• 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.

AMD Curve Optimizer any guides / experience

I'm confused lol.

Which temp sensor in hw info should we use for people with dual CCX?

I was looking at both CCX temp sensors when doing my tests.

And the spikes where while doing benchmarks, before I could get 89C while starting running CPU-Z, and then drop to 80C all cores. When I changed to normal voltage and positive offset the spikes were upto 77C and all core about 70C. But @mthembo and @Zach_01 told me that I was getting very high voltages over 1.5V and even tho temps were below 80C those voltages are not safe.
Up to 1.5v is NORMAL for low current workloads, ie. single core boost. Anyone who tells you different fundamentally misunderstands the architecture. You will never see 1.5v with all the cores loaded.

Again, seeing the high temp spikes is directly because you are watching individual core temps as they react to context switches. And by reducing voltage you are limiting what the chip can boost to. Thus yea, you'd then see lower boost temp and thus are also getting lowered single core performance along with it.

As I wrote before I don't even bother looking at the main temp sensor. I only pay attention to die average. If you have a motherboard with the error code display, you can set it to show temps. That temp off the motherboard effectively = socket temp which is close to die average.
 
I'm confused lol.

Which temp sensor in hw info should we use for people with dual CCX?

I was looking at both CCX temp sensors when doing my tests.

And the spikes where while doing benchmarks, before I could get 89C while starting running CPU-Z, and then drop to 80C all cores. When I changed to normal voltage and positive offset the spikes were upto 77C and all core about 70C. But @mthembo and @Zach_01 told me that I was getting very high voltages over 1.5V and even tho temps were below 80C those voltages are not safe.
Easy
 

Attachments

  • Capture12ccx.PNG
    Capture12ccx.PNG
    95.8 KB · Views: 378

Attachments

  • CaptureSTTEST.PNG
    CaptureSTTEST.PNG
    90 KB · Views: 241
  • CaptureST1.56.PNG
    CaptureST1.56.PNG
    42.1 KB · Views: 234
Another sensor to be aware of is the core voltage. That sensor shows voltages for all the individual cores and average voltage so you have to keep that in mind. It's annoying actually so we have to compare the voltage vs the load to know if the voltage we are seeing is per core or average voltage for the whole die.

Yes, that is what I always look. But @thesmokingman says that we have to pay attention to average.

@thesmokingman my voltages were going beyong 1.5V, like 1.56V.
You shouldn't be seeing 1.56v unless you raised your LLC or some other setting, or its a reading issue with your board.
 
Another sensor to be aware of is the core voltage. That sensor shows voltages for all the individual cores and average voltage so you have to keep that in mind. It's annoying actually so we have to compare the voltage vs the load to know if the voltage we are seeing is per core or average voltage for the whole die.


You shouldn't be seeing 1.56v unless you raised your LLC or some other setting, or its a reading issue with your board.
This was the HW info when I ran that test (-30 on the curve using cpu core voltage normal with dynamic +0.050, cpu llc in auto, 250/125/125 for PBO, and +50Mhz for max boost), so that why @Zach_01 and @mtcn77 got worried.

HWiNFO64 v6.40-4330 Sensor Status 2021-01-10 22.41.png
 
This was the HW info when I ran that test, so that why @Zach_01 and @mtcn77 got worried.

View attachment 183750
Ok, hmm you're highlighting the core clock perf which just shows what the boost clock is, or what is possible. It's not what the clocks really are running at. The Effective clock is where you see what the clocks really are and how you judge whether you're in a low current load or high current load. In your screen above I see 1.5v right? In the max you see 1.56v. It might be a reading inaccuracy cuz the cpu algorithms max out at 1.5v so any overage would either be an error in reading, settings or issue with your board.

Effective clock vs instant (discrete) clock | HWiNFO Forum
 
Ok, hmm you're highlighting the core clock perf which just shows what the boost clock is, or what is possible. It's not what the clocks really are running at. The Effective clock is where you see what the clocks really are and how you judge whether you're in a low current load or high current load. In your screen above I see 1.5v right? In the max you see 1.56v. It might be a reading inaccuracy cuz the cpu algorithms max out at 1.5v so any overage would either be an error in reading, settings or issue with your board.

Effective clock vs instant (discrete) clock | HWiNFO Forum
Oh no, I wasn't highlighting it for that, I was just showing that all my cores were able to pass 5Ghz, effective clocks were accurate when doing multicore. Also I think CPU Core VID is what the core is asking for voltage right? (I think) if you see most of them requested over 1.5V. Anyway this test was a couple of days ago, it not what I'm running right now.
 
Oh no, I wasn't highlighting it for that, I was just showing that all my cores were able to pass 5Ghz, effective clocks were accurate when doing multicore. Also I think CPU Core VID is what the core is asking for voltage right? (I think) if you see most of them requested over 1.5V. Anyway this test was a couple of days ago, it not what I'm running right now.
Yea, regarding VID. What board are you using and what version of hwinfo is that? It looks to be setup in a way that is asking for too much VID voltage. You can see the individual core VID above core clocks. Your chip is asking 1.544v VID on each core at max. Needless to say it should NOT be doing that.
 
X570 AORUS Master rev 1.2, latest BIOS F30 with AGESA 1.1.0.0 Patch D. HW info version is v6.40-4330 (Latest), the reason for the high VID and high Core Voltage SVI could be because I had a high PPT/TDC/EDC running with positive voltage offset.

But this is no longer the case. This is my current HW info.

HWiNFO64 v6.40-4330 Sensor Status 2021-01-12 14.40.png
 
X570 AORUS Master rev 1.2, latest BIOS F30 with AGESA 1.1.0.0 Patch D. HW info version is v6.40-4330 (Latest), the reason for the high VID and high Core Voltage SVI could be because I had a high PPT/TDC/EDC running with positive voltage offset.

But this is no longer the case. This is my current HW info.

View attachment 183759
It's probably when you had the positive offset?
 
I wasnt worried exactly for the 1.56V. Only pointing to it because the CPU would never do that by each own, and one should be aware of it. Of course, as @thesmokingman said, this high 1.45~1.55Vcore is only during single/reduced threaded loads. Even in games you can get an avg of 1.45+V avg. Its normal.
I only told @juanyunis to take it easy on voltage because at the time the temp was around 80~85C for all core load, indicating potential too high voltages/speeds than normal. And power draw was over 200W.

From what I see from the 5000 series the all core voltage is (and should be) well below 1.2V

----------------------------------

I've said these(below) a few times around multiple threads/topics.

By the author of HWiNFO
All readings under "CPU AMD Ryzen XXXX (Enhanced)" is pulled straight from CPU.
Individual VIDs are just a request probably based on a speed/load/current table.
"CPU Core VID (effective)" is the final CPU request for voltage.
"CPU Core Voltage (SVI2 TFN)" is the most accurate voltage reading of what the CPU is getting.

"CPU (Tctl/Tdie)" is the hotspot temp that switches through a large number of sensors to report always the highest.
"CPU CCD (Tdie)" is the "traditional" edge temp of each CCD. This a fixed located sensor reading on some side of CCD.
"CPU Die (average)" is the average of all sensors on both CCDs (could be over 100). At this point it is unclear if this avg includes IOD sensors also. No one but AMD knows this.

----------------------------------

On a full (100%) load scenario it doesnt matter which one you see. The heat is almost equal everywhere and all report about the same temp.
Its on single and reduced threaded loads that sensors start to differentiate. On idle, browsing, watching videos the hotspot can be 10~15C above the other. On gaming the difference is reduced to 5~10C or less.
 
I just had access to CTR Tuner 2.0 RC3. (from 1usmus patreon) and tried it with my system it reported my cpu as a bronze sample and it gave me very low scores, way lower than my manual tuned PBO. Probably with CTR 2.1 with curve optimization it will get better.
 
I just had access to CTR Tuner 2.0 RC3. (from 1usmus patreon) and tried it with my system it reported my cpu as a bronze sample and it gave me very low scores, way lower than my manual tuned PBO. Probably with CTR 2.1 with curve optimization it will get better.
Keep in mind that @1usmus may have tune the app in a more conservative way than your manual settings. We dont know if the app targets for 200W or 150W or whatever it takes account.
From my understanding it's supposed to tune the CPU and not decisively overclock/overvoltage to gain performace at all cost...

The rating (bronze, silver, gold, platinum) of the CPU is surely playing major role.
 
I just had access to CTR Tuner 2.0 RC3. (from 1usmus patreon) and tried it with my system it reported my cpu as a bronze sample and it gave me very low scores, way lower than my manual tuned PBO. Probably with CTR 2.1 with curve optimization it will get better.
That just goes to show you need more voltage.
It is not the same as in the bulldozer series when E meant lower voltage and normal meant high leakiness. Now it is about the resistance vs. the low voltage scale. It actually means you are safer, but high temperature can initiate electromigration anyway. Keep things in check.

TL;DR: I've been thinking, couldn't AMD make a single best case example like high PPT low EDC when the cpu temperature is cold and low PPT high EDC when it is hot? By low and high I mean in small steps of 5 units just until it frees, or falls within the FIT threshold limiter.
 
Yeah that would be a nice feature to have in a Bios @mtcn77

Anyway, today I spent some time tweaking my curve value, instead of using one value for everything I discovered a post in reddit about using OCCT to test curve stability per core and it does work!

So my curve started at -30 for per core, then I had to tune values for core 2, 3 and 4. For core 2 and 3, I had to use -25. For core 4, I had to use -15.

So at the end of the day my settings are:

1. Voltage: auto
2. CPU LLC: auto
3. PBO Advanced settings:
a. PPT: 180
b. TDC: 125
c. EDC: 145
d. Curve: see above
e. Scalar: Auto
f. Max Boost Override: 150Mhz
g. Platform Thermal Limit: 82C (thinking on leaving 85C)

After tuning the curve, my cinebench r20 / r23 scores went up (probably because of effective clocks were more stable).

For anyone interested in tuning the curve follow this guide: (you should only use this if you are using a curve value)
1. Download OCCT
2. Open the App
3. Open Task Manager, go to Details, find OCCT app, right click and select Set Affinity
4. Select the core (single) you want to test (for example for core 0 in CCX, you have to select Core 0 and Core 1 in Set Affinity window, this is because each core has 2 threads) and click OK.
4. In OCCT select small data set, SSE instruction set, 3 or 5 minutesand and thread count to 1
5. Click start and wait for the test to complete or find error.

Notes:
If the test finds an error you have to go to bios and change the curve value for that specific core until you don't get errors.

You have to repeat the process for each core.
 
Seems while my Negative 20 was stable during load, it reboots the PC without an error during idle due to too low voltage.
 
Seems while my Negative 20 was stable during load, it reboots the PC without an error during idle due to too low voltage.
My guess is that my motherboard plays a big part on it. Even with my high value for the curve -30 for all cores I never had a reboot. I did get a couple of WHEA errors reported but they didn't cause any crash/reboot, that's why I used OCCT to resolve those WHEAs

I have an X570 AOURUS Master with latest BIOS running CPU LLC in auto, this board is know for its great VRM and power delivery so maybe it has a way to compensate.

I used to have a X570 MSI Gaming Plus and it was giving me so many power issues with my 5950X and 4 memory sticks. I was unable to enable PBO because doing so my all core clocks in some apps/programs will get stuck at 546Mhz and also my memory was unable to run higher than 3400CL14. So that's why I decided on getting a better board, and buildzoid liked the new one I have.

What CPU/motherboard do you have?
 
Seems while my Negative 20 was stable during load, it reboots the PC without an error during idle due to too low voltage.
Try to disable eco mode in bios
 
@jesdals @harm9963 could you guys run an AIDA cache and memory benchmark. I have noticed that my L3 speed drops about half (370GB/s) when doing custom limits in PBO. As soon I put Auto with the curve L3 cache speed increases to 780GB/s. Wondering why. @Zach_01 @mtcn77 @thesmokingman you guys might know why.
 
@jesdals @harm9963 could you guys run an AIDA cache and memory benchmark. I have noticed that my L3 speed drops about half (370GB/s) when doing custom limits in PBO. As soon I put Auto with the curve L3 cache speed increases to 780GB/s. Wondering why. @Zach_01 @mtcn77 @thesmokingman you guys might know why.
BIOS vs BIOS for my system
Settings the same .
 

Attachments

  • Capturebvios.PNG
    Capturebvios.PNG
    354.8 KB · Views: 259
@harm9963 interesting! thank you for sharing, so you think that it might be a BIOS bug? in my case using the motherboard limits I get stunning results for Cinebench R20 / R23, great CPU-Z and excellent AIDA results at the cost of getting close 90C temp, but I guess is fine.

CPU-Z 2021-01-13 03.59.39.png
HWiNFO64 v6.40-4330 Sensor Status 2021-01-10 22.41.png
CINEBENCH R20.060 2021-01-13 03.46.09.png
CINEBENCH R23.200 2021-01-13 03.48.19.png
AIDA64 Cache & Memory Benchmark 2021-01-13 04.07.0.png
 
@harm9963 interesting! thank you for sharing, so you think that it might be a BIOS bug? in my case using the motherboard limits I get stunning results for Cinebench R20 / R23, great CPU-Z and excellent AIDA results at the cost of getting close 90C temp, but I guess is fine.

View attachment 183910 View attachment 183911 View attachment 183912 View attachment 183913 View attachment 183914

My guess is that my motherboard plays a big part on it. Even with my high value for the curve -30 for all cores I never had a reboot. I did get a couple of WHEA errors reported but they didn't cause any crash/reboot, that's why I used OCCT to resolve those WHEAs

I have an X570 AOURUS Master with latest BIOS running CPU LLC in auto, this board is know for its great VRM and power delivery so maybe it has a way to compensate.

I used to have a X570 MSI Gaming Plus and it was giving me so many power issues with my 5950X and 4 memory sticks. I was unable to enable PBO because doing so my all core clocks in some apps/programs will get stuck at 546Mhz and also my memory was unable to run higher than 3400CL14. So that's why I decided on getting a better board, and buildzoid liked the new one I have.

What CPU/motherboard do you have?
Pass all test with OCCT , when i did power test ,my UPS alarm went off , never had that happen , know it works , test once a month , will start pc and run for a while, 510 watts .
 
Pass all test with OCCT , when i did power test ,my UPS alarm went off , never had that happen , know it works , test once a month , will start pc and run for a while, 510 watts .
@harm9963 i foresaw that when I bought my UPS. Mine is 1050w with sinusoidal wave. I'm able to run my machine at max, I also have a Mac attached to it and a huge 43.4 super ultra wide monitor.
 
Back
Top