrade to 20.04 on my RX5700 system.  Thanks to the guide at https://foldingforum.org/viewtopic.php?f=81&t=35141, I was able to install the drivers to the point that F@H is able to recognize the graphics card and starts "working on" a WU:

 

 

But it just stops there.  I confirmed with radeontop (which I had to compile from source to get it working with a navi card), that the GPU is doing nothing.

 

At least, I'm not (yet) seeing any errors related to bad WU's ( last post of https://linustechtips.com/main/topic/1171449-rx5700xt-on-linux-works/page/2/ )

 

Is anyone aware of anything that I may be missing?

 

@Gorgon, didn't you do a guide for this?  I'm having trouble finding it to link though.

[viming_aint_easy]

24 minutes ago, justpoet said:

@Gorgon, didn't you do a guide for this?  I'm having trouble finding it to link though.

Is this it? 

 

It does seem to be focused on 18.04 rather than 20.04.  Going off of the first post that I linked ( https://foldingforum.org/viewtopic.php?f=81&t=35141 ), there was a weird step to change the OS version back to 18.04 before installing the drivers, which tells me that 20.04 is still not officially supported.  However, the post was made on 2020-05, and the drivers were updated in June, so it may be outdated.

 

An additional item I saw was one of the options where "--no-dkms" was provided in the Folding board guide, but Gorgon encountered some issues with that flag.  I'm going to see if that makes any difference.

[rkv_2401]

9 hours ago, yaboistar said:

you find out if it's unstable when you get a whole bunch of WU crashes and failures. i don't suggest overclocking as part of FaH, the difference is extremely minimal. as someone with a 2080Ti Lightning-Z I find that it makes perhaps 25-40kppd difference, compared to the 3.7-4.2 i can output regularly.

 

as a rule of thumb, if you have a permanently stable benchmark OC, and you back off by 100 mem and 25 core, that usually leads to a stable folding OC.

 

in the long run, it is actually better to set a power limit of about 75-80% which will reduce power consumption and thermal output, for much less of a hit to your PPD than you might expect. my 2080ti is set to 75% plimit and i still see 4 - 4.1mil crop up on some work units for a calculated PPD value so it's clearly not lost much of it's performance

Ah. For the 1060, there’s a noticeably bigger difference. Back in the Core21 days, leaving it at ~1850Mhz would get me 350,000PPD(off of the instantaneous PPD reading, I also play games on this machine so I can’t use long term averages)  but overclocking it to 2Ghz would get that figure upto around 420,000. Similarly, turning boost off and leaving it at base clock of 1506Mhz would make the PPD drop to around 280,000. 
 

I’ve actually created a new undervolted curve(which ends at 1847Mhz @ 900mV compared to the 1050 mV it normally requires under boost), and I have a performance profile where I lock the frequency at 2Ghz for 993mV. So yeah, I’ve already found the sweet spot for efficiency for my card.  

[cbigfoot]

WOOT TOP 100

[viming_aint_easy]

Good news!  After following the guide that @Gorgon provided, my RX5700 is hard at work!  Turns out the drivers needed DKMS to get them to fully function.  *sigh*  I was hoping that DKMS would be in my past, after moving off of AFS, but it just won't leave me alone - heh.

 

Thanks all!

[OIIOOIOI]

BTW Overcocking with increase of power is 2010 ... today state of the art is undervolting as low as possible before getting unstable and then increase the clocking to speed the card up again...with luck nearly to the default speed. Life has become greener last 10 years

[GOTSpectrum]

Damn loving these cooler temps atm after that heatwave over the folding sprint and last couple weeks or so.