I’ve been trying to optimize my setup but need some help. Unfortunately I can’t get the 7950X3D to boost properly. Max single core boost for CCD0 is 4.9 and for CCD1 it’s 5.2 GHz. Multi-core boost is lower.
Tried several settings on PBO but couldn’t get any noticable changes.
As soon as I try to undervolt with the curve optimizer (even with as little as -5) my system crashes at the Windows login screen with a BSOD (varying error messages)
Any ideas?
Thanks
System:
Ryzen 9 7950X3D
Asus RogStrix B650 A Gaming WiFi
64 GB Kingston DDR5-5600 @ 5400 DOCP I
Corsair 850W PSU
RTX 4090
Your CPU is limited by thermals. Look into what temperatures you are getting. If you cooler is decent quality then try looking into an under-volting guide.
At max values around 65°-70° that’s highly unlikely.
As I said above, even the slightest attempt at undervolting (-5) immediately crashes the system
Thought about that. When I bought the 4090 I used several PSU calculators by Corsair, BeQuiet etc, and they all recommended an 850 (or higher). Calculators NOT belonging to PSU manufacturers were even less conservative and even showed 750W PSUs as possible
The problems already start occurring without any significant GPU involvement (like on Windows startup and with CPU stress tests), however
So I tend to relegate that to the “possible but probably not” pile for the time being.
Could I just have a mediocre CPU specimen?
Will also try with an older firmware for the MB.
Update your chipset drivers and bios. He may be right about the power supply. It’s gotta be power draw or thermals. Did you disable global c state and set your system max temp to above 90?
I disabled C-State and installed the newest chipset drivers. Thanks for that. The chipset drivers were pretty old. Wasn’t familiar with C-State.
Also wanted to try the newest firmware, but it informed before installing that it wouldn’t be possible to go back to a previous firmware. That put me off.
So I went BACK through most of the previous firmwares. Now I’m on a firmware from April 2023 which ASUS recommended for maximum performance.
This firmware actually lets me set the CO to minus 10 and boost the CCD1 to 5.4 and the CCD to 5.0 on multiple cores and I saw the CCD1 peak once or twice at 5.6.
However the longer the computer runs, the lower the GHz gets and the CCD1 will be back to 5.2. The CCD0 will stay around max 5.0. Quite frustrating.
Temps are definitely NOT the problem. My case is very well ventilated. The highest CPU temp I got today was 72°. I also kept the case open today for testing to ensure that temperature is not an issue. (And before anyone asks: no there’s no dust blocking anything. I clean the interior regularly )
How can I set the system max temperature?
About the PSU issue:
I don’t want to dismiss it, but I can’t fathom how it can be, considering the CPU usually draws less than 100W and the GPU is basically idling. Even when letting Unigine Heaven Benchmark run in the background and the GPU draws about 280-300W (while the CPU get’s comfortable around 75W) this shouldn’t be a problem.
All the other components together (MB, DDR, SSDs) should draw less than 100W, so in the current situation I should have more than enough headroom. Unfortunately I have no idea how much power the board and chipset actually draw, so that’s the big unknown. But even if the MB WOULD actually draw 200W (which I’m sure it doesn’t) I would be in the green unless my PSU is faulty.
But I could be wrong of course. So if anyone has any further insight here that I miss, please feel free to educate me
@Archer374 , I am a little confused by your post. What exactly are you expecting to see? From what you posted, your CPU is working just fine. You will never see a max boost of 5.7GHz if you are running all of the cores. In a single core bench you may get near that but many factors come into play. And max boost as you already aluded to depends on the silicon lottery.
Bottom line is from your posted above picture your processor is working just fine. Also keep in mind CCD0 is going to automatically be throttled down due to the 3D vcache.
Start enjoying that CPU and stop worrying about nothing ;).
Edit: You should also consider updating your bios. Your old bios can be overvolting the CPU and also may not be measuring the voltage properly. That is why you cannot revert…CPUs were burning up due to high SOC voltages.
Thanks.
From my previous experiences with the 5950X and 5800X3D I was expecting regular boost rates much closer to the maximum. So I may have gotten a litte too invested. Especially since even on idle boosts of single cores to 5.5+ are extremely rare.
From what you’re saying it seems that this one simply behaves differently
About the BIOS I’m still on the fence, especially considering the changelog doesn’t list anything apart from the an AGESA update.
Also I think I’ve found the reason for the undervolting issues. Looks like cores 13 and 14 on my specimen seem to be extremely sensitive to undervolting. When I exclude those two completely I can ramp up the others to at least -25 (currently - still testing)
@Archer374 , I think you better go back and read all of the Bios updates since April. Each one makes a change to SOC voltages and the way they are handeled and can be adjusted. At a minimum you should upgrade to 1616.
First of all, I agree with @BoboSkypark and that you should update your bios.
With regards to getting the max boost from your cpu, I would rethink that. I used to have my boost override at +200 via PBO but have since scaled it back to 75. The reality is that those cpus hit their max boost rate very, very briefly off and on over time - mainly due to the heat generated at this high frequency. The typical cpu frequency that’s more or less maintained while stressing the cpu for long periods of time (i.e. MSFS) is several hundred Mhz less than max boost.
What I saw is that if I have my max boost at just 75 along with under volting, my 5900x of course didn’t briefly boost as high, however, it was able to maintain a higher frequency by several hundred mhz under stress. The key is the under volting reduces the heat factor allowing the cpu to maintain a higher frequency under stress which helps maintain the smoothness in the sim (for me at least).
I know you have a different cpu but the theory is still the same.
There’s also the school of thought whereby you set your override to a negative value so you can have a more consistent and longer lasting boost. I haven’t tried it, but there may be something to it.
with the old AGESA version from the pre-8000 generation my CPU boosts to 5250/>5750Mhz. I have also activated +200MHz PBO Boost additionally. The CPU fully utilized this as well. With a BCLK of 101Mhz you get a calculated 5302.5/60007.5Mhz.
Of course I also use the Curve Optimizer, 5.9-6Ghz were achieved even with load on a few cores.
With all later bios versions I only get 4.9/5.2Ghz even with a benchmark like Noobbench which only tests a single core with a very light SSE load which is supposed to get the max clock. Even in everyday life when HWInfo is running there is not a single sugar above the multiplier 52x to be seen, PBO is completely ineffective. it makes no difference whether +0/+25 or +200Mhz.
Messages in the MSI EN forum have been censored several times although it is always possible to verify that in extreme cases I am missing 800Mhz with the new bios versions.
The error still exists with AGESA 1.2.0.2b!
I’ve been testing every bios for ~1.5 years now and I’m not an OC virgin, the performance I used to have is not coming back. The whole misbehavior of course also affects load when all 8 or 16 cores are under load. It’s just difficult to show in absolute numbers because Windows distributes the threads as it pleases. In Cinebanch I can see that there is a real lack of performance.
So far, I assumed that MSI had messed up here because they censored messages about it.
btw. normally you could also influence the boost with the Load Line Calibration. With MSI, however, the options are just placebos without any function. It’s simply frightening that nobody really notices this because the masses leave everything on “Auto”.
No, there is no temperature throttling, absolutely everything is in the green range. HWinfo64 shows this in clear text as a flag. I am also able to see whether my CPU is at 85 degrees or just 50 degrees. I can also assure you that opening any image file, which should normally bring the CPU briefly into the max boost, has no effect and that it does not enter the temperature range where throttled for the 1sec load.
Otherwise, as I said, Nopbench uses the “X86 No Operation” function, so the CPU does nothing as quickly as possible
h__ps://skatterbencher.com/nopbench/
→ Maximum clock at zero load.
Don’t worry, I’ve been running OC/UC for ages, the fact that the CPU no longer goes into boost is solely due to the bios.
)
