PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : Number crunching : Rtx 30 series (Message 143492)
Posted 5 hours ago by Profile j.sheridanProject donor
Some may consider a used 2080ti with its lower power consumption a better buy.

Did you check power consumption?
Because of the 8 nm process and as FP32, RTX and Tensor Cores are not used, power usage may be lower than TPD = 320W.

My UPS shows 64% load with it running GFN and 25% load without. It's a Smart UPS-1000 which is supposed to have 700W maximum load which means the 3080 is using 210W.


Some actual runtimes to confirm the earlier benchmarks. The DYFL time is probably a bit slower than optimum as I had a few hours of doing some other tests without suspending it.

24s Genefer 15 v3.21
52s Genefer 16 v3.21
135s Genefer 17 Low v3.21
145s Genefer 17 Mega v3.21
455s Genefer 18 v3.21
1,020s Genefer 19 v3.21
4,262s Genefer 20 v3.21
65,973s Do You Feel Lucky? v3.21
2) Message boards : Number crunching : A "ghost" user !? (Message 143480)
Posted 10 hours ago by Profile j.sheridanProject donor
It's very strange behaviour but all the tasks eventually got dumped back after 4 days or so.
3) Message boards : Aggie The Pew message board (Message 143435)
Posted 1 day ago by Profile j.sheridanProject donor

On behalf of us all we feel your pain.

We do, sorry for your loss Crun-chi. This year is definitely one to forget.
4) Message boards : Aggie The Pew message board (Message 143434)
Posted 1 day ago by Profile j.sheridanProject donor

I think the warning should be a thankyou note: By entering the advanced section you will be viewing how ridiculous the settings can be and basically anything you could do couldn't possibly make it worse - maybe lower the settings a bit and we'll have a crack at only double digit temperature settings? Cheers from the CPU.


Agreed, with amd you also have to go into the overclocking "you're going to kill your cpu" settings to switch off PBO.
I thought I'd really screwed up when I changed my htpc over from intel to amd. The new chip claimed to be 35W just like the outgoing one but I found the temperatures on my fanless case went from mid 30s with the intel to mid 50s with the amd. It tooks several months of head scratching and searching before a post suddenly appeared saying "switch off pbo" and that was it, temps dropped to mid 30s.
Same on the 3700x, for PG tasks it was mid 80s and dropped into the 60s with PBO switched off and clocks actually improved as the chips will boost higher when cooler. I got another 10C drop by switching from the noctua nh-14 to an Arctic 280mm AIO despite all the claims that watercooling isn't worthwhile any more. It's also able to achieve those temperatures without having to run 100% fan so is much quieter.
5) Message boards : Aggie The Pew message board (Message 143427)
Posted 1 day ago by Profile j.sheridanProject donor
Weird as it sounds I think it was working by design.

Here's the amd equivalent


On the Gigabyte X570 Master with the 3900X,
PBO limits were as follows: PPT 1200W, TDC 540A, and EDC 600A.

On the MSI Godlike, the limits were: 1000W, 490A, and 630A.

With PBO disabled the limits were PPT 142W, TDC 95A, and EDC 140A



ie it's deliberately set to crazy limits then the chip decides what it can handle. All this appears to do is generate loads of extra temperature with little or no extra performance.
6) Message boards : Number crunching : Hyperthreading with AMD Threadripper (Message 143400)
Posted 2 days ago by Profile j.sheridanProject donor
For SoB 8 threads is better - the crossover point is around 1920K.

It's very easy to check using Prime95 (mprime on linux?).

Just download it and run the benchmark.

Put in the FFT size for the project you want to run (2880-3200 for SoB) and it will go off and test run 1,2,4,8 tasks etc.

For SoB on the 3950x I see this:

[Sep 18 18:42] Worker starting
[Sep 18 18:42] Your timings will be written to the results.bench.txt file.
[Sep 18 18:42] Compare your results to other computers at http://www.mersenne.org/report_benchmarks
[Sep 18 18:42] Benchmarking multiple workers to measure the impact of memory bandwidth
[Sep 18 18:42] Timing 3200K FFT, 16 cores, 1 worker. Average times: 1.30 ms. Total throughput: 772.02 iter/sec.
[Sep 18 18:42] Timing 3200K FFT, 16 cores, 2 workers. Average times: 2.05, 2.06 ms. Total throughput: 973.94 iter/sec.
[Sep 18 18:42] Timing 3200K FFT, 16 cores, 4 workers. Average times: 9.18, 8.39, 8.47, 8.75 ms. Total throughput: 460.34 iter/sec.
[Sep 18 18:42] Timing 3200K FFT, 16 cores, 8 workers. Average times: 23.31, 23.55, 20.48, 20.49, 21.70, 22.16, 21.55, 21.88 ms. Total throughput: 366.31 iter/sec.
[Sep 18 18:42]
[Sep 18 18:42] Throughput benchmark complete.
[Sep 18 18:42] Throughput benchmark complete.
[Sep 18 18:42] Worker stopped.


Which tells me the best performance is 2 tasks of (16/2) 8 threads each.
The 3970x is essentially 2x 3950x so I'd guess yours will show 4 tasks/workers as best but it's easy to check for yourself.

It's certainly possible to set affinity in linux but I'm not familiar enough with it to offer the best solution. Hopefully someone else will pop along to help with that. In the brief testing I did on smaller tasks I seem to recall that linux was around 10% slower vs affinity-set windows.
7) Message boards : Aggie The Pew message board (Message 143397)
Posted 2 days ago by Profile j.sheridanProject donor
Why am I suddenly hungry?

Nick, that sounds like you've reinvented parmentier potatoes.
8) Message boards : Number crunching : Hyperthreading with AMD Threadripper (Message 143396)
Posted 2 days ago by Profile j.sheridanProject donor
It shouldn't make that much difference. How many tasks are you running?

edit:
I can run 2 tasks of 8 threads each in under 24 hours and I have HT on.
With the 3970x you should be able to run 4.
What might upset things is if linux starts swapping things between CCXs and CCDs. It's vital to set the affinity for each task so they don't move. In windows there's a 20-30% performance hit if you don't do this. Linux is supposed to be slightly better but there's still a difference.
9) Message boards : Number crunching : Rtx 30 series (Message 143395)
Posted 2 days ago by Profile j.sheridanProject donor

Now, I have to find how to exploit these useless FP32 and Tensor Cores...

if anyone can do it, you can.
10) Message boards : Number crunching : Rtx 30 series (Message 143389)
Posted 2 days ago by Profile j.sheridanProject donor
I have an older window machine running with an i7 processor but it has an outdated 1080ti, would it be worth replacing the 1080ti with an 3080? Assuming no power supply issues, is there anything else making the upgrade to a 3080 not make sense?

depends on your funds. Some may consider a used 2080ti with its lower power consumption a better buy.



Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2020 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 2.91, 2.85, 2.44
Generated 20 Sep 2020 | 20:11:41 UTC