A question for anyone with a 5600X.....

[NoddyPirate]

Just a quick one here:

My 5600X boosts to 4.65 GHz for single threaded loads and sustains just under 4.3 GHz when I am using all cores - referring to effective CPU clock - which is great and exactly what I expected.

However, I am noticing that once the CPU temperature increases above about 60-65 degrees the frequency slowly reduces - to a little over 4.0 GHz when it's in the upper 70's - or about 4% or so. I haven't pushed it higher than that. The CPU power draw slightly reduces also although much less so. It looks like it begins throttling quite early - I kind of expected that it would have sustained those boost to much higher temperatures?

Is throttling actually what it is doing? And does this match others experience with the 5600X?

No big deal either way of course - and it's still way above it 3.7 GHz base clock - I'll also never notice a 4% drop in performance for my workflow. But thought I'd ask anyway.....

 

[NoddyPirate]

Surprised that there's been no insights here yet. Anyway, I fiddled around with this a bit yesterday. What idiot said I would never go messing around with Overclocking? 🤦‍♂️

Under stock conditions and an all core boost my chip sees 76W at the chip and 4.285 GHz across all cores. Core Voltage is around 1.18 or so but with occassional peaks to 1.375 - which is higher than I would have expected. Seems to be a Ryzen thing?

A stock single core boost sees a steady 1.33 V across all cores which is interesting and suggests a safe target if you were looking for a single core boost across all cores perhaps?

So, anyway to look at my reducing all core boost with increasing temperature, I set a manual OC at 4.3GHz and 1.1875 volts and it held steady to matter what happened temp wise. So I assume my stock behaviour is just normal power management going on behind the scenes.

Then I lost the run of myself and kept going! PBO enable gave minimal increase in speeds - around 4.4 GHz - but at 1.288 V and high power. Manually I comfortably got 4.625 GHz across all cores at 1.3 Volts. 4.65 GHz and 1.3V wasn't stable and I figured I wasn't going to push things any further than that voltage on a permanent basis, but I imagine more educated folk than I could do so easily enough. I limited the package power along with sustained and peak currents in BIOS to avoid any surprises for my motherboard - so it peaked at my limit of 100W.

This gave me around a 1,000 point increase in the Cinebench all core score, 750 point increase in PassMark CPU score, and surprisingly saw my LR and PS heavy workloads improve noticeably. A large export from LR would see the biggest help with 5 seconds cut from a 70 second export - about 7% better and close to the 9% increase in the all core frequencies.

But as has been mentioned before here many times - it's just not worth it for two reasons on my assessment at least:

All other performance reduced, 2D and 3D scores and so on. And the all core scenario only saw an improvement if my CPU was at about 95% usage or higher. Less than that and the benefit more or less disappeared completely in actual workloads, even if the benchmark scores still showed better numbers.

I also didn't like that my core voltages remained at 1.3V even when idle. A BIOS voltage offset might be the way to stop that happening - but the main issue that appears then, is that I now have no idea what I am talking about - so I shall look into that no further!

Suffice to say, the 5600X seemingly will OC easily and quite well, but if you don't have all cores at 100% more or less all the time, then there is absolutely no point whatsoever that I can see, save for boasting about your benchmark scores online somewhere.

(How this thread has morphed so much I don't know.....)

 

[NoddyPirate]

Forgot to mention that I have a GTX1650 so I am completely and utterly limited by that for graphics. My CPU might not even realise that I have a GPU installed to be honest.....

So the reduction in non full-multi-core performance was interesting to see - even if it wasn't a surprise from what I have read - I assume memory timings or something like that might be to blame??

 

[SpyderTracks]

Forgot to mention that I have a GTX1650 so I am completely and utterly limited by that for graphics. My CPU might not even realise that I have a GPU installed to be honest.....

So the reduction in non full-multi-core performance was interesting to see - even if it wasn't a surprise from what I have read - I assume memory timings or something like that might be to blame??

That’s expected with Ryzen because of how good auto boost is and how the cores are specced. With Ryzen they have a few high performance cores and then the rest will be weaker, the idea being auto boost is only there to help single core applications with high frequency. You’ll achieve around 5GHz through auto boost.

when applying an all core overclock you’ll only be able to overclock to the max of the weakest core which will be significantly less than the stronger cores are able to achieve. Therefor, any single threaded apps will be heavily compromised.

This is why it’s rarely suggested to overclock Ryzen, as in the main, you’ll actually reduce performance for anything outside of extreme rendering projects or scientific calculation farms.

 

[NoddyPirate]

That’s expected with Ryzen because of how good auto boost is and how the cores are specced. With Ryzen they have a few high performance cores and then the rest will be weaker, the idea being auto boost is only there to help single core applications with high frequency. You’ll achieve around 5GHz through auto boost.

when applying an all core overclock you’ll only be able to overclock to the max of the weakest core which will be significantly less than the stronger cores are able to achieve. Therefor, any single threaded apps will be heavily compromised.

This is why it’s rarely suggested to overclock Ryzen, as in the main, you’ll actually reduce performance for anything outside of extreme rendering projects or scientific calculation farms.

Thanks SpyderTracks - all makes sense.

My single core boosts are on all cores alternately on every core except core 3 - which doesn't seem to ever really get a single core workout - so perhaps that's the weaker of the 6.

However, the 5600X max single core auto boost is advertised as 4.6 GHz rather than 5.0 - I am seeing 4.65 GHz stock - slightly above the advertised - but not unusal.

So my all core OC to 4.625 GHz is limiting the single boosts to just 0.5% slower than stock - which doesn't seem to explain the performance reduction on it's own. Although maybe it's just the variances from test to test perhaps.....

I can get to 4.65 GHz with 1.3125 V but I just wasn't comfortable going above 1.3V permanently given my novice status - although again even at stock you might see well over 1.4V for short periods.....and stock settings use a higher voltage across all cores full time for single core boosts anyway.

I would say exceeding the single core boost frequency on all cores is perfectly manageable - at least in appearance......

 

[NoddyPirate]

Note to self - close Ryzen Master before testing anything. 🤦‍♂️ Standard performance back to normal with it out of the loop. Surprisingly it was also the cause of a significant drop in disk performance.

 

[SpyderTracks]

Thanks SpyderTracks - all makes sense.

My single core boosts are on all cores alternately on every core except core 3 - which doesn't seem to ever really get a single core workout - so perhaps that's the weaker of the 6.

However, the 5600X max single core auto boost is advertised as 4.6 GHz rather than 5.0 - I am seeing 4.65 GHz stock - slightly above the advertised - but not unusal.

So my all core OC to 4.625 GHz is limiting the single boosts to just 0.5% slower than stock - which doesn't seem to explain the performance reduction on it's own. Although maybe it's just the variances from test to test perhaps.....

I can get to 4.65 GHz with 1.3125 V but I just wasn't comfortable going above 1.3V permanently given my novice status - although again even at stock you might see well over 1.4V for short periods.....and stock settings use a higher voltage across all cores full time for single core boosts anyway.

I would say exceeding the single core boost frequency on all cores is perfectly manageable - at least in appearance......

Yeah, apologies, I think I was defaulting to the 5900x that was the cpu they showcased in the release videos.

 

[SpyderTracks]

There’s a really cool tool for achieving best per core overclocks on Ryzen called ClockTuner 2 if you want to experiment, this dynamically adjusts the overclock per core on all cores to achieve the absolute max possible overclock.

ClockTuner 2.0 for Ryzen (CTR) Guide

In this article, we will talk you through ClockTuner for Ryzen version 2.0, a helpful tool that allows you to further refine ZEN2, ZEN3 and Ryzen 3000/5000 performance....

www.guru3d.com

 

[NoddyPirate]

Yeah, apologies, I think I was defaulting to the 5900x that was the cpu they showcased in the release videos.

Ah yes of course - that makes sense - I couldn't remember which one of the series it was which was at 5 GHz.

With Ryzen Master closed my standard performance is normal again - but I still wouldn't want to leave the OC enabled at the 1.3V. My cores voltage do actually drop back to sleep baselines, but only when there is basically no activity at all of any notable sort. Once something happens, then it behaves exactly like the single core boost - every voltage comes up to the OC setting even for tiny workloads.

I believe that Ryzen can handle more than 1.3 on a permanent basis without degradation, but I'm not that comfortable with it being so far above it's stock values for basically 99% of the time.

 

[NoddyPirate]

There’s a really cool tool for achieving best per core overclocks on Ryzen called ClockTuner 2 if you want to experiment, this dynamically adjusts the overclock per core on all cores to achieve the absolute max possible overclock.

ClockTuner 2.0 for Ryzen (CTR) Guide

In this article, we will talk you through ClockTuner for Ryzen version 2.0, a helpful tool that allows you to further refine ZEN2, ZEN3 and Ryzen 3000/5000 performance....

www.guru3d.com

Ah dammit! Just crossed with you again! Sorry!

Cool - thanks for that - I'll look at it for some fun perhaps. Given an all core boost for me will only be when I do infrequent and odd stuff within LR or PS, any OC really just isn't worth the hassle. But learning is still fun!!

Playing with individual cores sounds a lot better than Ryzen Master I must say - but I don't really want to get to the max possible OC though - and I imagine you know why!!!

 

[SpyderTracks]

Ah dammit! Just crossed with you again! Sorry!

Cool - thanks for that - I'll look at it for some fun perhaps. Given an all core boost for me will only be when I do infrequent and odd stuff within LR or PS, any OC really just isn't worth the hassle. But learning is still fun!!

Playing with individual cores sounds a lot better than Ryzen Master I must say - but I don't really want to get to the max possible OC though - and I imagine you know why!!!

It’s good for learning how Ryzen behaves. It’s quite different to what we’ve been used to with Intel for so long.

 

[NoddyPirate]

It’s good for learning how Ryzen behaves. It’s quite different to what we’ve been used to with Intel for so long.

Just reading through that link SpyderTracks and it looks superb. As you say it looks like a great way to just better understand Ryzen behaviour. I'll have fun with that - thanks again.....

 

[NoddyPirate]

@SpyderTracks - I've been playing with ClockTuner for Ryzen today and it is great. Thanks again for passing it on!

It doesn't allow control of individual cores, but instead indiviudal core complexes - so since my 5600X has six cores in only one complex, I can't control any of them singly. However it is superb for understanding the structure and behaviour - exactly as you said. My 3rd and 4th cores are the weakest of the 6 - I figured number 3 was the poor brother but in turns out I have twins! 👶👶 They are not too far off un-overclockable (to coin a new term).

After studying the stock behaviour a bit more, I could push the whole lot together over the max single core boost of 4.6 GHz - to 4.7 GHz - at 1.360 V - which actually gave me voltages within 0.025 V of the stock behaviour. But the transient voltage spikes when unloading were getting a bit uncomfortable - around 1.4 V (EDIT - should read 1.44 V) - again you can actually get exactly the same under stock conditions too - but I would have no interest in seeing more than that without first reaching the point where I figured my CPU owed me nothing - so fun in a few years time maybe!

What I like about ClockTuner is that you set the P1 and P2 profiles but can adjust under what CPU load they kick in and how long for - all the while leaving your idle state totally stock. Scaling everthing back a little, what I have now is a normal idle (less than 0.97V) and low workload single core boost behaviour (up to 4.65 GHz on my chip), then as CPU load increases I have all the cores boost to join the single core target of 4.65 GHz, then when it gets really busy all cores drop back to 4.45 GHz at a lower voltage to keep the stress down.

In the end I have much better all core performance, the same single core performance, completely normal idle, and all with bascially the same voltages and roughly 15% higher power draw under a worst case scenario. But the best bit is that my power and temperatures are much more stable when the usage is varying - slightly higher on average perhaps, but more stable....making for less fan speed hunting all around.

But! A question for you if you are familiar with the software - It frequently gets stuck in my P2 Profile for no good reason. And it doesn't cancel it when it reaches my power limits either. So I end up with 100% load at 4.650 GHz and 1.35V - giving me peak power of around 110 W. It's perfectly manageable and isn't actually a problem, but it shouldn't be doing it. I can limit the power draw in BIOS if I have to, but any idea what's happening? 🤔

 

[SpyderTracks]

I can limit the power draw in BIOS if I have to, but any idea what's happening?

Not sure, it’s not something I’ve had the pleasure of playing with yet, just popped up in my tech feed shortly after the Ryzen 3 release and then Linus Tech Tips did a video on it which looked really promising.

 

[NoddyPirate]

Not sure, it’s not something I’ve had the pleasure of playing with yet, just popped up in my tech feed shortly after the Ryzen 3 release and then Linus Tech Tips did a video on it which looked really promising.

It is really interesting software - I’d say you’ll enjoy using it.

The tuning function is brill - you set an initial frequency and voltage - add power limits if you wish - then it will stress test at your f & V values, and then same f but V-0.06 - if it passes it will up the f and try again. If it fails it will drop the f by 25 MHz and keep going until it passes. At the end it gives a recommended combination with some headroom.

it also has a new feature called the Phoenix! Rises from the ashes in the event of a crash and carries on where it left off! 😀 I had one crash at 4.7 GHz and a low V and when it restarted it had a nice message waiting for me saying perhaps try something a little less unrealistic.... 🙂

If it sticks in P2 though I’ll just happily return to defaults - the cooling gets noisy enough to make me think I’m back at work again.....

 

[AgentCooper]

The way Ryzen processors seem to work certainly is interesting.

You technologically minded types probably read into it more than a enthusiast like myself but the way I’ve understood it (and using layman’s terms) is that they seem to have combinations of weaker and stronger cores which allow the processor to generate a peak across those strong cores and maximise utilisation. Thus creating the fabled AMD autoclock frequency boost. And that’s why manual overclocking can actually hinder performance chasing if it’s applied across all cores, it’s not allowing them to do the job Lisa Su assigned them.

 

[NoddyPirate]

The way Ryzen processors seem to work certainly is interesting.

You technologically minded types probably read into it more than a enthusiast like myself but the way I’ve understood it (and in layman’s terms) is that they seem to have combinations of weaker and stronger cores which allow the processor to generate a peak across those strong cores and maximise utilisation. Thus creating the fabled AMD autoclock frequency boost. And that’s why manual overclocking can actually hinder performance chasing if it’s applied across all cores, it’s not allowing them to do the job Lisa Su assigned them.

Yes that’s basically it in a nutshell @AgentCooper - the highest boost is reserved for the best cores only - one or two at a time.

My experience seems to show that exactly - to get all my cores to exceed the default max boost in any meaningful way would have required a lot of voltage - because the weakest cores need so much help - at a level I am sure would degrade the chip.

A standard all core boost is way below that level. What’s interesting is when and how stock behaviour moves from single to all core boost frequencies. I haven’t quite figured that out yet.

But I have managed to get a chip that is the same for single core use but better at multi core without too much trouble. The thing is that this only really helps me a very small percentage of the time. Large imports into LR is my most frequent case to see some help perhaps....

Ultimately, probably not worth it in the long term.

 

[AgentCooper]

Yes that’s basically it in a nutshell @AgentCooper - the highest boost is reserved for the best cores only - one or two at a time.

My experience seems to show that exactly - to get all my cores to exceed the default max boost in any meaningful way would have required a lot of voltage - because the weakest cores need so much help - at a level I am sure would degrade the chip.

A standard all core boost is way below that level. What’s interesting is when and how stock behaviour moves from single to all core boost frequencies. I haven’t quite figured that out yet.

But I have managed to get a chip that is the same for single core use but better at multi core without too much trouble. The thing is that this only really helps me a very small percentage of the time. Large imports into LR is my most frequent case to see some help perhaps....

Ultimately, probably not worth it in the long term.

It’s good that you’ve done this experiment, it’s a thread I’ll potentially be pointing people towards in future 👍

But it just seems to show how far AMD have come that they’ve got their processors working so well that outside of very specific use cases, no tweaking is required.

Compare that to an Intel CPU, where you’re recommending someone to go previous gen with a lower core count in order to suit their uses, achieve sustained performance and make sure you’re not going to burn their house down in the process 🙃

 

[Scott]

It sounds like you may not have got the best out of the silicone lottery. AMD actually changed their marketing in line with Intel for the 5000 series, in that they advertise the boost for the lowest core.

There have been numerous reports of the 5600X clocking in at 5.0ghz in stock form, with (I believe) GN showing single clock speeds of 5ghz in stock trim for benchmarks.

As much as the worst cores may not be selected, due to the AI noticing them as being such, they should be capable of the "Boost speed" on the box. The key is that, with appropriate cooling, you should be able to blitz right past the stock limits on a single core.

AMD chips are similar to GPUs in that they are very temperature sensitive. If you can keep the chip under 60C I think the results would amaze you.

For poops n giggles you should set a maximum cooling profile, take the side of the case etc, and absolutely go for as cool as you can.... and then run a single core test at ideal conditions. I think you would see some very nice numbers on the max single core.

 

[NoddyPirate]

It sounds like you may not have got the best out of the silicone lottery. AMD actually changed their marketing in line with Intel for the 5000 series, in that they advertise the boost for the lowest core.

There have been numerous reports of the 5600X clocking in at 5.0ghz in stock form, with (I believe) GN showing single clock speeds of 5ghz in stock trim for benchmarks.

As much as the worst cores may not be selected, due to the AI noticing them as being such, they should be capable of the "Boost speed" on the box. The key is that, with appropriate cooling, you should be able to blitz right past the stock limits on a single core.

AMD chips are similar to GPUs in that they are very temperature sensitive. If you can keep the chip under 60C I think the results would amaze you.

For poops n giggles you should set a maximum cooling profile, take the side of the case etc, and absolutely go for as cool as you can.... and then run a single core test at ideal conditions. I think you would see some very nice numbers on the max single core.

I’m not sure Scott if we still have the 5950 or 5900 on our minds?

AMD’s own ads say 4.6 GHz single core boost - GN saw 4.65 GHz in their review here - same as me:

5600X review

They could OC it to 4.8 GHz all core at 1.4 Vcore but that wasn’t stable for all tests so they have another entry for OC at 4.7 GHz.

And in their super tuned battle after 6 hours of tuning they again had it at 4.8 GHz all core with the occasional need to drop to 4.7 GHz for some tests.

5600X vs 10600K OC showdown

In terms of the silicon lottery however I most likely haven’t lucked out as I do have two cores with pretty low CPPC tags. But my results - stock or OC - still aren’t far off GN’s at all really.

The biggest difference is in the stock multi core frequency where they saw a higher average frequency than I do - by about 100 MHz. Given my stock multi core speeds are temperature dependent I assume their superior cooling for their tests drove the difference here? And perhaps that’s where your case lids off experiment would be an interesting one to look at!