Zen 2 for Distributed Computing: Any interest?

Page 5 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
@phoenicis , I believe as the race goes on, the llr tasks take longer so benchmarks are a moving target. Perhaps @Ken g6 can confirm.

My last 20 tasks @3.4 GHz/1.08 vcore/121 watts (underclock) are taking an average of 1875 seconds to complete (SMT off/8 tasks). It's good to know that task completion times are CPU frequency dependent.

I thought I read somewhere that HT on for intel processors takes more power. Maybe it's the same with SMT on Zen 2.
 
  • Like
Reactions: phoenicis

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
rying to think of a project that's easy to compare speeds........ Milkyway@home I think, as long as we choose the same credit WU results.

I can run milkyway@home tasks after the PG race is done in a few hours. FAH would be project dependent. TnGrid is another possibility I think.
 

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
I tried to replicate your 3.6Ghz approach but didn't do as well with PPSllr tasks taking 1850 secs (8 tasks SMT on) although power draw from the wall did drop to 139W

At 3.6 GHz with SMT on, 8 tasks, 124 watts, tasks are taking 1772 seconds. I have the wraith cooler fan running 100% 24/7 so I don't think my CPU freq varies due to thermal throttling.
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,218
3,797
75
@phoenicis , I believe as the race goes on, the llr tasks take longer so benchmarks are a moving target. Perhaps @Ken g6 can confirm.
That's likely to happen; also, different WUs will take different times in general. I see that there are benchmarks for mprime, but I'm not sure how to run or report them.
 
  • Like
Reactions: phoenicis

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
@Assimilator1 , here are the 3700X data for Milkyway@home.

OS: linux mint 19.1
Computer: 3700X @3.6 GHz (8 cores/16 threads, SMT on), 2933 DDR4 14, 14, 14, 31.
APP: Separation v1.46 (227 credit)
Average run time (20 tasks): 4056 seconds
Power draw from the wall: 115 watts
 

emoga

Member
May 13, 2018
187
301
136
I was thinking of adding a 3900x to the fleet.
It sounds like cheaping out on a older motherboard is causing some grief with overclocking? (albeit better with bios updates)
Since this would be a DC computer...Would the lack of overclocking options really matter if the cpu seems to run hot and the OC frequency might be limited by heat anyways? (from running all cores)
or should just I save myself the headache and just buy one of the newer, sexier X570 motherboards for 2-3x the price?
 

TennesseeTony

Elite Member
Aug 2, 2003
4,204
3,631
136
www.google.com
I vote no overclocking since this hobby is a bit expensive, and monthly ongoing energy costs should be kept to a minimum.

Edit: I also vote the cheap motherboard...that's 6 months or more you could put to the energy bill.
 
Last edited:
  • Like
Reactions: emoga

IEC

Elite Member
Super Moderator
Jun 10, 2004
14,323
4,904
136
I was thinking of adding a 3900x to the fleet.
It sounds like cheaping out on a older motherboard is causing some grief with overclocking? (albeit better with bios updates)
Since this would be a DC computer...Would the lack of overclocking options really matter if the cpu seems to run hot and the OC frequency might be limited by heat anyways? (from running all cores)
or should just I save myself the headache and just buy one of the newer, sexier X570 motherboards for 2-3x the price?

Save your money, you will hit limits from heat before anything else, especially for AVX2 workloads:
Stable OC @ 1.25V.PNG

This is with 1.25V fixed, per-CCX (actually, per-CCD) OC with best chiplet at 4350 and other chiplet at 4200. AVX2 stable. I can go to 4425/4325 with a little more voltage but then I'm pushing the limits of my cooling.
 
  • Like
Reactions: TennesseeTony

VirtualLarry

No Lifer
Aug 25, 2001
56,226
9,990
126
Anyone need a R7 2700 CPU? Does 4Ghz easy (with sufficient cooling).

I have it listed for $180 in my FS thread, but I would sell it (CPU ONLY) to a fellow DC'er for $165 shipped CONUS.

I really want to upgrade my laptop from a dual-core Bay Trail Atom, to this sweet (not 1080P though) Ryzen 3200U slim/light laptop that WM has for $269, marked down from $399. (Edit: My Mom loaned me $200, I picked up the 14" HP Slim Ryzen 3 3200U laptop 4GB / 128GB SSD from my local WM store.) See Hot Deals forum for some of my other WM finds.

Selling this chip would allow me to do that.

It's a really great WCG cruncher, 8C/16T.

Have some 32GB kits of RAM too, and boards. Just ask.
 
Last edited:

crashtech

Lifer
Jan 4, 2013
10,521
2,111
146
I only have eyes for Zen 2, @VirtualLarry , sorry! I'll be selling a 1700X here pretty soon myself if the 3700X works in the old board. The RAM sounds interesting, though...
 

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
Interesting point about X570 m/brd cost, didn't realise it would be that big a difference!

@Assimilator1 , here are the 3700X data for Milkyway@home.

OS: linux mint 19.1
Computer: 3700X @3.6 GHz (8 cores/16 threads, SMT on), 2933 DDR4 14, 14, 14, 31.
APP: Separation v1.46 (227 credit)
Average run time (20 tasks): 4056 seconds
Power draw from the wall: 115 watts

Cool, some very useful info there :), although which 227 credit WUs were you looking at? The decimal numbers matter too, see here, their can be quite a time variance between them.
I'll run MW on my 4930k & see what I get.

Looking at uallas5's valid tasks I see most are either 227.51 or 227.53 credit WUs, I guess your average was a mix of those?
 

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
Looking at uallas5's valid tasks I see most are either 227.51 or 227.53 credit WUs, I guess your average was a mix of those?

It was a mix of 227.51, 227.52 and 227.53 tasks. I didn't see much variation in completion times.
 

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
That is tight, just 0.2% variance! Thanks for extra stats :), maybe these collectively could be used as a new benchmark?? Although that depends if they stick around!
Now that I've run down the LHC WU cache a little I'll run MW on it for a bit.
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
@biodoc, by way of extrasensory perception I have come to know that a 3700X is completing 8.2 ESP-LLR tasks per day on PrimeGrid.

There were 3.6 GHz and 124...140 W power consumption reported for PPS-LLR here, so I suspect it's about the same with ESP-LLR. Any data available on this?

For reference, ESP-LLR on dual-socket E5-2696 v4 (Broadwell EP, 14 nm, dual 22C/44T) pegged at 2.6 GHz, runs 10 tasks at a time (5 per socket) with 7.5 h task duration, thus 32 tasks per day. This takes 510 W at the wall. That's
510e-6 MW * (24*3,600) s/d / 32 tasks/d = 1.4 MJ/task​
or
510 W * 27,000 s/task / 10 tasks at once = 1.4 MJ/task​

Given the power figures above, the higher clocked 7 nm Ryzen 3000 should be about as efficient with ESP-LLR as the much lower clocked 14 nm Xeon E5-2600.

Oh, and power consumption aside, merely four Ryzen 3700X systems beat the dual Xeon in ESP-LLR throughput.

--------
Marginalia:
-- The clock and the power consumption of the Xeon are due to the "frequency-optimized" BIOS of my Supermicro board, which never lets the processors go below all-core turbo under load. There is a platinum 750W PSU in the box, and an idling passive GCN 1.0 GPU.
-- I generally abhor mere run-time based performance comparisons of PrimeGrid LLR jobs. But from 10 completed ESP-LLR jobs, I am seeing only a coefficient of variation of 1.5 % in run times, plus all current ESP-LLR tasks have the same FFT length currently, so that should be sufficient for a rough power efficiency comparison.
 
Last edited:

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
There were 3.6 GHz and 124...140 W power consumption reported for PPS-LLR here, so I suspect it's about the same with ESP-LLR. Any data available on this?

I have a meter on 1 3700X running ESP-LLR and it's pulling 125 watts at the wall. I'll let you do the math. :)

EDIT: it's not overclocked so it's running @3.6 GHz
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
Thanks. So that's 1.32 MJ/task for the Ryzen 3700X, vs. 1.38 MJ/task for the 2696 v4.

(Or, using a unit which may be more familiar to those who pay power bills:
0.366 kWh/task for the Ryzen, vs. 0.383 kWh/task for the 2696 v4.)

Edit, PS,
this is with the Ryzen system's PSU on 110 V, but the Xeon system's PSU on 230 V.
 
Last edited:

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
No completed tasks yet, but my i7 4930k @4.1 GHz running 12 MW threads is pulling ~232w from the wall (230v). Result times to follow.......
 

biodoc

Diamond Member
Dec 29, 2005
6,257
2,238
136
One of the frustrating things about linux and Zen 2 is I have no idea what the temp the processor is running at. :( I set the cpu fan in the bios to run 100% all the time and have avoided any overclocking.
 

Howdy

Senior member
Nov 12, 2017
572
480
136
One of the frustrating things about linux and Zen 2 is I have no idea what the temp the processor is running at. :( I set the cpu fan in the bios to run 100% all the time and have avoided any overclocking.

I use PSensor to monitor CPU temps in Linux
 

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
@Assimilator1 , here are the 3700X data for Milkyway@home.

OS: linux mint 19.1
Computer: 3700X @3.6 GHz (8 cores/16 threads, SMT on), 2933 DDR4 14, 14, 14, 31.
APP: Separation v1.46 (227 credit)
Average run time (20 tasks): 4056 seconds
Power draw from the wall: 115 watts

So my i7 4930k @4.1 GHz running 12 MW@H threads is pulling ~232w from the wall (230v), average of it's 1st 12 tasks is 5378s.
That's a 50/50 mix of 227.51 & .53 credit results, varying from 5340-5474, all bar that latter one are from 5340-5387s, 227.51's mostly ranged from 5360-5370, 227.53 from 5367-5387 bar one.
So your rig's crunching WUs about 25% faster and using 1/2 as much power!!:openmouth::sunglasses:
Amazing! I know my rig's CPU is a few+ generations old, but even so, that's still impressive! I've got to get me a Zen 2 soon!

[edit] And I'd forgotten about the fact that your Zen2 has another 2 cores over my i7. So total output (for MW at least), would actually be about 58% higher! So even more impressive! (Obviously GPU crunching is faster for MW@H).

[edit2] Going from 232w to 116w 24/7, would save me £129/year!

Btw, I'm not seeing X470 at 1/2 the price of X570, although it's hard to get a direct comparison at my favourite etailer (Ebuyer.com), a
Asus TUF X470-PLUS GAMING AM4 DDR4 ATX Motherboard is £143, the Gigabyte X470 AORUS ULTRA GAMING AM4 ATX Motherboard is £133. Where as the Asus PRIME X570-P is £190, and the Gigabyte X570 GAMING X is £183. I've no idea what's the difference between the different models btw (chipset aside). Nor do I know if those X470 mbrds support Zen2.

Anything to watch out for with any of the above mbrds?
 
Last edited:
  • Like
Reactions: biodoc