My CPU after optimisation of PBO settings.

[Tech Reprise]

I posted few days back that my CPU (5700X) was reaching 80°C while gaming. Today I spent a few hours optimising it and here are my results.

Ambient: 36C

 

Stock PBO Auto

 

BF1- 72°C

Apex- 75°C

CBR23- 74°C (13684/1517)

Aida64- 75°C

 

Eco mode + Curve Optimiser

 

BF1- 67°C

Apex- 70.4°C

CBR23- 62°C (12905/1537)

Aida64- 63.5°C

 

Curve Optimiser

-17 all cores except core 3&5 which are -9 as they are best cores as oer Ryzen Master. So far stable.

 

While the temps didn't decrease as much I would expect but they are atleast under or around 70C. The voltage now barely touches 1.3V except in some sudden boosts where for a fraction it touches 1.35V. 

My clocks are around 3.9Ghz in benchmarks and 4.65Ghz in games and I didn't see any clock stretching so far.

Btw as per Ryzen master while stress testing my PPT, TDC is in red/yellow zone. So is this a cause for concern?

[DoctorNick]

45 minutes ago, Tech Reprise said:

I posted few days back that my CPU (5700X) was reaching 80°C while gaming. Today I spent a few hours optimising it and here are my results.

Ambient: 36C

 

Stock PBO Auto

 

BF1- 72°C

Apex- 75°C

CBR23- 74°C (13684/1517)

Aida64- 75°C

 

Eco mode + Curve Optimiser

 

BF1- 67°C

Apex- 70.4°C

CBR23- 62°C (12905/1537)

Aida64- 63.5°C

 

Curve Optimiser

-17 all cores except core 3&5 which are -9 as they are best cores as oer Ryzen Master. So far stable.

 

While the temps didn't decrease as much I would expect but they are atleast under or around 70C. The voltage now barely touches 1.3V except in some sudden boosts where for a fraction it touches 1.35V. 

My clocks are around 3.9Ghz in benchmarks and 4.65Ghz in games and I didn't see any clock stretching so far.

 

Btw as per Ryzen master while stress testing my PPT, TDC is in red/yellow zone. So is this a cause for concern?

If you get lower score in CB23 with higher clock speeds than stock, then its a fail

[PDifolco]

24 minutes ago, DoctorNick said:

If you get lower score in CB23 with higher clock speeds than stock, then its a fail

Yep

I suspect some limit is reached and caps the CPU 

@Tech Reprise you could test with limits PPT EDC TDC removed/ ultra high

Plus did you allow +200 OC?

[Tech Reprise]

2 hours ago, DoctorNick said:

If you get lower score in CB23 with higher clock speeds than stock, then its a fail

After eco mode clock speeds in cinebench have dropped to 3.8-3.9 from 4.1-4.2 and the occasional boost to 4.65 is not seen which might give the lower multicore speed. However while playing games it does boost to 4.6Ghz.

 

2 hours ago, PDifolco said:

Yep

I suspect some limit is reached and caps the CPU 

@Tech Reprise you could test with limits PPT EDC TDC removed/ ultra high

Plus did you allow +200 OC?

No, I didn't allow any OC as I wanted to undervolt and reduce temps with as much stability as possible which my previous offset undervolts and fixed core ratio weren't giving. Now I had tried -25 All core without any PPT/EDC/TDC set and it crashed.

[DoctorNick]

5 minutes ago, Tech Reprise said:

After eco mode clock speeds in cinebench have dropped to 3.8-3.9 from 4.1-4.2 and the occasional boost to 4.65 is not seen which might give the lower multicore speed. However while playing games it does boost to 4.6Ghz.

 

No, I didn't allow any OC as I wanted to undervolt and reduce temps with as much stability as possible which my previous offset undervolts and fixed core ratio weren't giving. Now I had tried -25 All core without any PPT/EDC/TDC set and it crashed.

Ok I missed that, then its good  

[PDifolco]

26 minutes ago, Tech Reprise said:

After eco mode clock speeds in cinebench have dropped to 3.8-3.9 from 4.1-4.2 and the occasional boost to 4.65 is not seen which might give the lower multicore speed. However while playing games it does boost to 4.6Ghz.

 

No, I didn't allow any OC as I wanted to undervolt and reduce temps with as much stability as possible which my previous offset undervolts and fixed core ratio weren't giving. Now I had tried -25 All core without any PPT/EDC/TDC set and it crashed.

It's cause of the -25 all cores, put back old values but with max  PL and 200 OC

It should be fine and faster

[YoungBlade]

4 hours ago, Tech Reprise said:

Btw as per Ryzen master while stress testing my PPT, TDC is in red/yellow zone. So is this a cause for concern?

All that means is that you are near or meeting the limits for PPT and TDC.

 

PPT is a hard power limit in Watts. TDC is a current limit in Amps. If you are seeing red or yellow, it means that the CPU wants to draw more power or push more current, but the limits prevent it.

 

Basically, it means that your PBO power limits are limiting the power draw, which is kinda the point.

[Tech Reprise]

2 hours ago, YoungBlade said:

All that means is that you are near or meeting the limits for PPT and TDC.

 

PPT is a hard power limit in Watts. TDC is a current limit in Amps. If you are seeing red or yellow, it means that the CPU wants to draw more power or push more current, but the limits prevent it.

 

Basically, it means that your PBO power limits are limiting the power draw, which is kinda the point.

Great. That is what I wanted.

I just wanted lower temperatures in summer and some stability. 

 

Btw is this safe for the CPU in the long run as I am restricting the CPU instead of overclocking it.

[YoungBlade]

1 hour ago, Tech Reprise said:

Great. That is what I wanted.

I just wanted lower temperatures in summer and some stability. 

 

Btw is this safe for the CPU in the long run as I am restricting the CPU instead of overclocking it.

Yes. Undervolting and reducing power limits is actually easier on the CPU than running it at stock settings - if it has any impact on the longevity of your components, it should only improve them. An undervolt and reduced power limits will not harm your hardware.

 

The only potential risk is system instability from the undervolt. Theoretically, the worst that could happen is that it could lead to data loss if the OS or a program crashed while in the middle of writing data, but as long as you were thorough with your stability testing, the odds of that are low.

[Tech Reprise]

14 minutes ago, YoungBlade said:

Yes. Undervolting and reducing power limits is actually easier on the CPU than running it at stock settings - if it has any impact on the longevity of your components, it should only improve them. An undervolt and reduced power limits will not harm your hardware.

 

The only potential risk is system instability from the undervolt. Theoretically, the worst that could happen is that it could lead to data loss if the OS or a program crashed while in the middle of writing data, but as long as you were thorough with your stability testing, the odds of that are low.

What stability tests would you recommend?
I already did P95 and Occt AVX2 Variable test for a few hours and there were no crashes or errors in Event viewer.

[YoungBlade]

13 minutes ago, Tech Reprise said:

What stability tests would you recommend?
I already did P95 and Occt AVX2 Variable test for a few hours and there were no crashes or errors in Event viewer.

Those would be my recommendations for dedicated software tests.

 

Beyond dedicated software tests, I'd recommend actually using whatever applications/games are a part of your regular workload to see if they are all stable. It sounds like you may have already done that, though.

 

If you want to do more tests, you may want to run OCCT in its mode that switches between instruction sets as the test goes on, as older sets can sometimes let the CPU clock higher, which could expose some instability in that particular scenario. Something with Ryzen undervolting in particular that I would recommend would be to use OCCT in its core cycle mode, having it run on just one core at a time, as the boost behavior for single core loads is also different, so could again be a source of instability.

[Tech Reprise]

10 hours ago, YoungBlade said:

Those would be my recommendations for dedicated software tests.

 

Beyond dedicated software tests, I'd recommend actually using whatever applications/games are a part of your regular workload to see if they are all stable. It sounds like you may have already done that, though.

 

If you want to do more tests, you may want to run OCCT in its mode that switches between instruction sets as the test goes on, as older sets can sometimes let the CPU clock higher, which could expose some instability in that particular scenario. Something with Ryzen undervolting in particular that I would recommend would be to use OCCT in its core cycle mode, having it run on just one core at a time, as the boost behavior for single core loads is also different, so could again be a source of instability.

Thank you, will try out OCCT with the parameters you mentioned. 

[Tech Reprise]

On 4/23/2024 at 12:31 AM, YoungBlade said:

Those would be my recommendations for dedicated software tests.

 

Beyond dedicated software tests, I'd recommend actually using whatever applications/games are a part of your regular workload to see if they are all stable. It sounds like you may have already done that, though.

 

If you want to do more tests, you may want to run OCCT in its mode that switches between instruction sets as the test goes on, as older sets can sometimes let the CPU clock higher, which could expose some instability in that particular scenario. Something with Ryzen undervolting in particular that I would recommend would be to use OCCT in its core cycle mode, having it run on just one core at a time, as the boost behavior for single core loads is also different, so could again be a source of instability.

System is stable but is this difference in Tctl/Tdie and Average temps because my 3rd core clocked higher and had a sudden spike in temps for a few seconds? Core 3 and 5 are my best and second best cores and they sustain the 4.65Ghz speed for the longest as is evident from the average clock speeds across all cores.

 

[YoungBlade]

1 minute ago, Tech Reprise said:

System is stable but is this difference in Tctl/Tdie and Average temps because my 3rd core clocked higher and had a sudden spike in temps for a few seconds?

Pretty much, yes.

 

Tctl/Tdie is a hotspot temperature, and should therefore report on the hottest relevant sensor. In practice, if the temperature spike is brief, Tctl/Tdie might not record it. So you may notice inconsistencies like that when looking at Max or Min values over a period of time.