Question Zen 2, clock speed vs power consumption vs ppd sweet spot?

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

TennesseeTony

Elite Member
Aug 2, 2003
3,801
2,617
136
www.google.com
It is crazy how much energy lessor tiers of PSUs waste. I have been feeling a bit like a zealot, preaching "GOLD! Titanium! Platinum, my brothers!" for the last few years. Thanks for the confirmation that even Silver PSUs are wasteful.
 
  • Like
Reactions: Assimilator1

Assimilator1

Elite Member
Nov 4, 1999
23,438
192
106
By chance I checked my PCs idle power earlier, it was ~80w.
Seems like a large part of that could be down to the 11yr old 80 white Corsair PSU I have! But that said, my PC almost never idles ;) (although a gold rated Corsair RM550x is in the pipeline for better loaded efficiency).
Oh Tony (et all), you might find this interesting, my PSU posts at OcUK (a thread I subscribed to donkey years ago! lol), efficiencies vs cost pay back.
 
Last edited:

Endgame124

Senior member
Feb 11, 2008
288
182
116
By chance I checked my PCs idle power earlier, it was ~80w.
Seems like a large part of that could be down to the 11yr old 80 white Corsair PSU I have! But that said, my PC almost never idles ;) (although a gold rated Corsair RM550x is in the pipeline for better loaded efficiency).
Oh Tony (et all), you might find this interesting, my PSU posts at OcUK (a thread I subscribed to donkey years ago! lol), efficiencies vs cost pay back.
One thing to note on power supplies greater than 5 years old, is that they start to lose efficiency. [H] did periodic reviews of their supplies that were 10 years old, and were often showing a 5% (or more) efficiency loss. I would direct link to one of the articles, but with the site being gone and all..
 

Endgame124

Senior member
Feb 11, 2008
288
182
116
Here is the matching forum thread - looks like the supply did well, but no great summary of the results:

 

Assimilator1

Elite Member
Nov 4, 1999
23,438
192
106
One thing to note on power supplies greater than 5 years old, is that they start to lose efficiency. [H] did periodic reviews of their supplies that were 10 years old, and were often showing a 5% (or more) efficiency loss. I would direct link to one of the articles, but with the site being gone and all..
The website's gone!? Damn! Didn't know that :(, admittedly I didn't go there often, but I did like it!

Hadn't even crossed my mind that they would go 'off' over time, that was really cool that H OCP did some 10yr testing! :)
I found page 5 btw! :) (DC output quality), graphs aren't clickable though. Well the voltage ripple had hardly changed, I guess we won't know how the rest of test went :(.
 
Last edited:

Endgame124

Senior member
Feb 11, 2008
288
182
116
The website's gone!? Damn! Didn't know that :(, admittedly I didn't go there often, but I did like it!

Hadn't even crossed my mind that they would go 'off' over time, that was really cool that H OCP did some 10yr testing! :)
I found page 5 btw! :) (DC output quality), graphs aren't clickable though. Well the voltage ripple had hardly changed, I guess we won't know how the rest of test went :(.
They did 3 or 4 10 year tests that I remember.

To the best of my memory:
A silver stone 1000w where the power supply literally blew up on the final stress test (max rated power draw in a hot box at like 50c)

The Seasonic one where the power supply was very close to its original specs

A cooler master one where the power supply passed but only met the 80+ standard instead of gold

And, I think, a pc power and cooling supply that was really old, and didn’t explode but had fairly poor performance.
 

Assimilator1

Elite Member
Nov 4, 1999
23,438
192
106
It's a shame that he shut his site down, reviews like that are extremely rare, if not unique. I can't understand why he shut his site down, sure he'd be too busy (or maybe no longer interested) to add to it, but why not just leave it open as it is? Unless it was costing him money to run rather than making money for him? Then I could understand it.
Anyway, good information lost :(, presumably (although I bet Kyle himself still has his site on a HDD somewhere)......

On a different note, it's a pity we've got no easy accurate way to benchmark any BOINC project! (AFAIK) I'm sure someone mentioned it could be done somehow ;)........

PS Did everyone get a PM from the community manager? including a link to this thread! ;)
 
Last edited:
  • Like
Reactions: biodoc

Assimilator1

Elite Member
Nov 4, 1999
23,438
192
106
I've pulled the trigger and bought the Seasonic Prime Ultra 550 :), although it's rather pricey for a 550w unit (£115 incl delivery), the 12yr warrantee and it's Platinum rating together made me decide on that one over the Corsair RM550x, which is gold rated & has a 10yr warr. I've worked out that with a gold rated (550w) PSU I'd be saving £20.32/yr over my basic80+ Corsair TX650 on electricity costs, and with a Platinum (650w, so a 550w would be a tiny bit better) I'll be saving £24.44/yr (if ya wanta check out the maths see here). So when/if I could've got the RM550x for £88 (it's currently £111) it would've paid itself back in ~4yrs 4mths, the Ultra 550 will pay itself back in ~4yrs 8mths (& have the aforementioned longer warrantee). That's with running my main rig 24/7, 9mths of full CPU load at 115w from the PSU, 3mths of full CPU+GPU load at 233w per yr.
 
Last edited:
  • Like
Reactions: biodoc

Assimilator1

Elite Member
Nov 4, 1999
23,438
192
106
New PSU arrived this morning, fitted it this afternoon, running Rosetta@home (same WUs) power draw from the wall went from 129w to 112w, a nice 17w reduction :cool:.
That's from my 11yr old Corsair TX650 80+basic rating to a Seasonic Prime Ultra 550w platinum. I calculated & hoped for a 15w reduction in wall power draw (that was with 650w platinum ratings), I guess the extra 2w is for getting a 550w platinum PSU ;). Btw, that also indicates that my old PSU's efficiencies are still pretty much spot on at that light load! Old faithful can have a peaceful retirement now, lol. Although I might put it into my 2nd rig.
 
Last edited:

Assimilator1

Elite Member
Nov 4, 1999
23,438
192
106
So for the sake of efficiency I should get 16C CPU & an RTX 2080 then? :D

Btw, does anyone else think the choice of colour in that graph is utterly stupid for the 2 different voltages? Blue, and err blue! Wth!? lol :rolleyes: :tearsofjoy:
 
  • Haha
Reactions: Endgame124

Endgame124

Senior member
Feb 11, 2008
288
182
116
So for the sake of efficiency I should get 16C CPU & an RTX 2080 then? :D
See, you understand! It’s really the efficient thing to do until Ryzen 4 and the 3000 series vid cards come out.

Btw, does anyone else think the choice of colour in that graph is utterly stupid for the 2 different voltages? Blue, and err blue! Wth!? lol :rolleyes: :tearsofjoy:
Didn’t even think of that. Should be like blue and orange or something.
 
  • Haha
Reactions: Assimilator1

Assimilator1

Elite Member
Nov 4, 1999
23,438
192
106
Lol :D
I forgot to do an idle test earlier, it's ~69w :cool: , not that we ever let our PCs idle though! ;)
 

StefanR5R

Diamond Member
Dec 10, 2016
3,433
3,674
106
So for the sake of efficiency I should get 16C CPU & an RTX 2080 then?
I am running CPU projects often and GPU projects rarely. Plus, I ended up with computers which are solely dedicated to Distributed Computing. I therefore built computers of two types:
  • high core count, low core clock, ECC memory, no GPU (or a little low-power GPU on boards without BMC)
  • low core count, high core clock, no ECC memory due to choice of platform, dual high-power GPUs
Two of my computers deviate: The "daily driver" (at home) was built before I got into DC. It is a Haswell Xeon E3 with ECC memory and iGPU. Then I have a former Broadwell-E computer, once used as a desktop at work, now with Xeon E5, re-used non-ECC RAM, and 3 high-power GPUs. This system doesn't fit in and so far it doesn't look like I will pursue triple-GPU systems further, but who knows.

Hmm... is it possible to power down discrete GPUs entirely when not needed? (My Pascals pull 8 W when idle, possibly a bit more if the X11 display is running on them. Early reviews of Turing GPUs showed even higher idle consumption.) GPU power down/ power up doesn't need to happen when the system is up and running, it'd be perfectly fine if I had to shut the computer down to make the switch. — E.g., unplug the 6/8 pin power cables but leave the GPU in the PCIe slot? Maybe I should just try this instead of asking around. :-) — This would make a combination of large CPUs with large GPUs more attractive. But still, when I think of a GPU use case like TeAm AnandTech's annual Folding race, the split computer types described above are a better fit.
 

Assimilator1

Elite Member
Nov 4, 1999
23,438
192
106
I suspect pulling the power on the GPUs might just make the system crash, if not I bet they'll still pull their idle power through the PCIE slot.

On a separate note, I think something's gone astray with my maths, I'd previously estimated that the DC draw from my old PSU was 115w, yet my rig is now pulling 112w from the wall! Unless my PSU is over 100% efficient, lol, then I've messed up somewhere......ah no, I remember now, the 1TB HDD died, so I removed it and didn't replace it, I think that cut power usage by about 8-10w, IIRC. And my calculations were based on the power draw with the 1TB being installed & a 650w Platinum PSU.

I'll start over, with my old Corsair TX650, wall draw with Rosetta (only) running was 129w. An 80+ basic rating should mean at least 82, 85, 82% efficient for 230v at 20, 50 & 100% load.
The 129w wall draw would translate to about 106w DC output, (assuming 82% eff. at this ~16% load. You'd think it'd be less at that load, but apparently this review of the TX650 shows that it can still hold 82% eff. @~14% load, and that's at 115v!, hmm.....).

A Platinum rated 550w PSU at 230v should have 92, 94, & 90% efficiency at 20, 50 & 100% load.
So a 106w DC load (19%, so ~92% eff.) is about 115w wall draw, (pretty close to the actual figure :)). That's a 14w power saving.
With the TX650 the GPU loaded (+10T CPU) wall draw was ~260w, less 10w for the removed HDD, so ~250w would work out to about 209w DC output (~32% load, so estimating 83.5% eff).
A Platinum 550w PSU @230v with 209w DC output (estimating ~93% eff. at the 38% load) should give a wall draw of 225w, a 25w saving.

So assuming 9 mths CPU crunching & 3 mths of combined gaming/GPU crunching per year 24/7, and my electricity unit charge is 13.44p/kWh, that's a saving of £12.33/9 mths + £7.35/3 mths, a total of £19.68/yr. The Seasonic PU 550 cost me £115 (incl delivery), so it will pay itself back in ~5yrs 10mths at current elec prices (which will likely only go up, cutting the payback time). Just thought, as the actual CPU crunching power reduction is 17w, so an extra whole 3w ;), it's actually saving about £3.53/yr on top of the above calculation (a total of £23.21/yr), so that changes payback to just under 5yrs :).
Not quite as good as I originally thought, but still worthwhile as this particular PSU with it's 12yr warrantee is EOL & will shortly be unavailable new, also the prices of the slightly lesser Seasonic PX 550 (Focus Platinum) are virtually the same atm and it has a slightly shorter 10yr warrantee (and higher prices for the EVGA P2 650, the only other similar power rated Platinum PSU with a 10yr warrantee).

Btw, in doing all this I discovered that I'd made (another) maths error at my original OcUK post, I had lumped 12mths/yr of CPU crunching on top of 3mths/yr of GPU crunching! Corrected now, which obviously pushed up the pay back time (by about a year).

Have I made any maths errors this time??lol
 
Last edited:

Endgame124

Senior member
Feb 11, 2008
288
182
116
Is your 3600 dedicated to Rosetta, or are you mixing projects? I’m hoping to get a feel for what a single Ryzen 3 CPU will do in average credit (I know there are plenty of issues with watching credit in Rosetta, but loose estimates are better than no estimates)
 

StefanR5R

Diamond Member
Dec 10, 2016
3,433
3,674
106
FWIW, here are my dual-EPYC 7452 (dual 32c/64t Zen 2) numbers with 100 % Rosetta@home load from early April again:
average core clock 2.6 GHz — 340 W power draw at the wall — 82,800 PPD host credits, averaged from >130 results

So that's ≈250 PPD/thread/GHz, and ≈240 PPD/W.
 

Endgame124

Senior member
Feb 11, 2008
288
182
116
FWIW, here are my dual-EPYC 7452 (dual 32c/64t Zen 2) numbers with 100 % Rosetta@home load from early April again:
average core clock 2.6 GHz — 340 W power draw at the wall — 82,800 PPD host credits, averaged from >130 results

So that's ≈250 PPD/thread/GHz, and ≈240 PPD/W.
I forgot that you had posted that. That is some very solid efficiency, and it’s just a single host to manage.
 

Assimilator1

Elite Member
Nov 4, 1999
23,438
192
106
After the race I was mixing LHC and Rosetta, but LHC ended up hogging CPU time so I stopped it getting new tasks.
Not sure when I stopped LHC, but looking at the stats threads it was somewhere between the 8th and 17th, since then it's been all Rosetta (bar occasionally throttling BOINC when playing Elite dangerous).
Bear in mind that my CPU is slightly underclocked via a 65w PPT, as per the op.
How long does RAC take to settle down?
I think I might've have mentioned the RAC of this rig (and my 2nd rig) during the race, but it's probably in the related mega-thread, so good luck finding that! ;)
[edit] This might be a better source, shows 7day average of 11.9k ppd.
 
Last edited:
  • Like
Reactions: Endgame124

StefanR5R

Diamond Member
Dec 10, 2016
3,433
3,674
106
Rosetta@home —
How long does RAC take to settle down?

[edit] This might be a better source, shows 7day average of 11.9k ppd.
Your FreeDC link as well as your hosts list at R@h show the host ID 1761696. Boincstats has got a record of the last 60 days of production: It is showing about 12 kPPD during the last few days.

My April 6 stats, divided by PPT: 82,800 PPD / (2 · 155 W) ≈ 270 PPD/W*
Your June 21 stats, divided by PPT: 12,300 PPD / 65 W ≈ 190 PPD/W*

Looks reasonable, but we have no way of knowing whether the jobs on April 6 and on June 21 are suitable to compare.

*) Note that these are Watts at the processor package, not at the wall. And not actual Watts, but the ceiling which the processor firmware attempts not to crack.
 
Last edited:

ASK THE COMMUNITY