Overclock.net banner
Cancel
Create thread New Forums

[Official] Zen 4 X3D Owner's Club (7800x3D / 7900x3D / 7950x3D)

Tags
3d v-cache 7800x3d 7900x3d 7950x3d amd
5 reading
Jump to Latest
3,381 - 3,400 of 9,076 Posts
#3,381 ·
Burnt2myCore said:
By the way I messed around a bit with Windbg last night but I am not sure how to read the crash dumps. I am using the analyze command. When it says the index of the failed core is "0000000006" or "0000000000007" does that mean core 6 or core 7? Also there is an "address" for the failed core but it is just a string of numbers and letters and I can't make sense of it.

I got to -15 on my frequency CCD Last night and 0 on cache CCD while remaining stable. Increasing one of the values on frequency ccd to 20 causes browser tabs to crash. If I set whole frequency CCD to -20/Cache CCD to 0 I will boot to windows but then my fans ramp up and I hard freeze and crash to reboot. While there is -15 on frequency CCD even -2 on cache CCD will cause ShA3 to BSOD. I am still looking for which clues to leave ar 0 or as others have suggested put a positive offset on.

Do the "fastest cores" shown in Ryzen Master have anything to do with that? If I set my "fastest" cores to 0 or positive offset in CO should that technically give me some stability? Im
Click to expand...
On Windbg you need to scroll up after pressing analyse to see what faulting core it is and remember that SMT if on counts one core as two, so core 0 and core 1 really is just core 0 and so forth, mine keeps saying faulting core 5 which really is core 2, remember as well that Ryzen master starts with core 01 which really is core 0 in the bios.

Going by what you said and taking that you have SMT on then your problem core is core 4.

Not fully understanding how you are testing there as it seems you are randomly selecting numbers for the CO (I may be wrong) if so, that’s not the way to do it as you will always be chasing your own tail.

Best to start by one core at time and work your way through, I did one CCD per day.

The quickest way was using SHA3 in single core, you can test this without process lasso just for core 0 by running AIDA64 then going to benchmark then press SHA3 then press parameters then select 1 CPU then select in your bios that core which would be core 0 and put -20 on it and all other cores should be on 0 no positive or negative offsets and run SHA3.

Once you do that keep repeating SHA3 about 20 times until it passes without BSOD if it does then go back into the bios and reduce the offset to by 5 every time, so you would put -15, if you passed at -20 then you can go bigger on the offset by 5 to -25.

Hope you understand that but going to the other cores requires process lasso to lock AIDA64 benchmark thread to the core you are testing because selecting it yourself through processer affinity is too slow and AIDA64 never runs the same thread twice, so you need to keep selecting it and it never works, process lasso does, it locks it to the core selected within a second or less.
 
#3,382 · (Edited)
BlueShark said:
On Windbg you need to scroll up after pressing analyse to see what faulting core it is and remember that SMT if on counts one core as two, so core 0 and core 1 really is just core 0 and so forth, mine keeps saying faulting core 5 which really is core 2, remember as well that Ryzen master starts with core 01 which really is core 0 in the bios.

Going by what you said and taking that you have SMT on then your problem core is core 4.

Not fully understanding how you are testing there as it seems you are randomly selecting numbers for the CO (I may be wrong) if so, that’s not the way to do it as you will always be chasing your own tail.

Best to start by one core at time and work your way through, I did one CCD per day.

The quickest way was using SHA3 in single core, you can test this without process lasso just for core 0 by running AIDA64 then going to benchmark then press SHA3 then press parameters then select 1 CPU then select in your bios that core which would be core 0 and put -20 on it and all other cores should be on 0 no positive or negative offsets and run SHA3.

Once you do that keep repeating SHA3 about 20 times until it passes without BSOD if it does then go back into the bios and reduce the offset to by 5 every time, so you would put -15, if you passed at -20 then you can go bigger on the offset by 5 to -25.

Hope you understand that but going to the other cores requires process lasso to lock AIDA64 benchmark thread to the core you are testing because selecting it yourself through processer affinity is too slow and AIDA64 never runs the same thread twice, so you need to keep selecting it and it never works, process lasso does, it locks it to the core selected within a second or less.
Click to expand...
Thank you so much for the fast response! I am not just picking random numbers. I was starting at 0 on the cache CCD, and then I would work my way up by increasing the negative offset in intervals of -1 per run. I was doing this by increasing all cores on a CCD at the same time and then when I ran in to instability, decreasing them individually until it became stable. That didn't work though because changing the value of one core changed the stability of another core, making it pretty much impossible to find which core was causing the instability.

I am going to give your method a shot now. It seems very promising. So, just to clarify your instructions.

07/26/2023 Edit:

DON'T FOLLOW THESE INSTRUCTIONS!!! THEY ARE INCORRECT!

If you are looking for a surefire way to get curve optimizer values that PASS the AIDA64 SHA3 benchmark without crashing @BlueShark posted a write up with a confirmed working method a few replies down in this thread.

Preparation:
  1. Download Process Lasso
  2. Download WinDbg
  3. Set all curve optimizer values back to 0
  4. Grab snacks, drinks, Nintendo Switch/Steam Deck because this is gonna take a while.​
Step 1:
  1. Begin by setting a negative offset of -2 to -5 for core 0 on CCD0/Cache CCD.
  2. Leave all other cores at 0.
  3. Log in to windows and open AIDA 64.
  4. Run SHA3 cpu benchmark*.
  5. If it crashes, go back in to curve optimizer and ease back the offset a bit
  6. Repeat sub steps 3&4
  7. *If it passes 20x runs of SHA3 without failing, go on to step 2.
Step 2:
  1. Repeat everything in main Step 1 for the next core. (Maintain negative offset on previous core).
  2. If it crashes, hope to god there is a crash dump.
  3. Open WinDbg and open the crash dump. Run the Analyze command. Find which core failed by looking at line 4 of the crash dump.
  4. Go back to curve optimizer and ease back the offset for the core that failed.
  5. Keep repeating these steps until every core is able to pass SHA3 benchmark 20x
  6. Move on to main step 3 once you find a level of stability you deem acceptable
Step 3:
  1. Repeat main steps 1 & 2 for CCD 1/Frequency CCD (Maintain negative offset for CCD0)
  2. After many failed attempts and hard crashes, you'll eventually find stable offsets for both CCD's (probably not though). REJOICE!
  3. Hope to god your "stable" under volt/overclock doesn't cause random crashes later on and potentially cause data loss.
  4. Cry because you only managed to get an extra 1–2 FPS in games.
  5. Question your existence.
  6. Upgrade to a Zen 5 processor next year anyway because you lack self-control and make very poor financial decisions. Lol. (just so you know, I am talking about myself).

    That should about cover it. Let me know if I missed anything.
 
#3,384 · (Edited)
AIDA64 SHA3 is a useless test case. Don't get mad trying to stabilise it, there will never be a videogame or app that goes even near that kind of AVX2/AVX512 multicore workload. SHA3 in itself is a useful hash algorithm for data validation, but the specific AIDA implementation is particularly efficient in its ability to trigger a crash. The reference OpenSSL implementation is probably faster and does not show the same issue.
This said, I wish AMD or AIDA still fixes it at a certain point, and anyone is free to waste his time trying to find progressively better ways to make it run successfully.
In the meantime you can use OCCT to stress test your stability on more reasonable AVX2/AVX512 single-core or multi-core workloads.
 
Save
Reply Quote Rep+ Rep+
#3,385 · (Edited)
Alemancio said:
Anybody has a TL;DR around SHA3 Stability?
1. Why is it important? (for 100% gaming pcs)
2. What tool do you use to test for it?
Click to expand...
1. I don't claim to be an expert. In fact, I am definitely a n00b when it comes to all of this. However, from what I gathered from others, is that there is no instability that is good instability. The fact that you can make your system crash by running a simple benchmark probably isn't a good sign and could lead to problems down the road. It is up to you if you want to run your machine like that, but you should know that there is a measurable risk involved. No one can predict what will cause your specific hardware to crash and when, so by running your own benchmarks you can get a loose idea of what you might be able to expect in everyday usage. There are plenty of games that use AVX2 instructions, and in the future there could be a piece of software that affects your system similarly to SHA3. On top of that, by running your CPU in any configuration other than stock, there is no telling what the long term effects will be. There hasn't been any long term testing done on X3D chips running with a per core negative offset that happens to crash instantly on a certain benchmark. That underlying instability could potentially shorten the shelf life of your silicon quite a bit.

Testing for MAXIMUM stability is as good as you're going to get at mitigating those risks. The more stable your chip is or the closer to stock stability you can get, the better. My chip passes SHA3 at stock settings. As soon as I mess with Curve optimizer on the cache CCD it will cause that test to BSOD. It doesn't matter that it's just one test. The chip is simply not stable if it can't pass it.

2. Aida 64 Extreme (30 day trial period)

Even on runs where I pass SHA3, I have had random restarts and browser tabs and windows crashing and the system just generally acting strange. I am probably just going to call it a day and stop messing with curve optimizer because at this point it is way more frustrating than it is fun. I am on my 2nd chip and have already done more than 3 weeks of messing around, swapping components, motherboard, ram etc. trying to get different results. It is insanity.
edit: That's just me too. There are some people in this thread who have been through 4 or more CPU's and motherboards trying to get a "golden" sample but getting the same or worse results every time.
 
#3,386 ·
Burnt2myCore said:

Thank you so much for the fast response! I am not just picking random numbers. I was starting at 0 on the cache CCD, and then I would work my way up by increasing the negative offset in intervals of -1 per run. I was doing this by increasing all cores on a CCD at the same time and then when I ran in to instability, decreasing them individually until it became stable. That didn't work though because changing the value of one core changed the stability of another core, making it pretty much impossible to find which core was causing the instability.

I am going to give your method a shot now. It seems very promising. So, just to clarify your instructions.

Preparation:
  1. Download Process Lasso
  2. Download WinDbg
  3. Set all curve optimizer values back to 0
  4. Grab snacks, drinks, Nintendo Switch/Steam Deck because this is gonna take a while.​
Step 1:
  1. Begin by setting a negative offset of -2 to -5 for core 0 on CCD0/Cache CCD.
  2. Leave all other cores at 0.
  3. Log in to windows and open AIDA 64.
  4. Run SHA3 cpu benchmark*.
  5. If it crashes, go back in to curve optimizer and ease back the offset a bit
  6. Repeat sub steps 3&4
  7. *If it passes 20x runs of SHA3 without failing, go on to step 2.
Step 2:
  1. Repeat everything in main Step 1 for the next core. (Maintain negative offset on previous core).
  2. If it crashes, hope to god there is a crash dump.
  3. Open WinDbg and open the crash dump. Run the Analyze command. Find which core failed by looking at line 4 of the crash dump.
  4. Go back to curve optimizer and ease back the offset for the core that failed.
  5. Keep repeating these steps until every core is able to pass SHA3 benchmark 20x
  6. Move on to main step 3 once you find a level of stability you deem acceptable
Step 3:
  1. Repeat main steps 1 & 2 for CCD 1/Frequency CCD (Maintain negative offset for CCD0)
  2. After many failed attempts and hard crashes, you'll eventually find stable offsets for both CCD's (probably not though). REJOICE!
  3. Hope to god your "stable" under volt/overclock doesn't cause random crashes later on and potentially cause data loss.
  4. Cry because you only managed to get an extra 1–2 FPS in games.
  5. Question your existence.
  6. Upgrade to a Zen 5 processor next year anyway because you lack self-control and make very poor financial decisions. Lol. (just so you know, I am talking about myself).

    That should about cover it. Let me know if I missed anything.
Click to expand...
Not trying to be rude or put you down but I think that's a waste of time to be honest for minimal performance gains. 7950x3d from experience is already fast enough i'd spend time tuning ram timings on the new beta bios than mess with CO and have it reboot or freeze or bsod randomly while running a 24 hour video encoding task.
 
#3,387 · (Edited)
Burnt2myCore said:
1. I don't claim to be an expert. In fact, I am definitely a n00b when it comes to all of this. However, from what I gathered from others, is that there is no instability that is good instability. The fact that you can make your system crash by running a simple benchmark probably isn't a good sign and could lead to problems down the road. It is up to you if you want to run your machine like that, but you should know that there is a measurable risk involved. No one can predict what will cause your specific hardware to crash and when, so by running your own benchmarks you can get a loose idea of what you might be able to expect in everyday usage. There are plenty of games that use AVX2 instructions, and in the future there could be a piece of software that affects your system similarly to SHA3. On top of that, by running your CPU in any configuration other than stock, there is no telling what the long term effects will be. There hasn't been any long term testing done on X3D chips running with a per core negative offset that happens to crash instantly on a certain benchmark. That underlying instability could potentially shorten the shelf life of your silicon quite a bit.

Testing for MAXIMUM stability is as good as you're going to get at mitigating those risks. The more stable your chip is or the closer to stock stability you can get, the better. My chip passes SHA3 at stock settings. As soon as I mess with Curve optimizer on the cache CCD it will cause that test to BSOD. It doesn't matter that it's just one test. The chip is simply not stable if it can't pass it.
Click to expand...
Not really imho. Stress test developers make mistakes. There are a few known patterns to make a RAM stress test or CPU stress test fail and crash, look for example at the rowhammer attack for the RAM. But because those patterns are widely known they are not used anywhere and later may be patched by a microcode update. So it is possible that the devs of AIDA64 randomly found something similar, or unknowingly coded one of the known exploit, and this issue once analysed and understood will never be used in a normal game or app anyway.
The fact that a very specific user mode pattern of code is able to crash an operating system on a stock cpu is really bad, but it just means that the OS or the cpu firmware have to be updated to avoid an attack based on that pattern, and it doesn't mean that your PC is really unstable, or will be unstable at a certain time, with real apps or games.
 
Save
Reply Quote Rep+ Rep+
#3,388 ·
til5qoah said:
Not trying to be rude or put you down but I think that's a waste of time to be honest for minimal performance gains. 7950x3d from experience is already fast enough i'd spend time tuning ram timings on the new beta bios than mess with CO and have it reboot or freeze or bsod randomly while running a 24 hour video encoding task.
Click to expand...
Not being rude. Definitely have a point, lol. Way more gains to be had from RAM Overclocking. Just by putting in buildzoid timings, I got a 15% uplift in performance over my kits expo values. I am going down that rabbit hole right now. I am finding it a lot more rewarding and interesting than trying to OC my 7950x3d with the limited paths we have. All I have got from 3 weeks of messing around with curve optimizer is instability. It is super fast at stock settings anyways and I think real world gains from the 200mhz or so potential uplift you can get from PBO/CO are extremely limited. Why then do so many people seem so bent on doing it? Is it those irresponsible influencers posting -40 curve offsets and then running 1 loop of cinebench and putting the "STABLE MAX OC" stamp on it that is throwing n00bs like me off? Probably. I actually went around asking a bunch of different youtubers to run actual stability tests, and a couple of them came back to me and admitted they weren't stable.

I must continue to tinker though, in the name of science.
 
#3,389 ·
Burnt2myCore said:

Thank you so much for the fast response! I am not just picking random numbers. I was starting at 0 on the cache CCD, and then I would work my way up by increasing the negative offset in intervals of -1 per run. I was doing this by increasing all cores on a CCD at the same time and then when I ran in to instability, decreasing them individually until it became stable. That didn't work though because changing the value of one core changed the stability of another core, making it pretty much impossible to find which core was causing the instability.

I am going to give your method a shot now. It seems very promising. So, just to clarify your instructions.

Preparation:
  1. Download Process Lasso
  2. Download WinDbg
  3. Set all curve optimizer values back to 0
  4. Grab snacks, drinks, Nintendo Switch/Steam Deck because this is gonna take a while.​
Step 1:
  1. Begin by setting a negative offset of -2 to -5 for core 0 on CCD0/Cache CCD.
  2. Leave all other cores at 0.
  3. Log in to windows and open AIDA 64.
  4. Run SHA3 cpu benchmark*.
  5. If it crashes, go back in to curve optimizer and ease back the offset a bit
  6. Repeat sub steps 3&4
  7. *If it passes 20x runs of SHA3 without failing, go on to step 2.
Step 2:
  1. Repeat everything in main Step 1 for the next core. (Maintain negative offset on previous core).
  2. If it crashes, hope to god there is a crash dump.
  3. Open WinDbg and open the crash dump. Run the Analyze command. Find which core failed by looking at line 4 of the crash dump.
  4. Go back to curve optimizer and ease back the offset for the core that failed.
  5. Keep repeating these steps until every core is able to pass SHA3 benchmark 20x
  6. Move on to main step 3 once you find a level of stability you deem acceptable
Step 3:
  1. Repeat main steps 1 & 2 for CCD 1/Frequency CCD (Maintain negative offset for CCD0)
  2. After many failed attempts and hard crashes, you'll eventually find stable offsets for both CCD's (probably not though). REJOICE!
  3. Hope to god your "stable" under volt/overclock doesn't cause random crashes later on and potentially cause data loss.
  4. Cry because you only managed to get an extra 1–2 FPS in games.
  5. Question your existence.
  6. Upgrade to a Zen 5 processor next year anyway because you lack self-control and make very poor financial decisions. Lol. (just so you know, I am talking about myself).

    That should about cover it. Let me know if I missed anything.
Click to expand...
Just off to bed and I just read your post and made me laugh, keep it upđź‘Ť.

I will detail it more tomorrow, but you are on the right track as I been testing this CPU since launch SHA3 while you may not get it to run a full multi core, what this does makes your PC 100% stable, well for me anyway and that was across the 3 CPUS I had.

The only part that is wrong is once you find the right offset for that given core you are testing and then you move on to the next core, put the core that you have already done back to zero so that it does not influence the outcome of the stress on the core you are testing because process lasso sometimes will not catch the process that AIDA64 starts.

Will detail it tomorrow for you and yes, in step 3: part 5 Question your existence. That I did halfway through lol.
 
#3,390 ·
Hi all, I can run aida SHA3 with negative all core CO -5 on my 7950X3D stable, anything higher will randomly result in a bsod. The frequency is 4965 on 3D cache CCD and 5133 on normal CCD, resulting score 10839MB/s
Product Rectangle Azure Font Line

Stable CO for anything else that I'm using:
Product Rectangle Font Parallel Screenshot
 
#3,391 · (Edited)
BlueShark said:
Months of testing with 4 CPUS and a spare SSD with clean OS with no AMD drivers and just the same, then installing only through windows update to get basic non 3D cache drivers and still the same.

I think we need someone with a good working CPU with all CO negative offsets that pass SHA3 to show there clock speed when running SHA3 in real time, because I did see one guy who had massive offsets and had a very high score from SHA3 but is clock speed for the 3D cache was around 5000mhz but is non 3D cache CCD was boosting why higher, but all my CPU’S that I had always boost more on the 3D CCD, I think this maybe the problem.

Headache for sure.
Click to expand...
I can pass SHA3 with -20 all core
The avg effect clock is ~4800 during the test
Edit. it was a lazy run while working so lot's of things running on my pc now. I can try to make a clean run latter in safe mode. Raised HWinfo priority to high otherwise the benchmark lock all cpus and couldn't get any valid data from HWinfo
Font Screenshot Software Technology Electronic device
 
  • Helpful
Reactions: BlueShark
Save
Reply Quote Rep+ Rep+
#3,392 ·
Kelutrel said:
Not really imho. Stress test developers make mistakes. There are a few known patterns to make a RAM stress test or CPU stress test fail and crash, look for example at the rowhammer attack for the RAM. But because those patterns are widely known they are not used anywhere and later may be patched by a microcode update. So it is possible that the devs of AIDA64 randomly found something similar, or unknowingly coded one of the known exploit, and this issue once analysed and understood will never be used in a normal game or app anyway.
The fact that a very specific user mode pattern of code is able to crash an operating system on a stock cpu is really bad, but it just means that the OS or the cpu firmware have to be updated to avoid an attack based on that pattern, and it doesn't mean that your PC is really unstable, or will be unstable at a certain time, with real apps or games.
Click to expand...
I'd agree with you if all zen4 cpus were crashing and that's not the case.
 
#3,393 ·
Burnt2myCore said:

Thank you so much for the fast response! I am not just picking random numbers. I was starting at 0 on the cache CCD, and then I would work my way up by increasing the negative offset in intervals of -1 per run. I was doing this by increasing all cores on a CCD at the same time and then when I ran in to instability, decreasing them individually until it became stable. That didn't work though because changing the value of one core changed the stability of another core, making it pretty much impossible to find which core was causing the instability.

I am going to give your method a shot now. It seems very promising. So, just to clarify your instructions.

Preparation:
  1. Download Process Lasso
  2. Download WinDbg
  3. Set all curve optimizer values back to 0
  4. Grab snacks, drinks, Nintendo Switch/Steam Deck because this is gonna take a while.​
Step 1:
  1. Begin by setting a negative offset of -2 to -5 for core 0 on CCD0/Cache CCD.
  2. Leave all other cores at 0.
  3. Log in to windows and open AIDA 64.
  4. Run SHA3 cpu benchmark*.
  5. If it crashes, go back in to curve optimizer and ease back the offset a bit
  6. Repeat sub steps 3&4
  7. *If it passes 20x runs of SHA3 without failing, go on to step 2.
Step 2:
  1. Repeat everything in main Step 1 for the next core. (Maintain negative offset on previous core).
  2. If it crashes, hope to god there is a crash dump.
  3. Open WinDbg and open the crash dump. Run the Analyze command. Find which core failed by looking at line 4 of the crash dump.
  4. Go back to curve optimizer and ease back the offset for the core that failed.
  5. Keep repeating these steps until every core is able to pass SHA3 benchmark 20x
  6. Move on to main step 3 once you find a level of stability you deem acceptable
Step 3:
  1. Repeat main steps 1 & 2 for CCD 1/Frequency CCD (Maintain negative offset for CCD0)
  2. After many failed attempts and hard crashes, you'll eventually find stable offsets for both CCD's (probably not though). REJOICE!
  3. Hope to god your "stable" under volt/overclock doesn't cause random crashes later on and potentially cause data loss.
  4. Cry because you only managed to get an extra 1–2 FPS in games.
  5. Question your existence.
  6. Upgrade to a Zen 5 processor next year anyway because you lack self-control and make very poor financial decisions. Lol. (just so you know, I am talking about myself).

    That should about cover it. Let me know if I missed anything.
Click to expand...
Okay let me start, start process lasso and click option, CPU, pro balance, enable pro balance and make sure this is disabled, then go to options, general, configure startup, then select do not startup at login and have it launched with the GUI. This is to stop it overriding AMD drivers.

Goto power plan options in windows and select the balanced power plan and if Bit sum power plan is there delete it.

You will need Ryzen Master and expand the CCD sections, so it looks like this.
Font Screenshot Software Technology Terrestrial plant


Now to start,

1) Make sure that core boost in the bios is disabled and select your core that you are going to test and select a offset where you want to start, I started at -20 and if it BSOD then I then I went down in -5 so I then selected -15, make sure every other core is zero. Do this part in the bios only and not through software.
2) Boot into windows and start AIDA64 and Ryzen Master and process lasso and arrange that you can see Ryzen master CCD windows.
3) Go to AIDA64 and select tools and system stability test, tick only CPU and press start, you will see on Ryzen master all cores activated.
4) Now go and select this thread for AIDA64 here
Rectangle Slope Font Line Parallel

Now right click it and then go to CPU affinity and select always and select CPU affinity.
5) Now the select CPU affinity screen will be open and then deselect all cores by pressing Node 0 twice and then select CPU 0 and CPU 1, this will be your core 0 from in the bios and the two CPU 0 and CPU 1 are SMT threads, select Ok.
6) Now you should see Ryzen Master and Process lasso CPU readout change with Ryzen Master showing core 1 only with its core speed and Process lasso top right showing the two SMT threads, if so, then you are set.
7) Now stop the system stability test and close it.
8) Now go to SHA3 benchmark and select parameters and select 2 CPU, I select 2 cores to have a slightly more load and it shows on Ryzen Master if the process lasso locks the thread to that core, there maybe a bit of a delay and sometimes it may miss it, but just wait for SHA3 to finish and press it again and it will get it right.
9) Keep pressing it at least 15 times or 20 if you are enjoying yourself.
10) That’s it, if you stable at any given offset say -20 and you passed 15 times or more then go back to the bios and increase by -5 if not then reduce by -5 and repeat , by opening all the above software and just go to AIDA64 SHA3 and start again as process lasso will remember to lock AIDA64 thread to the core you are testing.
11) When you are finished with a core then write down the offset and then repeat from steps 2, on step 3 you will now not see all cores activated but the last core you tested, on step 5 just deselect your CPU 0 and CPU 1 and select CPU 2 and CPU 3 and so forth.
When you are opening AIDA64 and clicking SHA3 you will need to always select parameters and 2 CPU.

That is it, once you get into the flow of it, it becomes easy to remember it, just try not to fall asleep lol.

Once you get to the end of it all, then check the offsets for your best cores to see if they are not too large, best way to test this would be OCCT and select stability test and CPU and select Thread settings and select fixed AVX512 and run that to see if you stable on you best cores, if not then reduce until you are.

Hope this helps and if you need and more information just let me know and good luck and keep us posted how you get on.
 
#3,394 ·
alexcheveau said:
I can pass SHA3 with -20 all core
The avg effect clock is ~4800 during the test
Edit. it was a lazy run while working so lot's of things running on my pc now. I can try to make a clean run latter in safe mode. Raised HWinfo priority to high otherwise the benchmark lock all cpus and couldn't get any valid data from HWinfo
View attachment 2622188
Click to expand...
Cheers mate, yes, when you get time I will be interested what clock speed you get as I did have a 7800X3D for testing but great timing my motherboard broke down so I sent it back and I always wondered if the core speed on SHA3 would be lower than 5050Mhz on a 7800X3D.
 
#3,395 ·
BlueShark said:
Okay let me start, start process lasso and click option, CPU, pro balance, enable pro balance and make sure this is disabled, then go to options, general, configure startup, then select do not startup at login and have it launched with the GUI. This is to stop it overriding AMD drivers.

Goto power plan options in windows and select the balanced power plan and if Bit sum power plan is there delete it.

You will need Ryzen Master and expand the CCD sections, so it looks like this.
View attachment 2622202

Now to start,

1) Make sure that core boost in the bios is disabled and select your core that you are going to test and select a offset where you want to start, I started at -20 and if it BSOD then I then I went down in -5 so I then selected -15, make sure every other core is zero. Do this part in the bios only and not through software.
2) Boot into windows and start AIDA64 and Ryzen Master and process lasso and arrange that you can see Ryzen master CCD windows.
3) Go to AIDA64 and select tools and system stability test, tick only CPU and press start, you will see on Ryzen master all cores activated.
4) Now go and select this thread for AIDA64 here
View attachment 2622203
Now right click it and then go to CPU affinity and select always and select CPU affinity.
5) Now the select CPU affinity screen will be open and then deselect all cores by pressing Node 0 twice and then select CPU 0 and CPU 1, this will be your core 0 from in the bios and the two CPU 0 and CPU 1 are SMT threads, select Ok.
6) Now you should see Ryzen Master and Process lasso CPU readout change with Ryzen Master showing core 1 only with its core speed and Process lasso top right showing the two SMT threads, if so, then you are set.
7) Now stop the system stability test and close it.
8) Now go to SHA3 benchmark and select parameters and select 2 CPU, I select 2 cores to have a slightly more load and it shows on Ryzen Master if the process lasso locks the thread to that core, there maybe a bit of a delay and sometimes it may miss it, but just wait for SHA3 to finish and press it again and it will get it right.
9) Keep pressing it at least 15 times or 20 if you are enjoying yourself.
10) That’s it, if you stable at any given offset say -20 and you passed 15 times or more then go back to the bios and increase by -5 if not then reduce by -5 and repeat , by opening all the above software and just go to AIDA64 SHA3 and start again as process lasso will remember to lock AIDA64 thread to the core you are testing.
11) When you are finished with a core then write down the offset and then repeat from steps 2, on step 3 you will now not see all cores activated but the last core you tested, on step 5 just deselect your CPU 0 and CPU 1 and select CPU 2 and CPU 3 and so forth.
When you are opening AIDA64 and clicking SHA3 you will need to always select parameters and 2 CPU.

That is it, once you get into the flow of it, it becomes easy to remember it, just try not to fall asleep lol.

Once you get to the end of it all, then check the offsets for your best cores to see if they are not too large, best way to test this would be OCCT and select stability test and CPU and select Thread settings and select fixed AVX512 and run that to see if you stable on you best cores, if not then reduce until you are.

Hope this helps and if you need and more information just let me know and good luck and keep us posted how you get on.
Click to expand...
That is quite the write-up. I appreciate you, man. Thanks. I just have a couple of questions.
When you say to make sure Core Boost is disabled in BIOS are you talking about the PBO setting Core Boost Override where you can set a positive or negative core boosting target? Also, I am a little curious as to why we are using process lasso here. I understand we are trying to isolate only the threads that we are trying to find stable Curve Optimizer values for each run, but is the end result actually any different from just going through each core individually and running SHA 3 without process lasso?
 
#3,396 · (Edited)
BlueShark said:
Okay let me start, start process lasso and click option, CPU, pro balance, enable pro balance and make sure this is disabled, then go to options, general, configure startup, then select do not startup at login and have it launched with the GUI. This is to stop it overriding AMD drivers.

Goto power plan options in windows and select the balanced power plan and if Bit sum power plan is there delete it.

You will need Ryzen Master and expand the CCD sections, so it looks like this.
View attachment 2622202

Now to start,

1) Make sure that core boost in the bios is disabled and select your core that you are going to test and select a offset where you want to start, I started at -20 and if it BSOD then I then I went down in -5 so I then selected -15, make sure every other core is zero. Do this part in the bios only and not through software.
2) Boot into windows and start AIDA64 and Ryzen Master and process lasso and arrange that you can see Ryzen master CCD windows.
3) Go to AIDA64 and select tools and system stability test, tick only CPU and press start, you will see on Ryzen master all cores activated.
4) Now go and select this thread for AIDA64 here
View attachment 2622203
Now right click it and then go to CPU affinity and select always and select CPU affinity.
5) Now the select CPU affinity screen will be open and then deselect all cores by pressing Node 0 twice and then select CPU 0 and CPU 1, this will be your core 0 from in the bios and the two CPU 0 and CPU 1 are SMT threads, select Ok.
6) Now you should see Ryzen Master and Process lasso CPU readout change with Ryzen Master showing core 1 only with its core speed and Process lasso top right showing the two SMT threads, if so, then you are set.
7) Now stop the system stability test and close it.
8) Now go to SHA3 benchmark and select parameters and select 2 CPU, I select 2 cores to have a slightly more load and it shows on Ryzen Master if the process lasso locks the thread to that core, there maybe a bit of a delay and sometimes it may miss it, but just wait for SHA3 to finish and press it again and it will get it right.
9) Keep pressing it at least 15 times or 20 if you are enjoying yourself.
10) That’s it, if you stable at any given offset say -20 and you passed 15 times or more then go back to the bios and increase by -5 if not then reduce by -5 and repeat , by opening all the above software and just go to AIDA64 SHA3 and start again as process lasso will remember to lock AIDA64 thread to the core you are testing.
11) When you are finished with a core then write down the offset and then repeat from steps 2, on step 3 you will now not see all cores activated but the last core you tested, on step 5 just deselect your CPU 0 and CPU 1 and select CPU 2 and CPU 3 and so forth.
When you are opening AIDA64 and clicking SHA3 you will need to always select parameters and 2 CPU.

That is it, once you get into the flow of it, it becomes easy to remember it, just try not to fall asleep lol.

Once you get to the end of it all, then check the offsets for your best cores to see if they are not too large, best way to test this would be OCCT and select stability test and CPU and select Thread settings and select fixed AVX512 and run that to see if you stable on you best cores, if not then reduce until you are.

Hope this helps and if you need and more information just let me know and good luck and keep us posted how you get on.
Click to expand...
Yo... So, I followed your instructions to the letter. I saw the PBO power limit values in ryzen master redline and then the program crashed. Ryzen master won't open now. It is telling me to reinstall it lol. Gonna reboot my computer and see what happens.

Edit: After restarting, Ryzen Master will open, but Aida 64 is now bugged, and I can't edit paramaters on benchmarks. I am getting the error Access of violation at address 00495584 in module 'aida64.exe'

Edit 2: It seems process lasso is causing some kind of conflict with Aida 64. I was able to get access back by setting affinity back to all 32 threads.

Edit 3: I repeated the steps and it now seems to be working as intended. No idea what happened the first time but it's all good now.
 
#3,397 ·
Burnt2myCore said:
That is quite the write-up. I appreciate you, man
Click to expand...
Not a problem, thanks.
Burnt2myCore said:
When you say to make sure Core Boost is disabled in BIOS are you talking about the PBO setting Core Boost Override where you can set a positive or negative core boosting target?
Click to expand...
Spot on.
Burnt2myCore said:
Also, I am a little curious as to why we are using process lasso here.
Click to expand...
We are using process lasso because it automatically keeps the target thread on the core we need as there is no other way to do this because all what AIDA64 can do is run on core 0 and not the other cores in order, plus if you get a BSOD then you can just run the program and it will restore what you selected.
Burnt2myCore said:
Edit 3: I repeated the steps and it now seems to be working as intended. No idea what happened the first time but it's all good now.
Click to expand...
I think you may of selected more threads or selected no CPU threads on the selection screen. But glad you got it all working, will be interested how you get on.
 
#3,398 ·
til5qoah said:
Not trying to be rude or put you down but I think that's a waste of time to be honest for minimal performance gains. 7950x3d from experience is already fast enough i'd spend time tuning ram timings on the new beta bios than mess with CO and have it reboot or freeze or bsod randomly while running a 24 hour video encoding task.
Click to expand...
CO on a good chip can get you around 10c less
 
#3,399 · (Edited)
CCD1 is parking now, but I don't know exactly what I did to make it work. LOL. I think it's a BIOS ACPI setting somewhere.

It parks with PBO/CO on plus boost. But I can't recreate Resource Meter and Cinebench running like a game, just like what the AMD guide says. It also doesn't use the non-vcache CCD when running Cinebench single core, which results in a low score.

It's a pain trying to make this CPU work. I hope AMD is working on a Windows installation app that will do everything for you.

I also wish AMD will make a Windows app that will let you prioritize either CCD0 or CCD1 on the fly, without rebooting and changing a BIOS setting. I used Process Lasso with my 5950x ( without core parking) and I know that's what it does and more. I will definitely look at it again.
 
#3,400 ·
BlueShark said:
Okay let me start, start process lasso and click option, CPU, pro balance, enable pro balance and make sure this is disabled, then go to options, general, configure startup, then select do not startup at login and have it launched with the GUI. This is to stop it overriding AMD drivers.

Goto power plan options in windows and select the balanced power plan and if Bit sum power plan is there delete it.

You will need Ryzen Master and expand the CCD sections, so it looks like this.
View attachment 2622202

Now to start,

1) Make sure that core boost in the bios is disabled and select your core that you are going to test and select a offset where you want to start, I started at -20 and if it BSOD then I then I went down in -5 so I then selected -15, make sure every other core is zero. Do this part in the bios only and not through software.
2) Boot into windows and start AIDA64 and Ryzen Master and process lasso and arrange that you can see Ryzen master CCD windows.
3) Go to AIDA64 and select tools and system stability test, tick only CPU and press start, you will see on Ryzen master all cores activated.
4) Now go and select this thread for AIDA64 here
View attachment 2622203
Now right click it and then go to CPU affinity and select always and select CPU affinity.
5) Now the select CPU affinity screen will be open and then deselect all cores by pressing Node 0 twice and then select CPU 0 and CPU 1, this will be your core 0 from in the bios and the two CPU 0 and CPU 1 are SMT threads, select Ok.
6) Now you should see Ryzen Master and Process lasso CPU readout change with Ryzen Master showing core 1 only with its core speed and Process lasso top right showing the two SMT threads, if so, then you are set.
7) Now stop the system stability test and close it.
8) Now go to SHA3 benchmark and select parameters and select 2 CPU, I select 2 cores to have a slightly more load and it shows on Ryzen Master if the process lasso locks the thread to that core, there maybe a bit of a delay and sometimes it may miss it, but just wait for SHA3 to finish and press it again and it will get it right.
9) Keep pressing it at least 15 times or 20 if you are enjoying yourself.
10) That’s it, if you stable at any given offset say -20 and you passed 15 times or more then go back to the bios and increase by -5 if not then reduce by -5 and repeat , by opening all the above software and just go to AIDA64 SHA3 and start again as process lasso will remember to lock AIDA64 thread to the core you are testing.
11) When you are finished with a core then write down the offset and then repeat from steps 2, on step 3 you will now not see all cores activated but the last core you tested, on step 5 just deselect your CPU 0 and CPU 1 and select CPU 2 and CPU 3 and so forth.
When you are opening AIDA64 and clicking SHA3 you will need to always select parameters and 2 CPU.

That is it, once you get into the flow of it, it becomes easy to remember it, just try not to fall asleep lol.

Once you get to the end of it all, then check the offsets for your best cores to see if they are not too large, best way to test this would be OCCT and select stability test and CPU and select Thread settings and select fixed AVX512 and run that to see if you stable on you best cores, if not then reduce until you are.

Hope this helps and if you need and more information just let me know and good luck and keep us posted how you get on.
Click to expand...
I got my CO values using Hydra. It's a good 12 hours of diagnostics but it's automated and gives you a good starting point. I can pass SHA3 with PBO/CO but I have to limit my boost, otherwise it hangs.
 
3,381 - 3,400 of 9,076 Posts
About this Discussion
9.1K Replies
540 Participants
Last post:
MarkPost
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling