16th Annual Folding@Home Holiday Season Race: The race is over and the Moonshots win!

[voodoo5_6k]

I guess my Ryzen isn't worth folding on? Seeing as no one answered

You have to test it, but I didn't bother with creating a tiny CPU slot with the remaining few threads of my rig. The last time I looked into it it wasn't worth it.

@Assimilator1, even on a 64-threaded server CPU, there are sometimes WUs which take almost a day to complete. You'll need to be a patient person if you turn to CPU folding on a desktop CPU.

Yeah, I've even seen three already that were close to 30h on my CPU slot (for 650,000 - 700,000 points).

 

[Markfw]

You have to test it, but I didn't bother with creating a tiny CPU slot with the remaining few threads of my rig. The last time I looked into it it wasn't worth it.


Yeah, I've even seen three already that were close to 30h on my CPU slot (for 650,000 - 700,000 points).

Well, I guess it depends on the CPU. I have a 5950x with 32 threads running under an hour at 670k ppd. unit 18210, 35% done and 38 minutes more.

 

[voodoo5_6k]

Well, I guess it depends on the CPU. I have a 5950x with 32 threads running under an hour at 670k ppd. unit 18210, 35% done and 38 minutes more.

Yes, and also on the WU. I've had WU's that would estimate at close to 1 million PPD. Like @StefanR5R already mentioned, CPU folding PPD is fluctuating extremely.

 

[voodoo5_6k]

Yeah don't count on me @voodoo5_6k I don't have all my GPUs in this, and of the ones I do most of them are rocking x1 slots.

Hey, you're definitely ramping up Just looked at your current stats because EOC listed you as a "threat" for overtaking me Thank you, @Skillz

 

[crashtech]

I added my Radeon VII and a few more CPU cores. I guess you could say that I thought I had it floored, but decided to cut a hole in the floor to get more pedal travel.

 

[Markfw]

WOW, not that I have the money for it, but $1175, a 2080TI RVGA FTW3 still under warranty for 479 more days ! (ebay)

 

[Endgame124]

Thanks for the stats biodoc
Close race!!

I guess my Ryzen isn't worth folding on? Seeing as no one answered

I’m folding with 28 CPU on my 5950 and 12 CPU on my 2700x. The 5950 is repeatedly crashing with WHEA errors - I may need to RMA it

 

[Markfw]

I’m folding with 28 CPU on my 5950 and 12 CPU on my 2700x. The 5950 is repeatedly crashing with WHEA errors - I may need to RMA it

I would say memory.... or settings, NOT CPU. I do have 3 of them.

Edit: what motherboard and what memory, and memory settings and voltage for memory ?

 

[Endgame124]

I would say memory.... or settings, NOT CPU. I do have 3 of them.

Edit: what motherboard and what memory, and memory settings and voltage for memory ?

This memory:

Are you a human?

www.newegg.com

Same Voltages as this post from March

Page 10 - PrimeGrid Challenges 2021

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Same Timings at this post:

Page 11 - PrimeGrid Challenges 2021

Page 11 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

System was and is rock solid using ECO mode. With PBO enabled however, the seems to whea when the CPU load suddenly drops to 0 - like when a WU completes. I can pretty much predict the errors by looking at the estimated completion time for a F@H WU.

 

[Icecold]

I'm pretty sure a number of people had issues with the early Ryzen 5xxxx series CPU's in terms of how high each core boosts before giving errors, etc. This post comes to mind - https://forums.anandtech.com/threads/5800x-fails-single-core-cycler-test.2591958/post-40467364

I have always run my 5950x's on ECO mode(without issues), so I never paid super close attention to it but AFAIK it's somewhat a known issue. Googling '5950x whea' comes up with a bunch of people RMA'ing their CPU's.

 

[Markfw]

This memory:

Are you a human?

www.newegg.com

Same Voltages as this post from March

Page 10 - PrimeGrid Challenges 2021

Page 10 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Same Timings at this post:

Page 11 - PrimeGrid Challenges 2021

Page 11 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

System was and is rock solid using ECO mode. With PBO enabled however, the seems to whea when the CPU load suddenly drops to 0 - like when a WU completes. I can pretty much predict the errors by looking at the estimated completion time for a F@H WU.

You didn;'t say what motherboard, but I would go back to the 3733/1866 and 1.4 volt. Mine are at 3800/1900 using 1.4v on an ASUS x570 pro WS

 

[Endgame124]

I'm pretty sure a number of people had issues with the early Ryzen 5xxxx series CPU's in terms of how high each core boosts before giving errors, etc. This post comes to mind - https://forums.anandtech.com/threads/5800x-fails-single-core-cycler-test.2591958/post-40467364

I have always run my 5950x's on ECO mode(without issues), so I never paid super close attention to it but AFAIK it's somewhat a known issue. Googling '5950x whea' comes up with a bunch of people RMA'ing their CPU's.

Yeah, got mine Jan 2020. From my reading, the issue is that the earliest 5950s are barely capable of the advertised clock speed. When they switch power states in PBO, the voltage doesn’t ramp as high as necessary to keep the system stable. The issue is especially common on 5950s where one chiplet has most of the issues. On many of them, the issue can be corrected by setting a positive voltage offset on the problematic cores. My system got a lot more stable (though not 100%) with an all core offset of +10. I’m actually working on a per core offset to see if I can get it stable just setting the offset for the 2 cores that have been generating most of the WHEA errors.

since I’ll likely move the system back to ECO mode full time, I haven’t committed to trying to RMA, mostly because I don’t want to drain the loop unless I have to.

 

[Endgame124]

You didn;'t say what motherboard, but I would go back to the 3733/1866 and 1.4 volt. Mine are at 3800/1900 using 1.4v on an ASUS x570 pro WS

I’m running an ASUS hero viii wifi, now with the newest bios.

 

[cellarnoise]

5950x I have found is very sensitive to memory and cpu overclock. Push too much and I get we errors and reboots. I have also found that F@H is more intensive than WCG. I need a lot more cpu voltage to push the limits than any other software so far when running (on 5950x) over 24 threads.

Though F@H on the CPU side really needs cpu cache. I can run all the way down to 4 cpu threads on F@H and does not hurt so much when also pushing a 3070 pgu.

I am currently running a 5950x at 1.1325 at 4.30 both ccx, with 4 cpu threads in F@H and the 3070 gpu. Ram is overvoltage 3200 32gb ram at 3466 not great timings but seems to be stable now over months. F@H cpu maxed out at 31 or 30 or down to 24 causes errors and reboots on this same system. I don't get errors with the gpu running underclocked and ram overclocked at 163Wand the cpu at 72 watts!

These new intel and amd cpus push the voltage and watts to the max for lower number of core clock speeds. Single core results are stupid anymore as modern O.S. seems to never run single core...

I'm tired, but Im running 164w 3070 gpu and 5950x at 4.3ghz at 72 watts at now 4 core F@H. Best points per total power so far. I can push like 2 percent higher for another total 50 watts. Not worth it!!!

Even in a challenge!

This is all with Chrome running with several windows on a Windows 10 system. It will produce another 200K to 300K once I shut chrome and HWinfo64 down. Crazy.

I am looking up Cache thrash monitoring programs. The "coin" producers seem to be realizing that large cpu caches do wonders also, so they will likely to continue to go the way of GPUs...

I bet the 5950x with large cache will be very tasty if any mortal can purchase...

 

[Markfw]

5950x I have found is very sensitive to memory and cpu overclock. Push too much and I get we errors and reboots. I have also found that F@H is more intensive than WCG. I need a lot more cpu voltage to push the limits than any other software so far when running (on 5950x) over 24 threads.

Though F@H on the CPU side really needs cpu cache. I can run all the way down to 4 cpu threads on F@H and does not hurt so much when also pushing a 3070 pgu.

I am currently running a 5950x at 1.1325 at 4.30 both ccx, with 4 cpu threads in F@H and the 3070 gpu. Ram is overvoltage 3200 32gb ram at 3466 not great timings but seems to be stable now over months. F@H cpu maxed out at 31 or 30 or down to 24 causes errors and reboots on this same system. I don't get errors with the gpu running underclocked and ram overclocked at 163Wand the cpu at 72 watts!

These new intel and amd cpus push the voltage and watts to the max for lower number of core clock speeds. Single core results are stupid anymore as modern O.S. seems to never run single core...

I'm tired, but Im running 164w 3070 gpu and 5950x at 4.3ghz at 72 watts at now 4 core F@H. Best points per total power so far. I can push like 2 percent higher for another total 50 watts. Not worth it!!!

Even in a challenge!

This is all with Chrome running with several windows on a Windows 10 system. It will produce another 200K to 300K once I shut chrome and HWinfo64 down. Crazy.

I am looking up Cache thrash monitoring programs. The "coin" producers seem to be realizing that large cpu caches do wonders also, so they will likely to continue to go the way of GPUs...

I bet the 5950x with large cache will be very tasty if any mortal can purchase...

Just to let you know... On a 5950x, F@H only needs one, or at the most 2 CPU threads for a GPU. So you can run your 3070 and 31 threads of F@H on the CPU. Try it, I bet it works. I only allow one thread, even on my EPYC systems, to the GPU support.

 

[crashtech]

Just to let you know... On a 5950x, F@H only needs one, or at the most 2 CPU threads for a GPU. So you can run your 3070 and 31 threads of F@H on the CPU. Try it, I bet it works. I only allow one thread, even on my EPYC systems, to the GPU support.

@Markfw , do you happen to be monitoring GPU utilization with and without a CPU slot? I have seen a pretty big performance hit when using over half the CPU threads...

 

[Markfw]

@Markfw , do you happen to be monitoring GPU utilization with and without a CPU slot? I have seen a pretty big performance hit when using over half the CPU threads...

Linux... Not sure of the exact utilization of everything, I just know after 15 years that you only need one thread for the CPU on linux to do F@H on a video card, and the rest can go to other things.

Windows could be an entirely different matter. Now under windows, here is this box, doing 23 threads of WCG(3900x), network monitoring of F@H and all BOINC, and chrome, and....

Looks like F@H does not take squat. And I am getting over 2 million ppd on a 1080TI on this box on a unit 18121.

 

[crashtech]

Something like GPU-Z or Afterburner is a good GPU monitor in Windows. It's not really the CPU use that is the problem, but other resources like PCIe bandwidth and cache usage. I have seen CPU intensive processes impact GPU usage even when the GPU app only uses one thread when nothing else is running. It's a thing.

 

[Icecold]

To test this out I just setup 31 thread CPU Slot on a 5950x (with a 3070 that was already running a GPU slot) running Linux Mint to see what happens. In Linux if you run nvidia-smi from a terminal it tells you the GPU utilization on the right hand side. I went from 99% utilization when running GPU only to bouncing around between 92 and 98% once I added the 31 thread CPU slot. My time per fold on the GPU slot dropped from 1 minutes, 16 seconds per fold, to 1 minutes, 21 seconds, and the PPD on the GPU slot dropped from 5,012,961 to 4,696,496. The CPU slot is doing 518,434 PPD, so it is still a net gain vs running the GPU alone, but it's definitely starving the GPU somewhat. I'm going to remove the CPU slot and go back to just Folding on the 3070.

I'm sure there are a lot of factors such as what project type the GPU Is running, what project type the CPU slot gets, etc. as far as how much it affects it, but that was the results on my quick test. In addition, if it was running BOINC on the CPU threads and not Folding@home, depending what project, etc. it may provide different results too.

 

[crashtech]

Afaict, you can run half the threads in a Folding slot without much impact on the GPU.

 

[StefanR5R]

a) about desktop CPUs and how they are configured by vendors and users

This is whole subject entirely alien to me (by now). I tried similar things when I had only a single computer for computationally heavy workloads. But these were workstation-like workloads with low parallelism and partial interactivity. When I purchased my first two dedicated computers for HPC-style applications which were better parallelized (still before I took on the DC hobby), that began to change my mind.

CPU clocks below 3 Ghz are good.

b) how to run F@h on GPU and whatever CPU workload alongside

I don't know about FahCore_22 on AMD GPUs. But on Nvidia GPUs, it is like many other GPGPU applications: A single-threaded process, consuming 100% CPU time if it can (that is, it certainly works in sort of a polling mode), evidently performing lots of PCIe I/O, presumably performing memory I/O as well.

That is, what Fahcore_22 evidently needs is 1 (one) hardware thread, preferably a fast one, and the I/O parts of the CPU. It is not obvious without some sort of profiling what else Fahcore_22 might need.

We have seen in the past that running PrimeGrid LLR based subprojects alongside Fahcore_2[12] impacts the performance of the latter especially. Whether this is due to LLR's cache footprint or due to LLR's memory I/O demands is not known to me.

Just going by the fact that Fahcore_a[78] is performing molecular dynamic simulation, I imagine it has got a sizable cache footprint and some degree of memory I/O demand too.

My own GPU-equipped computers were mainly built during times in which DC competitions were their sole purpose. Consequently, the dual-GPU computers among my GPU-equipped computers all have merely 4-core CPUs and are dedicated to GPU applications exclusively. Oh, I shall admit that I disregarded the "CPU clocks below 3 GHz are good" rule on these computers. On them, CPU clocks below 4 GHz are good.

 

[voodoo5_6k]

With the pinnacle of the holiday season slowly approaching, I'd like to take a short break from witty remarks, although the battle of F@H hills is still raging on...

<witty_remark_break>
I'm hoping that everybody who's going to celebrate Christmas is in good health and already in a "christmasy" spirit Everybody, have a great and merry Christmas and enjoy the time as much as possible And for those that don't care that much for Christmas, it's basically the same, health and a great time, just without the "christmasy" addition And while we do that, we enjoy a nice little folding race with a lot of fun and also teasing And whoever might win this race in the end, one important thing to keep in mind is that we're all doing a good deed for medical science So, have a good time everybody and stay safe
</witty_remark_break>

OK, and now back to the "battlefield" @Assimilator1, I do take it that the cheering gerbils are now wearing their holiday sweaters while cheering us to victory? I'm already excited to see, how much longer we can stay in front of the hard pushing Moonshot Gang... And if they already overtook us, how close we can dog them... Maybe, some of them already prefer nice and not too warm and/or loud living conditions during the holidays? We'll see, how long they want to bear their current situation

 

[StefanR5R]

Last year each team individually beat the prior years total points. Can we do that again this year?

At the current pace, it looks like we will miss this by about 4 days. Though in exchange we won an intense PrimeGrid challenge in parallel this time.

But at least TeAm AnandTech is on track to match the December 2020 points a few days from now (IOW surpass them by the end of the month).

 

[Pokey]

Hope everyone gets to enjoy the holidays at least a little, whatever your circumstance.



We are right at a week away from finishing this year’s event and I know everyone would like to affect the outcome, so do what you can, and let’s bring this one to a thunderous conclusion.

 

[biodoc]

Day 23 stats: The Moonshots take a slim lead.