Message boards :
Number crunching :
WU completion times
Message board moderation
Author | Message |
---|---|
Send message Joined: 22 Oct 18 Posts: 19 Credit: 3,186,503 RAC: 0 |
One of my hosts equipped with Xeon E5-2650L v4 @ 1.70GHz calculates tasks within 2.5 to 3 hrs on average (Linux OS) The other host whit i3-2100 @ 3.1GHz needs more than 6 hours to complete one task (WIndows 10 OS). Why such a big difference in calculation times? |
Send message Joined: 25 Mar 16 Posts: 12 Credit: 828,261,700 RAC: 0 |
WU run times on Windows are significantly longer than on Linux. Have a look at the top computer list: https://universeathome.pl/universe/top_hosts.php Dominated by Linux machines. My 3900X with Ubuntu 19.04 currently needs just a bit more than an hour per WU. 1-2 weeks back this was down to ~25 minutes, WUs must have been shorter / less intense then... Running Universe within a Linux-VM achieves almost native Linux speed, you could try that. |
Send message Joined: 22 Oct 18 Posts: 19 Credit: 3,186,503 RAC: 0 |
WU run times on Windows are significantly longer than on Linux. Have a look at the top computer list: I will try Linux then. |
Send message Joined: 28 Feb 15 Posts: 253 Credit: 200,562,581 RAC: 0 |
My 3900X with Ubuntu 19.04 currently needs just a bit more than an hour per WU. That seems fast compared to my various Ryzens. Are you running on all cores? Or else, I need a 3900x. |
Send message Joined: 25 Mar 16 Posts: 12 Credit: 828,261,700 RAC: 0 |
Hi Jim, I'm running the 3900x @95W power target with air cooling, 3000MHz RAM, threads 0-20 open for CPU WUs (SMT enabled, 100% usage), threads 21-23 locked on Einstein & SETI GPU apps. ZEN2 is a beast. I had an R7 1700 before, the 3900x easily doubled my CPU throughput. |
Send message Joined: 28 Feb 15 Posts: 253 Credit: 200,562,581 RAC: 0 |
Thanks. That makes sense. It is always good to be thinking of the next upgrade, especially with Ryzens. Intel is such a disappointment that I no longer bother to even look at their stuff. Let me know when they are back. |
Send message Joined: 2 Jun 16 Posts: 169 Credit: 317,253,046 RAC: 0 |
That is much faster than my 1950x and 2700x as well under Linux. Its also much faster than any of the other Zen systems I checked on the top hosts. |
Send message Joined: 25 Mar 16 Posts: 12 Credit: 828,261,700 RAC: 0 |
https://universeathome.pl/universe/results.php?hostid=526615 on #11 https://universeathome.pl/universe/results.php?hostid=491902 on #16 https://universeathome.pl/universe/results.php?hostid=539951 on #19 these are also fast, if not faster (the 3960X) Interesting though that the run times vary so much, even between the several 3900x with Linux My setup is untuned, using the distributions 5.0 kernel. Only thing I take care of is separating CPU and GPU Tasks onto dedicated threads. |
Send message Joined: 25 Mar 16 Posts: 12 Credit: 828,261,700 RAC: 0 |
Actually, the separation of CPU/GPU tasks might have something to do with my times. Through taskset -a -p -c 0-20 $2 I set the CPU thread PID in $2 to run on CPU threads 0-20 only. Then nail GPU threads to 21-23. I had assumed this would limited GPU work to the last 1.5 physical cores, but actually it puts them on the 2nd thread of all cores on the 4th CCX. Since CPU load from GPU WUs is low on 21-23, CPU threads 9, 10 and 11 would almost run with full (non-SMT) speed. WUs on these threads should complete faster than eg. those scheduled on 0 & 12. However, I noticed a bit of thread hopping. WUs get moved from thread to thread, so the times average out to some extend. There are no 3 WUs per hour that are fast while the rest are normal. If we had relatively stable WU times in this project, I'd compare running 21 vs. 24 Universe WUs once. But we don't have those, so comparing run times is even tricky on a single machine. I don't think the effect is huge, a full 24 WUs might be 20-30% slower at max... ~# lstopo Machine (31GB) Package L#0 L3 L#0 (16MB) L2 L#0 (512KB) + L1d L#0 (32KB) + L1i L#0 (32KB) + Core L#0 PU L#0 (P#0) PU L#1 (P#12) L2 L#1 (512KB) + L1d L#1 (32KB) + L1i L#1 (32KB) + Core L#1 PU L#2 (P#1) PU L#3 (P#13) L2 L#2 (512KB) + L1d L#2 (32KB) + L1i L#2 (32KB) + Core L#2 PU L#4 (P#2) PU L#5 (P#14) L3 L#1 (16MB) L2 L#3 (512KB) + L1d L#3 (32KB) + L1i L#3 (32KB) + Core L#3 PU L#6 (P#3) PU L#7 (P#15) L2 L#4 (512KB) + L1d L#4 (32KB) + L1i L#4 (32KB) + Core L#4 PU L#8 (P#4) PU L#9 (P#16) L2 L#5 (512KB) + L1d L#5 (32KB) + L1i L#5 (32KB) + Core L#5 PU L#10 (P#5) PU L#11 (P#17) L3 L#2 (16MB) L2 L#6 (512KB) + L1d L#6 (32KB) + L1i L#6 (32KB) + Core L#6 PU L#12 (P#6) PU L#13 (P#18) L2 L#7 (512KB) + L1d L#7 (32KB) + L1i L#7 (32KB) + Core L#7 PU L#14 (P#7) PU L#15 (P#19) L2 L#8 (512KB) + L1d L#8 (32KB) + L1i L#8 (32KB) + Core L#8 PU L#16 (P#8) PU L#17 (P#20) L3 L#3 (16MB) L2 L#9 (512KB) + L1d L#9 (32KB) + L1i L#9 (32KB) + Core L#9 PU L#18 (P#9) PU L#19 (P#21) L2 L#10 (512KB) + L1d L#10 (32KB) + L1i L#10 (32KB) + Core L#10 PU L#20 (P#10) PU L#21 (P#22) L2 L#11 (512KB) + L1d L#11 (32KB) + L1i L#11 (32KB) + Core L#11 PU L#22 (P#11) PU L#23 (P#23) |
Send message Joined: 25 Mar 16 Posts: 12 Credit: 828,261,700 RAC: 0 |
I just (16 Jan 2020, 13:54:51 UTC) reported 67 WUs, where all 24 slots were CPU WUs. Just two WUs were running longer than an hour, so that still seems faster than most other hosts in the top list. |
Send message Joined: 28 Feb 15 Posts: 253 Credit: 200,562,581 RAC: 0 |
I would not be surprised if they are sending out different batches of work units. I think I have seen one hour on my machine at one point. Your results seem consistent though. My own Ryzen 3000 series (3600 and 3700x) are maybe 30% faster than my Ryzen 2600 and 2700, but double the speed does not seem likely. Or maybe the large L3 cache of the 3900x does it? Some projects benefit from cache more than others. I will look around at the other projects and see how they do. I may have to put together a 3900x. The high efficiency of that chip is remarkable too, and would work through the summer. Thanks for your input. |
Send message Joined: 21 Feb 15 Posts: 15 Credit: 14,270,919 RAC: 0 |
WU's today are using almost double normal time, but giving normal credit. It would be nice to know why. |