WCG 24/7 and Overclocking question

[ArbitraryAffection]

Hi guys/gals ~

Just a quick question, is it worth (safe?) maybe overclocking my CPUs that are running 24/7 WCG? I don't mean heavy OC, but something like a 300-400Mhz increase on the 1700's (from 3.2 all core to 3.5-3.6) hopefully with minimal increase in power use. Is that worth it? From my understanding if one of my CPU's errors and messes up a result it will be compared to another bunch of the same work results from known stable PC's for validation then discarded. So it's not the end of the world. BUT obviously i want to never create bad results. All my CPU's are 100% stock atm, but i could maybe crunch that many more numbers by bumping the core speed up, especially on my 1700's.

Thoughts, opinions, advice greatly appreciated, thanks

 

[Deleted member 178884]

By overclocking, yes will you will increase output safely but you'll have to decide whether that's actually worthwhile being a 24/7 rig you'll want to optimize it for power efficiency, that 300-400mhz may lead to a 50w~ increase in power draw for example which might not be worthwhile in terms of performance-watt which is essential in a cruncher.
If you do pull off the OC make sure to push that vcore as low as possible to make it easier to cool / lower power draw a bit, and avoid an auto overclock of course.

 

[mstenholm]

Hi guys/gals ~

Just a quick question, is it worth (safe?) maybe overclocking my CPUs that are running 24/7 WCG? I don't mean heavy OC, but something like a 300-400Mhz increase on the 1700's (from 3.2 all core to 3.5-3.6) hopefully with minimal increase in power use. Is that worth it? From my understanding if one of my CPU's errors and messes up a result it will be compared to another bunch of the same work results from known stable PC's for validation then discarded. So it's not the end of the world. BUT obviously i want to never create bad results. All my CPU's are 100% stock atm, but i could maybe crunch that many more numbers by bumping the core speed up, especially on my 1700's.

Thoughts, opinions, advice greatly appreciated, thanks

Not worth if you have to pay for the electricity. These AMDs jump up quit a bit. Let's ask @thebluebumblebee and @Norton that I know uses previsous gen Ryzen if they OCed and measured theirs. I tried on mine but decided stock is fine.

 

[ArbitraryAffection]

I have a power meter installed so i might try a mild OC and see how much power use goes up. Stock voltage on the 1700 for 3.2 GHz all core seems to be around 1.03V to 1.04V (which from my understanding , ~1V is the optimal voltage in perf/watt for Zen CPU's) I don't think I would go over 1.1. My main worry is creating bad results. I don't wanna taint my record or anything :s

Kinda off-topic: What ISA do the various projects/tasks use? I assume it is SSE or AVX and heavy FPU use? Just out of curiosity, it's particle simulations right? I wonder if the wider FPU on Zen2 will massively increase WCG crunching performance. 7nm will for sure be a great boon for perf/watt though

Not worth if you have to pay for the electricity. These AMDs jump up quit a bit. Let's ask @thebluebumblebee and @Norton that I know uses previsous gen Ryzen if they OCed and measured theirs. I tried on mine but decided stock is fine.

I do have to contribute to the power bill, but I would honestly be OK paying for the increase if my GFLOPS went up and i could crunch more numbers

 

[thebluebumblebee]

I have to see a no risk benefit in order to OC and I'm not seeing that with Ryzen. w1zzard's review of the 2xxx Ryzens indicated that manually OC'ing is a waste of time. The only thing I which I could do with my Ryzen's is to run the Ryzen DRAM Calculator. That would require a temp install of W10 and I'm just too lazy.

 

[ArbitraryAffection]

I have to see a no risk benefit in order to OC and I'm not seeing that with Ryzen. w1zzard's review of the 2xxx Ryzens indicated that manually OC'ing is a waste of time. The only thing I which I could do with my Ryzen's is to run the Ryzen DRAM Calculator. That would require a temp install of W10 and I'm just too lazy.

What about my 1700's though? Quite a lot of oc head room. I mean these chips could potentially get a 700-800 MHz bump. Btw interesting note: my 2700X at stock is drawing 120-130W in software package power, for 3.95GHz all core. That's 100% more power use than the 1700 but only around 25% more performance. So maybe I won't OC them at all :X

 

[hat]

If this is the 1700 and not the 1700x (which includes XFR, which is automatic overclocking) then you can gain a few hundred MHz from bumping it up. The x chips though, are pretty much full tilt out of the box, so you're already past the point of diminishing returns the instant you mess with the settings.

As others have mentioned, though, either way you look at it you're increasing the power draw. That also means increased heat, so you have to have the cooling to handle that, and of course it also means increased heat generated in the house, so if you overclock everything it might get noticeably warmer than if you hadn't. Or your air conditioning is now going to be working harder to keep it cool, resulting is more money spent on the power bill.

As far as generating bad results, that definitely can happen, but if you do the proper stability testing first, it shouldn't. I have a really extreme approach to stability testing compared to the norms these days, bordering madness some people might say... but my rigs never crash (or turn out bad results)!

 

[ArbitraryAffection]

If this is the 1700 and not the 1700x (which includes XFR, which is automatic overclocking) then you can gain a few hundred MHz from bumping it up. The x chips though, are pretty much full tilt out of the box, so you're already past the point of diminishing returns the instant you mess with the settings.

As others have mentioned, though, either way you look at it you're increasing the power draw. That also means increased heat, so you have to have the cooling to handle that, and of course it also means increased heat generated in the house, so if you overclock everything it might get noticeably warmer than if you hadn't. Or your air conditioning is now going to be working harder to keep it cool, resulting is more money spent on the power bill.

As far as generating bad results, that definitely can happen, but if you do the proper stability testing first, it shouldn't. I have a really extreme approach to stability testing compared to the norms these days, bordering madness some people might say... but my rigs never crash (or turn out bad results)!

Yeah they are 1700 non X's. Btw we have no AC, most houses here in UK don't XD so I just keep my window open now. Actually I am considering getting a portable AC unit for the summer.

What would you suggest for stress testing? I usually use prime95 small ffts, CPUz and aida64 extreme stress test

I am tempted to increase core frequency on the 1700's to maybe 3.6 while keeping voltage below 1.1. the stock cooler should handle it

 

[hat]

Well, my standard stress test consists of running 100 passes of Linpack Xtreme (the Bootable Linux version). This test tends to weed out a bad OC very quickly, so 100 passes probably isn't necessary (the default 20? would probably work) but I like to do 100 just to be sure.

After that, it's the latest version of Prime95 (including the apparently terrible AVX everyone seems to hate) for 24 hours. I've had an OC pass Linpack Xtreme, only to fail P95 after a disappointingly long period of time (think it was somewhere between 6-8 hours). But if it passes Linpack Xtreme and Prime95 for 24 hours, I call it good.

Note that AVX Prime95 will crank out pretty high temps, higher than what WCG will do. Linpack Xtreme heats it up even more than AVX Prime95. So, whatever temps you see during Linpack Xtreme are likely to be the highest you'll ever see, no matter what you do with your rig. WCG load is lighter, and will produce lower temps than what you see here... so if your temps are acceptable during Linpack Xtreme, you know your WCG temps will be good. For example, I might allow temps between 80-90c during Linpack Xtreme, but only because I know it's the absolute max temp my chip will ever get to. I don't like temps that high, but I'm okay with it during the test. If I saw temps like that while running WCG though, I'd shut it down immediately and start over...

That's just my method, though. Many will tell you I'm insane for doing that... but, like I said before, at least my rigs don't crash and my temps are nice and low when I'm not torturing them.