You run a cleaner setup than Tom's hardware. Congrats!

pairing the cpu with a poor board will result in the board throttling when the vrms mosfets overheat

what board, and full specs?

post a cpuz validation link
http://www.cpuid.com/softwares/cpu-z.html
cpuz -> validate button -> submit button
it will open a browser, copy the url (address) and paste it here

Originally posted by _I_:

pairing the cpu with a poor board will result in the board throttling when the vrms mosfets overheat

what board, and full specs?

post a cpuz validation link
http://www.cpuid.com/softwares/cpu-z.html
cpuz -> validate button -> submit button
it will open a browser, copy the url (address) and paste it here

I don't think that's the issue, I can run cinabench all day and "Effective clocks" will stay pinned at 5.2 ghz with only very minor stretching (20 mhz) and cpu is under 100% load but when playing games and cpu is not getting 100% useage the "Effective clocks" drop. here is the validation link
https://valid.x86.fr/atj1nk [/quote]

Here is my result for comparison, I do have a 5090 though so that will push CPU a bit higher, I put same "1080p all ultra settings no up scaling no RT, no framegen" I had similar lows but max was about 270 FPS, I also don't use SMT as it's just a 100% gaming rig so that's why less cores.
https://gtm.steamproxy.vip/sharedfiles/filedetails/?id=3568042353

Now here are my results for a more typical gaming scenario, 1440P game maxed out with path tracing on, DLSS Quality Transformer Model, FPS locked at 70 with FG2X to take it to a solid 140 FPS, it just tickles the CPU Windows print screen doesn't like RT too as you can see, or maybe HDR.
https://gtm.steamproxy.vip/sharedfiles/filedetails/?id=3568043246

Originally posted by ΜΣ†ΛĿ:

but the VRM power phases are pathetic 2x4+2+1.

Iv'e dug into the board and here is what it has and it's theoretical max output.

8(80A) + 2 80A + 1 80A power stages using DrMOS.
8 phases 80A each for the Vcore main CPU power.
2 phases 80A each for the SOC voltage, which powers integrated GPU, memory controller, etc.
1 phase 80A for auxiliary power likely for other CPU-related rails like VCCSA or VCCIO.

8 phases x 80A = 640A.

AMD Ryzen processors under load is around 1.2V to 1.4V.

A typical CPU Vcore for AMD Ryzen under load is around 1.2V to 1.4V.

Using a a conservative Vcore of 1.3V a common value for Ryzen load we have Power = 1.3V x 640A = 832W.

Now a VRM can usually deliver 70/80% easy of it capability with decent cooling which the Tuff has so maybe 580-660W to the CPU max.

So long story short don't worry about your VRM it's solid dude, you would run out of power off the EPS12V rails before the VRM would give way.

Originally posted by wing0zero:

Originally posted by ΜΣ†ΛĿ:

but the VRM power phases are pathetic 2x4+2+1.

Iv'e dug into the board and here is what it has and it's theoretical max output.

8(80A) + 2 80A + 1 80A power stages using DrMOS.
8 phases 80A each for the Vcore main CPU power.
2 phases 80A each for the SOC voltage, which powers integrated GPU, memory controller, etc.
1 phase 80A for auxiliary power likely for other CPU-related rails like VCCSA or VCCIO.

8 phases x 80A = 640A.

AMD Ryzen processors under load is around 1.2V to 1.4V.

A typical CPU Vcore for AMD Ryzen under load is around 1.2V to 1.4V.

Using a a conservative Vcore of 1.3V a common value for Ryzen load we have Power = 1.3V x 640A = 832W.

Now a VRM can usually deliver 70/80% easy of it capability with decent cooling which the Tuff has so maybe 580-660W to the CPU max.

So long story short don't worry about your VRM it's solid dude, you would run out of power off the EPS12V rails before the VRM would give way.

Yeah I was sure he was trolling especially since he seems to know what he is doing but then sited Tom's hardware as a source and claimed an asus b650e had worse vrm than 90$ motherboards.

Originally posted by wing0zero:

Originally posted by ΜΣ†ΛĿ:

but the VRM power phases are pathetic 2x4+2+1.

Iv'e dug into the board and here is what it has and it's theoretical max output.

8(80A) + 2 80A + 1 80A power stages using DrMOS.
8 phases 80A each for the Vcore main CPU power.
2 phases 80A each for the SOC voltage, which powers integrated GPU, memory controller, etc.
1 phase 80A for auxiliary power likely for other CPU-related rails like VCCSA or VCCIO.

8 phases x 80A = 640A.

AMD Ryzen processors under load is around 1.2V to 1.4V.

A typical CPU Vcore for AMD Ryzen under load is around 1.2V to 1.4V.

Using a a conservative Vcore of 1.3V a common value for Ryzen load we have Power = 1.3V x 640A = 832W.

Now a VRM can usually deliver 70/80% easy of it capability with decent cooling which the Tuff has so maybe 580-660W to the CPU max.

So long story short don't worry about your VRM it's solid dude, you would run out of power off the EPS12V rails before the VRM would give way.

Thank you for confirming, So that's just how effective clocks behave on the 9800x3d then, clocks only ramp up when there is significant load on the cores. governed by load and c-states. I was expecting the clocks in games to boost to max even on low/medium load. this is a new CPU for me I skipped zen4 and and didn't have a chance to see haow they behave until now. My last CPU was a zen 3 5800x and effective clocks boosted to max clocks in game on that CPU.
Either way it's performing as it should. Thanks for your help and confirming the VRM's on the mobo are sufficient for this CPU. I actually benchmarked and stress tested this for hours this morning and was able to do -40 all core on the curve optimizer undervolt. I played around with adding more frequency but it's not worth it for me since the core voltage goes up with more frequency for not much more performance increase (single digit FPS% improvements) but adds a whole bunch of unnecessary voltage & heat.
I'm surprised this CPU is so efficient and runs super cool and I'm only using a Kraken x63 280mm AIO.

Originally posted by Arbiter of mediocrity:

Yeah I was sure he was trolling especially since he seems to know what he is doing but then sited Tom's hardware as a source and claimed an asus b650e had worse vrm than 90$ motherboards.

Trolling? I just wanted to confirm that this is how the effective clock is supposed to behave while playing games and not under 100% load. The board is actually an Asus B650e-E (has PCIE 5.0 support for GPU & SSD) it has the lowest VRM phases out of any of these boards, The previous TUF B650e (PCIE 4.0 GPU) had better VRMs 2+6+2+1 and this new one only has 4x2+2+1, The x870 & x870e boards which were made for the Zen5 even the cheapest ones have 2x7+2+1 or 2x8+2+1 VRMs and the better ones are 2x10+2+1 or 2x12+2+1 so I was wondering if the low power phases could be causing a problem. And I only Cited Toms Hardware for his CP2077 9800x3d CPU score to have something to compare to (obviously something is wrong with Tom's setup)

Anyway, here are some screenshots for the -40 all core PBO. Nice low core temps for Cinabench2024 under 100% load (aroud 60c), and the CP2077 is with SMT disabled temps around mid 40's, The PC is whisper quiet even under load, the 4090 also doesn't really got past 60c either.
https://gtm.steamproxy.vip/sharedfiles/filedetails/?id=3568230555
https://gtm.steamproxy.vip/sharedfiles/filedetails/?id=3568231224

Originally posted by ΜΣ†ΛĿ:

The previous TUF B650e (PCIE 4.0 GPU) had better VRMs 2+6+2+1 and this new one only has 4x2+2+1, The x870 & x870e boards which were made for the Zen5 even the cheapest ones have 2x7+2+1 or 2x8+2+1 VRMs and the better ones are 2x10+2+1 or 2x12+2+1 so I was wondering if the low power phases could be causing a problem.

All VRM's are overkill and more a marketing ploy that came quite a bit back, BOOM look at our VRM kind of thing when in fairness you could even get rid of half of it on your board and still deliver enough power, if you do the math on my X870 board it's something silly like capable of 1500W in theory.

https://www.asus.com/us/motherboards-components/motherboards/tuf-gaming/tuf-gaming-b650e-e-wifi

8+2+1 vrm config with heatsinks

8 x25w = 200w to cpu cores, 2 x25w to igpu, 1 x25w to imc

200w should be ok to the cores, but may not be enough to hold max clocks at load

amd says its a 120w cpu, but they can draw about 2x their rated tdp

Originally posted by _I_:

https://www.asus.com/us/motherboards-components/motherboards/tuf-gaming/tuf-gaming-b650e-e-wifi

8+2+1 vrm config with heatsinks

8 x25w = 200w to cpu cores, 2 x25w to igpu, 1 x25w to imc

200w should be ok to the cores, but may not be enough to hold max clocks at load

amd says its a 120w cpu, but they can draw about 2x their rated tdp

I disabled the IGPU and undervolted the CPU, with -40 PBO Under 100% load vcore is around 1.05v and I don't see any clock stretching when benchmaking. Seems to be enough to not cause any problems. It's not the cheapest board either. I believe it's priced at around $200 and at that price it should really have a more robust VRM's configuration. I mean the x870 boards at the same price have twice the VRM's not to mention all the other connectivity extras. I only got it because it was part of the Microcenter AMD bundle. I'm going to keep it since there are no issues. I would hate to have to rip it out and get a better board. I dread redoing the whole build again, setting up and tweaking everything not to mention all of the win 11 tweaks. It takes way to much time to tweak and stress test everything to be worth upgrading the board just to get the same performance. I have lost enough sleep already on this build.

the igpu has its own vrm mosfets

test with prim95 small fft test
that will give it an unrealistic load at 100% on all cores
then watch the core clocks with hwmonitor
if they do drop below stock, the board is throttling it

P95 is what I used to use all the time when overclocking back in the day before PBO and undervolts, had to pass 12 hr test to consider stable. I haven't used it in a while and forgot how it tortures the CPU. Anyway my -40 was unstable it froze the PC within the 1st few minutes, Is it unrealistic to expect to pass the Small FFT with -40 on the voltage curve, I enabled LLC and it went a bit longer but locked up again. Is LLC even usefull with PBO? I know when doing manual overclocks with fixed voltage it's necessary but I'm not sure how it works with PBO on the voltage curve. I backed it down to -30 and it seems good also left LLC on level 4. CPU hits 90 degrees really quick and throttles down from max possible clock about 200mhz and stays there, voltages don't drop and core effective clocks are aligned with core clocks.
see screenshot below, does this look good? screenshot was taken after more than 10 minutes of starting the test
https://gtm.steamproxy.vip/sharedfiles/filedetails/?id=3568384699

looks fine then

are games stuttering or anything?
software could just be reading the low clocks at those samples

Originally posted by _I_:

looks fine then

are games stuttering or anything?
software could just be reading the low clocks at those samples

no, everything seems to be working perfectly fine and in line with other peoples results

after messing around in the bios for a bit more I was able to get the -40 all core on the voltage curve to not crash by setting the PBO scaler to 10x, temps seem the exact same except it's more aggresively pushing the higher clocks with the small voltage bump. in the end I gained 100mhz average on the small FFT test, however over a 15 or so minute run I noticed that the vcore had spiked to 1.1v for a second (might be caused by LLC L4)

Does anyone know if LLC is usefull when using PBO and VC?

here is the screenshot of the -40 run around 15 minutes into it.
https://gtm.steamproxy.vip/sharedfiles/filedetails/?id=3568407403