Ryzen 5 5600X - raising limits lowers performance?

[wymdrnik]

Hi there.

After a 7-8 years I finally upgraded my piece of hardware and switched from Intel to AMD.
I'm trying to find out what is my piece of silicon capable of (undervolting, overclocking, finding sweet-spot, ...).

To begin with, I understand how Precision Boost algorithm handles the boosting thing on my processor.
There are 3 main limits (TDC, EDC, PTT) and then maximum boost frequency.
Started with a default voltage of 1.300V, LLC level 4/5, PBO disabled, I get 648/4134 points SC/MC in CPU-Z benchmark v.17...
I saw in HWiNFO that PPT limit was at 100% (it is set to 76W by default) and hitting EDC limit was also close (90A).
Lowered the Vcore to 1,281V, scores were: 648/4489. Makes sense, saw all cores boosting up to 4050 MHz in multi-core part of the benchmark.
Single core max boost was hitting 4650 MHz, which was already achieved during my first test - that's PB frequency limit. I get it, totally, so far.

Then, enabled PBO to manual mode, Fmax disabled, maximum freq. override +125 MHz, scalar 3X, PPT set to 105W, EDC to 130A and TDC to 80A.
Voltage still at 1,281V. For unknown reason, I get much lower SC score - just 616 points. Multi-core score is fine - 4839 points.
I can see in HWiNFO that EDC limit was hit, that's why cores stopped boosting after 4600 MHz in multi-core part of the benchmark.

Could someone explain me, why do I get lower score in single-core part of the benchmark when I push the limits upwards?
Saw in HWiNFO, that PPT consumption was at 86W the software actually reported that cores (at some time) boosted to 4775 MHz.

And same thing (lower SC score) happened when I change PBO mode to Motherboard limits (which are 190A/395W EDC/PPT, don't remeber the TDC one).
I'm not using XFR Enhancement nor AMD Overdrive, just playing around with Ai Tweaker section.

(MB is ASUS B550-E Gaming with latest BIOS, 2x16 GB DDR4 RAM G.Skill 3466MHz 16-18-18-38 @ 1.350V, 2x 512GB M.2 SSD, Radeon RX 5700XT, other stuff ...)

 

[freeagent]

Sounds like it is starving for power.. that cpu can pull a little more than 140w if you let it. I have only played a little bit with limits.. so I cant precisely answer your question. I did set limits once to that of what a big cpu would pull and it hurt performance..

 

[wymdrnik]

Makes no sense to me that it's starving for power in SC benchmark. No limit is hit (I reset the HWiNFO stats right after CPU-Z switches to SC part).
Reported PPT 86W vs. 105W set as a limit by me ... there's still 19W of unused power left. EDC limit was hit during MC test only, but all cores were at 4600~4625 MHz.

 

[Zach_01]

Hi there.

After a 7-8 years I finally upgraded my piece of hardware and switched from Intel to AMD.
I'm trying to find out what is my piece of silicon capable of (undervolting, overclocking, finding sweet-spot, ...).

To begin with, I understand how Precision Boost algorithm handles the boosting thing on my processor.
There are 3 main limits (TDC, EDC, PTT) and then maximum boost frequency.
Started with a default voltage of 1.300V, LLC level 4/5, PBO disabled, I get 648/4134 points SC/MC in CPU-Z benchmark v.17...
I saw in HWiNFO that PPT limit was at 100% (it is set to 76W by default) and hitting EDC limit was also close (90A).
Lowered the Vcore to 1,281V, scores were: 648/4489. Makes sense, saw all cores boosting up to 4050 MHz in multi-core par of the benchmark.
Single core max boost was hitting 4650 MHz, which was already achieved during my first test - that's PB frequency limit. I get it, totally, so far.

Then, enabled PBO to manual mode, Fmax disabled, maximum freq. override +125 MHz, scalar 3X, PPT set to 105W, EDC to 130A and TDC to 80A.
Voltage still at 1,281V. For unknown reason, I get much lower SC score - just 616 points. Multi-core score is fine - 4839 points.
I can see in HWiNFO that EDC limit was hit, that's why cores stopped boosting after 4600 MHz in multi-core part of the benchmark.

Could someone explain me, why do I get lower score in single-core part of the benchmark when I push the limits upwards?
Saw in HWiNFO, that PPT consumption was at 86W the software actually reported that cores (at some time) boosted to 4775 MHz.

And same thing (lower SC score) happened when I change PBO mode to Motherboard limits (which are 190A/395W EDC/PPT, don't remeber the TDC one).
I'm not using XFR Enhancement nor AMD Overdrive, just playing up with Ai Tweaker section.

(MB is ASUS B550-E Gaming with latest BIOS, 2x16 GB DDR4 RAM G.Skill 3466MHz 16-18-18-36 @ 1.350V, 2x 512GB M.2 SSD, Radeon RX 5700XT, other stuff ...)

Try with lower scalar and not X3. Leave it on auto/default(X1) or X2.
Other than the default (X1) higher numbers just raise the voltage across all loads and frequencies.
This would help raising frequency across all workloads but only if temp remains the same. If you do not increase cooling along with scalar then the increased voltage = increase temperature = counteracts with raising frequency

Also when trying PBO enabled, use the advanced method (not MB limit) or just leave it Enabled.
Auto = Disabled.

 

[freeagent]

Makes no sense to me that it's starving for power in SC benchmark. No limit is hit (I reset the HWiNFO stats right after CPU-Z switches to SC part).
Reported PPT 86W vs. 105W set as a limit by me ... there's still 19W of unused power left. EDC limit was hit during MC test only, but all cores were at 4600~4625 MHz.

Yup, sorry like I said I haven't bothered tuning that way yet. I didn't really like what I saw for results and prefer an all core oc.. one day I will work up to it.. maybe after the summer. And starving for power could have been a poor choice in words.

 

[blu3dragon]

"Lowered the Vcore". Where and how are you doing this? Normally if you are using PBO you should leave vcore on auto since overriding this can lead to lower performance.

 

[wymdrnik]

OK, sorry no time to reply sooner ...

So, after further investigation, blu3dragon, you were right - changing Vcore to anything from "Auto" disables PBO. That was the reason of poor performance.

When I leave Vcore at Auto, PBO at disabled, I get 648 SC and 4880 MC score in CPU-Z benchmark v.17.

When I enable PBO, leave Vcore at Auto and just raise the limits (except for max. freq.), performance raises in MC to 5016,
in SC stays at 648 (seems obvious as of +0 MHz freq. overdrive). Voltage not worth it.

With freq. overdrive set to +150 MHz, holy shit, voltage set at Auto raised to 1.444V while in SC benchmark. Definitely not worth it.

My best result while doing manual OC (PBO disabled, Vcore manual) was with all-core 4.8 GHz @ 1.3250V (669/5304). Need to re-test with HWiNFO.

I haven't done Prime95 or OCCT test yet, but it was stable in CPU-Z stress test for 1 hour and I played Borderlands 3 with my cousin few times and no crashes so far, knock knock.