Ryzen and Threadripper

Message boards : Number crunching : Ryzen and Threadripper
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 31 · 32 · 33 · 34 · 35 · 36 · 37 . . . 69 · Next

AuthorMessage
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2015878 - Posted: 18 Oct 2019, 19:44:00 UTC - in response to Message 2015875.  

I have a Noctua NH-D15 SE AM4 sitting in the closet I could put on for comparison after I pulled all the custom loop bits out.

Can you tell me what the delta is on your system from ambient temp to the reported cpu temp under load?
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2015878 · Report as offensive     Reply Quote
Profile -= Vyper =-
Volunteer tester
Avatar

Send message
Joined: 5 Sep 99
Posts: 1652
Credit: 1,065,191,981
RAC: 2,537
Sweden
Message 2015894 - Posted: 18 Oct 2019, 22:22:13 UTC - in response to Message 2015806.  

Question for @Vyper, which of the cpu temp sensor readouts do you use? The motherboard SIO CPU Temp or the cpu internal Tdie temp? The motherboard CPU temp is only polled every couple of seconds and is an average over time. It only reports whole integers and the accuracy is limited by the LSB conversion of the ADC.

The Tdie temp is the millisecond polled, random, internal sampling of each cpu core and has accuracy to the half degree. That is the one I use. So depending on which cpu core gets sampled, the reported Tdie temp can be 68° C. if the core that is looked at has just become unloaded or it can report 83° C. if the core that is currently polled is under heavy AVX loading from a cpu task. So I see quite a bit of variance in the Tdie reported temp depending on which core is being looked at.


I was testing with prime in Torture mode.. Small FFTs maximum heat. The software reports the Tdie and it's around 69 degrees Celsius, got a sporadic peak to 78 degrees Celsius and i could hear the fans rev and then it was back to 68-69 degrees C again with low noise.

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group
ID: 2015894 · Report as offensive     Reply Quote
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13854
Credit: 208,696,464
RAC: 304
Australia
Message 2015900 - Posted: 18 Oct 2019, 23:12:25 UTC - in response to Message 2015894.  

got a sporadic peak to 78 degrees Celsius and i could hear the fans rev and then it was back to 68-69 degrees C again with low noise.
When I first got my present systems that was driving me crazy. Slight increase in temperature, the fans would ramp up, temperature drops, fans ramp down. Temperature climbs, fans ramp up, temperature drops, fans ramp down. Ad infinitum.
Took me a while to figure out the BIOS setting for the fan speed control to dampen the fan behaviour, so a short sharp change in temperature doesn't result in a change in fan speed, but if that change in temperature isn't brief then the fan speeds change accordingly.
Grant
Darwin NT
ID: 2015900 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2015904 - Posted: 18 Oct 2019, 23:38:58 UTC

I was testing with prime in Torture mode.. Small FFTs maximum heat. The software reports the Tdie and it's around 69 degrees Celsius, got a sporadic peak to 78 degrees Celsius and i could hear the fans rev and then it was back to 68-69 degrees C again with low noise.

I get lower temps in Prime95 Small FFT torture test running the same number of cores that I do in BOINC. About 3° C. lower than when running the actual BOINC load.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2015904 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2015920 - Posted: 19 Oct 2019, 2:21:40 UTC

Well I was testing various settings that involved rebooting the system multiple times to change things in the BIOS and the last time into the BIOS once I got back in because the reboots were getting flaky and I thought it was because of my changes, I saw that I was at 88° C. in the BIOS. The flakiness was because of cpu overtemp. Believe the pump has died finally or I have a massive blockage in the system.

So I pulled the custom loop off and installed the Noctua NH-D15 SE AM4 twin tower air cooler. Will start tearing the custom loop apart tomorrow and diagnose the problem. Take the block apart and check obviously and check for pump flow. May change back to the Raystorm Pro block which is one of the least restrictive blocks on the market. I was going to have to open the loop up anyway to install the flowmeter I should get tomorrow. Probably will have to RMA the pump/reservoir but the Noctua will suffice in the meantime.

Ran Prime95 Small FFT test on my normal amount of cores and was hitting 83° C. That was the same temp I was getting on the custom loop. Now on my normal configuration as before and the normal loading I am sitting around 85-90° C. on the Noctua air cooler. I'm about 10° C. better on the loop. So not as good as the custom loop when it is working correctly. Which is what I expect to be so.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2015920 · Report as offensive     Reply Quote
Profile M_M
Avatar

Send message
Joined: 20 May 04
Posts: 76
Credit: 45,752,966
RAC: 8
Serbia
Message 2015934 - Posted: 19 Oct 2019, 7:09:31 UTC - in response to Message 2015920.  
Last modified: 19 Oct 2019, 7:24:17 UTC


Ran Prime95 Small FFT test on my normal amount of cores and was hitting 83° C. That was the same temp I was getting on the custom loop. Now on my normal configuration as before and the normal loading I am sitting around 85-90° C. on the Noctua air cooler. I'm about 10° C. better on the loop. So not as good as the custom loop when it is working correctly. Which is what I expect to be so.


Good custom loop is for sure better then best air coolers, incl. NH-D14/15. One of the main reasons is much bigger cooling area, but also that heat sink with custom loop is typically outside of the case, where the air temperature is at least 5-10C lower comparing to cases even with excellent ventilation. So for apple2apple comparison of cooling efficiency, open case should be used.

Anyhow, back to your case,
1) Are you sure that vents on NH-D15 are running on their full speed when under full load?
2) Have you try to resit a cooler few times and also use a different amount and spread pattern of some good thermal paste, I use NT-H2. Maybe the contact area between NH-D15 base and CPU is not so good in your particular situation, could be related to surface shapes, orientation, or way how you tighten the cooler screws?

BTW, Also we me, all core SETI load with AVX is pushing few C higher temps then SmallFFT in Prime95 with AVX.

And one more thing I forgot to mention, I replaced my middle fan on NH-D14 to high static pressure one.
ID: 2015934 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2015946 - Posted: 19 Oct 2019, 13:51:59 UTC - in response to Message 2015934.  

1) Fans are plugged into the fan hub controller and are powered from SATA 12V connection with no speed control. Fans are running at their max 1500 rpm's.

2) No I have not remounted the cooler. The paste is MX-4 and is spread across the entire surface of the IHS before applying any cooling block. The mounting screws are fully tightened.

3) Even with NO top or side panels on, the temps remain the same.

4) I just have a hot running chip I think. I reduced the core clock and voltage down and it made no difference really.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2015946 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2015993 - Posted: 20 Oct 2019, 0:18:22 UTC

Back on the custom loop with a working pump finally. Flowmeter installed and showing 17 liters per minute flowrate. Temps under BOINC load around 72° C.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2015993 · Report as offensive     Reply Quote
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 2015998 - Posted: 20 Oct 2019, 0:47:14 UTC - in response to Message 2015993.  

Did you replace the pump? Or change something else?
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 2015998 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2016007 - Posted: 20 Oct 2019, 6:03:55 UTC - in response to Message 2015998.  

Did you replace the pump? Or change something else?

No I took the pump apart and figured what was wrong with it. It was binding up because the impeller got knocked too far down the bore during my first construction. Problem caused by me and I didn't realize what I had done since I didn't know how the pump was constructed and designed until I took it apart.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2016007 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2016029 - Posted: 20 Oct 2019, 13:02:49 UTC

So what is the consensus? Is an AMD 3900x (12c/24t) a good buy for cpu crunching? Especially now that you can get it at list price?

I know Keith originally posted that others should wait while he and others experimented/struggled to get theirs "working" properly.

If I did buy it I would be upgrading from an Amd 2700 (8c/16t).

Tom
A proud member of the OFA (Old Farts Association).
ID: 2016029 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2016055 - Posted: 20 Oct 2019, 14:55:54 UTC - in response to Message 2016029.  

Yes I think any Ryzen 3000 is a good buy compared to a 2700X simply because of the architecture improvement. Better maths all around with the larger caches lead to better crunch times. More cores lead to higher production. The current BIOS' are good enough and the rumor that the next 1.0.0.4 releases will be better still.
You don't have to get the X570 motherboards to use it if you have a decent X470 motherboard which you do with the Crosshair VII Hero. Either use the BIOS as designed or just use a fixed multiplier and set a reasonable offset to keep the temps under control. Most of the early BIOS teething issues have been worked out.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2016055 · Report as offensive     Reply Quote
Profile Bill Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 30 Nov 05
Posts: 282
Credit: 6,916,194
RAC: 60
United States
Message 2016057 - Posted: 20 Oct 2019, 15:07:28 UTC - in response to Message 2016055.  

Do you feel that there is an advantage with the X470 chipset over the B450 with respect to crunching?
Seti@home classic: 1,456 results, 1.613 years CPU time
ID: 2016057 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2016059 - Posted: 20 Oct 2019, 15:09:51 UTC - in response to Message 2016057.  
Last modified: 20 Oct 2019, 15:10:36 UTC

Do you feel that there is an advantage with the X470 chipset over the B450 with respect to crunching?

No not really. As long as the VRM design of the B450 motherboard is adequate, it makes no difference to cpu crunching for the chipset type.

The BIOS' of the B450 boards will not be as full featured for cpu control as the X470 boards though.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2016059 · Report as offensive     Reply Quote
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13854
Credit: 208,696,464
RAC: 304
Australia
Message 2016130 - Posted: 21 Oct 2019, 5:53:33 UTC

Next-gen Threadripper rumored to arrive on November 19
on November 5, AMD will announce the 3960X and 3970X with their accompanying TRX40 motherboards, and two weeks later they’ll arrive in-store. The mysterious 3990X will get a mention but won’t be fully revealed until January 2020.

Grant
Darwin NT
ID: 2016130 · Report as offensive     Reply Quote
Profile -= Vyper =-
Volunteer tester
Avatar

Send message
Joined: 5 Sep 99
Posts: 1652
Credit: 1,065,191,981
RAC: 2,537
Sweden
Message 2016132 - Posted: 21 Oct 2019, 6:30:00 UTC - in response to Message 2016059.  

Do you feel that there is an advantage with the X470 chipset over the B450 with respect to crunching?

No not really. As long as the VRM design of the B450 motherboard is adequate, it makes no difference to cpu crunching for the chipset type.

The BIOS' of the B450 boards will not be as full featured for cpu control as the X470 boards though.


Here is a list of motherboards and their capabilities for running Ryzen 3Gen, the page is in Swedish so you might need translate.google.com to make it understandable.

https://www.sweclockers.com/forum/post/17951749

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group
ID: 2016132 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2016134 - Posted: 21 Oct 2019, 6:50:45 UTC

Thanks Vyper, that list is great for showing the VRM capabilities and entirely reasonable considerations for each cpu class. Should answer Bill's question.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2016134 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2016308 - Posted: 22 Oct 2019, 23:52:36 UTC

I ran across a table of Epyc cpus that had an older version of a 16c/32t cpu that turboed to maybe 3.3Ghz but was cheaper than the list price for a 3950x. It made me "hmmmmmmmmm"

Yes, I know I am ignoring the cost of the MB.... but I had to ignore the cost of the MB when I tried out a TR2 cpu..... :)

Tom
A proud member of the OFA (Old Farts Association).
ID: 2016308 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 5126
Credit: 276,046,078
RAC: 462
Message 2016309 - Posted: 22 Oct 2019, 23:58:48 UTC

And here is a TR2 16c/32t) (2950x) for less than the list price of a 3950x.

https://smile.amazon.com/AMD-Ryzen-Threadripper-Processor-YD295XA8AFWOF/dp/B07GFN6CVF/ref=sr_1_12?keywords=amd+3900x+cpu&qid=1571788447&sr=8-12

So the question is are you buying it for the most efficient crunching or to drive a lot of gpus?

Tom
A proud member of the OFA (Old Farts Association).
ID: 2016309 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2016310 - Posted: 23 Oct 2019, 0:19:22 UTC - in response to Message 2016309.  

If you are just pushing gpus, there is no reason to purchase a multi core cpu that has more cores than the number of gpus you are going to run. If you are going to run BOTH cpu and gpu tasks, then the formula changes. The new Zen2 cores are much more productive on the type of cpu tasks we get at Seti compared to TR2 cores. The FFT math we do is much faster on Zen2 cores simply because of the full size FP registers it has compared to TR2 cores and the much higher amount of L1, L2 and L3 caches which all speed up math calculations. The IPC increase in Zen2 cores are another benefit. Add in the ability to run much faster memory adds again another performance benefit of the the Zen2 core Threadripper 3900X or 3950X.

There isn't going to be any more efficient crunching if you are referring to power usage.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2016310 · Report as offensive     Reply Quote
Previous · 1 . . . 31 · 32 · 33 · 34 · 35 · 36 · 37 . . . 69 · Next

Message boards : Number crunching : Ryzen and Threadripper


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.