Anandtech vs Tom's Hardware Folding@Home Coronavirus Race thread

[StefanR5R]

What is more effective for my old computer rosseta@home or F@h
I have 2500k @3.3 Ghz, AMD Radeon 460 RX
or both rosseta for CPU, F@h for GPU ?

Welcome @Ladicek!

The GPU can be used at F@h but not at Rosetta. Here is a sheet which estimates PPD/Watt of RX 460 being at the same level as larger RX 400/ RX 500 cards, and same as GTX 1050/ 1050Ti, and still quite a bit more efficient than several older but bigger GPUs.

The CPU can be used at F@h or at Rosetta. But F@h needs (and awards) quick return of individual work units because new work is created based on returned results of previous work units. A smaller CPU is disadvantaged in this respect, therefore Rosetta@home may be more suited to run on this CPU.

If you switch the CPU to Rosetta but use the GPU for F@h, then watch how much CPU the FahCore process uses. If it has to drive larger GPUs, especially Nvidia GPUs, it easily wants a full CPU core for itself for each GPU. Smaller GPUs and AMD GPUs put lower demand by FahCore on the CPU. That is, you may need to restrict BOINC to "Use at most 75% of the CPUs" to avoid bottlenecking the CPU portion of F@h's GPU application.

 

[borandi]

Nice
But why is the assignment rate zero for that 1 along with 2 of the azure servers?.......... has 'No CS', whatever that means ( no explanations there)

To add in my 2p, I believe they're still setting them up. Azure 4+5 are relatively new. They're added into the stats page before they are fully ready to be turned on

 

[Assimilator1]

Stefan
I thought anytime you ran F@H on the GPU a spare CPU core is needed regardless of GPU type?
Short of looking at crunching times afterwards, how would we know?

Ladicek
To change BOINC CPU usage you need to change from simple view to advanced view by hitting view>advanced view, then options>computing preferences, under 'Computing, Usage limits', 'Use at most xx% of the CPUs'. If you need to lower it .

 

[StefanR5R]

@Assimilator1, a few years ago I had a FirePro W7000, which is almost the same as an HD 7850. At times I had a GTX 1070 in the same computer. If I remember correctly, CPU usage of the process which drove the W7000 was not as high as that which fed the GTX 1070. Granted, I had the Pro driver installed with the W7000, not the gaming driver, but I suspect this didn't make the difference.

You can overcommit the CPU if you want to. After all, the process which drives the GPU is most of the time merely polling the GPU for results, and now and then copying data to/ from VRAM (more so on Windows than on Linux). If FahCore_21/22 is like several other GPGPU applications of the BOINC world, then it may periodically perform some double precision calculations for NVidia GPUs but not for AMD GPUs. Just at those times, you'd prefer the execution units of the CPU to busy themselves with FahCore_21/22's needs, the rest of the time you primarily need the CPU's I/O capabilities there.

Note that Rosetta@home does take away some of the precious RAM bandwidth of the CPU. But it is not as bad as the notorious PrimeGrid LLR applications in this regard. Running PrimeGrid LLR alongside GPGPU applications has been seen to impact the latter negatively even if the "spare one core per GPU" rule was observed.

Short of looking at crunching times afterwards, how would we know?

Crunching times are not a good measure in Folding@home, because they vary so widely between WUs even if you keep the CPU free entirely. It is better to monitor GPU shader utilization while you run both Folding@home and the CPU workload. Then, reduce the CPU workload until you get almost as high shader utilization as without CPU workload.

 

[StefanR5R]

Day 10: Saturday, March 28
14:00 UTC stats
(10 AM EDT, 7 AM PDT)
links: official race stats, TeAm AnandTech at EOC, Tom's Hardware at EOC

Daily production


Our advantage


Some of us have good task supply, others not so much... The latest three new workservers are not in full operation yet, but will be soon, hopefully.

 

[Assimilator1]

Thanks for the stats Stefan

Re shader utilisation, I don't see it in GPU-Z (only GPU load), what do you use to monitor it?

 

[StefanR5R]

In GPU-Z, the "GPU Load" sensor should be the right one.

I am using nvidia-smi dmon -d # in the command line, with # being the reporting interval in seconds.

 

[Markfw]

20m last 24 hours for me !

 

[Pokey]

Another FYI:
I tried client-type advanced on one rig but it got stuck when contacting work server so I switched back to beta and it got work immediately.
So beta is the only way to go as far as I can tell.

 

[Markfw]

Per Ken's graph, I think we have this under control now.

 

[Endgame124]

One of my good friends has joined on Tom’s side adding his sli 1070s to the mix. It’s a given that I Can’t let him generate more points than me, I fired up an old core 2quad with a rx250 in it. I suspect this will not really allow me to close the gap, so what is the preferred direction for building a folding rig?

Cheap cpu and minimal ram and a Pair of larger cards (2070?), bigger cpu (threadripper?) and single large card? Bigger cpu and like 4x 1660s?

 

[Markfw]

One of my good friends has joined on Tom’s side adding his sli 1070s to the mix. It’s a given that I Can’t let him generate more points than me, I fired up an old core 2quad with a rx250 in it. I suspect this will not really allow me to close the gap, so what is the preferred direction for building a folding rig?

Cheap cpu and minimal ram and a Pair of larger cards (2070?), bigger cpu (threadripper?) and single large card? Bigger cpu and like 4x 1660s?

Is money a consideration ? best bang/buck are a pair of 2060's. And since they only need on 6 or 8 pin power, you don't need a huge PSU. CPU ? I would do a 3600 or 3600x (not sure the pricing right now).

With more budget, twin 2070 supers might be OK.

 

[Endgame124]

Is money a consideration ? best bang/buck are a pair of 2060's. And since they only need on 6 or 8 pin power, you don't need a huge PSU. CPU ? I would do a 3600 or 3600x (not sure the pricing right now).

With more budget, twin 2070 supers might be OK.

We are talking about it now. We might each give ourselves a set budget 2500(?) and then see who can generate more points.

i think I have the edge- he only has the 1 system (OCd intel 8700k and the sli 1070). I have a Ryzen 7 2700x, a trinity apu, a Kaveri apu, and the old core 2 quad 9650. I think I could just spend most of my budget on video cards, drop 1 in each system, and be set. However, I’m not sure if those old, low powered cpus would actually hinder video card perf? Maybe I’m better to just get another new system?

 

[Markfw]

We are talking about it now. We might each give ourselves a set budget 2500(?) and then see who can generate more points.

i think I have the edge- he only has the 1 system (OCd intel 8700k and the sli 1070). I have a Ryzen 7 2700x, a trinity apu, a Kaveri apu, and the old core 2 quad 9650. I think I could just spend most of my budget on video cards, drop 1 in each system, and be set. However, I’m not sure if those old, low powered cpus would actually hinder video card perf? Maybe I’m better to just get another new system?

OK, with that budget... Get 2 of these: https://www.newegg.com/evga-geforce-rtx-2080-ti-11g-p4-2281-kr/p/N82E16814487418?&quicklink=true
Put them in the 2700x system. You may need a new PSI, at least an 850, and a platinum one. That 5-6 million ppd in one box.. all of that may be under 2500. If so, make sure you have plenty of fans, and you might even leave the side cover off as well.

 

[Endgame124]

OK, with that budget... Get 2 of these: https://www.newegg.com/evga-geforce-rtx-2080-ti-11g-p4-2281-kr/p/N82E16814487418?&quicklink=true
Put them in the 2700x system. You may need a new PSI, at least an 850, and a platinum one. That 5-6 million ppd in one box.. all of that may be under 2500. If so, make sure you have plenty of fans, and you might even leave the side cover off as well.

I would be tempted to add them to my WC loop. I have 2x 3 fan rads with just the processor on the loop at the moment. Not Sure water blocks would have a roi in ppd though... originally I was going to add the 1080ti to the loop, but never got around to it. The system already has a seasonic 1kw power supply in it, so I’m likely set for power.

in this case, the 1080ti would move to the A10-7870K. If there was any money left, I would probably add a cheap card to the A10-5800K

 

[Markfw]

I would be tempted to add them to my WC loop. I have 2x 3 fan rads with just the processor on the loop at the moment. Not Sure water blocks would have a roi in ppd though... originally I was going to add the 1080ti to the loop, but never got around to it. The system already has a seasonic 1kw power supply in it, so I’m likely set for power.

in this case, the 1080ti would move to the A10-7870K. If there was any money left, I would probably add a cheap card to the A10-5800K

No, for those video cards, don't touch them, don't void the warranty. I have 5 EVGA 2080TI's, and they are great. They cool just fine. Put your 1080TI in core2 quad 9650, it has enough to keep that 1080TI fed. And the 2700x ? even with those 2 cards, you will have at least 12 threads free. 1 cpu and maybe even one smt cpu is enough for those video cards. so 2-4 threads leave available for the cards.

Edit, with that 1080TI that 6-7 million ppd if you do what I suggest.

 

[Endgame124]

No, for those video cards, don't touch them, don't void the warranty. I have 5 EVGA 2080TI's, and they are great. They cool just fine. Put your 1080TI in core2 quad 9650, it has enough to keep that 1080TI fed. And the 2700x ? even with those 2 cards, you will have at least 12 threads free. 1 cpu and maybe even one smt cpu is enough for those video cards. so 2-4 threads leave available for the cards.

Edit, with that 1080TI that 6-7 million ppd if you do what I suggest.

So it’s better to consolidate the GPUs into one system instead of spreading them around? If I put the 1080ti in the core 2 quad, what do I do with the apu hosts? The integrated video is in the realm of a RX250 if I remember correctly... just power them off?

 

[Markfw]

So it’s better to consolidate the GPUs into one system instead of spreading them around? If I put the 1080ti in the core 2 quad, what do I do with the apu hosts? The integrated video is in the realm of a RX250 if I remember correctly... just power them off?

Since F@H only uses GPUs, and those APU are so slow, its not worth powering them on for the CPU or the GPU portion. So, the 1080ti going in the one with no apu, and thats over 1 million ppd. Then add the 2 new cards to the one that has plenty of cores to supply the 2080TIs with stuff.. 2 PC's and 6-7 million ppd.

 

[Endgame124]

Since F@H only uses GPUs, and those APU are so slow, its not worth powering them on for the CPU or the GPU portion. So, the 1080ti going in the one with no apu, and thats over 1 million ppd. Then add the 2 new cards to the one that has plenty of cores to supply the 2080TIs with stuff.. 2 PC's and 6-7 million ppd.

My gut tells me the 7870k is faster than the q9650. I’ve got all 3 of them running Rosetta right now, so we see where they land in a few days. None the less, I see where you’re coming from, and saving the power from shutting down two systems would be good - I’m pretty sure what I’ve got running now will overwhelm the excess my solar is generating... at least it will when the 1080ti gets consistent work.

 

[Markfw]

My gut tells me the 7870k is faster than the q9650. I’ve got all 3 of them running Rosetta right now, so we see where they land in a few days. None the less, I see where you’re coming from, and saving the power from shutting down two systems would be good - I’m pretty sure what I’ve got running now will overwhelm the excess my solar is generating... at least it will when the 1080ti gets consistent work.

Solar ? Those 2 2080TIs and CPU loaded will take 700 watts, maybe a little more.

 

[Endgame124]

Solar ? Those 2 2080TIs and CPU loaded will take 700 watts, maybe a little more.

I have 9kw of solar panels on one side of my roof. Last month (feb 13-mar 13) I produced 336 kWh more than I used. I’ve been tracking my usage as I add more computers and while I’m generating more power as the days get longer, I’m also using more of the excess. I’ve been considering putting in another 8.5 kw of panels, but the west side of my house isn’t as good as the East side, (and my wife doesn’t want them on that side, but who is counting that )

700watt sustained load would crush my winter production, but that is what summer is for, right?

 

[JeffCos]

Hello, it's been a while since I've been here, but it's great to see some familiar names from when I was a regular.

Maybe someone knows the answer to this: I remember reading on one of the forums years ago, probably when GPUs were first added to F@H, that there were CPU specific work units that were vital to the projects and that everyone should still keep their CPUs folding even though the points were very low. Does anyone know if these WUs that CPUs can do and GPUs can't still exist or has everything been reworked so that any work unit can be deployed to either CPU or GPU?

F@H ~ DainBramaged

 

[StefanR5R]

So it’s better to consolidate the GPUs into one system instead of spreading them around?

Back at a time when I still had air-cooled GPUs, I found cooling a single GPU per ATX computer case hard enough, and two near impossible. But then I am somebody who is not even considering to operate high speed fans.

I remember reading on one of the forums years ago, probably when GPUs were first added to F@H, that there were CPU specific work units that were vital to the projects and that everyone should still keep their CPUs folding even though the points were very low. Does anyone know if these WUs that CPUs can do and GPUs can't still exist or has everything been reworked so that any work unit can be deployed to either CPU or GPU?

I am not sure about this, but I found this recent comment from a user in the foldingforum:

At present work is generated to use the available resource -- most project have big compute requirements and are aimed at GPUs, those with smaller requirements are aimed at CPUs. The split can be adjusted slightly to keep both resources loaded. There is more work to do than resource to do it (the recent inability to serve the work fast enough is a different matter, and is being addressed urgently. Even then it was NOT a lack of work but of server capacity.)

 

[Assimilator1]

Hey JeffCos, good to see you about. (as a side note, in your early days of the team I remember telling my then young son about your team name, we had a good chuckle about it ).

 

[mopardude87]

Been running so long online folding on the cpu and gpu that LOL i think the L2 or L3 cache must be choked or something? Idk i could be wrong but till i reboot i get 86% or so cpu usage during folding and entire rig is now choking HA. I been wanting to game too but been pretty dedicated to keeping the 2 rigs running as long as possible for the cause. Prob need to reboot before i can fold on the cpu. Shame this rig refuses to pick up a gpu wu, the other rig is getting both non stop it seems.

Edit: hehe speak of the devil it picked up the wu for a gpu on a cause and continue, i figured out the single pause option just for cpu I forgot who told me this trick but thank you again Been working like a charm every time it gets stuck.