Question BOTTOM 1% 5900x CPU Profile Score?

[the7goonies]

When I run the CPU Profile score on 3DMark, I get BOTTOM 1%, which I know isn't good haha.
Not only that, but I also get a really poor Time Spy score (I'm guessing because my cpu score is so low).

I'm not into specifics when it comes to PCs, although I did build this one (and 4 others) but they are all cookie-cutter builds. I have an older 7700k 3080 pc that runs just as well, if not better than this (as far as CPU goes). So in reality, I have no idea what to look for other than the basics.

User Benchmark: https://www.userbenchmark.com/UserRun/55471960
CPU Profile 3DMark Test: https://www.3dmark.com/cpu/728343
Time Spy 3DMark Test: https://www.3dmark.com/spy/30839362

CPU-Z CPU Tab: https://pasteboard.co/hasLgJ6mLxMD.png
CPU-Z Memory Tab: https://pasteboard.co/BQEuDqwas9yu.png
CPU-Z SPD Tab: https://pasteboard.co/yLfewQF0ProW.png

Specs:
ASUS ROG Crosshair VIII Dark Hero
AMD Ryzen 9 5900X 12-Core Processor
EVGA 3090 FTW3 Ultra Graming
iCUE H150i ELITE CAPELLIX Liquid CPU Cooler
Corsair CMH32GX4M2D3600C18 4x16GB (I use 64gb because rendering)
Samsung 980 Pro NVMe PCIe M.2 1TB

-Shows running mem frequency as 3200mhz ddr4
-AGESA Version: ComboPIv2_1200
-BIOS Version is default from 1/2021, I'm afraid to update it (could this be the reason?)

--
Target CPU Speed: 3700mhz
Target DRAM Frequency: 3200mhz
Target FCLK Frequency: 1600mhz

CPU Frequency: 3700mhz
Temp: 48C
Core Voltage: 1.481v
Memory Frequency: 3200mhz
Voltage: 1.344v
--

I have tried looking up specifics about RAM and whatnot, but it's all fairly confusing to me from an overview standpoint.

Any help is appreciated!

 

[drea.drechsler]

You may not like to hear it, but you should have saved your money and bought a different motherboard. That Dark Hero is a massive beast with a boatload of awesome features that are only useful for extreme overclocking, especially on LN2. Your post does not sound like you are one of those. You'd probably be just as well served with a B550 Strix-F or similar and could have used the money to go towards a 5950X, something that would help a lot with your rendering tasks.

First thing first: update BIOS to the latest for your motherboard. That's pretty important as updates have fixed a lot of issues.

Would be worthwhile running a CPU specific benchmark...Cinebench R20 is very good for this. Check scores on this web site to get an idea how it compares.

Things to look out for: disable motherboard utilities if you can and Corsair's iCUE software. Many of those are drags on performance; it's really best not to install them if you can. Also check performance with BIOS settings in defaults...find a setting to "load optimized defaults" in BIOS and do it. Either that or reset CMOS or maybe even both. Then set XMP to enabled, but that's all, to set your memory to run optimally.

If you are on the latest BIOS it will probably enable UEFI mode, the fTPM and Secure Boot by default so you don't need to mess with that.

 

[the7goonies]

You may not like to hear it, but you should have saved your money and bought a different motherboard. That Dark Hero is a massive beast with a boatload of awesome features that are only useful for extreme overclocking, especially on LN2. Your post does not sound like you are one of those. You'd probably be just as well served with a B550 Strix-F or similar and could have used the money to go towards a 5950X, something that would help a lot with your rendering tasks.

First thing first: update BIOS to the latest for your motherboard. That's pretty important as updates have fixed a lot of issues.

Would be worthwhile running a CPU specific benchmark...Cinebench R20 is very good for this. Check scores on this web site to get an idea how it compares.

Things to look out for: disable motherboard utilities if you can and Corsair's iCUE software. Many of those are drags on performance; it's really best not to install them if you can. Also check performance with BIOS settings in defaults...find a setting to "load optimized defaults" in BIOS and do it. Either that or reset CMOS or maybe even both. Then set XMP to enabled, but that's all, to set your memory to run optimally.

If you are on the latest BIOS it will probably enable UEFI mode, the fTPM and Secure Boot by default so you don't need to mess with that.

Hm okay - at the time I bought the mobo it was on sale at microcenter, so I figured it was worth the take. I also have a lot of peripherals so I figured the USB count was worth it I actually don't have iCUE nor any of that armory crate crap. I'll go ahead and update the BIOS and report back. I'll reset default settings and make sure XMP is enabled (it's currently enabled). I have no idea what fTPM or Secure Boot is so I'll have to make sure about that once the update happens. Even with the motherboard being a beast for overclocking, you'd think it could handle normal build?

 

[drea.drechsler]

Hm okay - at the time I bought the mobo it was on sale at microcenter, so I figured it was worth the take. I also have a lot of peripherals so I figured the USB count was worth it I actually don't have iCUE nor any of that armory crate crap. I'll go ahead and update the BIOS and report back. I'll reset default settings and make sure XMP is enabled (it's currently enabled). I have no idea what fTPM or Secure Boot is so I'll have to make sure about that once the update happens. Even with the motherboard being a beast for overclocking, you'd think it could handle normal build?

fTPM and Secure Boot are security features "required" for Windows 11. They are pretty good to have...secure boot along with the way UEFI works protects against the root kits that caused so many headaches several years ago. fTPM will securely store passwords and keys...like the keys to access a bitlocker protected drive. Some games require an fTPM for anti-cheat protection. Microsoft wants them enabled by default in BIOS so mfr's have been doing it. But there have been other fixes, like USB drop-out, and compatibility improvements in later BIOS's.

Yes, it can definitely handle a normal build LOL. It's just overkill in the extreme.

 

[the7goonies]

fTPM and Secure Boot are security features "required" for Windows 11. They are pretty good to have...secure boot along with the way UEFI works protects against the root kits that caused so many headaches several years ago. fTPM will securely store passwords and keys...like the keys to access a bitlocker protected drive. Some games require an fTPM for anti-cheat protection. Microsoft wants them enabled by default in BIOS so mfr's have been doing it. But there have been other fixes, like USB drop-out, and compatibility improvements in later BIOS's.

Yes, it can definitely handle a normal build LOL. It's just overkill in the extreme.

So I ran the Cinemark test and compared it to the scores shown on the website:
Website Scores: Single-1670 Multi-22046
My Scores: Single-1544 Multi-20025

I'm going to update the BIOS now, then I'll load optimized defaults and change the XMP. And I'll post those updated scores.

 

[drea.drechsler]

So I ran the Cinemark test and compared it to the scores shown on the website:
Website Scores: Single-1670 Multi-22046
My Scores: Single-1544 Multi-20025
...

Those scores are low but it at least they are "in range". Hopefully getting the BIOS updated with optimized defaults will help close the gap.

 

[Roland Of Gilead]

Your ram is running at 3200mhz, when it's 3600mhz rated! That is extra performance lying right there. Get your mem clock/infinity fabric/mem controller running at 1:1:1. Will give better performance.

 

[the7goonies]

Those scores are low but it at least they are "in range". Hopefully getting the BIOS updated with optimized defaults will help close the gap.

BIOS has been updated successfully.
I went in and set everything to default, then proceeded with the DRAM changes:
AI Overclock Tuner [Auto]->[DOCP Standard]
Memory Frequency [Auto]->[DDR4-3600mhz]
There were some other changes that it automatically did when I changed it to DOCP, such as trcdrd, dram ras, dram voltage, etc.

Ran the Cinebench test again:
My Results: Single-1527 Multi-20072 (technically better than before, but not by much.. and nowhere near the 22k mark)

CPU Profile 3DMark Test: https://pasteboard.co/CKeDpiak0iaD.png
TimeSpy Score: https://www.3dmark.com/spy/30858384

You're ram is running at 3200mhz, when it's 3600mhz rated! That is extra performance lying right there. Get your mem clock/infinity fabric/mem controller running at 1:1:1. Will give better peformance.

Yup it has been updated with the updated BIOS. I'll have to look up what infinity fabric / mem controller are because I have no idea lol.

 

[drea.drechsler]

BIOS has been updated successfully.
I went in and set everything to default, then proceeded with the DRAM changes:
AI Overclock Tuner [Auto]->[DOCP Standard]
Memory Frequency [Auto]->[DDR4-3600mhz]
There were some other changes that it automatically did when I changed it to DOCP, such as trcdrd, dram ras, dram voltage, etc.

Ran the Cinebench test again:
My Results: Single-1527 Multi-20072 (technically better than before, but not by much.. and nowhere near the 22k mark)

CPU Profile 3DMark Test: https://pasteboard.co/CKeDpiak0iaD.png
TimeSpy Score: https://www.3dmark.com/spy/30858384



Yup it has been updated with the updated BIOS. I'll have to look up what infinity fabric / mem controller are because I have no idea lol.

I'd expect a better CB score too. What temperature is your CPU running during a Cinebench run? I suggest getting HWINfo64 to check that since it's the best monitoring tool for Ryzen.

Also check voltages, the SVI2 TFN CPU core voltage is the one to look for.

If temps are running OK something to do is go in and enable PBO so you can go into Core Optimizer and set all cores to a negative curve with a value of -10. That's really simple and should be stable but you should also see an improvement in scores.

If you like to tweak, you could find the optimal curve for each core. Most of them should run down to a -25 or -30 value, but some probably won't and remain stable.

What this does is undervolt the cores at high boosts so the core runs cooler. You won't likely see a lower temperature but the algorithm can hold a high boost clock longer before it gets to a temp where it has to lower it with a result being better performance.

 

[Roland Of Gilead]

BIOS has been updated successfully.
I went in and set everything to default, then proceeded with the DRAM changes:
AI Overclock Tuner [Auto]->[DOCP Standard]
Memory Frequency [Auto]->[DDR4-3600mhz]
There were some other changes that it automatically did when I changed it to DOCP, such as trcdrd, dram ras, dram voltage, etc.

Ran the Cinebench test again:
My Results: Single-1527 Multi-20072 (technically better than before, but not by much.. and nowhere near the 22k mark)

CPU Profile 3DMark Test: https://pasteboard.co/CKeDpiak0iaD.png
TimeSpy Score: https://www.3dmark.com/spy/30858384



Yup it has been updated with the updated BIOS. I'll have to look up what infinity fabric / mem controller are because I have no idea lol.

You can change the settings in the Asus AI Tweaker tab in the bios. Looks like this:

https://imgur.com/VYMr3Xg

View: https://imgur.com/VYMr3Xg

 

[the7goonies]

I'd expect a better CB score too. What temperature is your CPU running during a Cinebench run? I suggest getting HWINfo64 to check that since it's the best monitoring tool for Ryzen.

Also check voltages, the SVI2 TFN CPU core voltage is the one to look for.

If temps are running OK something to do is go in and enable PBO so you can go into Core Optimizer and set all cores to a negative curve with a value of -10. That's really simple and should be stable but you should also see an improvement in scores.

If you like to tweak, you could find the optimal curve for each core. Most of them should run down to a -25 or -30 value, but some probably won't and remain stable.

What this does is undervolt the cores at high boosts so the core runs cooler. You won't likely see a lower temperature but the algorithm can hold a high boost clock longer before it gets to a temp where it has to lower it with a result being better performance.

Here are the HWInfo64 temps/voltages during the test (about 5 minutes into the test, so halfway):
https://pasteboard.co/8rkzU1XplVct.png

I'm not sure what I'm looking for here as far as good temps/voltages. I've heard of PBO before but never looked into it.

You can change the settings in the Asus AI Tweaker tab in the bios. Looks like this:

Yup I've changed the DRAM settings there to match the DOCP and the 3600mhz, but still no improvement.

 

[drea.drechsler]

Here are the HWInfo64 temps/voltages during the test (about 5 minutes into the test, so halfway):
https://pasteboard.co/8rkzU1XplVct.png

I'm not sure what I'm looking for here as far as good temps/voltages. I've heard of PBO before but never looked into it.



Yup I've changed the DRAM settings there to match the DOCP and the 3600mhz, but still no improvement.

About half way into which test? A Cinebench20 test shouldn't take your processor more than a minute...not even that long. Don't run the extended test only the single pass test when looking for a comparison score. And when you're running it don't do anything else and have nothing else running...not even HWinfo if you're looking for high scores.

But I'm still wondering just what test if that HWInfo64 screen was taken in the middle of it. 70C max (Tdie/Tctrl) is pretty good temp for a 5900X.

 

[the7goonies]

About half way into which test? A Cinebench20 test shouldn't take your processor more than a minute...not even that long. Don't run the extended test only the single pass test when looking for a comparison score. And when you're running it don't do anything else and have nothing else running...not even HWinfo if you're looking for high scores.

But I'm still wondering just what test if that HWInfo64 screen was taken in the middle of it. 70C max (Tdie/Tctrl) is pretty good temp for a 5900X.

Sorry, I'm on the Cinebench23 test and I was on the "multi-core" test that runs for 10 minutes (each core test runs 10 minutes apparently) so about halfway is when I captured it. If I need to download the Cinebench20 instead of the 23, I'll do that. I ran the HWinfo during the test show you could see temps (as you requested). My previous tests I make sure nothing else on the computer is running.

Update: I went ahead and got the 20 and the results are:
My Results - 7824pts
Website Results - 8545pts

 

[drea.drechsler]

...If I need to download the Cinebench20 instead of the 23, I'll do that. ...

You don't need to...just make sure you're comparing CB23 scores. Also, run the single pass to compare. The multi-pass will help assess cooling system performance but the comparisons for CPU performance are generally based only the single pass score. All cores (24 threads, one for each physical and virtual core with a 5900X) are running in the single pass. It shouldn't take that long.

Because there is much greater cooling system heating a long, multipass test that lasts 10 min's will be lower than a short, single pass test.

 

[the7goonies]

You don't need to...just make sure you're comparing CB23 scores. Also, run the single pass to compare. The multi-pass will help assess cooling system performance but the comparisons for CPU performance are generally based only the single pass score. All cores (24 threads, one for each physical and virtual core with a 5900X) are running in the single pass. It shouldn't take that long.

Because there is much greater cooling system heating a long, multipass test that lasts 10 min's will be lower than a short, single pass test.

Yup my previous results were based on that (I did both single and multi for the C23).

For C23:
My Results: Single-1527 Multi-20072
Website Scores: Single-1670 Multi-22046

For C20:
My Results - 7824pts
Website Results - 8545pts

 

[drea.drechsler]

Yup my previous results were based on that (I did both single and multi for the C23).

For C23:
My Results: Single-1527 Multi-20072
Website Scores: Single-1670 Multi-22046

For C20:
My Results - 7824pts
Website Results - 8545pts

I think we're misunderstanding...

By single pass I mean rendering the scene one time only with all cores/threads, not rendering the scene with only one single core.

The test can be run for 10 min's, rendering the scene multiple times if you're running it with all cores/threads.

 

[the7goonies]

I think we're misunderstanding...

By single pass I mean rendering the scene one time only with all cores/threads, not rendering the scene with only one single core.

The test can be run for 10 min's, rendering the scene multiple times if you're running it with all cores/threads.

Gotcha okay, C20 is single pass (completing in seconds) and those scores remain the same.
My Results - 7824pts
Website Results - 8545pts

 

[drea.drechsler]

Gotcha okay, C20 is single pass (completing in seconds) and those scores remain the same.
My Results - 7824pts
Website Results - 8545pts

Well, at least your scores are in-range even if a bit low. The web-site scores are probably the result of some optimization but I can't believe they're the result of extensive tweaking. My 5800X scores are considerably higher than those listed but I have tweaked my PBO, Curve Optimizer and memory timings quite a bit.

I can't blame your cooling, BTW. Your posted HWInfo screen (supposedly taken in the middle of a Cinebench run) showed CPU temps of 70C max and that's quite good. But it also suggest there's room for optimizing performance with PBO and Curve Optimizer..

 

[Karadjgne]

Those website scores are almost certainly run with PBO enabled. The 5900x can actually hit its TDP ceiling, and running stock can go beyond those limits if cooling allows. It's technically not an overclock, enabling PBO just raises the limits of what is allowed by AMD, which are only in place for eco reasons. Set it free.

You'll find PBO in the AMD Overclocking selection, bottom of Advanced tab, enable PBO, set the core boost to 200. Leave the other choices as is.

Also, in the main page, look for FClock. Change that from Auto to 1800MHz, since that's the data rate of 3600MHz DDR ram.

 

[the7goonies]

Well, at least your scores are in-range even if a bit low. The web-site scores are probably the result of some optimization but I can't believe they're the result of extensive tweaking. My 5800X scores are considerably higher than those listed but I have tweaked my PBO, Curve Optimizer and memory timings quite a bit.

I can't blame your cooling, BTW. Your posted HWInfo screen (supposedly taken in the middle of a Cinebench run) showed CPU temps of 70C max and that's quite good. But it also suggest there's room for optimizing performance with PBO and Curve Optimizer..

Those website scores are almost certainly run with PBO enabled. The 5900x can actually hit its TDP ceiling, and running stock can go beyond those limits if cooling allows. It's technically not an overclock, enabling PBO just raises the limits of what is allowed by AMD, which are only in place for eco reasons. Set it free.

You'll find PBO in the AMD Overclocking selection, bottom of Advanced tab, enable PBO, set the core boost to 200. Leave the other choices as is.

Also, in the main page, look for FClock. Change that from Auto to 1800MHz, since that's the data rate of 3600MHz DDR ram.

I'll have to look into PBO and Curve Optimizer then, is it appropriate to use the BIOS only, or should I try something like Ryzen Master?

EDIT: I decided to get ryzen master just to see what I could do with PBO in there, set PPT to 200, TDC 175, EDC 180 and already achieved better results on the C20 benchmark from 7824 to 8360. The only concern is that the CPU reached around 81C, but I believe that's still stable. Also, any idea why my voltage is so high? I'm reading my average voltage is always above 1.2?

UPDATE: I decided to run the C23 test (the one that goes for 10 minutes) and achieved another better result from the previous C23 test. The average CPU temp was around 82C after 10 minutes of load. The CPUs seem to max out around 4.4ghz when running all of the tests. I read a few threads in regards to doing everything in the BIOS but it's a lot to take in lol.

 

[drea.drechsler]

I'll have to look into PBO and Curve Optimizer then, is it appropriate to use the BIOS only, or should I try something like Ryzen Master?

EDIT: I decided to get ryzen master just to see what I could do with PBO in there, set PPT to 200, TDC 175, EDC 180 and already achieved better results on the C20 benchmark from 7824 to 8360. The only concern is that the CPU reached around 81C, but I believe that's still stable. Also, any idea why my voltage is so high? I'm reading my average voltage is always above 1.2?

UPDATE: I decided to run the C23 test (the one that goes for 10 minutes) and achieved another better result from the previous C23 test. The average CPU temp was around 82C after 10 minutes of load. The CPUs seem to max out around 4.4ghz when running all of the tests. I read a few threads in regards to doing everything in the BIOS but it's a lot to take in lol.

Temps in the 80's isn't unexpected. In fact, it's normal and expected to reach as high as 90C in heavy all-core workloads. The CPU is designed for it.

The CPU will hit voltages as high as 1.5V in light bursty workloads (maybe even a bit more if it's 'chill', as after first boot-up). Bouncing around between 1.2-1.3 in heavy all-core workloads (like CineBench) is normal.

The processor works very much different from Intel CPU's you may be used to. It's very dynamic, constantly boosting a single core to max clocks, and over if well cooled, during light work. It needs the high voltage (1.5V) to boost, it's designed for it.

When a core is completely idle it's put into a C6, deep sleep state. In that state it's completely off and voltage is at zero. When a core goes into C6 it reports it's last voltage before dropping into the state since reporting that it's sleeping would wake it up at which point it wouldn't be in C6 deep sleep anymore. So it rarely reports anything less than about 1.1V or so.

As far as using Ryzenmaster goes: it's definitely a useful tool but is intrusive and can hurt performance. Some have even found it makes their system unstable when gaming. That's OK since it's intended purpose is for extreme overclocking demonstrations and not 24/7 use. Once you've figured out a decent setup I'd uninstall it (to remove the service) and use BIOS only to make the settings, for stability.

As far as PBO setups go: I'd suggest checking out some youtube videos on the process for 5900's. Curve Optimizer is very effective but it can be a bit tedious to get it right.

 

[the7goonies]

The processor works very much different from Intel CPU's you may be used to. It's very dynamic, constantly boosting a single core to max clocks, and over if well cooled, during light work. It needs the high voltage (1.5V) to boost, it's designed for it.

Hit the nail on the head there - was about to ask why it would max clock randomly as well as voltage. I've never owned an AMD CPU until this.

As far as using Ryzenmaster goes: it's definitely a useful tool but is intrusive and can hurt performance. Some have even found it makes their system unstable when gaming. That's OK since it's intended purpose is for extreme overclocking demonstrations and not 24/7 use. Once you've figured out a decent setup I'd uninstall it (to remove the service) and use BIOS only to make the settings, for stability.

Perfect okay - this is what I thought. I hated the fact that having it open 24/7 is already painful enough (I can't even find a way to minimize it to tray when using).

As far as PBO setups go: I'd suggest checking out some youtube videos on the process for 5900's. Curve Optimizer is very effective but it can be a bit tedious to get it right.

Thanks! I'll definitely take a look further.. I think the hardest part for me is nobody truly explains every setting they are changing. For instance, two videos I watched you disable PBO but then enable it in a different tab (?)

 

[drea.drechsler]

....
Thanks! I'll definitely take a look further.. I think the hardest part for me is nobody truly explains every setting they are changing. For instance, two videos I watched you disable PBO but then enable it in a different tab (?)

Yahhh....well...in BIOS it can be a bit strange. The BIOS's I'm familiar with, at least (Gigabyte, MSI and Asus boards). Most mfr's have a proprietary section for their overclock tuning in BIOS and give settings brand-familiar names that match closer to their branded feature sets. In that area you can do many things but there's usually another area (called AMD Overclocking) where things are presented differently...but pretty much the same in all the BIOS' I've seen. I suspect AMD requires this user interface. As far as Ryzenmaster goes...I only used it early on, I understand a lot's been added. So I'm not familiar with what it can do now.

 

[the7goonies]

Yahhh....well...in BIOS it can be a bit strange. The BIOS's I'm familiar with, at least (Gigabyte, MSI and Asus boards). Most mfr's have a proprietary section for their overclock tuning in BIOS and give settings brand-familiar names that match closer to their branded feature sets. In that area you can do many things but there's usually another area (called AMD Overclocking) where things are presented differently...but pretty much the same in all the BIOS' I've seen. I suspect AMD requires this user interface. As far as Ryzenmaster goes...I only used it early on, I understand a lot's been added. So I'm not familiar with what it can do now.

Gotcha okay.

Here's what the "curve optimizer" did in ryzen master (it ran an optimization test on all cores separately for almost 2 hours) and gave me this result:
https://pasteboard.co/yxWqtSYCyLBC.png

From what I've read.. -30 seems to be pretty steep, and surprisingly there's no difference between values per core (I know in 99% of the videos it shows varying results per core). Do I cinebench this and see if it's stable? Do I run any specific programs that would tell me otherwise?

EDIT:
I ran the C20 benchmark test and scored the highest I've ever had at 8667pts, which is actually above the website results of 8545.
I ran the C23 benchmark test and scored the highest as well at 22306 (more than 2.3k score from previous), which is above the website results of 22046. During this test, I hit 85C but nothing higher.

Even if it doesn't matter anymore I still ran the CPU Profile, I'm no longer in last place, but moved into the bottom 20%.. mainly because of my 16-thread score being crap: https://pasteboard.co/FOPXvhngZ2cP.png

 

[drea.drechsler]

Gotcha okay.

Here's what the "curve optimizer" did in ryzen master (it ran an optimization test on all cores separately for almost 2 hours) and gave me this result:
https://pasteboard.co/yxWqtSYCyLBC.png

From what I've read.. -30 seems to be pretty steep, and surprisingly there's no difference between values per core (I know in 99% of the videos it shows varying results per core). Do I cinebench this and see if it's stable? Do I run any specific programs that would tell me otherwise?

EDIT: I ran the C20 benchmark test and scored the highest I've ever had at 8667pts, which is actually above the website results of 8545.

What curve optimizer does is to lower the voltage at the high end of the V/F performance curve of the CPU core, the highest frequency end. It might do nothing at the low end and very little in the middle.

Yes, -30 is VERY high, probably too high for your gold star cores which already have a steep negative voltage offset from the factory. But since you're testing with a heavy load it's not changing the voltage much since the CPU is operating in the middle of it's performance curve. You have to test with a light, bursty load that makes each core boost to it's maximum clocks repeatedly. There's a tool called Core Cycler, go looking for that. It sets up a light, bursty load and cycles it through each core multiple times. When it crashes on a core, that's the one you have to reduce the negative offset a few points.

It's one of the better tools to see if you've lowered too much, but another one is simply gaming. Gaming is inherently light and bursty so it will crash an unstable core, usually resulting in a BSOD. The problem with that is figuring out which core crashed, but there is a way.

Tweaking in extreme CO undervolts can be hard, that's why most folks just dial in a -10 or -12 on all cores and call it a day. But also, pay attention to the PPT, TDC and especially EDC settings for best performance. EDC in particular can be counter-intuitive in that lowering it 10A or 20A from stock can improve performance in heavy workloads.