5900x, running hot

Joined
Nov 25, 2020
Messages
28
I'd like to invite both of you to joining me on the darkside and buying a D15. I have a 5900X and am doing a lot better than both of your outcomes. Blowing those Corsairs out of the water.. I can't get mine to go over 61C after 10 minutes of a Prime95 24-thread torture test blend in a 19.6 liter mITX case, holding 4.5 to 4.6GHz on all cores. In games as you're describing, I'm not going much over 50C. It does sound like temps you're getting on that AIO are within expected range.

Could be, but my original budget for this rig was $1,000 less than what it became. I sold off my 1080ti for a net return of $400, so that helped cover some of that extra cost, but I can't justify additional expenses unless there is a clear reason to do so for stability or preventing damage to parts. Neither of those appear to be in play.

By the way, I have temperature data for Jedi Fallen Order. During a 3 hour session it ran at 65C-70C.
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
Could be, but my original budget for this rig was $1,000 less than what it became. I sold off my 1080ti for a net return of $400, so that helped cover some of that extra cost, but I can't justify additional expenses unless there is a clear reason to do so for stability or preventing damage to parts. Neither of those appear to be in play.

By the way, I have temperature data for Jedi Fallen Order. During a 3 hour session it ran at 65C-70C.
That's within range of most cooling solutions I'd think. You guys are fine with those temps, as long as it's not throttling (85C at least on my last CPU, 2700X) then it really doesn't matter. But if you do another build, consider big air. I just pieced together this 5900X rig of course, was always curious about the D15, and am shocked at how it performs.
 

Island

Gawd
Joined
Sep 28, 2005
Messages
880
I'd like to invite both of you to joining me on the darkside and buying a D15. I have a 5900X and am doing a lot better than both of your outcomes. Blowing those Corsairs out of the water.. I can't get mine to go over 61C after 10 minutes of a Prime95 24-thread torture test blend in a 19.6 liter mITX case, holding 4.5 to 4.6GHz on all cores. In games as you're describing, I'm not going much over 50C. It does sound like temps you're getting on that AIO are within expected range.
Thanks for the reply. I'm sure your Noctua is a great air cooler but I seriously doubt it will cool better then my Capellix 360 AIO. Are u running the Newest Prime95 with AVX instruction test? The temps we are seeing are with AVX enabled. My CPU idles in the 20's-30's with Ambient room temp at 72-75F. I stay in the 50's with most tests/tasks but if It's something with AVX, that is where the CPU will spike in temp
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
Thanks for the reply. I'm sure your Noctua is a great air cooler but I seriously doubt it will cool better then my Capellix 360 AIO. Are u running the Newest Prime95 with AVX instruction test? The temps we are seeing are with AVX enabled. My CPU idles in the 20's-30's with Ambient room temp at 72-75F. I stay in the 50's with most tests/tasks but if It's something with AVX, that is where the CPU will spike in temp
I just checked to be sure on my settings, yes those temps are with AVX enabled. I don't want to attack any AIO but my numbers are in a constrained case. Granted I have 5 Noctua fans in there, but some are less than ideal (15x120mm), nothing is setup for any trickery. Stock bios settings etc. Mine are legit temps on what I consider a proper D15 rig. I've seen hwmon record a 71C max, but that's momentarily. I've never caught it hitting over 71C, even for a second, and my fans are set to ramp to 100% after 255 seconds, giving the heatsink a lot of time to get saturated with heat. I do that to prevent tiny blips from making any noise on the desktop. Under load, I can't get this system to go over 61C. At least the hardest workload I know of, Prime95 with 24-thread torture test.
 

sabrewolf732

Supreme [H]ardness
Joined
Dec 6, 2004
Messages
4,633
This starts to get into the temp range where I'd normally see 8c+ consumer cpus pegged from synthetic benchmarks.

Is AMD trying to close the IPC gap by keeping their cpus pegged on zen3?

OP I'd worry about fsilures when they happen.
Figure an aio goes 3 years, you got your $ out of it.

Maybe increase intake volume vs aio curve so you aren't creating a psi problem at full boost.


you don't decrease IPC gap by pushing MHz. You can mitigate an IPC gap by clocking higher.

Also, AMD has had superior IPC since zen 2. Sorry for being annoying, so many people use the term ipc incorrectly.
 

Island

Gawd
Joined
Sep 28, 2005
Messages
880
I just checked to be sure on my settings, yes those temps are with AVX enabled. I don't want to attack any AIO but my numbers are in a constrained case. Granted I have 5 Noctua fans in there, but some are less than ideal (15x120mm), nothing is setup for any trickery. Stock bios settings etc. Mine are legit temps on what I consider a proper D15 rig. I've seen hwmon record a 71C max, but that's momentarily. I've never caught it hitting over 71C, even for a second, and my fans are set to ramp to 100% after 255 seconds, giving the heatsink a lot of time to get saturated with heat. I do that to prevent tiny blips from making any noise on the desktop. Under load, I can't get this system to go over 61C. At least the hardest workload I know of, Prime95 with 24-thread torture test.
Well then that Noctua is a great cooler, especially in a tight case. I will run some prime95 AVX later and get back to u with the temps I am seeing. I also have a Lian li Dynamic XL so my case is huge, and if you are getting those temps, it is putting my 360 AIO to shame
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
Well then that Noctua is a great cooler, especially in a tight case. I will run some prime95 AVX later and get back to u with the temps I am seeing. I also have a Lian li Dynamic XL so my case is huge, and if you are getting those temps, it is putting my 360 AIO to shame
Yeah I didn't post it to show off or be a dweeb about it. If I knew how to help you guys get those temps down, I would. I was just happy- I have been so impressed with this system that I came to the forum to see what people were talking about in regards to cooling. I always wanted a D15 and this worked out better than I imagined. I'd use an AIO but I know there's a massive difference between different choices and there's a lot going on there, I'm kind of lazy, I want to set things up and forget it. The only ones I've looked at though were the NZXT and I'm unfamiliar with the rest. I took a screenshot of it last time I did a 10 minute run, I did this a couple times to burn it all in. Proof (as good as you can produce at least) because talk is cheap and people lie, but I don't.

It does get worse. This is with an IC Diamond graphite thermal pad. One that has been moved from a previous build and it's cracked/torn, which probably hurts nothing, which is why I used it again. No paste in there at all, I wanted zero maintenance. I remember a Jayz2cents video or similar that said water is always better for short-burst cooling, and they take longer to saturate. I'm hitting 5050MHz on two of my twelve cores everyday though, so I'm not complaining about the burst performance.

1611094448248.png
 

KickAssCop

Supreme [H]ardness
Joined
Mar 19, 2003
Messages
6,887
Yeah I didn't post it to show off or be a dweeb about it. If I knew how to help you guys get those temps down, I would. I was just happy- I have been so impressed with this system that I came to the forum to see what people were talking about in regards to cooling. I always wanted a D15 and this worked out better than I imagined. I'd use an AIO but I know there's a massive difference between different choices and there's a lot going on there, I'm kind of lazy, I want to set things up and forget it. The only ones I've looked at though were the NZXT and I'm unfamiliar with the rest. I took a screenshot of it last time I did a 10 minute run, I did this a couple times to burn it all in. Proof (as good as you can produce at least) because talk is cheap and people lie, but I don't.

It does get worse. This is with an IC Diamond graphite thermal pad. One that has been moved from a previous build and it's cracked/torn, which probably hurts nothing, which is why I used it again. No paste in there at all, I wanted zero maintenance. I remember a Jayz2cents video or similar that said water is always better for short-burst cooling, and they take longer to saturate. I'm hitting 5050MHz on two of my twelve cores everyday though, so I'm not complaining about the burst performance.

View attachment 320554
Can you post results from CBR20 and HWInfo.
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
Can you post results from CBR20 and HWInfo.
I don't have CBR installed right now but will be happy to put in the effort to get you those. Just eager to see Island's results, and wouldn't mind if you shared yours too- you also have a 5900X.
 
Joined
Nov 25, 2020
Messages
28
Did some more temp analysis. R6 ranged from 65C to 75C over 3 hours. Just for fun I tracked it during Halo 2, and it came back in the 50s over an extended play time.

Path of Exile Stayed between 55C and 60C most of the time.
 
Last edited:

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
Thought I'd check back in on this thread, see if Island had his results yet.. I haven't done my CBR testing but I'd like to after updating to AGESA 1.2.0.0.
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
I'm guessing Island did not get great results and is pissed, or he'd be back here by now.

I ran this for ya. I don't know if my score is great or not but everything is stock here, nothing is tweaked out. Temps while this ran varied from 56C-59C. Peak of 62C at some blip that I didn't catch as you can see here. While performance could be tweaked out, only running spec 3200MHz on my RAM, but my temps are still (apparantly) phenomenal on ye ole D15! Hail Noctua!

1612411756194.png
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
May as well put up CPU-Z since I'm here, and just ran it.
1612414361933.png


I found the single core CBR20 bench too, thought I'd toss that in. Temp gets higher at peak here because in the normal multicore bench that I posted above, I ran it on a loop to get my CPU fans closer to full speed since I use a 255 second timer to spin those up so it takes a while. For the single core, I was too lazy and just ran it so it blipped at 66C while the fan was spinning up. It was about 62C the whole time. Seems kinda weird that this was a "worse" result than the multithreaded, no idea. Those are the results.

1612415063473.png
 
Last edited:

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
You've definitely got your rig well optimized. I never was the type to care too much, just want stability and if i have that, life is good. I've never seen instability on my system, been rock solid the whole time. With my 2700X I did have issues, but it was because the memory controller spec was 2933 and I was running 3200, which was randomly failing. It wouldn't hold up under a night of memtest86 until I dropped it down to 2933.
But if you have any general bios settings that I should be enabling, let me know. I think for a stock setup, my scores are correct and fine. I'd like to try your hardware as configured, with my cooling and see how it does.
 
Last edited:

housecat

Gawd
Joined
Jan 26, 2005
Messages
741

New results for me now too I guess. I tried these 3 settings. Didn't think they'd make any difference being on auto, but it seems to have.
1612501368740.png


The PBO Fmax Enhancer seemed to have hurt my single core performance, so I disabled that. Dropped it to 595. The other two bumped my multicore from 7709 to 8301. Could just be variance in runs though? I don't know. Both PBO enabled (I think one is for older Zen revisions), but PBO Fmax off.
1612501625655.png


My temps did rise though from before, it didn't throttle, but it did hit 80C finally. Only the multicore test, the single core stays at 60-61C the entire time. So while my old scores are still pretty good from what I know, these are a lot better just brought up my temps. If I were running the system at nearly 100% on all cores daily I'd probably put it back to how it was. But rendering at 100% on all cores isn't a usecase I'm going to be doing.

Pretty good scores though I think. I'm relatively ignorant as I don't look around much. I'm chalking this up to the trusty Asus X470, Asus has never done me wrong now for 21 years and counting.
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
Well then that Noctua is a great cooler, especially in a tight case. I will run some prime95 AVX later and get back to u with the temps I am seeing. I also have a Lian li Dynamic XL so my case is huge, and if you are getting those temps, it is putting my 360 AIO to shame
We going to hear back from you? Curious on more benchmark results myself. My rig like all of yours is a new 5900X build so comparing is good. I'm happy with my 8301/612 CBR20 score, but my previously amazing temps appear to be gone now- so rest easy on that front! I'm probably closer to the rest of you. Unleashing the additional 600 points in CBR seems to have unleashed some power limitation that was in place with PBO being set to auto in my BIOS.
 

evhvis

n00b
Joined
Feb 12, 2021
Messages
31
I'd like to invite both of you to joining me on the darkside and buying a D15. I have a 5900X and am doing a lot better than both of your outcomes. Blowing those Corsairs out of the water.. I can't get mine to go over 61C after 10 minutes of a Prime95 24-thread torture test blend in a 19.6 liter mITX case, holding 4.5 to 4.6GHz on all cores. In games as you're describing, I'm not going much over 50C. It does sound like temps you're getting on that AIO are within expected range.
Keep in mind that when playing games you will be GPU limited. E.g. running at 4k with a 2060 will make sure the CPU barely does any work so there isn't as much heat produced. If you change the res to 720p and everything to low then your temps will go up so comparing games without everything else the same is apples to oranges.

Running 24 threads prime95 is actually very kind to the 5900x as all the cores will be put down to low wattage and heat will be distributed in over a large area. Worst temps in prime95 will be achieved with 4-6 threads (first point which you reach 142 watts). This will put all the load on one CCD and wont have switching between threads so you get a lot of heat in a tiny area. For reference I get temps around 57C with 24 threads small fft in prime95 and with 5 threads I am at around 80C with ambient around 23C. For the torture test I average around 60 degrees as it doesn't even fully load the CPU (only 133 watts average). In cinebench 24-thread I have around 62C average with some minor peaks into 65C as it fully loads the CPU to 142 watts. Keep in mind all my temps are CPU-package which are about 10 degrees above CPU reported by the motherboard. The one noted as CPU in hwinfo is afaik motherboard and reports around 10-12 degrees lower than CPU-package, depending on how high the temp is, so deduct 10 degrees from all my numbers if you compare with the one that says CPU. Keep in mind that there is a noise factor as well. Case fans are running at 650rpm and low noise fans on rad are making noise equivalent to running two 140mm noctua fans at 800rpm on inntake.
 
Last edited:

evhvis

n00b
Joined
Feb 12, 2021
Messages
31
We going to hear back from you? Curious on more benchmark results myself. My rig like all of yours is a new 5900X build so comparing is good. I'm happy with my 8301/612 CBR20 score, but my previously amazing temps appear to be gone now- so rest easy on that front! I'm probably closer to the rest of you. Unleashing the additional 600 points in CBR seems to have unleashed some power limitation that was in place with PBO being set to auto in my BIOS.
Posting my cinebench scores in case you are interested. The score of 8359 is stock and 8770 is with PBO and 627 single core with PBO off. Single core score drops when PBO is on, not sure why. Not much gain for a whole lot of watts. Keep in mind that the CPU will throttle on watts. Mine stays at at 195w average. My guess is yours throttles a bit more. The D15 is a great cooler, but once you run PBO with a 200w limit you are close if not above the thermal limit of the D15. At stock 142w it should work very well if noise is not a factor (needs high speed on case fans to provide fresh air vs. a front mounted rad).
cinebenchRes.jpg
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
I never understood the people that worry about noise. I have my rig to turn all 5 fans up to 100% at 50C+. I think that's why I'm getting such good results. Noctua fans are still definitely bearable at full speed. I also have a pair of Sony WH1000XM4 noise cancelling headphones, but I didn't buy them for the fans. :) I'm pretty happy with my results, I'm within margin of error with everyone else's 5900X, and my multicore is only suffering a bit due to zero memory optimization being done. I'm also on a mini-ITX board, which is surprisingly adequate. To be clear, I like watercooling, I just don't think the additional performance is worth it for me. If you have the space as I do, and you refuse any case that doesn't have a horizontal motherboard layout (GPU sag and any form of strain from components is eliminated), a big chunk of metal and fans makes sense for me.. someone who wants to set it and forget it. I highly doubt in real world use it makes any difference between the best coolers, especially in gaming. My machine only really gets any sort of intensive use when in-home streaming (8-threads in use and the game, Steam Link) or code compilation.
 

evhvis

n00b
Joined
Feb 12, 2021
Messages
31
I never understood the people that worry about noise. I have my rig to turn all 5 fans up to 100% at 50C+. I think that's why I'm getting such good results. Noctua fans are still definitely bearable at full speed. I also have a pair of Sony WH1000XM4 noise cancelling headphones, but I didn't buy them for the fans. :) I'm pretty happy with my results, I'm within margin of error with everyone else's 5900X, and my multicore is only suffering a bit due to zero memory optimization being done. I'm also on a mini-ITX board, which is surprisingly adequate. To be clear, I like watercooling, I just don't think the additional performance is worth it for me. If you have the space as I do, and you refuse any case that doesn't have a horizontal motherboard layout (GPU sag and any form of strain from components is eliminated), a big chunk of metal and fans makes sense for me.. someone who wants to set it and forget it. I highly doubt in real world use it makes any difference between the best coolers, especially in gaming. My machine only really gets any sort of intensive use when in-home streaming (8-threads in use and the game, Steam Link) or code compilation.
Margin of error is generally 1% or at most 2% so around 5% is well outside. How much real world impact it would have depends on the use. My setup is built in memory profile and all else stock, outside of turning on PBO in one run. Manual OC and increasing temp and power limits just dont't bring real world gains that are large enough on the 5900x and it is just better to control the temps as the CPU will OC by itself. That said, I do agree with you that real world gains are generally small outside of noise between air and water on stock settings as there just isn't enough heat at 142 watts or less.

People have different threshold for noise. I get annoyed with 140mm noctua fans at 800rpm on desktop use and even went as far as to get 7tb of SSDs to avoid the noise. At most I will tolerate 1200 rpm for the case fans for gaming, but I usually just leave them well below 1000rpm. I use open back headphones for gaming and the fans would be heard during quiet parts of games. Photo editing is another use, which will run at 85% CPU load when converting images and I can't stand loud noise while spending hours editing photos. Others are fine with fans at full blast speed which is OK as well. We each get to choose what we want and build our system accordingly :)

For real world use it depends on what you are using it for and how the system is set up. Sticking an AIO with rad at the top of my case would never work as my GPU dumps 300+ watts of heat (running on air) into the case so the air coming into the rad would raise the temps quite a bit. The issue would be similar though with air. Temps are at 65c on GPU at full load and noise is minimal so will probably keep it on air, but it means there shouldn't be any cooling for the CPU above it.
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
I haven't really looked hard at your results or anyone elses, I just know I'm in the ballpark at 8490 multicore. It's margin of error in my book, as a small tweak or memory overclock would put me around 8770 as you had there. I'm not going to make any changes though as my system is fine as it is. I'm running AMD spec/3200MHz. My machine isn't noisy at full fanspeed because I have all-Noctua. Not a lot of room in my case for heat to build up.
1613256117356.png
 

evhvis

n00b
Joined
Feb 12, 2021
Messages
31
I haven't really looked hard at your results or anyone elses, I just know I'm in the ballpark at 8490 multicore. It's margin of error in my book, as a small tweak or memory overclock would put me around 8770 as you had there. I'm not going to make any changes though as my system is fine as it is. I'm running AMD spec/3200MHz. My machine isn't noisy at full fanspeed because I have all-Noctua. Not a lot of room in my case for heat to build up.
Nice small-form factor computer setup :) That setup will massively improve the performance of the air cooler compared to a normal case though, so it will not be comparable if running a midi tower or full tower.

We do disagree on the definition of noisy though. My computer is all noctua as well. 3xNF-A12 on rad, 1x NF-F12 bottom inntake, 1xNF-A14 rear and 1xNF-A14 at the top. Should be around 25db on idle and around 30-32db on full gaming load for my setup. Fan RPM that annoys me was for the NF-A14 fans. The NF-A12 are quite a bit lower noise for a given fan RPM, but they do become annoying at around 1350rpm in games and around 800rpm on idle.
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
My loudest fans are those two A9s at about 2000RPM. The annoyance for me is high pitched, like the old ~60mm 8000RPM fans that I ran on my Alpha heatsink back in the Athlon Thunderbird era. I haven't had any Noctua fans do that.
I don't doubt that SFF could potentially be better for air, I think the original author (Island) mentioned that in this thread at a point. My setup is pretty swift for getting air in, and getting it back out. I'm not too worried about putting a 3000 series card in there. My setup is nearly my perfect setup if I were to design my own case. Just wish the back was designed in a way to fit a 25mm fan with the D15, there was a bit more space between the PSU and D15, and the front had a single 140mm. For me, an anti-RGB/bling type of old school user that does appreciate SFF, it's ideal.
 

evhvis

n00b
Joined
Feb 12, 2021
Messages
31
Don't know the a9 fans or your case so these numbers are guesses. If you put your a9 fans to 700rpm and the rest to 650 then you would get an idea what my idle noise is. Should be barely audible in a quiet room and PSU-fans stop is needed. PSU fan is twice as loud as the rest of my system if I turn off idle stop (fan doesn't run until around 300w powerdraw). Increase to 1000 rpm on a-9 and rest to around 800 for the rest and you will probably get an idea what my gaming noise is like. My case does damp high pitched noise though and quite a bit of the fan noise so my estimates could be slightly off, but it should give you an idea.
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
I'll take your word for it, not in the mood for testing, but I can say I tested my own setup and I set my normal operating speeds at the most I want to tolerate in desktop use. Which for me is 50% fanspeed. Which comes to about 1000RPM on my A9/A12x15s and 750RPM on my A15. Any more than that and I notice it. This machine is right on my desk to my left, so it's hardly tucked away. That ramps up to 2K on 4 fans, and 1.5K on the remaining A15. My PSU (Corsair SF600) does shut the fan off, but I never notice that fan either way- which might change when I get a 3000 series card.
 

evhvis

n00b
Joined
Feb 12, 2021
Messages
31
Housecat I am just really curious about your D15 performance. Could you do a loop run/long test with cinebench with stock settings, let it soak for 3-4 minutes and then take a 5 minutes measurement and a similar screenshot to below which includes both average temps and average watts? Will need HW-Info or similar tool that keeps averages and Tdie is the interesting temp. Remember to reset measurements after 5 minutes or so and don't stop the test. The temps drop significantly in the short time it takes to init of the next run, even when it is on loop. Cinebench 23 and 20 give equal temps, just different scoring.

Intersting thing is that PBO only gives 6-7% at the expense of a massive temp and power gap and no I have no real world cases that benefit from it so I just leave it off unless testing thermals or benchmarking for fun.

I did put my fans on full for an unbareable 40+ db at 1m.

12 minutes pre-soak stock with ambient temp of 21C continuous loop with 5 minutes measurement. As you can see, my memory isn't anything special, just about average (D.O.C.P with no additional tweaking).
cinebenchSoakedStock.jpg


14 minutes pre-soak PBO with continuous loop with 5 minutes measurement with 21c ambient temp. Do keep in mind that the watts must be equal for it to be truly comparable. Just 10 watts difference could mean 5 degrees, depending on the cooler, especially since this is throttling on thermals
temps_pbo.jpg
 

kensiko

Weaksauce
Joined
Oct 5, 2011
Messages
124
The smaller the process goes the higher the temp will increase. In a way I miss my Threadripper, I might have a look again when Threadripper Zen3 arrives.

I confirm that PBO gives you a insignificant performance boost for a huge amount of additional power.

I've have had a similar gain overclocking my RAM to 1800 instead of 1600 without all that increased heat.
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
Intersting thing is that PBO only gives 6-7% at the expense of a massive temp and power gap and no I have no real world cases that benefit from it so I just leave it off unless testing thermals or benchmarking for fun.
That's what I found through the course of this thread. PBO while increasing my CBR20 score by about 800 points, shot temps through the roof. Probably power too, I didn't look at that. It basically just takes the CPU out of any reasonable binning limits for not a lot of gain. I assume it's better to leave enabled as most tasks are shorter bursts and not as heavily threaded as CBR20. It does look like you're at 3600MHz on your RAM while I'm at 3200, which is going to make up the sort of difference we're seeing between my system and yours in multithreaded results. I could bump mine up probably, just don't want to.
To AMD's credit, they've always said pretty clearly that PBO is to "take advantage of extra cooling and power in enthusiast systems". It's freebie overclocking without the work.
 

evhvis

n00b
Joined
Feb 12, 2021
Messages
31
That's what I found through the course of this thread. PBO while increasing my CBR20 score by about 800 points, shot temps through the roof. Probably power too, I didn't look at that. It basically just takes the CPU out of any reasonable binning limits for not a lot of gain. I assume it's better to leave enabled as most tasks are shorter bursts and not as heavily threaded as CBR20. It does look like you're at 3600MHz on your RAM while I'm at 3200, which is going to make up the sort of difference we're seeing between my system and yours in multithreaded results. I could bump mine up probably, just don't want to.
To AMD's credit, they've always said pretty clearly that PBO is to "take advantage of extra cooling and power in enthusiast systems". It's freebie overclocking without the work.
Running 3200mhz at 14-14-14-36 or something like that would generally perform about equal or slightly better in most cases as long as it is 2x 2rank. Only some cases where 3600mhz with higher latency has a major benefit over 3200mhz with lower latency. If I were to run with OC then I would probably OC once a certain threshold is reached in amps and run "stock" the rest of the way. None of the games I've tried are even close to the 142 watt limit without PBO so makes no sense for me, especially with the possability of the added heat on occasion. However even with cyberpunk running 12-14 threads the boost clocks are still 4650+ so I doubt there are any games where overclocking would benefit and for other tasks it doesn't matter much if it takes 55 seconds or 1 minute. The single core performance drops about 3% in my system though with PBO on so stock seems the most sensible.
 

evhvis

n00b
Joined
Feb 12, 2021
Messages
31
The smaller the process goes the higher the temp will increase. In a way I miss my Threadripper, I might have a look again when Threadripper Zen3 arrives.

I confirm that PBO gives you a insignificant performance boost for a huge amount of additional power.

I've have had a similar gain overclocking my RAM to 1800 instead of 1600 without all that increased heat.
From my testing it seems to be the insane voltage they run (up to 1.5v in single or low core loads) which causes temp issues at stock. I have seen as high as 22w on one core and that can easily cause temps to spike to 75+ even if I can easily cool the entire CPU at 142 watts. The CPU is designed to run this hot though and it isn't much worse than what intel has been doing so it isn't something I worry about now, but it was a bit of a shock when first running the CPU in synthetic loads.
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
From my testing it seems to be the insane voltage they run (up to 1.5v in single or low core loads) which causes temp issues at stock. I have seen as high as 22w on one core and that can easily cause temps to spike to 75+ even if I can easily cool the entire CPU at 142 watts. The CPU is designed to run this hot though and it isn't much worse than what intel has been doing so it isn't something I worry about now, but it was a bit of a shock when first running the CPU in synthetic loads.
It's hard to know the best settings with these things. This is another area where Intel just has you taken care of. I was getting acceptable performance with PBO on "auto" (off). I feel like I've seen single/dualcore speeds drop a bit with PBO on though. I was seeing 5050MHz regularly and I haven't been catching that for a while. Your single threaded point may be accurate and I'm losing a bit of ST performance for the multicore uplift, and using a lot more electricity and generating a lot more heat. In the end, likely I go back to how I entered this thread, PBO auto/off. Hard to know which is the way to go! Only thing we can do is benchmark our own individual usecases, whether it's games or work.

The smaller the process goes the higher the temp will increase. In a way I miss my Threadripper, I might have a look again when Threadripper Zen3 arrives.

I confirm that PBO gives you a insignificant performance boost for a huge amount of additional power.

I've have had a similar gain overclocking my RAM to 1800 instead of 1600 without all that increased heat.
The one issue with this is that the #1 instability generating change for people is fiddling with their RAM. Few to none do the effort of running their machine all night on memtest86 to check for memory stability. Since it plays their little games, they think all is well. It's not. Random issues pop up later on and they wonder why their "software" is so buggy.
 
Last edited:

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
From my testing it seems to be the insane voltage they run (up to 1.5v in single or low core loads) which causes temp issues at stock. I have seen as high as 22w on one core and that can easily cause temps to spike to 75+ even if I can easily cool the entire CPU at 142 watts. The CPU is designed to run this hot though and it isn't much worse than what intel has been doing so it isn't something I worry about now, but it was a bit of a shock when first running the CPU in synthetic loads.
Welp, just got a random reboot tonight while doing 8 thread libx264 encoding, so I've disabled PBO. Nothing in the event logs that's useful, no crash dump to sift through. So I can't blame a driver on it, I usually like to point the finger at Nvidia. Probably PBO pushing a bit too hard in some way.
 

evhvis

n00b
Joined
Feb 12, 2021
Messages
31
Welp, just got a random reboot tonight while doing 8 thread libx264 encoding, so I've disabled PBO. Nothing in the event logs that's useful, no crash dump to sift through. So I can't blame a driver on it, I usually like to point the finger at Nvidia. Probably PBO pushing a bit too hard in some way.
You could try setting your TRC to XMP spec instead of the autosetting that you get with D.O.C.P. Got between 1 and 1.5 percent gains from that in benchmarks. Mine was set to 85 on auto and XMP spec for mine is 58. Did stability testing similar to what I would do with a permanent OC with no issues (about 10 hours worth of tests total). Also made some minor changes to my cooling as well so those two improvements combined has me up to 635single core and 8568 in multicore in cinebench 20 (no hardware monitors running) with only D.O.C.P+trc and rest on default settings. Most of the improvement was from the TRC.
 

housecat

Gawd
Joined
Jan 26, 2005
Messages
741
You could try setting your TRC to XMP spec instead of the autosetting that you get with D.O.C.P. Got between 1 and 1.5 percent gains from that in benchmarks. Mine was set to 85 on auto and XMP spec for mine is 58. Did stability testing similar to what I would do with a permanent OC with no issues (about 10 hours worth of tests total). Also made some minor changes to my cooling as well so those two improvements combined has me up to 635single core and 8568 in multicore in cinebench 20 (no hardware monitors running) with only D.O.C.P+trc and rest on default settings. Most of the improvement was from the TRC.
I'll take a look at that. What has bugged me is that there's 3 PBO options in my BIOS.

Enabling this one is the one that works, bumps my CBR20 multicore score up 700+ points.
1614542497956.png


This one does nothing in CBR20 without the above option being set to enabled. I think it's just granular control of it, but hidden away.
1614542555301.png


Then there's this one for the chipset PBO, or so I assumed based off the reference to "Northbridge IO". Right now I'm trying it set to auto since that means "off" from what I've read. My crash was with this enabled as well, maybe it's just too much PB Overdrive boosting going on in the system. 🤔
1614542647311.png


I'm not going to set the optimum PBO boost per-core, I was just looking for the easy and free PBO boost. But if it's not going to be stable, I'm happy with standard PB performance too.
 
Last edited:

evhvis

n00b
Joined
Feb 12, 2021
Messages
31
Screen #1 is an Asus specific one/shortcut, while the 2 others are part of the AMD AGESA. There are some duplicates where Asus seems to give shortcuts or it's own version of the setting and then there is the AMD AGESA version, which I believe is in all bioses. I only tested the advanced tweaker version briefly, as I don't have any scenarios that truly benefit from extra overclocking, but it was enough to affect performance. Generally auto means disabled/stock in the Asus bios. Not sure if you should need to touch the xfr enhancement, but it is supposed to be to allow the CPU to boost beyond advertised boost frequencies and the AMD overclocking is where most of the OC features should be. Those who have spent lots of time overclocking the ryzen platform will have better knowledge of this.

Did you check your VRM, chipset and other temps while running PBO? It puts a lot more stress on components so it could be other things than just the CPU itself. I usually keep hwinfo open as it records both max and average and monitor much more than my CPU temp when I am testing PBO or other things that are well outside of stock. Voltage drops, VRM temps etc. can also be an issue when running high amounts of current and large voltages. Max temp on CPU can be kind of random at +/- 5 degrees if the load is not constant, while the averages are usually about the same for the same workload.
 
Top