Question Zen2 vs Zen3 improvements for DC?

[StefanR5R]

The VirtualBox based "rosetta python projects" application took a lot of RAM, I would have to look up how much exactly. This one is not active at Rosetta@Home now, but there was a time during which they had only work for this one and not for the classic Rosetta application.

NFS@home has got five applications which work on differently large numbers. They take 0.5/ 0.8/ 1.0/ 1.0/ 1.25 GB per task respectively. Work for the 0.5 GB type is not always available. And PPD is higher for the larger applications, and of course highest for the 1.25 GB one. SMT is beneficial for these applications, so a 12c/24t CPU could take up to 24× 1.25 GB = 30 GB if the host has got this much plus what is needed for the OS and everything else.

yoyo@home's ECM P2, which gives higher PPD than other applications there, can take several GB per task, much dependent on the particular workunit. Last time I ran it myself there were frequently workunits which took 8 GB individually, randomly mixed with "smaller" ones. There have been, or still are, ECM P2 workunits with even larger RAM footprint.

 

[Assimilator1]

So I finally installed the new CPU yesterday, a 5800X (updating the BIOS beforehand, from v1.8 to v1.H4, latest beta).
Set the PPT to 65w (same as the 5800 and 5700X according to wiki) and ran BOINC.
Base clock is 3.8 GHz for the 5800X and 3.4 GHz for the 5800 & 5700X (all 8C/16T CPUs).

Running Einstein it was fine, clocking at about 3.8-3.9 GHz, but running Asteroids it only clocked to about 2.8 GHz! (with an unsurprisingly cool 54C CPU temp).
Now although the 3600 also ran below its base clock of 3.6 GHz with Asteroids (only), it was only 100 MHz under (which interestingly the beta bios closed the gap by about 50 MHz it seemed). But of course that CPU was intended to run at 65w PPT, vs my 5800Xs 105w spec.
That said I'm somewhat surprised by how far under clock it was, considering the 5800 & 5700s base clock, perhaps their later stepping helps with low power high load, high core number work?

What I found remarkable was how consistent the wall draw is with the PPT setting, with either CPU and either DC project, at the 65w setting the wall draw was from 110-115w.
Anyway, upping the PPT to 70w :-
Asteroids it clocked at 3.2-3.3 GHz, 55-57C, (Core VID 0.96v, core utility 85%), 118w wall draw.
Einstein, 3.8-4.05 GHz, 55-57C, (Core VID 1.05-1.1v, core utility 100-108%), 117-123w. Big swings in clock speed!

PPT 75w
Einstein, 3.9-4.15 GHz, 57-60C, Core VID 1.06-1.15v, core utility 103-110%, 121-128w wall draw (was seeing bigger swings in clock speed, vcore and wall draw with this project and PPT setting).
Asteroids, ~3.55 GHz, 58-59C, Core VID 0.965v, core utility 93%, 125w wall draw.

PPT 80w
Einstein, 4-4.3 GHz, 60-63C, Core VID 1.1-1.2v, wall draw 127-134w (wide swings in all but temp here again).
Asteroids, ~3.65 GHz, 61-62C, Core VID ~1v, wall draw 130-132w.
(Core util not noted for this setting).

PPT 105w (default)
Einstein, 4.4-4.55 GHz, 68-73C, Core VID 1.25-1.35v, core utility 117%, 156-163w (wide swings again, but less so).
Asteroids, 4 GHz, 71-72C, Core VID ~1.1v, core utility 105%, 165-166w

Going from 65w PPT to 70w gained ~16% clock speed and consumed 4.4% more power (assuming 3.25 GHz, tested with Asteroids only).
From 70w to 75w, +9% clock speed and 6% more power.
75 to 80w, +3% clock speed and 5% more power.
80w to 105w, +9.6% clock speed and 26% more power!

Clearly the sweet spot here between clock gains and small power increase here is at 75w PPT, so that's what I've set it to.
So vs my old 3600, with Asteroids, the 5800X @75w PPT is running at pretty much the same clock speed (3.55 GHz), has an extra 2C/4T, IPC gains of 10-15% (for programs in general), and is using only about 13w more (+5.4%) power .
So roughly that's a 43-48% performance increase for just a ~5% increase in power usage, I call that a win win!

To roughly emulate a gaming load (assuming 2 cores being used), I ran Einstein on 4 threads at 75w PPT and got 4-4.2 GHz, VID 1.3-1.4v, 65-69C.
For 4C/8T, 4.2-4.4 GHz, VID 1.17-1.28v, 59-62C. It's actually getting better speeds with 4 cores loaded than 2! I'm not sure what's going on there.....
Factory boost speed is 4.7 GHz with 105w PPT.

All figures, except power draw, were taken from HWinfo64 v8.16, rig spec in sig.
Power figures taken from a Maplin plug in the wall meter (230v ac).
Room temp ~20C.

 

[Assimilator1]

And now comparing times between the Ryzen 5 3600 (Zen 2) and Ryzen 7 5800X (Zen 3).
Asteroids :-
3600 average run time for 15 results ranging from 50.4-54 credits is 3681s (most ~3550-3700s)
5800X average run time for 15 results ranging from ~51.1-54 credits is 3519s

The 5800X appears to be ~4.4% faster than the 3600, assuming that the run times don't simply vary by that much anyway!
That of course ignores the fact the 5800X has 1/3 more cores .
Although the credit range is slightly different for each, the times were varying so much for the 3600 that it's not going to make much difference.
I would've posted results now for Einstein too, but I had so much grief simply getting Open Office Calc to actually give me an average that I'm out of time!

 

[Assimilator1]

Well, I've got the averages for Einstein, but only realised it's pretty pointless as Einstein gives the same 693 credits for all WUs! And AFAIK there's no other way to tell the workload difference between WUs.
Anyway, FWIW, here are the average CPU run times for my 2 CPUs for 20 WUs each.
Ryzen 5 3600 - 19,494s
Ryzen 7 5800X - 17,356s

The 5800X is 11% faster here, but who knows whether that's down to the CPU or WUs!
Although the WU times captured were just before and just after the CPU swap, so the WUs are probably broadly similar.

I forgot to mention earlier that CPU times could also vary due to fluctuations in clock speed due to different WUs.
Although with the 3600 and the old v1.8 bios, whenever I looked at the clock speed when running Asteroids, it was running very close to 3.5 GHz.
With the few times I looked on the updated beta bios (1.H4) it was at 3.55 GHz, a 1.4% clock speed increase. And I did check just before and after the bios update, so I believe that is with the same WUs.
And as I mentioned earlier, the 5800X ran at 3.55 GHz too with Asteroids.