• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

Is something wrong with my CPU?

  • Thread starter Thread starter Deleted member 212040
  • Start date Start date
D

Deleted member 212040

Guest
Back to paranoia hell. Specs under my profile. I am on the latest AGESA 1.2.0.3 Patch A BIOS for my board, latest chipset drivers, latest OS updates (19043.1110).

I've seen people get their 5900X to 9000 multi-core and 650 single core in Cinebench R20. But no matter what I do, I can't get those scores. On stock, I get around 8200 multicore and 620 single core. And only a 4.15 GHz all core boost in Cinebench R20.

After setting these settings in the BIOS:

PBO -> Advanced
PPT -> 165
TDC -> 115
EDC -> 150
Max Boost Override -> +200
Scalar -> Auto
Curve Optimizer -> Per Core (cores 0 to 11), Negative offset on all. Best cores (1 and 5) set to 11, second best cores (6 and 11) to 15, all other cores to 17, as indicated by Ryzen Master.

That brought me to 8650~ multicore and 631 single core. Good enough I guess. Finally hit that 631 that AMD was talking about... after optimizing, not on stock. Getting 4.41 GHz all core in Cinebench R20 now.

I then went further, to these settings:

PBO -> Advanced
PPT -> 165
EDC & TDC -> 150
Max Boost Override -> +200
Scalar -> Auto
Curve Optimizer -> Per Core (cores 0 to 11), Negative offset on all. Best cores (1 and 5) set to 15, second best cores (6 and 11) to 17, all other cores to 20, as indicated by Ryzen Master.

This made my multicore go up by 100 to 8720, and single core went to 633. Hurray. Cinebench clocks are now 4.45 GHz all core.

With the latest settings above, I got these results in CPU-Z and Passmark:

NnRvZLJ.png


Xu1Y2Ek.png


CPU-Z link: https://valid.x86.fr/jik1tj

Userbenchmark run: https://www.userbenchmark.com/UserRun/44709128

On stock, Passmark single thread was about 3494. CPU-Z single thread was about 650-660, multicore was hovering around 9500-9600.

So, after all of this, my Cinebench scores still didn't seem to increase. No matter what settings I used. I tried a multitude of optimizer settings, sometimes surprised how low I can go, and sometimes being hit by WHEA UNCORRECTABLE FUN TIMES. But no matter what, the latest settings above seem to be my best bet.

Is this just a silicon limitation at this point or is my CPU just bad? Or have I done something wrong? Why am I not able to reach that 9000 and 650?

My temperatures are fine, 77-80C in Cinebench. Temps are not the problem here.
 
Your micro optimizing at this point. Looks like you just hit the limits of your particular setup.
 
Update: HWiNFO window after running CoreCycler. Stable, but as you can see, the max effective clocks aren't great.

etV9AQj.png


Should also mention that my highest boost in Cinebench R20 was 4.95 GHz on Core 1. That's about it. No more than that.

Your micro optimizing at this point. Looks like you just hit the limits of your particular setup.
I really hope that's the case, but looking at other's scores and effective clocks, I'm concerned.

Update: these are my R23 scores.


ksmKG5I.png


Gonna call it a day here. The CPU runs well enough for me. Sorry for the pointless thread lol. Although, low R20 scores still bother me. The CPU performs perfectly in every other benchmark, except R20. It's so bizzare.
 
Update: HWiNFO window after running CoreCycler. Stable, but as you can see, the max effective clocks aren't great.

etV9AQj.png


Should also mention that my highest boost in Cinebench R20 was 4.95 GHz on Core 1. That's about it. No more than that.


I really hope that's the case, but looking at other's scores and effective clocks, I'm concerned.

Update: these are my R23 scores.


ksmKG5I.png


Gonna call it a day here. The CPU runs well enough for me. Sorry for the pointless thread lol. Although, low R20 scores still bother me. The CPU performs perfectly in every other benchmark, except R20. It's so bizzare.

If I remember, I'll try to run my 5900x on R20 when I get home. I haven't done any kind of tinkering with my 5900x, she's just running stock. Just looking at your top clocks when running multi core runs, it seems similar to what I recall seeing on my CPU - right around the 4850 speed. Also, I know for a fact the BIOS I'm running is only using AGESA 1.2.0.0 - I haven't bothered with any more recent BIOS since things right properly for me. As they say, if it ain't broken, don't fix it.
 
You can't directly compare your specs to someone else's by running synthetic benchmarks. Your CPU's silicon is different than the one that scored 9000/650.
 
If I remember, I'll try to run my 5900x on R20 when I get home. I haven't done any kind of tinkering with my 5900x, she's just running stock. Just looking at your top clocks when running multi core runs, it seems similar to what I recall seeing on my CPU - right around the 4850 speed. Also, I know for a fact the BIOS I'm running is only using AGESA 1.2.0.0 - I haven't bothered with any more recent BIOS since things right properly for me. As they say, if it ain't broken, don't fix it.
You should probably mess with the curve optimizer and PBO2. It's free performance with no tradeoffs

You can't directly compare your specs to someone else's by running synthetic benchmarks. Your CPU's silicon is different than the one that scored 9000/650.
Eh, I dunno. Too many people reaching 650 and over, and here i am with 631.
 
You should probably mess with the curve optimizer and PBO2. It's free performance with no tradeoffs
See the attached image. This is with no tweaking. It's not the best run I've had on R20. Best single core was around 640 and best multicore was around 8600, then again, I was streaming a show while it ran so maybe that made a small impact....? Also posted a cpuz benchmark shot I ran, too.

Looks like you're running things just fine.

Capture.PNG
 
I've seen people get their 5900X to 9000 multi-core and 650 single core in Cinebench R20. But no matter what I do, I can't get those scores. On stock, I get around 8200 multicore and 620 single core. And only a 4.15 GHz all core boost in Cinebench R20.

So, after all of this, my Cinebench scores still didn't seem to increase. No matter what settings I used. I tried a multitude of optimizer settings, sometimes surprised how low I can go, and sometimes being hit by WHEA UNCORRECTABLE FUN TIMES. But no matter what, the latest settings above seem to be my best bet.

Is this just a silicon limitation at this point or is my CPU just bad? Or have I done something wrong? Why am I not able to reach that 9000 and 650?

My temperatures are fine, 77-80C in Cinebench. Temps are not the problem here.

A few people on Reddit boasting about 650+ R20 (or 1700+ R23) isn't even remotely indicative of the average 5900X rolling off the assembly line. Silicon lottery is a thing, and it affects Ryzen perhaps even more so than Intel due to the random distribution of core quality. Best not to pay too much attention to those posts - silicon quality is something you simply cannot change.

ST performance is entirely limited to the two preferred cores on CCD1, you are entirely at the mercy of their V-F characteristics and their undervolting headroom. It does not matter if your chip is like mine and another core can outperform them after CO, a ST benchmark will always default to either one of the two, or alternate constantly between the two.

Take a look at HWInfo's new per-core temperatures after a run of Corecycler. There is no valid comparison to be had to the cores on "someone else's" chip, there's way too much variance. Every core behaves differently as to the Vcore it needs, power it draws, and temp it reaches.

curve optimizer.png

5900X, no PBO, stock power limits, with a stable curve:

cb r20 stable curve.png 5900x full fans pbo2 r23 high score crop.png 5900x cpuz final co settings 680 9843.png


Bottom line - your 5900X is where the average 5900X is. Fretting about various small things to gain 10-15pts in CB R20 isn't going to get you any real performance, not even close. I think you'd be hard-pressed to see any tangible difference outside benchmarks, even if you had a 5.1 vs 4.9 effective chip. I have never seen a game that will continuously and indefinitely sustain maximum possible ST clocks - not even Corecycler can do that. ST freq will fluctuate constantly and running any sort of normal background processes will bring it down.

As for MT performance, it's pretty much entirely decided by the PPT/TDC/EDC you can push, the cooling you can afford, and the temps you find acceptable.
 
Last edited:
Your infinity setting is at 1800mhz - some run 1900 or 2000mhz and mem settings means a lot
 
If you want the big multi core scores you have to take it off auto and run a manual all core oc. 1 multi 1 voltage.. usually ends up with some good heat so be prepared for it. I have hit OTP a few times with CB.
 
Back to paranoia hell. Specs under my profile. I am on the latest AGESA 1.2.0.3 Patch A BIOS for my board, latest chipset drivers, latest OS updates (19043.1110).

I've seen people get their 5900X to 9000 multi-core and 650 single core in Cinebench R20. But no matter what I do, I can't get those scores. On stock, I get around 8200 multicore and 620 single core. And only a 4.15 GHz all core boost in Cinebench R20.

After setting these settings in the BIOS:

PBO -> Advanced
PPT -> 165
TDC -> 115
EDC -> 150
Max Boost Override -> +200
Scalar -> Auto
Curve Optimizer -> Per Core (cores 0 to 11), Negative offset on all. Best cores (1 and 5) set to 11, second best cores (6 and 11) to 15, all other cores to 17, as indicated by Ryzen Master.

That brought me to 8650~ multicore and 631 single core. Good enough I guess. Finally hit that 631 that AMD was talking about... after optimizing, not on stock. Getting 4.41 GHz all core in Cinebench R20 now.

I then went further, to these settings:

PBO -> Advanced
PPT -> 165
EDC & TDC -> 150
Max Boost Override -> +200
Scalar -> Auto
Curve Optimizer -> Per Core (cores 0 to 11), Negative offset on all. Best cores (1 and 5) set to 15, second best cores (6 and 11) to 17, all other cores to 20, as indicated by Ryzen Master.

This made my multicore go up by 100 to 8720, and single core went to 633. Hurray. Cinebench clocks are now 4.45 GHz all core.

With the latest settings above, I got these results in CPU-Z and Passmark:

NnRvZLJ.png


Xu1Y2Ek.png


CPU-Z link: https://valid.x86.fr/jik1tj

Userbenchmark run: https://www.userbenchmark.com/UserRun/44709128

On stock, Passmark single thread was about 3494. CPU-Z single thread was about 650-660, multicore was hovering around 9500-9600.

So, after all of this, my Cinebench scores still didn't seem to

Back to paranoia hell. Specs under my profile. I am on the latest AGESA 1.2.0.3 Patch A BIOS for my board, latest chipset drivers, latest OS updates (19043.1110).

I've seen people get their 5900X to 9000 multi-core and 650 single core in Cinebench R20. But no matter what I do, I can't get those scores. On stock, I get around 8200 multicore and 620 single core. And only a 4.15 GHz all core boost in Cinebench R20.

After setting these settings in the BIOS:

PBO -> Advanced
PPT -> 165
TDC -> 115
EDC -> 150
Max Boost Override -> +200
Scalar -> Auto
Curve Optimizer -> Per Core (cores 0 to 11), Negative offset on all. Best cores (1 and 5) set to 11, second best cores (6 and 11) to 15, all other cores to 17, as indicated by Ryzen Master.

That brought me to 8650~ multicore and 631 single core. Good enough I guess. Finally hit that 631 that AMD was talking about... after optimizing, not on stock. Getting 4.41 GHz all core in Cinebench R20 now.

I then went further, to these settings:

PBO -> Advanced
PPT -> 165
EDC & TDC -> 150
Max Boost Override -> +200
Scalar -> Auto
Curve Optimizer -> Per Core (cores 0 to 11), Negative offset on all. Best cores (1 and 5) set to 15, second best cores (6 and 11) to 17, all other cores to 20, as indicated by Ryzen Master.

This made my multicore go up by 100 to 8720, and single core went to 633. Hurray. Cinebench clocks are now 4.45 GHz all core.

With the latest settings above, I got these results in CPU-Z and Passmark:

NnRvZLJ.png


Xu1Y2Ek.png


CPU-Z link: https://valid.x86.fr/jik1tj

Userbenchmark run: https://www.userbenchmark.com/UserRun/44709128

On stock, Passmark single thread was about 3494. CPU-Z single thread was about 650-660, multicore was hovering around 9500-9600.

So, after all of this, my Cinebench scores still didn't seem to increase. No matter what settings I used. I tried a multitude of optimizer settings, sometimes surprised how low I can go, and sometimes being hit by WHEA UNCORRECTABLE FUN TIMES. But no matter what, the latest settings above seem to be my best bet.

Is this just a silicon limitation at this point or is my CPU just bad? Or have I done something wrong? Why am I not able to reach that 9000 and 650?

My temperatures are fine, 77-80C in Cinebench. Temps are not the problem here.
Hello. For CPU's over 6 cores, like my 5800X, it is not worth messing with the Advanced PBO settings no matter what anyone else tells you. I had to simply enable PBO and nothing else. I replaced my cooler to a Corsair H150 AIO, plugged in my fans to the board and set my fan curves in bios. Do not let the Corsair software control the fan curves. The pump and fans go off the water temp rather than actual CPU temp.

Now the CPU boosts to 4.85Ghz during games, and 4.5Ghz during all core loads, like Blender video editing. Keep your CPU as cool as possible and if your motherboard has a good quality VRM, like a 10 phase or more vrm and rated to 50 amps or more, you should hit your advertised speeds. This has been the case with all my Ryzen 5000 cpu's. 5600x, 5800x, 5950x. Hope this helps.
 
I'm good. I raised limits a bit and managed pretty significant negative optimizer offsets on all my cores. 4.5 all cores in benchmarks, 4.5 - 4.65 all cores in multithreaded games, and boosting to 4.95 - 5.00 in single thread heavy games.
 
Your infinity setting is at 1800mhz - some run 1900 or 2000mhz and mem settings means a lot
Running your memory over 3800Mhz is guarranteed to result in WHEA. This does affect your system stability and could hinder performance. I have had these issues myself. I'm now at 3800Mhz and everything is smooth. No sudden frame drops in games or games crashing any more. Tested with MemtestPro 400% at 4000Mhz, FCLK 1:1 and passed with not errors. And still I got WHEA. Just stick to 3800Mhz and tweak out your timings. Best results :)

I'm good. I raised limits a bit and managed pretty significant negative optimizer offsets on all my cores. 4.5 all cores in benchmarks, 4.5 - 4.65 all cores in multithreaded games, and boosting to 4.95 - 5.00 in single thread heavy games.
Glad you got it under control then :)
 
Running your memory over 3800Mhz is guarranteed to result in WHEA.
This is highly dependent on your CPU. My 5600X will spit no errors at 2000 1:1 but it will at 2033 up to 2100 1:1. My 5900X will spit errors at 1933 and up. I have heard of some guys not being able to run 1800 1:1 without errors on 5900/5950.
 
oh god, dont tempt me to go through ram OC again. it was pure hell
I have a youtube channel and a few good videos on memory overclocking. I can't post that stuff here because it is against the rules. I respect that. But I can send you a link if you ever want to take that fun adventure again LOL
 
I have a youtube channel and a few good videos on memory overclocking. I can't post that stuff here because it is against the rules. I respect that. But I can send you a link if you ever want to take that fun adventure again LOL
Nope, I'm good at 3600/1800 lol. The only modifications I've done to my RAM was setting tRFC/2/4 manually to 324/241/148 and setting my tRC manually from the high Auto setting to the rated 52. This brought me down a few ns, but other than that, I'm not touching RAM OC again because honestly, not worth it. I want to start using this rig and not tweaking it for the next 5 years.
 
This is highly dependent on your CPU. My 5600X will spit no errors at 2000 1:1 but it will at 2033 up to 2100 1:1. My 5900X will spit errors at 1933 and up. I have heard of some guys not being able to run 1800 1:1 without errors on 5900/5950.
Strange. I have the same exact issues with all 3 of my CPU's no matter the ram or board i use. As soon as I go over 3800Mhz I get errors. Sometimes immediately, most times in a few hours of use.

Nope, I'm good at 3600/1800 lol. The only modifications I've done to my RAM was setting tRFC/2/4 manually to 324/241/148 and setting my tRC manually from the high Auto setting to the rated 52. This brought me down a few ns, but other than that, I'm not touching RAM OC again because honestly, not worth it. I want to start using this rig and not tweaking it for the next 5 years.
That sounds like the sensible thing to do LOL As long as one is happy, that is all that matters.
 
Could be a few things, we have a thread here somewhere where a bunch of us were posting high fclk/mclk occt completion screens. Some guys were able to nail down 2033 too I think.
 
Could be a few things, we have a thread here somewhere where a bunch of us were posting high fclk/mclk occt completion screens. Some guys were able to nail down 2033 too I think.
I'm good man. Really. I have tried all I can think of and I think I'm "okei" at overclocking ram. Not the best certainly. But no matter what voltage I change, ODT, driver strengths, blah blah...issue comes up sooner or later. I have spoken to Buildzoid and it's the same for him. No way around it.
 
Indeed.. well I have a new board coming for my 5600, she will be running at 2K as usual.. stable as a table..
 
Back
Top