5800x (and other Zen 3 chips) PBO settings/Temperature fix

[Mussels]

I've been running Intel Burn test for about 8 hours or more. It never goes above that 77.5C
I'm just wondering how consistent Hwinfo is for something like cinebenchR23 cause I've noticed all threw my single core test for 90% of the test it shows the current speed to be only 4,022mhz I end up with a score around 1598-1612 for single thread. If the chip isn't hitting it's single thread load more consistently it's lacking a lot of speed. My multi core test I've had a low of 11,450ish and a high of 12,278. I see max cpu in HWinfo says 4850mhz sometimes 4875mhz but it just doesn't seem to pin it there for the cinebench single test ever. On the current in HWinfo I'll sometimes see a spike of 4,335mhz. The all core load it's consistently at 4,700-4694mhz. I'm running pbo with +200.

Clock stretching
Zen 2 and 3 poll at 1ms (1,000 times a second) - if you hit a thermal or power limit, they can lower down in clock speed to stay within the limits, then be right back up to max clock before it's polled again.
Remember that ST can bounce from core to core to core, within those periods and therefore not get super accurate results either - try locking it to a single core in task manager and see if the results are more consistent
Cinebench is a heavy single threaded load, you just wont see single core max off that. You see it during lighter loads, like windows use.

 

[tabascosauz]

I've been running Intel Burn test for about 8 hours or more. It never goes above that 77.5C
I'm just wondering how consistent Hwinfo is for something like cinebenchR23 cause I've noticed all threw my single core test for 90% of the test it shows the current speed to be only 4,022mhz I end up with a score around 1598-1612 for single thread. If the chip isn't hitting it's single thread load more consistently it's lacking a lot of speed. My multi core test I've had a low of 11,450ish and a high of 12,278. I see max cpu in HWinfo says 4850mhz sometimes 4875mhz but it just doesn't seem to pin it there for the cinebench single test ever. On the current in HWinfo I'll sometimes see a spike of 4,335mhz. The all core load it's consistently at 4,700-4694mhz. I'm running pbo with +200.

What's your polling rate? Something's not right here because your scores are fine but it's definitely not running at 4022MHz. ~1600 ST is around 4.7-4.8GHz iirc

Yes, Zen 3 does bounce the load around, but if your chip is functioning correctly it will only bounce between 2 cores on CB R23.

Try using Snapshot Polling in HWInfo settings, and try upping your polling rate. Faster polling rate should cost you a tiny bit of performance but let's see what say 200ms or 500ms looks like. Cinebench R23 ST is a pretty consistent ST test, so I honestly don't know why it's happening to you.

How hard are you pushing this 5600X that you're at 77C on a 360mm rad?

 

[Valantar]

Thought I'd post my results here as I'm in the process of tuning my 5800X's BPO and Curve Optimizer settings. Currently at +100 offset, -11 for cores 0 and 1 ("best" cores), -20 for the rest, PPT 120, TDC 95, EDC 120. Cooling is my custom loop, single Corsair XR5 280mm rad in a Meshlicious with 2x Arctic P14 fans (intake, pull), Laing DDC 1T Plus PWM in a Nouvolo Aquanaut pump+block combo, PowerColor Liquid Devil Ultimate 6900 XT. "RAM+VRM" fan is a Noctua NF-A12x15 stuck inside of one side panel blowing over the pump, VRMs and RAM to aid in RAM overclocking. An EK flow indicator with the rotor removed as a micro reservoir/bubble catcher. Quick disconnects between all components, so some added restriction here. No idea about flow rates. Measurements in HWinfo were started about two minutes into the CB23 run. Running my "dynamic fast" fan/pump profile which is a bit more reactive than my typical "really quiet" profile that only reacts to changes in liquid temperatures. Still pretty quiet.

I'm not quite done tuning my CO; I was testing out -12 on the best cores yesterday but had a game crash to desktop out of the blue, so I'm back to -11 which seems stable (neither gave errors in Corecycler). Next up is increasing the offset on the remaining cores.

 

[tabascosauz]

I'm not quite done tuning my CO; I was testing out -12 on the best cores yesterday but had a game crash to desktop out of the blue, so I'm back to -11 which seems stable (neither gave errors in Corecycler). Next up is increasing the offset on the remaining cores.

Dunno when exactly I'll move to a custom loop whether it's in this case or the next, but it looks like I will have to consult your knowledge frequently but may require your expertise much much sooner for Aquasuite, fingers crossed

What's your setup for Corecycler? Are you still using the out of the box config without changing anything? Default config can generally get you pretty close to stable/generally indistinguishable or not problematic, but from long hours spent on Corecycler I'd be very hesitant to trust the default config with anything more than light web browsing. So far I haven't really encountered a need to test AVX, but I believe there is a lot of value in testing a much broader range of FFTs (All FFTs) for a much longer amount of time per iteration (1hr+). If you value complete stability for your daily system, for instance. But chances are if -12 can already pass default config repeatedly, then knocking 1-2 off that number should be just fine, if you don't want to spend endless hours on comprehensive corecycler testing.

 

[Valantar]

Dunno when exactly I'll move to a custom loop whether it's in this case or the next, but it looks like I will have to consult your knowledge frequently but may require your expertise much much sooner for Aquasuite, fingers crossed

What's your setup for Corecycler? Are you still using the out of the box config without changing anything? Default config can generally get you pretty close to stable/generally indistinguishable or not problematic, but from long hours spent on Corecycler I'd be very hesitant to trust the default config with anything more than light web browsing. So far I haven't really encountered a need to test AVX, but I believe there is a lot of value in testing a much broader range of FFTs (All FFTs) for a much longer amount of time per iteration (1hr+). If you value complete stability for your daily system, for instance. But chances are if -12 can already pass default config repeatedly, then knocking 1-2 off that number should be just fine, if you don't want to spend endless hours on comprehensive corecycler testing.

Let me know if I can be of any help My Aquasuite knowledge is of the "I don't really know what I'm doing, but my hacky solutions work for me" type, which ... well, as I said, it works for me I'm still figuring out how to use profiles and all the various logic you can set up for automatically switching between them - if anything this application has a bit too many options. But that also makes it really cool. There's also a ton of customization you can do if you are a programmer (at lest in the visualizations) which I understand exactly nothing of, and I'm constantly bumping up against limitations in the preset widgets (like there being no native way of having two datasets display in the same graph), leading me to some ... well, questionable solutions - but again, they kind of work. My per-core CPU clock+load indicator at the bottom is just a bunch of variously transparent and colored bar graphs stacked on top of each other, for example (and probably the part of the whole setup I'm the least happy with). Oh, and for some reason the default display page for the Quadro comes with some really good visualizations that don't exist as presets in the software itself, so I've been duplicating and modifying them to fit my needs. Pretty happy with its current state though!

I'm just running the stock Corecycler config, given that I started trying it out yesterday or the day before I really haven't had the time to consider whether there are better configurations. I've been running -10 all core for a couple of months, so I know that is stable as a baseline, yesterday was spent figuring out which cores were failing at -15 and where the threshold for stability for those cores lies. Seems like -11 is it, at least for now at +100. Tbh I don't have much of an idea of what I want to test for beyond "I hope nothing crashes" My previous, entirely uninformed CO experiments showed me that instability can be really unpredictable with this, so for now I'm just running Corecycler to identify any glaring issues and then using the PC as normal, hoping for the best.

 

[DemonicRyzen666]

What's your polling rate? Something's not right here because your scores are fine but it's definitely not running at 4022MHz. ~1600 ST is around 4.7-4.8GHz iirc

Yes, Zen 3 does bounce the load around, but if your chip is functioning correctly it will only bounce between 2 cores on CB R23.

Try using Snapshot Polling in HWInfo settings, and try upping your polling rate. Faster polling rate should cost you a tiny bit of performance but let's see what say 200ms or 500ms looks like. Cinebench R23 ST is a pretty consistent ST test, so I honestly don't know why it's happening to you.

How hard are you pushing this 5600X that you're at 77C on a 360mm rad?

I had some stuff really high, but one thing I know this chip has improved speeds just with voltage bump. The voltage bump on the board here is says cpu NB/Soc voltage. The single thread jumped up about 200-600 points. Just from bumping that buy +1 setting.
High like up in the 335 PPT 175 TCD 165 ECD . I doubt it did anything past 125PPT. Also and curve optimizer at like negative 15-17. My all core was sustained 4,700mhz-4,705mhz My single thread was something like 4,850mhz-4875mhz. I did throw in a bump in my bus speed. Because for some reason this boards bus speed is never what it states even on auto it's like 99.8-99.5. If I set 100.065. it comes up 99.9 :/ and that's only the minimum bump for the bus speed.
Went back to what's recommended on here as a start now. The radiator is in the case it's corsair 4000D air flow. How ever I noticed behind the radiator when the fans are at full blast there isn't much air going threw the radiator, with my hand behind it, doesn't feel like a lot was being forced threw. I might just take the fans from push and put them as pull fans instead. only trouble is how to mount the radiator to the front as the fans are what's mounted to the front. Last I could use better thermal paste too, stuff I got isn't' too great.

 

[outpt]

Need brains
ppt:120,tdc:75,edc:108.
cores 2,3,4, -30 other cores set to -0.
Sottr will run theses cores if i set the affinity to do so with no problems. if I set the affinity to all cores I get a system restart. these cores have been thoroughly tested. backing off the curve adds no help. does this make any sense?
btw I finally got 5.025ghz out of this thing

 

[tabascosauz]

Let me know if I can be of any help My Aquasuite knowledge is of the "I don't really know what I'm doing, but my hacky solutions work for me" type, which ... well, as I said, it works for me I'm still figuring out how to use profiles and all the various logic you can set up for automatically switching between them - if anything this application has a bit too many options. But that also makes it really cool. There's also a ton of customization you can do if you are a programmer (at lest in the visualizations) which I understand exactly nothing of, and I'm constantly bumping up against limitations in the preset widgets (like there being no native way of having two datasets display in the same graph), leading me to some ... well, questionable solutions - but again, they kind of work. My per-core CPU clock+load indicator at the bottom is just a bunch of variously transparent and colored bar graphs stacked on top of each other, for example (and probably the part of the whole setup I'm the least happy with). Oh, and for some reason the default display page for the Quadro comes with some really good visualizations that don't exist as presets in the software itself, so I've been duplicating and modifying them to fit my needs. Pretty happy with its current state though!

I'm just running the stock Corecycler config, given that I started trying it out yesterday or the day before I really haven't had the time to consider whether there are better configurations. I've been running -10 all core for a couple of months, so I know that is stable as a baseline, yesterday was spent figuring out which cores were failing at -15 and where the threshold for stability for those cores lies. Seems like -11 is it, at least for now at +100. Tbh I don't have much of an idea of what I want to test for beyond "I hope nothing crashes" My previous, entirely uninformed CO experiments showed me that instability can be really unpredictable with this, so for now I'm just running Corecycler to identify any glaring issues and then using the PC as normal, hoping for the best.

I'm probably just gonna use it for a few fans at first to get used to Aquasuite as I don't plan on going to water just yet. VIII Impact has some really weird fan header placements, hopefully the Aquaero can take over for some of the case fans so that I don't have to use the stupid PUMP header for DC case fans (only >60% speed for DC).

The default corecycler config is fine for a ballpark estimate of stability. I was a little harsh in my previous post. If all the cores can pass 5-10 iterations each, then it's fine to use that as a baseline.

After that you can go to a wider range of FFTs. Just make sure you adjust the iteration time to be appropriate for your choice of FFT. All FFT is very time-consuming, but it's hard to beat if you have the time for it. Otherwise, Heavy (30min) and HeavyShort (10min) seem to be recommended and seem to cover a good range, they're in the config file.

Better to focus on one or two cores at a time, lest you want a wack journey that looks like this (not mine):

 

[Mussels]

Dunno when exactly I'll move to a custom loop whether it's in this case or the next, but it looks like I will have to consult your knowledge frequently but may require your expertise much much sooner for Aquasuite, fingers crossed

What's your setup for Corecycler? Are you still using the out of the box config without changing anything? Default config can generally get you pretty close to stable/generally indistinguishable or not problematic, but from long hours spent on Corecycler I'd be very hesitant to trust the default config with anything more than light web browsing. So far I haven't really encountered a need to test AVX, but I believe there is a lot of value in testing a much broader range of FFTs (All FFTs) for a much longer amount of time per iteration (1hr+). If you value complete stability for your daily system, for instance. But chances are if -12 can already pass default config repeatedly, then knocking 1-2 off that number should be just fine, if you don't want to spend endless hours on comprehensive corecycler testing.

Custom loops are silly fun.

Be prepared for a lot of system downtime, and ordering in extra pieces you didn't think you'd need like angled fittings, extenders, joiners, face huggers and so on.

Just remember that on ryzen you wont really get much lower temps, but you may get more sustained boosts

 

[Valantar]

I'm probably just gonna use it for a few fans at first to get used to Aquasuite as I don't plan on going to water just yet. VIII Impact has some really weird fan header placements, hopefully the Aquaero can take over for some of the case fans so that I don't have to use the stupid PUMP header for DC case fans (only >60% speed for DC).

The default corecycler config is fine for a ballpark estimate of stability. I was a little harsh in my previous post. If all the cores can pass 5-10 iterations each, then it's fine to use that as a baseline.

After that you can go to a wider range of FFTs. Just make sure you adjust the iteration time to be appropriate for your choice of FFT. All FFT is very time-consuming, but it's hard to beat if you have the time for it. Otherwise, Heavy (30min) and HeavyShort (10min) seem to be recommended and seem to cover a good range, they're in the config file.

View attachment 225068

Better to focus on one or two cores at a time, lest you want a wack journey that looks like this (not mine):

View attachment 225067

Wow, that is some mess. And, what, a month and a half of testing according to the dates (and likely as much before that)? No thanks! At least I seem to have a stable baseline at my current -11/-20 setup, and I'm already pretty happy with that. We'll see how much more effort I'm willing to put into it.

The Aquaero series always kind of scared me - it has so many features and functions I wouldn't know where to begin with setting one up. Luckily the Quadro covers exactly what I need (DC fan control would be nice, but it's not a necessity). If you're just using it for a few fans it shouldn't be too overwhelming though, and I think it's a good idea to start out simple and get used to the software. That way you can have more of an idea how you want to run things once you move to water.

 

[tabascosauz]

Custom loops are silly fun.

Be prepared for a lot of system downtime, and ordering in extra pieces you didn't think you'd need like angled fittings, extenders, joiners, face huggers and so on.

Just remember that on ryzen you wont really get much lower temps, but you may get more sustained boosts

I'm counting on it being out of commission for a long time. Whenever there's downtime I just swap with the tiny PC upstairs

I have some idea of the main components that I would use if I were to do it right now (Optimus/EK block, alphacool rads, 10/16 zmt tubes, D5 if no airflow DDC if yes airflow, etc) but the little things are really offputting for me bc I realize that I really know nothing. Fittings, choice of additives, leak testing, drain plug, etc. Which is why I'm putting it on the backburner. Not to mention the 2060 Super FE is a pain in the ASS for waterblocking, iirc Corsair and Barrow kinda can work but idk if it's worth the money on a 2-year-old midrange GPU like this

Wow, that is some mess. And, what, a month and a half of testing according to the dates (and likely as much before that)? No thanks! At least I seem to have a stable baseline at my current -11/-20 setup, and I'm already pretty happy with that. We'll see how much more effort I'm willing to put into it.

The Aquaero series always kind of scared me - it has so many features and functions I wouldn't know where to begin with setting one up. Luckily the Quadro covers exactly what I need (DC fan control would be nice, but it's not a necessity). If you're just using it for a few fans it shouldn't be too overwhelming though, and I think it's a good idea to start out simple and get used to the software. That way you can have more of an idea how you want to run things once you move to water.

Unless you get Windows corruption or crashing/WHEAs, you can probably just leave it and be fine as long as you get past that default config baseline. Past default config corecycler, I've never had any BSODs/reboots, but I did get one Cache Hierarchy from trying something stupid.

How's the resource usage on Aquasuite? Light like HWInfo or cancer like iCue/CAM?

 

[Mussels]

So make a CPU only loop for now, with the plan for adding in your next GPU.
Gives you time to learn how it all works, with a less complicated setup initially.

I used a spare test bench system and made a practise loop with no PC attached, initially - and then my budget AM4 rig after that
Key was learning how the fittings i'd chosen worked, and what made them grip tightly/come undone before going anywhere near my main system
(Main soft tubing tip: Dont get lazy and twist the tubing to tighten a fitting. it unscrews the other end... always unscrew, push hose on, re-screw)

 

[Valantar]

I'm counting on it being out of commission for a long time. Whenever there's downtime I just swap with the tiny PC upstairs

View attachment 225099

That's a good plan - a first time custom loop will inevitably result in you missing something and needing to wait on some random fitting, especially if you're buildning somewhat small. That's the reason why I have a mix of EK, Alphacool and Barrow fittings - had to get what was available with decent delivery times while building.

I have some idea of the main components that I would use if I were to do it right now (Optimus/EK block, alphacool rads, 10/16 zmt tubes, D5 if no airflow DDC if yes airflow, etc) but the little things are really offputting for me bc I realize that I really know nothing. Fittings, choice of additives, leak testing, drain plug, etc. Which is why I'm putting it on the backburner. Not to mention the 2060 Super FE is a pain in the ASS for waterblocking, iirc Corsair and Barrow kinda can work but idk if it's worth the money on a 2-year-old midrange GPU like this

Some general advice: I'd strongly advise getting an air pump leak tester. It's an extra expense, but makes things so much easier. ZMT tubing is awesome, but be aware that you won't be doing any tight bends with it. For filling and draining, my choice would be to just add some QDCs to the loop - it's no more of a hassle than adding fill/drain ports, valves, etc., and you get the added advantage of modularity. It also makes for drastically easier filling and draining as you can just attach an external reservoir if needed (or just stick the ends of some spare tubing into your coolant bottle. I use Alphacool's industrial nylon QDCs, which while not drip-free (they lose 1-2ml of liquid per connection cycle, so keep some paper towels handy) are affordable, secure, not very restrictive, lightweight, durable, easy to install, and generally just great to work with. I exclusively use clear premix coolant - it's not necessary, but at ~€10/L and my loop requiring less than half a liter, it's not much of a cost. I left one of my previous setups for two years without draining and had zero corrosion, gunk, growth, or anything really. Didn't clean the loop at all for more than four years, and it was still pretty much spotless. I've used both EK Cryofuel and Aquacomputer DP Ultra, and both have been great. Premix saves me the hassle of figuring out which additives are worth my attention, which I like. For fittings, I have good experiences with pretty much everything I've used, though my Barrow 90° rotary connectors have stiffened to such a degree that they can't really be rotated any longer. Alphacool's rotaries feel rather loose/wobbly but don't leak, while my EK AFs are in-between, less loose and a bit harder to turn than Alphacool but still very usable after several years in service. I use barb fittings for the most part (with zip-ties), but I have some EK compression fittings for places where there's a sharp bend near the fitting, as the ZMT does have a tendency to slip off of shorter barbs if under tension. Compression fittings are a pain to screw on (always rubs my fingers raw), but are extremely secure. If you plan your loop reasonably you can avoid compression fittings entirely tough, which is a significant cost savings.

Unless you get Windows corruption or crashing/WHEAs, you can probably just leave it and be fine as long as you get past that default config baseline. Past default config corecycler, I've never had any BSODs/reboots, but I did get one Cache Hierarchy from trying something stupid.

Sounds good. I'll keep an eye out for anything weird.

How's the resource usage on Aquasuite? Light like HWInfo or cancer like iCue/CAM?

The application window when open uses ~250MB of RAM and .5-1% CPU, the main service sits between 70-120MB of RAM and .1-.7% CPU (mostly in the lower part of the range), and the profile switcher service is another 35MB of RAM and .2% CPU. Perfectly fine IMO.

 

[ViperXTR]

Im good with this for now i think, tested on a hot afternoon, on fairly decent weather, max temps are 75C. I have yet to get an ACU on my new apartment

 

[mattferris]

Just chiming in here with an endorsement for useful software from Mussels' original post to help tune your 5800x.

The core cycler utility is really nice for ironing out any lingering "stability" issues that might be hiding from you. It's customizable (via a well-documented ini file) and it will tell you which cores had errors. I set mine up to cycle through each core for a long time. I left, did stuff outside the house, came back and hit ctrl+C to kill the script and it spit out a message that I had errors on cores 1 and 3. I backed off my curve offsets by 2 in each core and reran without errors.

This is a lot better than (but not necessarily a substitute for) setting a custom filter in the Windows Event Viewer and waiting for a WHEA error (which might never happen).

Obviously,I had nice soft errors in my Prime95 runs. You could get a more catastrophic error and black screen reboot. In this case, you wouldn't get to see the errors spit out by core cycler and you'd have to check event viewer for the APIC ID when you get back into Windows.

 

[Mussels]

Honestly i'd forgotten to use corecycler after my latest rounds of tweaking - Mattferris reminded me

Core 3 is my highest clocking one that windows prefers a lot, kinda logical it wants more juice

Brainfart time: BIOS counts from 0, is this 3 the BIOS 3, or would it be BIOS 2 (since 1 is 0?)

 

[mattferris]

Honestly i'd forgotten to use corecycler after my latest rounds of tweaking - Mattferris reminded me

View attachment 225608

Core 3 is my highest clocking one that windows prefers a lot, kinda logical it wants more juice

Brainfart time: BIOS counts from 0, is this 3 the BIOS 3, or would it be BIOS 2 (since 1 is 0?)

It’s 3 in your BIOS. Core cycled counts from 0

 

[St54Kevin]

Hey guys, just joined the chat with my 5800X.
I started optimizing my CPU some weeks ago because of very bad Temps with default bios settings hitting 88-90C on a prime run. I even RMA'd my first 5800X because of it and got a new MB but nothing changed.

First of all, my specs:
ASUS Prime X570-Pro latest BIOS v4021
R7 5800X
4x8GB 3200 CL18 Vengance
Artic Liquid Freezer II 280mm
All Fans maxing out above 75C

These are the settings I came up with:
LLC Auto
PBO +50Mhz
Limits set to keep max 80C in my CB tests:
PPT 140W
EDC 105
TDC 100
CO (-20,-30,-30,-30,-30,-25,-30,-15)
CO values upped by 5 on the WHEA core (couple of weeks 8h/day usage, values above), I need to try this CoreCycler tool

And these are the results I get with them (room temp 23C):
- Idle 37C
- CB20 Multi 6152, max. 80C (average 4,692Ghz, 1,291v)
- CB20 Single 627, max. 56C (average 4,900Ghz, 1,325v)
- CB23 10min Multi 15839, max. 80C (average 4,692Ghz, 1,290v)
- Prime95 3min smallestFFT, max. 85C (average 4,560Ghz, 1,192v)
- Prime95 3min smallFFT, max. 84C (average 4,532Ghz, 1,176v)

I tried to play with the PBO Freq offset to get my single thread higher than 4900MHz, but it will either crash or skyrocket my temps to uncomfortable territories again.

BTW, I just tried the new HYDRA, but it gives me worse results at higher Temps not even running stable (reboots).

I'm open for any tips I can improve Cheers

 

[Mussels]

your curve optimiser is really aggressive, that's why you're crashing out with higher than +25Mhz


Use corecycler since you say you havent already

 

[St54Kevin]

Yeah it's running for an hour now. So far no core crashes. I will let it run through the night and see wether there are errors in the morning.

 

[Nordic]

I have been running all cores -10 despite ore cycler suggesting otherwise. It gives really good performance and stability is adequate for my needs. I he had one crash in 6 months.

 

[St54Kevin]

Do I understand it correctly? I need to raise the CO values in order to be able to raise the boost? How much per 25MHz?

 

[ViperXTR]

negative values for CO are basically undervolting steps, tone them down to be able to get more stable boost. Also those aggressive negative values may be stable on some stress test but to much CO can crash your system on low load or idle as well, find the balance.
Also, are you using Rev4 offset mount?

 

[Mussels]

Do I understand it correctly? I need to raise the CO values in order to be able to raise the boost? How much per 25MHz?

Well of course, higher clocks need higher voltages

No one knows, every chip is different - and every core on every chip is different

 

[St54Kevin]

negative values for CO are basically undervolting steps, tone them down to be able to get more stable boost. Also those aggressive negative values may be stable on some stress test but to much CO can crash your system on low load or idle as well, find the balance.
Also, are you using Rev4 offset mount?

I'm running this config for months now. Had to adjust two times because of a WHEA but never again. It's running fine and stable, I ran long cb23 and prime sessions to double check. No problems at all. I do not know what Rev4 offset is, so I guess not

Also when I up the values I get into the 80s C again, despite my top of the line AIO. Because my system seems to have temp problems in general I don't want that

BTW my overnight corecycler didn't give any errors