What's new
Carbonite

South Africa's Top Online Tech Classifieds!
Register a free account today to become a member! (No Under 18's)
Home of C.U.D.

High Temps Ryzen 9 5950X?

Asetek makes all the AIO's that get rebranded by the third party guys (Asus, Corsair etc)

Arctic is a brand that is renowned at making excellent cooling products.. Not sure if they use Asetek as the base though. Have a feeling they build their own AIO's from the ground up.
Not all. Corsair stopped using them. Not sure how long ago.
 
From looking at your package power draw from your cpu (130w) on a cinebench run and still hitting 85c
I think there is perhaps poor contact to your cpu with the block.

I can't imagine that the cpu would just 'run hot'
Have you got another cooler to test with?
 
It will hit about 82.5 starting Cinbench R23 in the beginning but it will then drop to 74-75


What you said here.. It spikes to 82.5c and then drops down?
From my experience that means poor contact to the cpu.
You're getting a Hotspot immediately and then it's cooling down as the heat spreads.

It should be a gradual (especially with an aio) climb to peak temp.
 
What you said here.. It spikes to 82.5c and then drops down?
From my experience that means poor contact to the cpu.
You're getting a Hotspot immediately and then it's cooling down as the heat spreads.

It should be a gradual (especially with an aio) climb to peak temp.
Will try another remount and see what's what then. Tried to tighten the screws in a cross pattern for better distribution but it sounds like it didn't work :(
 
@Duck good morning, in all honesty I want to give you some pointers for AMD Ryzen. I posted a thread in the overclocking section about how to fine tune your PBO on AMD Ryzen so you can go have a look. I was surprised yesterday when my cpu temps never reached 80c while running Cinebench R23, but ambient was a tad bit cooler. Now if your performance dropped with the negative vcore offset, try to gradually reduce or basically up the vcore and see if there is any change in the score. I found that when I put my vcore offset at -0.1250 my results tank a lot even though temps are in the 60C range. It is a balancing act to get the most without oveheating the processor.
 
Oh and just to add in PBO my processor is set to negative 25 all cores.
 
Oh and just to add in PBO my processor is set to negative 25 all cores.
I'll definitely check out your OC post a bit later once I'm home. In terms of the negative offset, whenever I try going higher than say 18 (20+) I'm met with a lot of system instability with odd crashes here and there so I've kept it at negative 16 all core.
 
You managed to get it to work correctly now?


Sent from my iPhone using Tapatalk
 
I haven't read through this thread, so apologies if I missed something .

I did a 5950X build for a client last year and was getting just over 30k in Cinebench 23 with PBO enabled. I fiddled a lot with settings and got it stable at a negative offset of 10 if i recall. Didn't mess with voltages though.

Was on a MSI X570 Tomahawk board. 24k with Stock settings, so PBO and tweaking made a huge difference. Temps were too high with just PBO enabled and no tweaking, so finding the right negative offset was critical to maintaining performance and temps.
Was using a Fractal Design 360mm AIO which I wasn't very impressed with. Temps were maxing out at about 82C with a max fan curve speed of 75%.

This was acceptable to the client and thankfully he has not had any issues. I made 2 profiles in his BIOS so he could choose the stock one for a quieter system or the PBO one when he wanted max perfomance and a bit more noise .

Anyway just general info and nothing too specific but thought i would share.
 
You managed to get it to work correctly now?


Sent from my iPhone using Tapatalk
Sorry for the delay, I didn't get any notifications of the thread being updated.

Unfortunately still having issues on this end, I've ordered an H150i Capellix 360mm AIO from Evetech that should be coming in on Monday that I'm hoping makes things better.

If it sorts out these temps then ill be sending in the H100 for RMA
 
I haven't read through this thread, so apologies if I missed something .

I did a 5950X build for a client last year and was getting just over 30k in Cinebench 23 with PBO enabled. I fiddled a lot with settings and got it stable at a negative offset of 10 if i recall. Didn't mess with voltages though.

Was on a MSI X570 Tomahawk board. 24k with Stock settings, so PBO and tweaking made a huge difference. Temps were too high with just PBO enabled and no tweaking, so finding the right negative offset was critical to maintaining performance and temps.
Was using a Fractal Design 360mm AIO which I wasn't very impressed with. Temps were maxing out at about 82C with a max fan curve speed of 75%.

This was acceptable to the client and thankfully he has not had any issues. I made 2 profiles in his BIOS so he could choose the stock one for a quieter system or the PBO one when he wanted max perfomance and a bit more noise .

Anyway just general info and nothing too specific but thought i would share.
I would definitely like to fiddle with the PBO settings but currently, as it stands I'm just trying to keep the CPU cool with the negative 16 offset lol
 
Soooo back again!

Got a 360mm iCUE H150i Elite Capellix on Monday and have been faffing around...the temps are slightly better.

What I've noticed is that when I run R23 Cinebench with PBO on plus a negative 20 offset undervolt it still hits 90C almost immediately though I see it climbs a lot slower above that than with the 240mm.

When I turn PBO off and keep the undervolt I'm getting a max of 77C but the score is at about 24000 points (PBO OFF), whereas with PBO ON I'm seeing 27000-28000 but terrible thermals that spike quick.

Not sure if there is something wrong with the CPU at this point?
 
Dude, that's very strange.
I'm on air cooling with pbo on and a negative offset of 10 and I don't see my cpu hit 90.

I know it's a mission and a schlep, but if you do take the cpu out of the socket again, please try put a dark (non transparent) ruler on the IHS, then look at it with a bright light source behind it to check if the IHS is flat. It sounds like your cpu's IHS is not very flat and that may be causing you compromised contact to the block.

Might need a good old lapping (which will void warranty)
 
I'm not familiar with other boards, but I can't go more than a negative offset of 12 on my asus board. Anything more and it's super unstable.
Is that kind of offset normal on your type of board?
 
So I have some good news

Just faffed about for the last hour and tried changing PBO settings here...

So with the Aorus Master when setting things to "motherboard" or "Auto" when it comes to regulating the PBO limit in BIOS it seems like those modes should be called "Open the doors to the fiery pits of hell and increase performance".

I've loaded optimized defaults for the 512th time but when toggling PBO limits I set it to "disabled" as opposed to "auto" or "motherboard", negative 18 offset for undervolt and now temps are looking great in comparison. Cinebench R23 would hit 90 immediately whereas now it's chilling at 70-75C maxing out.

The only downside here is that performance is not as good as before. Previously was hitting about 27000-28000 on R23 whereas now it's around 24900-25000 pts.

At this point, I'm just going to settle for the good temps as I hardly ever do rendering work so a loss in that department is acceptable.

Regarding the offsets, I checked them up on Youtube to see what was recommended and they advised 16-22 for the negative offset (ill see if I can find the link).

Not too clued up about these types of things but the lower the number the lower the voltage or would it be the higher the number?





Dude, that's very strange.
I'm on air cooling with pbo on and a negative offset of 10 and I don't see my cpu hit 90.

I know it's a mission and a schlep, but if you do take the cpu out of the socket again, please try put a dark (non transparent) ruler on the IHS, then look at it with a bright light source behind it to check if the IHS is flat. It sounds like your cpu's IHS is not very flat and that may be causing you compromised contact to the block.

Might need a good old lapping (which will void warranty)
 
Very important: always set CCX0 to 50-100mV above CCX1 for 5900Xs and 5950Xs. I.e., 4.7GHz = 4.75 on CCX0 and 4.65 on CCX1.

The Ryzen 5950X is made up of two CCXs. One CCX is excellent and the other is kinda trash. The trash one is what you’d find in the 5800X. They cut down the 8-core CCX to make the 5900X and 5600X, and so on.

The reason the 5950X needs to have an uber-good CCX0 and a fairly decent (better than a 5800X) CCX1 is to keep temps within reason. Basically the CPU has better silicon so it can run less voltage to hit its absolute max clocks.

I spoke to a person who’s an expert on Ryzen for around 8 hours on NAAF’s discord. He’s tuned hundreds of Ryzen systems the same way each time and I’m inclined to agree with his methodology, although I deviate slightly.

Now I don’t know how often Ryzen voltage spikes at idle, but what I do know is that AMD’s algorithm is pretty weird and it’ll feed far too much voltage that what is required for most workloads, especially when dealing with all-core workloads.

To put in perspective, my old 5900X while playing COD MW was boosting itself to 4.55GHz at 1.5V because the temps were sufficiently low and current draw was sufficiently low. I was able to daily 4.6GHz all-core at 1.25V set with 1.2V at load, which yielded 80c-ish in IBT V2, and idle temps that were fairly chilly.

A better way of doing what I did is simply dialling in your RAM settings and voltages rails (manual or just base XMP), disabling stuff like ‘Core Performance Boost,’ ‘CPPC,’ ‘CPPC Preferred Cores,’ and ‘Cool and Quiet.’ Thereafter you need to pick an LLC setting that you’ll end up testing later on. You can grab Ryzen Master on desktop and dial in a fixed core ratio and a set voltage (that’ll be mediated by the LLC setting you picked in BIOS).

Don’t daily anything above 1.3V (TSMC’s white paper limit that I’m inclined to believe after reading numerous accounts online), and also because you’re not going to be cooling any sort of stress test or R23 if you’re running anything above 1.3V set. In fact, I’d be surprised if you could even run 1.25V set on a 5950X and not hit well over 85c in something like IBT V2 or R23.

Even 1.2V load voltage is super high for the R9 parts due to the concentration of heat and the fact that only the Arctic coolers have an offset mount that caters to the CCX placement on the substrate.

In that regard my recommendation to you would be to set a fixed-core ratio at something like 1.25V (you should hit 4.6-4.7 depending on your CPU bin), and set something like LLC medium in your Aorus BIOS (+-40mV) of Vdroop. For some reason my friend had an issue setting a fixed voltage in the BIOS and booting from cold, so I’d recommend using the Ryzen Master method (it also means your CPU will down-clock at idle). Your CPU should be able to run a stress test at +-1.2V load voltage, and in R23 your temps should be reasonable.

Something to keep in mind is that it’s totally normal for your CPU to hit 90-100c in R23. The purpose of AMD’s algorithm is to squeeze as much juice out of your CPU given an envelope of power and heat. If your CPU is under-volted, all that happens is the decreased voltage will allow for more boosting headroom.

My 12900K hits 90-95c in R23 because my daily (and maximum usable) OC is 5.0 P, 4.0 E, 4.0 R at 1.35V set with around 1.3V under heavy load due to the LLC setting on my Asus board. It’s totally normal to have your CPU burn up in a stress test unless you’re using a decked out loop or exotic cooling. At least it is normal when you’re talking about R9’s and i9’s. The current draw of R23 is tremendous, and to run a high-ish all-core on both Intel and AMD requires a fair bit of voltage. Around 1.2V at load for a 5950X at 4.6GHz and around 1.3V at load for most i9’s at between 5.0-5.2 depending on how good your bin is. Multiple either of those numbers by the current draw of R23 (not R20, that’s fairly light).

But yeah… P95, Linpack Extreme, IBT V2, R23, and TimeSpy Extreme Physics; will all push your CPU above and beyond 75c. P95 at the right (or wrong) voltage, could see your CPU hit 110c pretty easily. That’s kinda just the nature of running your CPU at its all-core maximum in a heavy AVX workload. You can’t run the CPU at its max all-core with less than adequate voltage, so unless your cooling solution is insane, 90s are something one just has to get used to under the circumstances mentioned above.

Before someone types a reply:
AMD’s data sheets for Ryzen clearly stipulate an ‘up to’ spec for the boost. The only thing you are promised is base clock. This is in stark contrast to Intel’s fairly strict specifications for what your CPU should be capable of during the warranty period (not accounting for degradation due to manual OCing). I.e., my 12900K from mid-December 2021 till three years after the fact would’ve been guaranteed 4.9 all-core on the P-cores (technically it isn’t anymore).

The Individual I spoke to alarmed me when he explained how AMD’s CPU VID logic works, and it all makes sense when I read about the negative voltage offsets outlined in this thread. AMD’s CPUs know how much they can boost to given a certain amount of voltage. If the chip degrades it simply boosts to a lesser degree than before at the same voltage.

This individual also made a case in which he described an experience he had with two identical sealed 3600X’s. Basically he sent the one with the better silicon to his friend and his friend forgot to enable his OC profile, so his friend ran the chip at stock for 6 months, where after the Individual found that the max all-core dropped from 4.225 to 4.15 I believe.

This is anecdotal so you don’t have to trust it but it makes some level of sense to me judging by the accounts online from some users. Silicon degradation also depends on how perfect your silicon was to begin with.

The fact that under-volting by something like 0.125V is because you’re shaving 125mV off of the CPU’s voltage supply, and it will thus result in significantly decreased load voltage, but also a significantly nerfed boost clock. Instead of setting an under-volt, you’re better off setting something like 1.25V and finding out what your CPU’s silicon quality is like and how high you can manually set the all-core. Other than that you could set 1.2V and have it Vdroop down to 1.1-1.15V under heavy load, and set something like 4.4-4.5 all-core. Clock stability is much better at lower temps and 4.5 @ 1.125V or something to that effect might be achievable on something like a 5950X. That’ll result in something like 78c max in R23 if I had to spitball it.
 
Last edited:
Very important: always set CCX0 to 50-100mV above CCX1 for 5900Xs and 5950Xs. I.e., 4.7GHz = 4.75 on CCX0 and 4.65 on CCX1.

The Ryzen 5950X is made up of two CCXs. One CCX is excellent and the other is kinda trash. The trash one is what you’d find in the 5800X. They cut down the 8-core CCX to make the 5900X and 5600X, and so on.

The reason the 5950X needs to have an uber-good CCX0 and a fairly decent (better than a 5800X) CCX1 is to keep temps within reason. Basically the CPU has better silicon so it can run less voltage to hit its absolute max clocks.

I spoke to a person who’s an expert on Ryzen for around 8 hours on NAAF’s discord. He’s tuned hundreds of Ryzen systems the same way each time and I’m inclined to agree with his methodology, although I deviate slightly.

Now I don’t know how often Ryzen voltage spikes at idle, but what I do know is that AMD’s algorithm is pretty weird and it’ll feed far too much voltage that what is required for most workloads, especially when dealing with all-core workloads.

To put in perspective, my old 5900X while playing COD MW was boosting itself to 4.55GHz at 1.5V because the temps were sufficiently low and current draw was sufficiently low. I was able to daily 4.6GHz all-core at 1.25V set with 1.2V at load, which yielded 80c-ish in IBT V2, and idle temps that were fairly chilly.

A better way of doing what I did is simply dialling in your RAM settings and voltages rails (manual or just base XMP), disabling stuff like ‘Core Performance Boost,’ ‘CPPC,’ ‘CPPC Preferred Cores,’ and ‘Cool and Quiet.’ Thereafter you need to pick an LLC setting that you’ll end up testing later on. You can grab Ryzen Master on desktop and dial in a fixed core ratio and a set voltage (that’ll be mediated by the LLC setting you picked in BIOS).

Don’t daily anything above 1.3V (TSMC’s white paper limit that I’m inclined to believe after reading numerous accounts online), and also because you’re not going to be cooling any sort of stress test or R23 if you’re running anything above 1.3V set. In fact, I’d be surprised if you could even run 1.25V set on a 5950X and not hit well over 85c in something like IBT V2 or R23.

Even 1.2V load voltage is super high for the R9 parts due to the concentration of heat and the fact that only the Arctic coolers have an offset mount that caters to the CCX placement on the substrate.

In that regard my recommendation to you would be to set a fixed-core ratio at something like 1.25V (you should hit 4.6-4.7 depending on your CPU bin), and set something like LLC medium in your Aorus BIOS (+-40mV) of Vdroop. For some reason my friend had an issue setting a fixed voltage in the BIOS and booting from cold, so I’d recommend using the Ryzen Master method (it also means your CPU will down-clock at idle). Your CPU should be able to run a stress test at +-1.2V load voltage, and in R23 your temps should be reasonable.

Something to keep in mind is that it’s totally normal for your CPU to hit 90-100c in R23. The purpose of AMD’s algorithm is to squeeze as much juice out of your CPU given an envelope of power and heat. If your CPU is under-volted, all that happens is the decreased voltage will allow for more boosting headroom.

My 12900K hits 90-95c in R23 because my daily (and maximum usable) OC is 5.0 P, 4.0 E, 4.0 R at 1.35V set with around 1.3V under heavy load due to the LLC setting on my Asus board. It’s totally normal to have your CPU burn up in a stress test unless you’re using a decked out loop or exotic cooling. At least it is normal when you’re talking about R9’s and i9’s. The current draw of R23 is tremendous, and to run a high-ish all-core on both Intel and AMD requires a fair bit of voltage. Around 1.2V at load for a 5950X at 4.6GHz and around 1.3V at load for most i9’s at between 5.0-5.2 depending on how good your bin is. Multiple either of those numbers by the current draw of R23 (not R20, that’s fairly light).

But yeah… P95, Linpack Extreme, IBT V2, R23, and TimeSpy Extreme Physics; will all push your CPU above and beyond 75c. P95 at the right (or wrong) voltage, could see your CPU hit 110c pretty easily. That’s kinda just the nature of running your CPU at its all-core maximum in a heavy AVX workload. You can’t run the CPU at its max all-core with less than adequate voltage, so unless your cooling solution is insane, 90s are something one just has to get used to under the circumstances mentioned above.

Before someone types a reply:
AMD’s data sheets for Ryzen clearly stipulate an ‘up to’ spec for the boost. The only thing you are promised is base clock. This is in stark contrast to Intel’s fairly strict specifications for what your CPU should be capable of during the warranty period (not accounting for degradation due to manual OCing). I.e., my 12900K from mid-December 2021 till three years after the fact would’ve been guaranteed 4.9 all-core on the P-cores (technically it isn’t anymore).

The Individual I spoke to alarmed me when he explained how AMD’s CPU VID logic works, and it all makes sense when I read about the negative voltage offsets outlined in this thread. AMD’s CPUs know how much they can boost to given a certain amount of voltage. If the chip degrades it simply boosts to a lesser degree than before at the same voltage.

This individual also made a case in which he described an experience he had with two identical sealed 3600X’s. Basically he sent the one with the better silicon to his friend and his friend forgot to enable his OC profile, so his friend ran the chip at stock for 6 months, where after the Individual found that the max all-core dropped from 4.225 to 4.15 I believe.

This is anecdotal so you don’t have to trust it but it makes some level of sense to me judging by the accounts online from some users. Silicon degradation also depends on how perfect your silicon was to begin with.

The fact that under-volting by something like 0.125V is because you’re shaving 125mV off of the CPU’s voltage supply, and it will thus result in significantly decreased load voltage, but also a significantly nerfed boost clock. Instead of setting an under-volt, you’re better off setting something like 1.25V and finding out what your CPU’s silicon quality is like and how high you can manually set the all-core. Other than that you could set 1.2V and have it Vdroop down to 1.1-1.15V under heavy load, and set something like 4.4-4.5 all-core. Clock stability is much better at lower temps and 4.5 @ 1.125V or something to that effect might be achievable on something like a 5950X. That’ll result in something like 78c max in R23 if I had to spitball it.
1.25v for 4.6 -4.7ghz is not possible for me and I have a pretty decent 5950X. For 4.6ghz I need 1.3vcore set in the bios. Not that I am saying it is not possible, but the processors we get here in South Africa is not the cream of the crop and we have to make do with what we have. Now I will say this again as I said before, my processor is running with a -25 as in minus 25 all core vcurve thingy, but I also set the vcore is the bios with a -minus 0.0875 which took down all core load temps by 10c or so.

Now running Prime95 with my setup as stipulated I see just over 85c with the processor running 4.25ghz. Vcore around1.09v. Now I can set the vcore to minus 0.125v but when doing so performance drops by a fair bit. Now for manual overclocking I can run my processor at 4.7ghz but it needs 1.4vcore but that is for 3dmark time spy etc, which bare loads the cpu as much as say cinebench or not to mention prime95. I posted a thread in the overclock section about how to get the most out of PBO for AMD Ryzen.
 
1.25v for 4.6 -4.7ghz is not possible for me and I have a pretty decent 5950X. For 4.6ghz I need 1.3vcore set in the bios. Not that I am saying it is not possible, but the processors we get here in South Africa is not the cream of the crop and we have to make do with what we have. Now I will say this again as I said before, my processor is running with a -25 as in minus 25 all core vcurve thingy, but I also set the vcore is the bios with a -minus 0.0875 which took down all core load temps by 10c or so.

Now running Prime95 with my setup as stipulated I see just over 85c with the processor running 4.25ghz. Vcore around1.09v. Now I can set the vcore to minus 0.125v but when doing so performance drops by a fair bit. Now for manual overclocking I can run my processor at 4.7ghz but it needs 1.4vcore but that is for 3dmark time spy etc, which bare loads the cpu as much as say cinebench or not to mention prime95. I posted a thread in the overclock section about how to get the most out of PBO for AMD Ryzen.
I agree with most of what you’re saying. The 4.6 thing is a bit odd though. I was able to achieve that at 1.2V at load on my 5900X and 1.24V at load on a 5600X. Both IBT V2 stable. I’m not sure why your particular chip does 1.3V.

Your end result is fairly good, but I still think one would be better off ditching PBO as a manual OC will always be more voltage-efficient than AMD’s algorithm. With what you’re currently doing, the curve optimiser definitely helps up your max boosts as there is more voltage headroom available when the cores are using less to boost to what they were before, but with a negative 80mV offset you’re still shaving off performance directly and doing so whilst having the algorithm determine the all-core.

I’m not saying 4.25 is bad, but it’s definitely not optimal for maximum performance. The individual I spoke to on Discord has a bin that does 4.6875 at 1.3V and FrameChasers was doing 4.7 at 1.35V on his wife’s. I know what you said about ZA CPU binning but I’ve genuinely come to the conclusion it isn’t anything as sinister as that. I think you just got a bit of bad luck on that one, which is definitely something I know about.

I didn’t do a hell of a lot of testing, but in my humble opinion, the single-core boosts are largely irrelevant and setting a fixed all-core is always going to result in a better outcome. If low low temps are your things you could easily set 1.2V and have Vdroop down to between 1.1-1.15 and still outdo AMD’s PBO all-core. It does take some extra testing but I assure you one can outdo PBO by a significant margin and forego the whole PBO 1.5V silicon destroyer spike thing.
 
I agree with most of what you’re saying. The 4.6 thing is a bit odd though. I was able to achieve that at 1.2V at load on my 5900X and 1.24V at load on a 5600X. Both IBT V2 stable. I’m not sure why your particular chip does 1.3V.

Your end result is fairly good, but I still think one would be better off ditching PBO as a manual OC will always be more voltage-efficient than AMD’s algorithm. With what you’re currently doing, the curve optimiser definitely helps up your max boosts as there is more voltage headroom available when the cores are using less to boost to what they were before, but with a negative 80mV offset you’re still shaving off performance directly and doing so whilst having the algorithm determine the all-core.

I’m not saying 4.25 is bad, but it’s definitely not optimal for maximum performance. The individual I spoke to on Discord has a bin that does 4.6875 at 1.3V and FrameChasers was doing 4.7 at 1.35V on his wife’s. I know what you said about ZA CPU binning but I’ve genuinely come to the conclusion it isn’t anything as sinister as that. I think you just got a bit of bad luck on that one, which is definitely something I know about.

I didn’t do a hell of a lot of testing, but in my humble opinion, the single-core boosts are largely irrelevant and setting a fixed all-core is always going to result in a better outcome. If low low temps are your things you could easily set 1.2V and have Vdroop down to between 1.1-1.15 and still outdo AMD’s PBO all-core. It does take some extra testing but I assure you one can outdo PBO by a significant margin and forego the whole PBO 1.5V silicon destroyer spike thing.
Okay, my cpu is not that great but then again on Dry Ice it did 5.250ghz all core using only 1.525vcore which I felt was pretty good. I did not go higher as my mobo had an issue upping the vcore.

You can check my hwbot uploads to confirm.

Sent from my Hisense Infinity H40 Lite using Tapatalk
 
Currently I'm running the stock 120mm exhaust fan at the back and I have the other 2 120s pulling in air in on the front of the case attached to the AIO radiator. I've even resorted to taking off the front panel of the case just to get more airflow too which has improved things to a slight extent XD
Is your case not the culprit?
Do you have fans venting the top of the case as well?
I would suggest a re-paste with some better thermal paste and then getting all the fans in the correct orientation.
 
Okay, my cpu is not that great but then again on Dry Ice it did 5.250ghz all core using only 1.525vcore which I felt was pretty good. I did not go higher as my mobo had an issue upping the vcore.

You can check my hwbot uploads to confirm.

Sent from my Hisense Infinity H40 Lite using Tapatalk
Oh not this…

I’m not sure whether your CPU bin is good or not. You’re giving me a reason my advice is bad by explaining how ZA CPUs have bad bins and then tell me you can’t do a number that I’ve seen easily done, and you expect me to not conclude that you have a bad bin… I was going to ask, did you separate your CCX clocks? I assume you did, but just to check.

Also, I don’t want to be that guy, but your CPU can’t do 4.6 at under 1.3V set with normal cooling, so the logical conclusion is the bin isn’t good. If your CPU does really well on dry ice it can’t really say much though? My CPU at 30c at load would be able to do at least 200-300MHz higher on P, E, and Ring with the exact same voltage and LLC settings. Who’s to say a good bin 5950X wouldn’t do 5.4? Do you see where I’m going with this…
 
Last edited:
Oh not this…

I’m not sure whether your CPU bin is good or not. You’re giving me a reason my advice is bad by explaining how ZA CPUs have bad bins and then tell me you can’t do a number that I’ve seen easily done, and you expect me to not conclude that you have a bad bin… I was going to ask, did you separate your CCX clocks? I assume you did, but just to check.

Also, I don’t want to be that guy, but your CPU can’t do 4.6 at under 1.3V set with normal cooling, so the logical conclusion is the bin isn’t good. If your CPU does really well on dry ice it can’t really say much though? My CPU at 30c at load would be able to do at least 200-300MHz higher on P, E, and Ring with the exact same voltage and LLC settings. Who’s to say a good bin 5950X wouldn’t do 5.4? Do you see where I’m going with this…
Hi I am not saying what you say is wrong. I just think with Amd Ryzen it is not an easy thing to determine.

My whole experience with Amd Ryzen from 1700 to 2700 to 3600 to 3900 to 3950 to now 5950x is that determining the ability of the processor is more difficult at times than building a time machine. But that is why I love them.

Sent from my Hisense Infinity H40 Lite using Tapatalk
 
Hi I am not saying what you say is wrong. I just think with Amd Ryzen it is not an easy thing to determine.

My whole experience with Amd Ryzen from 1700 to 2700 to 3600 to 3900 to 3950 to now 5950x is that determining the ability of the processor is more difficult at times than building a time machine. But that is why I love them.

Sent from my Hisense Infinity H40 Lite using Tapatalk
if I can word it a bit better, my premise behind my entire thing is that the person I spoke to who has done testing for most of his non-working hours over the last (8 months) I believe came to a similar conclusion that I did, and along with the account of his friend’s degradation paired with accounts online of the same occurrence, I no longer view AMD’s stock algorithm as safe or reasonable.

I did some rough testing that I deleted, but my PBO was over-shooting voltage requirements heavily. With a set voltage once can affectively achieve whatever all-core and whatever temps they desire. I still do however believe that 90c in heavy AVX is totally fine, unless you run something that uses heavy AVX for long durations.

I think anything sub 1.3V with an adequately modest LLC and adapted to chip quality, cooling solution, and desired/tolerated heat at whatever use your deploy your chip for is ideal. I just think that the stock algorithm as a whole has no efficacy unless you desire single-core boosts that don’t really affect your experience all too much. This is coming from someone who used to be chuffed with their CPU boosting to 5.2GHz in benchmarks for insignificant periods of time.
 
To put in perspective, my old 5900X while playing COD MW was boosting itself to 4.55GHz at 1.5V because the temps were sufficiently low and current draw was sufficiently low. I was able to daily 4.6GHz all-core at 1.25V set with 1.2V at load, which yielded 80c-ish in IBT V2, and idle temps that were fairly chilly.

Don’t daily anything above 1.3V (TSMC’s white paper limit that I’m inclined to believe after reading numerous accounts online), and also because you’re not going to be cooling any sort of stress test or R23 if you’re running anything above 1.3V set. In fact, I’d be surprised if you could even run 1.25V set on a 5950X and not hit well over 85c in something like IBT V2 or R23.

In that regard my recommendation to you would be to set a fixed-core ratio at something like 1.25V (you should hit 4.6-4.7 depending on your CPU bin), and set something like LLC medium in your Aorus BIOS (+-40mV) of Vdroop. For some reason my friend had an issue setting a fixed voltage in the BIOS and booting from cold, so I’d recommend using the Ryzen Master method (it also means your CPU will down-clock at idle). Your CPU should be able to run a stress test at +-1.2V load voltage, and in R23 your temps should be reasonable.

Something to keep in mind is that it’s totally normal for your CPU to hit 90-100c in R23. The purpose of AMD’s algorithm is to squeeze as much juice out of your CPU given an envelope of power and heat. If your CPU is under-volted, all that happens is the decreased voltage will allow for more boosting headroom.

But yeah… P95, Linpack Extreme, IBT V2, R23, and TimeSpy Extreme Physics; will all push your CPU above and beyond 75c. P95 at the right (or wrong) voltage, could see your CPU hit 110c pretty easily. That’s kinda just the nature of running your CPU at its all-core maximum in a heavy AVX workload. You can’t run the CPU at its max all-core with less than adequate voltage, so unless your cooling solution is insane, 90s are something one just has to get used to under the circumstances mentioned above
So to touch on this, thanks for the detailed explanation there. While some of the points go a tad over my head due to my lack of an in-depth understanding I would like to run a few things by you just for a lil clarity.

So from what you advised, I can understand hitting the temps of 90-100C for R23 as it essentially utilises all cores and stresses the CPU. My problem is that when I was gaming, not really using more than 6 cores I was hitting 90C in certain cases which I feel is a tad odd due to it only using 1/3ish of the CPU cores (although I understand the clock rate would be boosting higher).

Right now as I have things set up I feel like I've struck a balance between thermals and performance to an extent. Currently, I am running a negative 18 offset for undervolt, PBO max CPU boost override set to 0Mhz and CPU vCore set to auto. Cinebench R23 I'm not going over 70C, results are approx 24900-25000 (which is quite low from what I'm told to expect), but in CPU-Z (see attached) when benching I'm now seeing 680pts for single-core which I'm told has more effect when playing games and such.

I've also experimented with turning PBO off and set a productivity profile in the BIOS that I can switch to that is 4.6Ghz with a 1.25v set voltage with auto LLC but as you say thermals stabilized around 91-93C.

Would it be possible to get to the best of both? As online I'm seeing scores far better compared to what I'm getting (see below link)

Link: Cinebench R23 Scores [Updated Results]
 

Attachments

  • CPUZ.jpg
    CPUZ.jpg
    24.4 KB · Views: 20
Is your case not the culprit?
Do you have fans venting the top of the case as well?
I would suggest a re-paste with some better thermal paste and then getting all the fans in the correct orientation.
Was originally thinking this may have been the "case" (excuse the pun) but after adding additional case fans, repasting and ensuring that orientation was fine it seems like the issue was in part due to a specific setting in the BIOS for PBO limits lol.

Now I just need to find the balance for thermals to perfomance
 
So to touch on this, thanks for the detailed explanation there. While some of the points go a tad over my head due to my lack of an in-depth understanding I would like to run a few things by you just for a lil clarity.

So from what you advised, I can understand hitting the temps of 90-100C for R23 as it essentially utilises all cores and stresses the CPU. My problem is that when I was gaming, not really using more than 6 cores I was hitting 90C in certain cases which I feel is a tad odd due to it only using 1/3ish of the CPU cores (although I understand the clock rate would be boosting higher).

Right now as I have things set up I feel like I've struck a balance between thermals and performance to an extent. Currently, I am running a negative 18 offset for undervolt, PBO max CPU boost override set to 0Mhz and CPU vCore set to auto. Cinebench R23 I'm not going over 70C, results are approx 24900-25000 (which is quite low from what I'm told to expect), but in CPU-Z (see attached) when benching I'm now seeing 680pts for single-core which I'm told has more effect when playing games and such.

I've also experimented with turning PBO off and set a productivity profile in the BIOS that I can switch to that is 4.6Ghz with a 1.25v set voltage with auto LLC but as you say thermals stabilized around 91-93C.

Would it be possible to get to the best of both? As online I'm seeing scores far better compared to what I'm getting (see below link)

Link: Cinebench R23 Scores [Updated Results]
1.25V with Auto LLC would be something like 1.15V under load. You’re going to need to pull up HWiNFO on a second screen for sure. I don’t know everything, but what I do know is that a 5950X at 1.15V under heavy load shouldn’t result in 90c. That can’t be right. It’s entirely possible that Auto LLC is like Turbo or something.

In that regard you’re definitely going to want to test anywhere from standard to medium. It’s hard to say if you could do 4.6GHz with 1.15V load voltage, but 1.15 X 180A (for instance), would be 207W. In that regard any heavy AVX such as R23 or Intel Burn Test V2 shouldn’t pin you near 90c. 210W is a lot of heat but it’s not otherworldly. In terms of gaming, you definitely shouldn’t be hitting 90c.

Again, you’re going to need to monitor the current draw and operating voltage. This sounds peculiar to me. The PBO route is pointless, the all-core makes far more sense. But while gaming, the all-core cannot be hitting 90c. Like no way. There’s just no way. It’s a math thing. It just can’t be that a game puts enough Amps through your chip to yield enough heat output to do that.
 
Last edited:
The Ryzen 5950X is made up of two CCXs. One CCX is excellent and the other is kinda trash. The trash one is what you’d find in the 5800X. They cut down the 8-core CCX to make the 5900X and 5600X, and so on.
Can confirm, the 5800X is absolutely terrible at heat transfer up to the contact point with the coldplate, because of the single trash CCX. Do anything semi intensive at all and watch temps hit 80+
 
Can confirm, the 5800X is absolutely terrible at heat transfer up to the contact point with the coldplate, because of the single trash CCX. Do anything semi intensive at all and watch temps hit 80+
I completely understand that and it’s exactly what I saw even on my 5900X with a D15 Chromax. Not quite to the same degree, but anything heavy like IBT V2 would destroy temps. It’s just a bit weird because I’m seeing remarkable numbers and I don’t know how some CPUs are spiking this badly.
 

Users who are viewing this thread

Latest posts

Back
Top Bottom