YOLO?

[Xyzzy]

TL;DR - This CPU is a tiny bit faster than a Vega 64 GPU for PRP work.



GPU data: https://drive.google.com/file/d/1Tim...l57X_RWO0/view

CPU data:

[Main thread Apr 10 20:26] Mersenne number primality test program version 29.8 
[Main thread Apr 10 20:26] Optimizing for CPU architecture: Core i3/i5/i7, L2 cache size: 18x1 MB, L3 cache size: 25344 KB 
[Work thread Apr 10 20:26] Starting Gerbicz error-checking PRP test of M77936863 using AVX-512 FFT length 4200K, Pass1=1920, Pass2=2240, clm=1, 18 threads 
[Work thread Apr 10 20:28] Iteration: 100000 / 77936863 [0.12%], ms/iter:  1.124, ETA: 24:17:50 
[Work thread Apr 10 20:30] Iteration: 200000 / 77936863 [0.25%], ms/iter:  1.124, ETA: 24:16:07 
[Work thread Apr 10 20:32] Iteration: 300000 / 77936863 [0.38%], ms/iter:  1.123, ETA: 24:12:45 
[Work thread Apr 10 20:34] Iteration: 400000 / 77936863 [0.51%], ms/iter:  1.122, ETA: 24:10:08 
[Work thread Apr 10 20:35] Iteration: 500000 / 77936863 [0.64%], ms/iter:  1.120, ETA: 24:05:56 
[Work thread Apr 10 20:37] Iteration: 600000 / 77936863 [0.76%], ms/iter:  1.127, ETA: 24:12:45 
[Work thread Apr 10 20:39] Iteration: 700000 / 77936863 [0.89%], ms/iter:  1.128, ETA: 24:12:14 
[Work thread Apr 10 20:41] Iteration: 800000 / 77936863 [1.02%], ms/iter:  1.119, ETA: 23:58:37 
[Work thread Apr 10 20:43] Iteration: 900000 / 77936863 [1.15%], ms/iter:  1.121, ETA: 23:59:15 
[Work thread Apr 10 20:45] Iteration: 1000000 / 77936863 [1.28%], ms/iter:  1.122, ETA: 23:58:56 
[Work thread Apr 10 20:45] Gerbicz error check passed at iteration 1000000.

During this test the CPU runs at 2.8GHz all-core (AVX-512) and uses ~200W. The temps stabilize at 57°C with a Noctua UH-12S HSF with the computer out in the open on a test bench.

[MattcAnderson]

I bought one computer with liquid cooling from Best Buy in Oregon. It was designed for gaming. I would not recommend liquid cooling. Too many things can go wrong.

[Xyzzy]

In this thread we will be randomly posting thoughts and findings in no particular order. Ten years from now this might be entertaining to read.

In the BIOS the (poorly documented) SP rating for the CPU is 85.

We need to figure out if we have the proper Intel "PL1" (165W) and "PL2" (198W) power targets set properly.

We upgraded the BIOS to the newest version.

The memory passed one pass of MemTest86. We will test it further tonight. We are surprised at how easily the system accepted the XMP profile.

Our Quadro RTX 4000 video card works great in Linux. We are using the beta Fedora 34 distro. We get mildly accelerated performance with the nouveau free driver and the system defaults to the Wayland graphics thingy.

With the current drought of available video cards it ended up being cheaper to buy a "workstation" card than a gaming card. We always wanted a Quadro anyways so that worked out well for us, although this card is a very low-end Quadro. (It does not have ECC memory.) WRT performance it is right about the same as a GTX 2060.

[LaurV]

Originally Posted by Xyzzy

.

Thanks for the benchmark Mike!

This is very useful for us in the light of this adventure, because we were always wondering what difference would have made a "TOP" mobo (like Rampage 6, or other RoG/Whatever mobo, or even a water-cooled one) to that CPU/GPU/RAM combination, but we decided to go for a "mid-range" mobo due to the cheaper price and the 4 PCIE slots (I really don't like the arrangement of the slots and M.2s for Rampage, sorry ).

So, now we (collective we) will have the opportunity to see (and post here for eternitycomparison, too) if the better chipset you got has much of a contribution to the turnover. Right now, the first two items arrived (the CPU is earlier than expected, but I didn't get the T-shirts yet ) and I am in this stage:



(probably the build will be finished in a week or two, still dealing with some parts and time issues)

[Xyzzy]

We were forced to temporarily install Windows 10 so we could update the "management engine".

This tool tells you if it is out of date: https://downloadcenter.intel.com/dow...Intel-CSMEVDT-

The GUI update tool didn't work so we followed these instructions: https://rog.asus.com/forum/showthrea...asIO-sys-error

We also noticed that Windows 10 (Home) only supports 128GB of memory. (W10 Pro allows 2TB of memory.)

We are currently waiting for some M.2 SSDs to be delivered so for now we are using a bootable Linux USB stick.

We figured out how to limit the PL1 and PL2 power targets in the BIOS. The lowest they will go is 59W. We set both to 165W. When we get a better HSF we might be able to increase the power limits.

Overnight we ran six concurrent P-1 tests with mprime. We allocated three cores per test and gave mprime 192GB of memory to work with.

In the top output below note that the "%CPU" is 1788, or basically almost all 18 cores being fully utilized.

top - 09:13:06 up 19:55,  1 user,  load average: 18.15, 18.08, 18.02 
Tasks: 509 total,   1 running, 508 sleeping,   0 stopped,   0 zombie 
%Cpu(s):  1.2 us,  1.0 sy, 48.5 ni, 48.7 id,  0.0 wa,  0.4 hi,  0.1 si,  0.0 st 
GiB Mem :    251.4 total,     51.2 free,    194.4 used,      5.8 buff/cache 
GiB Swap:      7.9 total,      7.9 free,      0.0 used.     53.2 avail Mem 
 
    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND 
  52353 liveuser  30  10  193.4g 192.2g   7.0m S  1788  76.5  12671:30 mprime 
   2010 liveuser  20   0 7001.5m 233.0m 128.7m S  23.5   0.1  29:30.76 gnome-shell 
   2064 liveuser   9 -11  129.1m  44.0m   5.7m S   0.0   0.0  12:03.08 pipewire-pulse 
   3351 liveuser  20   0  645.0m  79.4m  56.1m S  17.6   0.0   4:41.28 gnome-terminal- 
   2159 root      20   0 1096.1m 406.8m  23.1m S   0.0   0.2   2:49.32 packagekitd 
   2213 liveuser   9 -11  119.1m  19.1m   7.7m S   0.0   0.0   2:24.31 pipewire 
     14 root      20   0    0.0m   0.0m   0.0m I   0.0   0.0   0:56.81 rcu_sched 
   2154 root      20   0  253.2m  30.1m   9.2m S   0.0   0.0   0:54.09 sssd_kcm 
    944 root       0 -20    0.0m   0.0m   0.0m S   0.0   0.0   0:53.52 loop1 
   1551 systemd+  20   0   17.1m   8.4m   7.6m S   0.0   0.0   0:43.10 systemd-oomd 
   8358 liveuser  39  19  860.6m  34.0m  19.9m S   0.0   0.0   0:39.83 tracker-miner-f 
   1303 root     -51   0    0.0m   0.0m   0.0m S   0.0   0.0   0:38.17 irq/78-iwlwifi: 
  49609 root      20   0  217.1m   3.8m   2.5m S   0.0   0.0   0:37.91 watch 
   9575 liveuser  20   0  458.9m  66.3m  44.7m S   0.0   0.0   0:36.86 Xwayland

We haven't (properly) measured how much of an efficiency loss there is from running separate jobs but a quick look indicates the loss might be around 3%. If true, that would be really great! Maybe the quad-channel memory and the mesh interconnect architecture and the low frequency we are running yields this kind of efficiency?

Finally, one of the more interesting things about this platform is the number of PCI lanes you get. (See attached datasheet.)

[retina]

With 256GB RAM, I wonder about the purpose of the 8GB swap?

[Xyzzy]

Originally Posted by retina

With 256GB RAM, I wonder about the purpose of the 8GB swap?

We haven't turned it off yet.

https://fedoraproject.org/wiki/Changes/SwapOnZRAM

[Xyzzy]

1× https://www.samsung.com/semiconducto...nsumer/970pro/

https://www.tomshardware.com/reviews...view,5572.html
https://www.storagereview.com/review...pro-1tb-review
https://www.tweaktown.com/reviews/86...iew/index.html

$249.99 including shipping and taxes.

[James Heinrich]

Originally Posted by Xyzzy

We haven't built a long-term computer in a very long time, so we will keep this setup for five years.

My current main computer pre-dates your linked thread by more than half a year (Dec 2011). Was going to replace it last year, but AMD forgot to ship any 5950X to Canada. Still.

[Xyzzy]

There is a msieve benchmark thread here.

The following tests were run with the power limit set to 165W.

18 threads = 4h32m23s
36 threads = 4h05m09s

[Xyzzy]

1× https://www.microsoft.com/en-us/p/windows-10-pro/df77x4d43rkt

$199.00 including shipping and taxes.