Question Setting a long term Thermal Limit for an overclocked Ryzen 9 5950X ?

[Luk6655]

I've recently upgraded my Ryzen Zen 2 3700x for a Ryzen 9 (Zen3) 5950X and as I have a Corsair H150 AIO and an Asus tuf gaming x570-plus MB I wanted to take advantage of some overclocking.

First let me say I'm using a 32GB 3200MHZ cl14 RAM, I'd love to upgrade to 64GB with 2x32GB, but any sticks or RAM I can find at 32GB per stick and 3600MHz are cl18+ 😵 I briefly considered buying another pair of exact same modules I have now t up my RAM to 64GB, but they (G.skill cl14 3200MHZ) cost more than what I paid when I bought them 3+ years ago...

So I'm staying with 3200MHz RAM. I upped my PBO2 to 260/tdc 160/edc 180, I set my thermal limit at 85C and I created a per core under volt that makes my cores boost to 5050MHz when one is used and 4600 when all are used. The CPU draws 230W of power and stays at around 85 package temp (~82C core temps).

And I wonder. These CPUs have factory set 95C as throttle limit. Would I be loosing lots of potential longevity by setting it at 90C or even the original 95C? That should give ma a good 100MHz of boost at least hopefully. Going from 4300 to 4600 upped my cinebench R23 score from ~27k to ~29.5k. I'd love to get to 30k (my previous CPU was a bit under 12k).

So is anyone using heavy multicore loads ona ryzen 5000 series with high EDC and power almost double the spec for a long time? There is a lot of conflicting info online, but if AMD set the default thermal throttle limit at 95 it should be safe for the CPU?

What do you think?

 

[CountMike]

I've recently upgraded my Ryzen Zen 2 3700x for a Ryzen 9 (Zen3) 5950X and as I have a Corsair H150 AIO and an Asus tuf gaming x570-plus MB I wanted to take advantage of some overclocking.

First let me say I'm using a 32GB 3200MHZ cl14 RAM, I'd love to upgrade to 64GB with 2x32GB, but any sticks or RAM I can find at 32GB per stick and 3600MHz are cl18+ 😵 I briefly considered buying another pair of exact same modules I have now t up my RAM to 64GB, but they (G.skill cl14 3200MHZ) cost more than what I paid when I bought them 3+ years ago...

So I'm staying with 3200MHz RAM. I upped my PBO2 to 260/tdc 160/edc 180, I set my thermal limit at 85C and I created a per core under volt that makes my cores boost to 5050MHz when one is used and 4600 when all are used. The CPU draws 230W of power and stays at around 85 package temp (~82C core temps).

And I wonder. These CPUs have factory set 95C as throttle limit. Would I be loosing lots of potential longevity by setting it at 90C or even the original 95C? That should give ma a good 100MHz of boost at least hopefully. Going from 4300 to 4600 upped my cinebench R23 score from ~27k to ~29.5k. I'd love to get to 30k (my previous CPU was a bit under 12k).

So is anyone using heavy multicore loads ona ryzen 5000 series with high EDC and power almost double the spec for a long time? There is a lot of conflicting info online, but if AMD set the default thermal throttle limit at 95 it should be safe for the CPU?

What do you think?

If that cooling of yours works properly you shouldn't se over 90c anyway. CPU should boost to 4.7 to 4.8 GHz on all core load and 4.9 to 5GHz single core. Don't force 4,6 manually as you lose performance, Just control voltage thru CO.

 

[drea.drechsler]

....
And I wonder. These CPUs have factory set 95C as throttle limit. Would I be loosing lots of potential longevity by setting it at 90C or even the original 95C? That should give ma a good 100MHz of boost at least hopefully. ...

Since any semiconductor's life is determined as much (or more) by it's operating temperature then it's obvious you will lose some useful life by operating at a higher temperature with all else equal. But then remember that AMD did the math to determine the processor will have a decently long service life even when operating at it's rated Tjmax of 90C (not 95C).

But also remember that assumes it's operating as a typical user operates it, which is to say running typical desktop workloads and adequate cooling, and only sees 90C in intermittent peak processing periods and not pegged there nearly constantly, 24/7. That suggests that if you keep your system busy for 24/7 processing extremely complex all-core workloads (rendering batches of videos as a rendering workstation for a development studio, for instance) then the 85C thermal limit might help with a longer life. But it's more likely to become obsolete and upgraded well before then anyway.

One other thing using a lower (85C) thermal limit might help with is allow the processor to keep boosting individual cores at higher clocks than if allowed to run up to 90C. Whether or not that actually helps with performance probably depends on the specific workload and how the scheduler loads up processor cores for it. Probably a benefit for lightly threaded workloads, like gaming.

 

[Luk6655]

If that cooling of yours works properly you shouldn't se over 90c anyway.

What do you mean by that "if my cooling works I shouldn't see 90C". Did you read the bit where I wrote it's OC'ed to over 220W? Yes, the cooling works great, buy it is still possible to hit 90C in a heavy all core workload.


Heck I've been hitting 80C in single core workloads running at 80W with prime95 with AVX2. And before you say "your cooling must be crap", let me say this is the temperature within the core. The die itself is 3C less and the moment workload drops the temperature drops under45C. I believe this is perfectly normal for a non de-lidded cpu like this. If I max out my AIO I can get about 3C better. This tells me I'm essentially limited by the heat transfer of the ihs.

CPU should boost to 4.7 to 4.8 GHz on all core load and 4.9 to 5GHz single core. Don't force 4,6 manually as you lose performance, Just control voltage thru CO.

That's what I'm doing (Co) , but 4.7 to 4.8 on all core workloads on a 16 core cpu it must be for rather lite-weight all core workloads. Surely not Cinebench r23? Right?

If you're getting 4.7 to 4.8 in all core workload in Cinebench r23 you must be getting close to 32k points. That's like within the 50 best in the world... If you have a cpu like that, I envy you 🙂

Since any semiconductor's life is determined as much (or more) by it's operating temperature then it's obvious you will lose some useful life by operating at a higher temperature with all else equal. But then remember that AMD did the math to determine the processor will have a decently long service life even when operating at it's rated Tjmax of 90C (not 95C).

But also remember that assumes it's operating as a typical user operates it, which is to say running typical desktop workloads and adequate cooling, and only sees 90C in intermittent peak processing periods and not pegged there nearly constantly, 24/7. That suggests that if you keep your system busy for 24/7 processing extremely complex all-core workloads (rendering batches of videos as a rendering workstation for a development studio, for instance) then the 85C thermal limit might help with a longer life. But it's more likely to become obsolete and upgraded well before then anyway.

One other thing using a lower (85C) thermal limit might help with is allow the processor to keep boosting individual cores at higher clocks than if allowed to run up to 90C. Whether or not that actually helps with performance probably depends on the specific workload and how the scheduler loads up processor cores for it. Probably a benefit for lightly threaded workloads, like gaming.

Thanks, yes I might have to think about it as a balance between longevity and performance... I wish there were more data points about longevity.


Regarding tjmax of 90C. Yes, it is correct the tjmax is 90C, but still the default platform throttle limit is set at 95C on my motherboard. Weird, isn't it? That's why I was wondering if there is any harm in raising it.

Unfortunately all the "oc results" one sees online have to be taken with a grain of salt. If you see someone's getting 33k Cinebench R23 points you don't know if that system is just stable enough to complete one test, or can it run 24/7 like this?

 

[drea.drechsler]

...
Thanks, yes I might have to think about it as a balance between longevity and performance... I wish there were more data points about longevity.

Regarding tjmax of 90C. Yes, it is correct the tjmax is 90C, but still the default platform throttle limit is set at 95C on my motherboard. Weird, isn't it? That's why I was wondering if there is any harm in raising it.

Unfortunately all the "oc results" one sees online have to be taken with a grain of salt. If you see someone's getting 33k Cinebench R23 points you don't know if that system is just stable enough to complete one test, or can it run 24/7 like this?

With respect to reasonable operation, e.g., respecting AMD's 90C limit or lower, I don't think you'll see many "real world" data points because these processors based on geometries smaller than 10nm are just too new. But AMD has plenty of test data since they developed the FIT (Failures In Time) parameters that govern the boost algorithm which uses core temp as (one of) the primary inputs. AMD's data is obviously going to be proprietary but they have enough confidence in it to release literally 100's of millions of them into the world-wide marketplace with these spec's.

I also approach internet overclocking claims with low confidence. Many people indeed are not concerned with long-term stability, just that it boot Windows and game for a few hours. They can get away with that since gaming is not nearly as demanding of modern processors as you might think since they don't extensively employ instructions such as AVX and only one or two threads very intensively at that. If they tried a long video encode (h.264, heavy with AVX instructions using all available cores/threads) it may crash or generate garbage. Some people say h.264 doesn't adequately utilize a 16 core processor. OK then, launch two instances which will get two videos rendered simultaneously in less time than doing them sequentially (that's what I do ).

I wonder if the reason the motherboard defaults to 95C thermal limit is because only 5800X/X3d, 5900X and 5950X processors (on AM4) have the 90C Tjmax spec. All other Ryzen 5000 processors (and 3000 series processors) have 95C...at least the ones I've checked out. Still, it's an oversight on the board mfr's part none-the-less.

 

[Luk6655]

With respect to reasonable operation, e.g., respecting AMD's 90C limit or lower, I don't think you'll see many "real world" data points because these processors based on geometries smaller than 10nm are just too new. But AMD has plenty of test data since they developed the FIT (Failures In Time) parameters that govern the boost algorithm which uses core temp as (one of) the primary inputs. AMD's data is obviously going to be proprietary but they have enough confidence in it to release literally 100's of millions of them into the world-wide marketplace with these spec's.

Indeed, they even have a 3 year long warranty. I'm very tempted to grind that ihs flatter (I have equipment to achieve about half a micron across the size of a ryzen ihs, and to measure flatness down to 100nm - on account of a different hobby) or delid it, but I don't have much confidence in it not failing during these 3 years and then I'll regret I can't get a replacement... Especially that this is the best am4 cpu one can buy and odds of finding one in a year or two from now may be slim.

I don't even think my delidding/grinding would increase the odds of failure. I have a suspicion as you said previously, AMD is not expecting most of the users to load their cpus at 100% 24/7 at max temps and if you do they might just accept higher failure rate for such users.

I wish long term failure data with thermals was available, like Google's data for spinning drives they published some years ago. If I knew the typical chance of failure at stock settings in a heavy 24/7 workload, near tjmax are well under 1% I'd feel the chances are good enough to grind/delid too.

Someone may ask, why? Isn't this cpu "not fast enough"? If so, why haveeI upgraded to am5? Well, yes it is "fast enough" even without oc, but I do get extra pleasure from extracting as much as possible(within subjectively defined "reason") from the hardware I own.

I also approach internet overclocking claims with low confidence. Many people indeed are not concerned with long-term stability, just that it boot Windows and game for a few hours. They can get away with that since gaming is not nearly as demanding of modern processors as you might think since they don't extensively employ instructions such as AVX and only one or two threads very intensively at that.

Exactly. Although I'm hearing on YouTube some modern games are starting to utilise AVX2 and many cores when available. I'm not sure to what extent though.

I wonder if the reason the motherboard defaults to 95C thermal limit is because only 5800X/X3d, 5900X and 5950X processors (on AM4) have the 90C Tjmax spec. All other Ryzen 5000 processors (and 3000 series processors) have 95C...at least the ones I've checked out. Still, it's an oversight on the board mfr's part none-the-less.

I was mistaken with that 95C. I set it again to the default and ryzen master clearly shows the limit to be 90C. I'm not sure if I misread it before (very likely) or I somehow managed to set it to 95C without realising.

 

[drea.drechsler]

....
Someone may ask, why? Isn't this cpu "not fast enough"?
...
Exactly. Although I'm hearing on YouTube some modern games are starting to utilise AVX2 and many cores when available. I'm not sure to what extent though.
...

I'm not sure how much it will gain from polishing the IHS you but if your temp is running into the 80's then just about anything you can do to cool the processor better (everything else unchanged) will help performance because of how the boost algorithm works by using temperature as one of it's primary inputs to limit maximum boosting. Even though it might perform better, I wouldn't necessarily look for temperature to run a lot lower (again because of how the boost algorithm works) but it might.

Definitely explore undervolting with PBO2 and Curve Optimizer. CO works with the boost algorithm so you can get better performance from a cooler running processor while helping with service life by lowering voltage. Don't try simply lowering Vcore, though, as that confuses the boost algorithm and leads to lower performance if anything.

Most modern games are indeed much more heavily threaded than before. But the nature of any simulation is there is one thread that paces everything, the other threads are left constantly waiting for it so they can do something. That's not to say they aren't useful since anything that can be off-loaded from the main thread will help it's performance. Some turn-based strategy type games also make heavier use of multiple threads, my guess is for simultaneously calculating outcomes for the myriad of game pieces after moves.

While games may also use AVX and AVX2 instructions it's sporadic and not nearly as heavy as something like a rendering app (Cinebench, Handbrake/h.264) that hits all available threads with tightly looped AVX rich code. The heaviest thread loading I've noticed with the games I play is while de-compressing game assets taken off disk before loading into GPU memory. De-compressing uses AVX instructions.