Did I Degrade or Destroy my 5950x? Now with a response from Asus on page 2!

You also had erroneous registers in Ryzen master which leads me to belive there's more to it than software.
And those errors have been there in Ryzen Master from day one with this board on multiple cpus. I always assumed that was just a motherboard incompatibility.
 
Perhaps an email to asus support with screenshots of the anamalous readings and see what they say.
I checked out Ryzen Master on my server which is using an Asus Tuf Gaming x570 board and a 3900x. It shows up the same way. Zeros in the same places. I might draft them up an email tonight.
 
This is Ryzen Master on my computer, Asus CH8 Dark Hero and R9 5950X. Only enabled PBO and set manual limits. No other tweaks for the processor. Notice the empty MEM VDDIO and MEM VTT fields.

ryzenmaster.png


Your screenshot says OC Mode: Default. Which would mean stock frequencies with no PBO. Resulting in a 4GHz-ish all core boost, or even less. You need PBO and increased limits to get much further in my experience.
 
This is Ryzen Master on my computer, Asus CH8 Dark Hero and R9 5950X. Only enabled PBO and set manual limits. No other tweaks for the processor. Notice the empty MEM VDDIO and MEM VTT fields.

View attachment 473672

Your screenshot says OC Mode: Default. Which would mean stock frequencies with no PBO. Resulting in a 4GHz-ish all core boost, or even less. You need PBO and increased limits to get much further in my experience.
Thanks for confirming the Asus reporting bug in Ryzen Master. As for PBO, my problem is that I can set any level PBO and still not score as high as I was at stock. The best I can score is around 10600 in CB20, using over 200 watts for the cpu. I was scoring 11500 or so at stock wattage of around 130 watts.

Last night, trying to figure out if there is one bad core holding things back, I tested some in gaming mode and the 8 cores would boost up close to 4500 MHz and score 5600 in CB20. I then enabled only 4 cores from each CCX and it boosted similarly and scored 5800 in CB20, roughly half of what I was getting before at stock all core. Don't know if this tells me anything or not.
 
I finally got around to sending an email to asus support. We'll see how this goes.
 
Here is the response from Asus, about what I figured....

I’m writing this email as an update regarding the escalation case. In coordination with our service center, The specified AMD all core frequency is around 3.5-3.8Ghz range. So, if the customer tries to go above 3.8Ghz, which will be considered overclock territory, the customer has to adjust the Core voltage much higher for stability. When the customer enables PBO, please ensure to set Core voltage to 1.4V or higher like 1.41.V for stability If it still doesn't work well, the customer got not so good processor silicon lottery lot. If that is the case, the customer has to lower the frequency to a stable frequency or replace the processor for a better silicon lottery.

So, according to Asus, don't expect an all core boost on your 5950x greater than 3800 without "overclocking".
 
Here is the response from Asus, about what I figured....



So, according to Asus, don't expect an all core boost on your 5950x greater than 3800 without "overclocking".
That's BS. With good cooling and a decent motherboard, you should expect 4.2ghz all core----in an AVX load. Potentially higher, with non-AVX.

I had one for awhile and saw it myself. In an Asus board.

Here is Techpowerup's clock validation:
https://www.techpowerup.com/review/amd-ryzen-9-5950x/21.html
 
Last edited:
Without PBO working, he is correct. PBO will get you 4.x, non PBO is 3.7
Well, I got 4.2+ all core with stock settings and no PBO for quite a while. It would thermal throttle at 90 degrees even when water cooled pulling the same amount of power it does now except with much higher performance.

I'm sure AMD would love to post all over the Internet that they expect a max 3.7 all core boost under load on their flagship processor at stock earning a score of 10,000 in Cinebench R20. That would really get all the content creators excited.
 
Last edited:
Well, I got 4.2+ all core with stock settings and no PBO for quite a while. It would thermal throttle at 90 degrees even when water cooled pulling the same amount of power it does now except with much higher performance.

I'm sure AMD would love to post all over the Internet that they expect a max 3.7 all core boost under load on their flagship processor at stock earning a score of 10,000 in Cinebench R20. That would really get all the content creators excited.
Pretty sure that is with PBO. I'm 99.9 % sure it will use the CPUs stock TDP setting of 105w(unless you manually change it) which is very limiting for 16 cores. PBO extends those limits to a much higher # which allows the CPU to stretch its legs more.
 
Pretty sure that is with PBO. I'm 99.9 % sure it will use the CPUs stock TDP setting of 105w(unless you manually change it) which is very limiting for 16 cores. PBO extends those limits to a much higher # which allows the CPU to stretch its legs more.
Stock settings would push between 120 and 130 max through my cpu. It was enough for good boost, then one day, no more. I can hit it with 200 now and not achieve the same boost.
 
That's BS. With good cooling and a decent motherboard, you should expect 4.2ghz all core----in an AVX load. Potentially higher, with non-AVX.

I had one for awhile and saw it myself. In an Asus board.

Here is Techpowerup's clock validation:
https://www.techpowerup.com/review/amd-ryzen-9-5950x/21.html
It will always depend on the type of load. Some loads use a lot of the CPU and will result in lower clocks while lighter loads will cause the CPU to downclock more. There is also some downclocking in certain types of loads. You would have to use the exact same software for clock testing under load to have a valid comparison. According to techpowerup it is a "light" load compared to some of the benchmarks, stress tests etc. so clocks are probably on the high side for the thread count.
 
It will always depend on the type of load. Some loads use a lot of the CPU and will result in lower clocks while lighter loads will cause the CPU to downclock more. There is also some downclocking in certain types of loads. You would have to use the exact same software for clock testing under load to have a valid comparison. According to techpowerup it is a "light" load compared to some of the benchmarks, stress tests etc. so clocks are probably on the high side for the thread count.
Almost all of my testing for this has been using Cinebench R20 with HWInfo64 and/or Ryzen Master being the only other app(s) running. I did run 3dmark timespy once. With Cinebench, it was a notable drop going from scoring above 11500 to just under 10000 with the same settings. Power numbers were pretty much identical. The chip is using the same amount of power, but running much cooler while boosting lower and scoring lower. With timespy, the cpu score was lower, but not as big of a difference as what is shown in Cinebench. CPU temps (Tdie/Tctl) never get out of the 60s anymore. I used to peg 90 or close too it when benchmarking. Since I control my pumps and fans by liquid temp, the CPU should be hitting high temps until the liquid temps rise and up the pump and fan speed.
 
Just to take Windows out of the equation, I ran Ubuntu from a USB drive and ran Geekbench 5. I got a score of 1794 single core and 17626 multicore. My Geekbench 5 Score According to Geekbench, the 5950x should score around 1686 single core and 16516 multicore. Average Geekbench 5 Score for 5950x. What is strange is that the cpu never got warm at all during the test. All I had to go by was the LCD readout on the motherboard, but it never got out of the low 50s and was mostly in the upper 40s. That translates to Tctl/Tdie temps of low 60s for max. Could Windows be my problem? Does Geekbench 5 really test the limits at all the same way Cinebench 20 can?
 
Just to take Windows out of the equation, I ran Ubuntu from a USB drive and ran Geekbench 5. I got a score of 1794 single core and 17626 multicore. My Geekbench 5 Score According to Geekbench, the 5950x should score around 1686 single core and 16516 multicore. Average Geekbench 5 Score for 5950x. What is strange is that the cpu never got warm at all during the test. All I had to go by was the LCD readout on the motherboard, but it never got out of the low 50s and was mostly in the upper 40s. That translates to Tctl/Tdie temps of low 60s for max. Could Windows be my problem? Does Geekbench 5 really test the limits at all the same way Cinebench 20 can?
Try to uninstall ryzen/amd drivers then reboot and install the newest drivers and check your powerplan. I use balanced which seems to be good with ryzen 5000 series.
 
Try to uninstall ryzen/amd drivers then reboot and install the newest drivers and check your powerplan. I use balanced which seems to be good with ryzen 5000 series.
I did that when I first discovered the problem and it didn't help. There is a windows version of geekbench, so, when I have time, I'll run that and see what happens. If it scores poorly, I'll know where the problem is.

ETA: So, I just tested Geekbench 5 in Windows. Windows Geekbench 5 Score

Windows
Single Core: 1745
Multi Core: 16576

Linux Live USB
Single Core: 1794
Multi Core: 17626

So, the Windows score for single core is still up there, but the multi core score is right at the average according to geekbench 5. The linux score is much better. It could be due to less stuff running in the background I guess. I did see some decent all core boosting with geekbench in Windows. I saw some 4500+ all core every so often for a second or two depending on the test that was running.

Max TDC: 94.79A
Max EDC: 140.0A
Max CPU Package Power: 144.774W
Max Tctl/Tdie: 66.6C

I then ran Cinebench R20 again will monitoring everything. Score was 9895

Max TDC: 95.017A
Max EDC: 140.00A
Max CPU Package Power: 132.403W
Max Tctl/Tdie: 63.8C

So power draws are roughly equivalent, but max all core boost is around 3850. I'm guessing that it's not all that different from geekbench except geekbench's individual tests are so short that it only appears to hold the boost higher.
 
Last edited:
What PBO limits do you have enabled? It sounds like it was enabled previously and is now off (leaving stock power limits, which will have somewhat minimal gains with CO). Try enabling PBO with board limits and see if the scores return to where they were - that will not over-volt the cpu, just remove the power and current limits letting it boost based on temp/voltage. As long as you don't touch (positive) voltage offset or scalar, it will not operate outside of factory silicon reliability targets.
 
Package power seems low for cinebench, it should be 140+ while the benchmark is running which leads me to believe your CPU is not running full voltage and therefore artificially limits clock speed. I only have the 5800x and 5900x CPUs but both run above 140w in cinebench when water cooled. Could make sense to remove any overclocking software (ryzen master etc.) and make sure nothing starts up that can impact the CPU (check task scheduler, services and windows startup).

Make sure you don't have lots of background processes running etc. as something like asus's lighting software can consume as much as 1% CPU at times. Running monitoring software will also impact the benchscore, but should only be 1-2% difference.
 
What PBO limits do you have enabled? It sounds like it was enabled previously and is now off (leaving stock power limits, which will have somewhat minimal gains with CO). Try enabling PBO with board limits and see if the scores return to where they were - that will not over-volt the cpu, just remove the power and current limits letting it boost based on temp/voltage. As long as you don't touch (positive) voltage offset or scalar, it will not operate outside of factory silicon reliability targets.
I've done it before in this thread, but I did it again and enabled PBO with board limits. CPU package hits 220 watts. Cinebench R20 scores 10977, about 1000 higher than without PBO. Still 500 less than previous stock.
 
Package power seems low for cinebench, it should be 140+ while the benchmark is running which leads me to believe your CPU is not running full voltage and therefore artificially limits clock speed. I only have the 5800x and 5900x CPUs but both run above 140w in cinebench when water cooled. Could make sense to remove any overclocking software (ryzen master etc.) and make sure nothing starts up that can impact the CPU (check task scheduler, services and windows startup).

Make sure you don't have lots of background processes running etc. as something like asus's lighting software can consume as much as 1% CPU at times. Running monitoring software will also impact the benchscore, but should only be 1-2% difference.
I uninstalled Ryzen Master again today. Unfortunately, it doesn't make a difference.
 
I've done it before in this thread, but I did it again and enabled PBO with board limits. CPU package hits 220 watts. Cinebench R20 scores 10977, about 1000 higher than without PBO. Still 500 less than previous stock.
That score is not bad at all dude.
My max on PBO with a lot of LLC wasn't much higher, and my chip hit high 80s in temp. (maybe 4-500 higher) so it looks like your chip is working just fine with PBO enabled.
 
I've done it before in this thread, but I did it again and enabled PBO with board limits. CPU package hits 220 watts. Cinebench R20 scores 10977, about 1000 higher than without PBO. Still 500 less than previous stock.

Have you done any further CO work? It seems odd, but you are getting consistent results and the scores are pretty good. I have aggressively CO optimize my 5950X and don't get much higher scores than that (my board limits are at ~205W), I get 4.4-4.5 ghz all core, and hit around 85C. Basically I'm -30 on all cores except the top 2, which are -5. I think maybe there was some other setting that got flipped that was giving you a bit higher scores, but it doesn't appear anything is not working properly tbh...
 
Have you done any further CO work? It seems odd, but you are getting consistent results and the scores are pretty good. I have aggressively CO optimize my 5950X and don't get much higher scores than that (my board limits are at ~205W), I get 4.4-4.5 ghz all core, and hit around 85C. Basically I'm -30 on all cores except the top 2, which are -5. I think maybe there was some other setting that got flipped that was giving you a bit higher scores, but it doesn't appear anything is not working properly tbh...
No, I stopped messing with curve optimizer once everything went to crap.
 
I'm getting ready to rebuild soon and I did a little more research on the current state of my 5950x with low all core boost clocks. I'm not the only one and I'm really starting to lean more and more into this being a motherboard issue. When I rebuild, I'm going to throw my 5950x in to my Asus Tuf X570 that is powering my server and see what happens. The cooler is just a Silver Arrow, but I'm betting I get better clocks.

https://linustechtips.com/topic/1348521-5950x-low-clock-speeds/
https://linustechtips.com/topic/1424515-5950x-will-not-boost-past-36-ghz/
https://www.reddit.com/r/Amd/comments/jtzh6x/low_allcore_boost_clock_with_5950x_and_gigabyte/

The boost is not limited by temperature, no one is maxing that out, just low clocks.

The guy and Optimum Tech has the same issue in one of his builds that regards as failed. He noted that the 5950x boosted much higher on his prior review on another board. He also eliminated heat as a cause.:



What's funny is that he rolled with it and uses is as a comparison in his 7950x review showing the 5950x boosting to 3853 all core:

 
My motherboard is an ASRock x570 Taichi on the most recent (still Beta!) Bios but its performance hasn't changed much in the last few bios.

I have only a single datapoint with my own 5950X and an Artic Liquid Freezer 420. At full all core load I use 230w, around 4.4ghz and temp over 85c. That's undervolted. Things *can* go faster and hotter on my chip, but it's about 20 watts for 50mhz and that's not worth it IMO.
 
I'm getting ready to rebuild soon and I did a little more research on the current state of my 5950x with low all core boost clocks. I'm not the only one and I'm really starting to lean more and more into this being a motherboard issue. When I rebuild, I'm going to throw my 5950x in to my Asus Tuf X570 that is powering my server and see what happens. The cooler is just a Silver Arrow, but I'm betting I get better clocks.

https://linustechtips.com/topic/1348521-5950x-low-clock-speeds/
https://linustechtips.com/topic/1424515-5950x-will-not-boost-past-36-ghz/
https://www.reddit.com/r/Amd/comments/jtzh6x/low_allcore_boost_clock_with_5950x_and_gigabyte/

The boost is not limited by temperature, no one is maxing that out, just low clocks.

The guy and Optimum Tech has the same issue in one of his builds that regards as failed. He noted that the 5950x boosted much higher on his prior review on another board. He also eliminated heat as a cause.:



What's funny is that he rolled with it and uses is as a comparison in his 7950x review showing the 5950x boosting to 3853 all core:


That's nuts... I'd be pissed for sure. My 5950X boosts to call core 4.8~4.9Ghz gaming all the time; Prime95 will sit it around 4.6Ghz (sometimes 4.5Ghz). Single core typically 5.1~5,2Ghz depending on temps. If all of that stopped one day, I'd be looking for answers quick. In this situation, I'd even ponder if it was worth building with AMD again.
 
That's nuts... I'd be pissed for sure. My 5950X boosts to call core 4.8~4.9Ghz gaming all the time; Prime95 will sit it around 4.6Ghz (sometimes 4.5Ghz). Single core typically 5.1~5,2Ghz depending on temps. If all of that stopped one day, I'd be looking for answers quick. In this situation, I'd even ponder if it was worth building with AMD again.
Asus blew me off when I submitted a ticket and just said I had bad silicon. That motherboard is out of warranty as of August, so I'm screwed there. It worked fine one day, then no workie the next. There was a power blip that rebooted the computer. No amount of different BIOS or CMOS resets has done any good. It still runs games fine and it's not terrible in 3DMark CPU test. Geekbench scores normally as well. Cinebench though, different story.
 
Asus blew me off when I submitted a ticket and just said I had bad silicon. That motherboard is out of warranty as of August, so I'm screwed there. It worked fine one day, then no workie the next. There was a power blip that rebooted the computer. No amount of different BIOS or CMOS resets has done any good. It still runs games fine and it's not terrible in 3DMark CPU test. Geekbench scores normally as well. Cinebench though, different story.
I guess one would have to know the board design to understand this behavior, but I thought boosting was handled by the AGESA software running on the mobo and not really any hardware beyond what limitations you could have with power, temps, etc. I guess if the mobo is messed up, it could be feeding the AGESA software incorrect values tricking it into thinking a power limit is reached, over current, temps, etc. But I would also think you would be able to see that with monitoring software in windows. Seems so weird to me. If you had a local MicroCenter, I'd say try a board swap just to see (as you can return no questions asked).
 
I guess one would have to know the board design to understand this behavior, but I thought boosting was handled by the AGESA software running on the mobo and not really any hardware beyond what limitations you could have with power, temps, etc. I guess if the mobo is messed up, it could be feeding the AGESA software incorrect values tricking it into thinking a power limit is reached, over current, temps, etc. But I would also think you would be able to see that with monitoring software in windows. Seems so weird to me. If you had a local MicroCenter, I'd say try a board swap just to see (as you can return no questions asked).
Yeah, it's very strange, doesn't make a whole lot of sense to me. The change occurred abruptly. I can hammer it with PBO and let the chip draw 220 watts and it only improves CB20 scores by 500 points.
 
That comparison video is hilarious though between the 5950X and 7950X that he used that broken mobo/chip combo for the 5950X. I can tell you, I get over 30K in C23 on my setup, so yeah, 25K is messed up for that chip.
 
That comparison video is hilarious though between the 5950X and 7950X that he used that broken mobo/chip combo for the 5950X. I can tell you, I get over 30K in C23 on my setup, so yeah, 25K is messed up for that chip.
Shoot, I just tested my server with a 3900X and it was running all core between 3900 and 3950 in Cinebench R20 and R23. It scored 7102 in CB R20 and 18291 in CB R23.
 
Last edited:
I had figured that 25k in CB23 was like a baseline for the 5950x. Mine hovers around that when I don't mess with PBO.
 
It started that way and was working fine, then it didn't. I flashed back to the prior standard BIOS 4006 and get the same behavior.

I know you've flashed the BIOS a couple times- did you load the setup defaults after the flash? The reason I as kis a buddy of mine had a similar problem after a power issue where the PC went off and rebooted. It started randomly crashing and he spent a couple days trying all kinds of things. Eventually I suggested it seemed like a BIOS issue and had him load the defaults- problem solved. You've probably already done that but I figured I would mention it. Somehow the power outage corrupted his BIOS settings.
 
I know you've flashed the BIOS a couple times- did you load the setup defaults after the flash? The reason I as kis a buddy of mine had a similar problem after a power issue where the PC went off and rebooted. It started randomly crashing and he spent a couple days trying all kinds of things. Eventually I suggested it seemed like a BIOS issue and had him load the defaults- problem solved. You've probably already done that but I figured I would mention it. Somehow the power outage corrupted his BIOS settings.
Thanks, it's been a while, but I did clear cmos several times and try the defaults.
 
1664809350037.png


This is where I am sitting with just minor PBO tweaks. What I have noticed, is if you set some of your PBO settings too high like EDC, PPT they will "Bug" and your chip will just not preform. Try THM Limit of 90, TDC 210, EDC 215, and PPT 500. PPT is way higher than it needs to be, BUT I have noticed this number is funky with AGESA for whatever reason. Do a - Offset on the curve optimizer of 10. Leave the FIT Scaler on Auto or manually set it to 1. The fit scaler is really the thing that will hurt the CPU. Set your CPU Boost Max Freq to +200. Just let me know what happens there.
 
View attachment 515838

This is where I am sitting with just minor PBO tweaks. What I have noticed, is if you set some of your PBO settings too high like EDC, PPT they will "Bug" and your chip will just not preform. Try THM Limit of 90, TDC 210, EDC 215, and PPT 500. PPT is way higher than it needs to be, BUT I have noticed this number is funky with AGESA for whatever reason. Do a - Offset on the curve optimizer of 10. Leave the FIT Scaler on Auto or manually set it to 1. The fit scaler is really the thing that will hurt the CPU. Set your CPU Boost Max Freq to +200. Just let me know what happens there.
I'll give it a shot tonight.
 
I'm about 28k with a power limited CPU (Limited mine to 200w) and background programs running. All core is around 4.3ghz.

When I let it get all the way up to around 230w+, it would boost up to 4.4ghz all core, but temps also went up a good 10c, and I didn't feel that was worth it.
 
I'm about 28k with a power limited CPU (Limited mine to 200w) and background programs running. All core is around 4.3ghz.

When I let it get all the way up to around 230w+, it would boost up to 4.4ghz all core, but temps also went up a good 10c, and I didn't feel that was worth it.
Yeah I generally run an all OC on mine. I just find it more responsive. It does get closer to that 237 watt mark.

1664828242877.png
1664828273261.png
 
Back
Top