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

[Scott]

Thats really fantastic work. Really nice to read you taking the time to understand the settings and not just putting in random values and hoping for the best. Tonnes of work but I feel that the end result is well worth it.

I would personally be keeping the oc on as you have it. There is no down side when your temps are under 90c imo.

One thing I would query.... how are you stress testing? Have you tried prime95 small fft? If you can run that for an hour under 80 with the OC on I would say you have cracked it. I could barely get the 9900k under 90 at 5.1ghz all core with the h115i 🤣🤣🤣

 

[SpyderTracks]

Thats really fantastic work. Really nice to read you taking the time to understand the settings and not just putting in random values and hoping for the best. Tonnes of work but I feel that the end result is well worth it.

I would personally be keeping the oc on as you have it. There is no down side when your temps are under 90c imo.

One thing I would query.... how are you stress testing? Have you tried prime95 small fft? If you can run that for an hour under 80 with the OC on I would say you have cracked it. I could barely get the 9900k under 90 at 5.1ghz all core with the h115i 🤣🤣🤣

I do have to agree entirely, the way you've approached this and picked it up so quickly is quite phenomenal, your engineering background shining through.

Overclocking is a bit of an art form, and to grasp it so well so quickly is nothing short of remarkable.

 

[NoddyPirate]

Wowsers! I wasn’t expecting those replies! Thanks guys!

Yes, I think my background pre-disposes me to enjoy figuring it all out more than the final result. Learning is fun!

But I love theory too - trying to understand something before you do it - then doing it and seeing it behave the way you thought it would - is a nice feeling. Some like to dive in and figure it out afterwards, but I get more of a kick the other way around.

I didn’t use Prime95 - I’ll look at that - thanks!

If I was happy with the outcome of a tweak and thought I could live with it (unlike the high single core voltages which just had me restart straight back into BIOS again as soon as I saw them) then my routine was:

With HWInfo open all the time to monitor parameters at 500ms intervals - first Cinebench multi-core and single core single passes to confirm improvements - then the PassMark CPU tests because they seem to involve a sudden dramatic shot of core load at the end of their compression tests which can create a short term challenge for my cooler - then a 30 minute multi core Cinebench run (they call it a stability test - not sure how accurate that description is) to get the cooler fully soaked. I’d usually have the CPU Package Temp open as a graph so I could better see the small variations over time. Once I saw a flat line for a decent 20 mins I was happy the temp was stable.

I certainly didn’t stress it for a full 60 mins - but it certainly sounds like a good plan to do it to be sure!

 

[NoddyPirate]

Oh also - as I mentioned under the RAM OC thread - I would also add my own tests for performance - like a 150 image export from Lightroom and the same 15 image panorama merge - as they’re both heavily multi core tasks - and time them for comparison.

Benchmarks seem to be all the rage - but I figured it’s real world results that matter rather than some points system in some obscure prime number cycle or whatever that some benchmark program decided is useful!

 

[Ash_]

My H150i keeps my 5800X below 65 degrees, so if you want to crack it, i suggest getting one 🤣

 

[NoddyPirate]

My H150i keeps my 5800X below 65 degrees, so if you want to crack it, i suggest getting one 🤣

That’s the cooler I want to be able to tinker with a 5950X! If only someone would give me a loan of theirs for a few hours.... 🤔

 

[NoddyPirate]

Thats really fantastic work. Really nice to read you taking the time to understand the settings and not just putting in random values and hoping for the best. Tonnes of work but I feel that the end result is well worth it.

I would personally be keeping the oc on as you have it. There is no down side when your temps are under 90c imo.

One thing I would query.... how are you stress testing? Have you tried prime95 small fft? If you can run that for an hour under 80 with the OC on I would say you have cracked it. I could barely get the 9900k under 90 at 5.1ghz all core with the h115i 🤣🤣🤣

Just running Prime95 now - it certainly generates heat I tell ya!! 🔥

I'll report back in an hour or two if I haven't used the fire extinguisher in the meantime.....

 

[NoddyPirate]

Just running Prime95 now - it certainly generates heat I tell ya!! 🔥

I'll report back in an hour or two if I haven't used the fire extinguisher in the meantime.....

@Scott - Thanks for mentioning Prime95 - it helped me understand what was happening with that odd spike in heat load when running the PassMark bench I think - a sudden voltage spike that I think I missed with HWInfo becuase of having to high an interval.

Prime95 for an hour with AVX on a peak of 77 degrees, average of 71. With AVX off - highest peak of 71 degrees, average of 67 - The voltages that program pushes you to with AVX enabled are a bit mad really. I'm happy I have the curve offset in place I tell ya! You'd also want to be sitting watching it permanently rather than leaving it run and having a cuppa. Running that test with a severe OC could cause some serious damage I would think.... 🤔

 

[Scott]

I never take my eyes off it for the first 20 minutes haha.

It's a literal stress test. You can take the peak of that program as being the peak your CPU will put out as it absolutely crushes all instructions and takes it to the max.

I would be delighted with those results mate. I would have no issues whatsoever leaving that OC on full time

You can't really do too much damage with the chips nowadays. They tend to just reboot rather than fry. As long as you are reasonable with the settings you should always have the chance to recover. It's the reason it's not recommended. Most people don't have the patience to take the steps required to get to a decent level of tune.

Ironically, when you see experienced people applying an OC you will see them jump right in the deep end of tune, proper hash bash. This is purely down to experience and having had the patience in the beginning to understand what they were doing

I started the OC on my 9900k at 5.3ghz while the wrapper was still making crinkling noises in the bin, a bit different from my approach the first time around.

 

[NoddyPirate]

I never take my eyes off it for the first 20 minutes haha.

It's a literal stress test. You can take the peak of that program as being the peak your CPU will put out as it absolutely crushes all instructions and takes it to the max.

I would be delighted with those results mate. I would have no issues whatsoever leaving that OC on full time

You can't really do too much damage with the chips nowadays. They tend to just reboot rather than fry. As long as you are reasonable with the settings you should always have the chance to recover. It's the reason it's not recommended. Most people don't have the patience to take the steps required to get to a decent level of tune.

Ironically, when you see experienced people applying an OC you will see them jump right in the deep end of tune, proper hash bash. This is purely down to experience and having had the patience in the beginning to understand what they were doing

I started the OC on my 9900k at 5.3ghz while the wrapper was still making crinkling noises in the bin, a bit different from my approach the first time around.

In fairness I am dealing with the lowest TDP 5000 chip so the thermals were always less likely to be a major headache. But I am happy that I think I found a sweetspot above which I could easily add a LOT of extra power and heat without much actual gain in performance. But at the same time there is likely more fun hiding in the Curves Optimiser if I get brave enough! I kind of feel though like it would be easy to quickly push everything to the absolute limit and not realise that you might not have added much gain for a large amount of extra silicon distress....

I guess you're right about the CPU protecting itself - things have changed a lot from the days of sparklers going off on your motherboard I imagine! I love the idea of you perhaps never having seen your 9900k in stock condition - boot first time straight into BIOS and start tweaking! Update Windows afterwards......

Thanks again for your help @Scott - a fun journey and yes having watched my parameters for the past couple of days I'll think I'll leave the OC in. The majority of the time when just browsing and gaming things are actually cooler than they were stock as a result of the lower V's (being totally GPU bound of course). It's only LR at the moment that uses the tweaks fully as far as I can tell.

 

[Scott]

I installed Windows, updates, GPU drivers, 3DMark & ran Firestrike & Timespy..... just for a baseline

Only took about 20 minutes and then I opened the bonnet.

 

[NoddyPirate]

I installed Windows, updates, GPU drivers, 3DMark & ran Firestrike & Timespy..... just for a baseline

Only took about 20 minutes and then I opened the bonnet.

Ha ha! That's great!

 

[NoddyPirate]

Another add on from me:

I've been playing the EDC setting on my OC - and it is not behaving the way I expected. EDC is bascially a limit on the peak amount of power (in Amps) that the motherboard VRM can supply to your CPU.

At stock on my 65W TDP Rzyen CPU, the values are 76W for PPT and 90A for EDC. When I increase the EDC limit on it's own - the Amps supplied does indeed increase, but the power consumed in Watts remains the same - so I expected there would be a corresponding difference in Voltages somewhere - because W=A x V - so if the Watts are the same, then as the Amps go up, the Voltage must come down. But alas - I can see no difference on any power parameter of any sort across the entire HWIfno suite of parameters nor in the more basic layout of RM. No change to VCore, no change to individual Core VID's, no changes to thermals, nothing of note. (Although, how I wish I had proper VRM monitoring on my mobo.... )

If I keep increasing it, then ithe actual EDC power eventually maxes out at 107A drawn even if I enter a EDC limit higher than this. The PPT limit is the limiting one now and I need to increase the PPT limit to allow the EDC Amps to increase higher.

If I reduce the EDC, there is again no change in parameters, but eventually as you reduce it further then everything starts to throttle back, including the PPT draw - which is the reverse of the above and makes sense.

So with a given PPT - there is sort of a range of usable EDC limits you can enter.

However, all core boost performance is affected slightly. As the EDC goes up above the default value, the max allowed boost clock reduces. Fidling around with it - it turns out that the stock 90A setting for EDC is a sweet spot for my chip at that PPT power draw. Higher or lower values for EDC results in lower boost frequencies.

As an example, previously I had EDC set to 101A for my OC - and was boosting to around 4.550 GHz all core give or take. With it back at 90A, I now have them just under 4.60 GHz or so. That includes the effective clocks also - so it isn't just a misleading visual presentation of better speeds like you get with the ridiculous ASUS F-Max Enchancer

Ultimately, I have no idea what is going on - if anyone knows I would love more info 🙏 - but for now it seems if you are fiddling with these things, start with the PPT setting alone at first and work from there. Don't change them all at once as I initially did.

I think the EDC is behaving like the Curve Optimiser in some ways - reducing it can allow more headroom for clock speed boost within the same voltage and thermal limits - to a point - beyond which everything throttles back down again. I just don't yet know why!

 

[NoddyPirate]

Right so - the plot thinkens some more.

The only logical thing left to do was to strap a power meter onto the whole thing and see what the system was pulling from the wall.

With the EDC limit set to 90A - and HWInfo comfirming the CPU EDC draw is at that limit - my system pulls 140W total. (Pathetically low really! )

With the EDC limit set to 105A - and HWInfo confirming the CPU EDC draw is at that limit - my system plulls 140W total. (Wait, what? 🤔 )

So - either:

My assumption and understanding of what HWInfo is showing me is wrong, or

While the overall power is the same, the distribution of that power through the VRM or to the CPU socket may be different but beyond the capabilities of HWInfo to properly indicate how, or

HWInfo is just giving me incorrect reported values.

Answers on a postcard please.......

 

[NoddyPirate]

Hmmmm.... more ponderings.

I couldn't help myself and went playing with LLC, CPU Power Phase Control and VRM Switching Frequency a little bit. The latter two really didn't make any noticeable difference to anything in particular - but increasing the Switching Freq did cause a minor loss of clock speeds - in the region of 10-20 MHz or so. This suggests, somewhat counter-intuitively, that reducing the switching frequency would actually improve stability on a marginal OC - the opposite way around to my guess. This does seems to be supported by some of the limited info I can find about the subject. Another days work to figure all that out perhaps.

So, LLC then - the available settings are Auto, then - Regular, Medium, High & Extreme - I assume because 1, 2, 3, & 4 just weren't considered exciting enough by ASUS. 🤔

Different settings made no discernable difference to the spike voltages reportable by RM or HWI - I'm sure there is a difference, I just can't see it. It is obvious too that Auto really just means Regular - at least for my current set of tweaks - as these two settings give identical results across the board when it came to other parameters.

@Scott - I'll need your eye on this next bit to try to help me understand it! What was a total surprise was that as I increased the LLC beyond Regular, the average sustained core voltage under load went down, and dragged the boost clocks down with it (as you'd expect really with less voltage under a PBO OC specifically - it wouldn't reduce clocks with a manual OC I expect, it would just make it more marginal.) This was true for both all core and single core boosts - although more clearly for the all core case as the single core voltage jumps around that bit more and it's harder to quantify.

All told, the difference in voltage during all core boost between Regular and Extreme LLC was 30mV in favour of the Extreme setting - which is quite a lot - making LLC's overall effect on everything totally backwards to what my head expected. What is going on here? 😲

What it does mean though, is that with the CPU LLC set at Extreme, I could go back in to BIOS, and up the PPT limit to let the CPU draw more power and get the boost voltages back up to where they were before the LLC was changed. This unexpectedly has let me recover more boost frequency than had been given up through the LLC increase in the first place and the net result:

My stock 5600X which orginally out of the box giving me 4.25 GHz all core and 4.65 GHz single core at 76W, is now at 4.625 GHz all core and 4.825 GHz single core at 100W. Not bad me thinks!

(In case Ubuysa is reading this 😬, I have reset everything back to where it was for now and will continue to let the machine settle some more after the recent BSOD dramas. There is clearly much more to understand in all this stuff before I go fiddling some more!)

 

[AgentCooper]

Hmmmm.... more ponderings.

I couldn't help myself and went playing with LLC, CPU Power Phase Control and VRM Switching Frequency a little bit. The latter two really didn't make any noticeable difference to anything in particular - but increasing the Switching Freq did cause a minor loss of clock speeds - in the region of 10-20 MHz or so. This suggests, somewhat counter-intuitively, that reducing the switching frequency would actually improve stability on a marginal OC - the opposite way around to my guess. This does seems to be supported by some of the limited info I can find about the subject. Another days work to figure all that out perhaps.

So, LLC then - the available settings are Auto, then - Regular, Medium, High & Extreme - I assume because 1, 2, 3, & 4 just weren't considered exciting enough by ASUS. 🤔

Different settings made no discernable difference to the spike voltages reportable by RM or HWI - I'm sure there is a difference, I just can't see it. It is ovious too that Auto really just means Regular - at least for my current set of tweaks - as these two settings give identical results across the board when it came to other parameters.

@Scott - I'll need your eye on this next bit to try to help me understand it! What was a total surprise was that as I increased the LLC beyond Regular, the average sustained core voltage under load went down, and dragged the boost clocks down with it (as you'd expect really with less voltage under a PBO OC specifically - it wouldn't reduce clocks with a manual OC I expect, it would just make it more marginal.). This was true for both all core and single core boosts - although more clearly for the all core case as the single core voltage jumps around that bit more and it's harder to quantify.

All told, the difference in voltage during all core boost between Regular and Extreme LLC was 30mV in favour of the Extreme setting - which is quite a lot - making LLC's overall effect on everything totally backwards to what my head expected. What is going on here? 😲

What it does mean though, is that with the CPU LLC set at Extreme, I could go back in to BIOS, and up the PPT limit to let the CPU draw more power and get the boost voltages back up to where they were before the LLC was changed. This unexpectedely has let me recover more boost frequency than had been given up through the LLC increase in the first place and the net result:

My stock 5600X which orginally out of the box giving me 4.25 GHz all core and 4.65 GHz single core at 76W, is now at 4.625 GHz all core and 4.825 GHz single core at 100W. Not bad me thinks!

(In case Ubuysa is reading this 😬, I have reset everything back to where it was for now and will continue to let the machine settle some more after the recent BSOD dramas. There is clearly much more to undersatnd in all this stuff before I go fiddling some more!)

I love your fearless and ‘can do’ attitude to this kind of thing. I’m glad you’ve made the decision to let your system have a rest though. It’s probably all pale and jaded, like a Vietnam veteran who’s seen bad things.

 

[NoddyPirate]

Just writing all that made me wonder if the higher LLC setting simply gives a more useful and stable power supply to the CPU which means it doesn't need as much voltage to operate under the similar set of conditions. This then let's it have more headroom again at higher power.

If this were true then it would probably more than compensate for any higher spikes that might come with it.

Also - the nature of the PBO Overclock most definitely makes everything behave almost completely the opposite to what you would see with a Manual Overclock. The benefits are all still there - they just present themselves the other way around so to speak......

 

[NoddyPirate]

I love your fearless and ‘can do’ attitude to this kind of thing. I’m glad you’ve made the decision to let your system have a rest though. It’s probably all pale and jaded, like a Vietnam veteran who’s seen bad things.

View attachment 25234

Ha ha! Yes!

"Ah great! I'm out of the box on that shelf in PCS and finally in a PC doing what I was made for. A few trillion calculations before lunch will be fun. And this owner guy seems nice enough. A bit odd, but then he is Irish...."

.............one month later..............

"WHAT THE HECK IS HE DOING!!!! Save me GPU I beg you! I didn't mean those things I said to you in the FedEx warehouse!!"

 

[Scott]

I'm surprised by your findings as well. Perhaps that's why I had next to no joy with it when I fettled with it too. I was expecting it to allow a more comfortable overshoot when spiking a load but it didn't help me much either.

Perhaps as you have found, it's actually just a much tighter control of the voltages. This would be a good thing, obviously, but I can't for the life of me think why you would want anything but extreme? I can't see the down-side of the behaviour you have exhibited with the LLC setting on max.

I believe, and I could be completely wrong here, that the spikes are sometimes noted when the load drops off the CPU..... rather than when it's being pushed. Not sure if your polling data will catch that or not though. Perhaps that's the catch. Again, no-load voltage spike wouldn't really worry me either.

 

[NoddyPirate]

I'm surprised by your findings as well. Perhaps that's why I had next to no joy with it when I fettled with it too. I was expecting it to allow a more comfortable overshoot when spiking a load but it didn't help me much either.

Perhaps as you have found, it's actually just a much tighter control of the voltages. This would be a good thing, obviously, but I can't for the life of me think why you would want anything but extreme? I can't see the down-side of the behaviour you have exhibited with the LLC setting on max.

I believe, and I could be completely wrong here, that the spikes are sometimes noted when the load drops off the CPU..... rather than when it's being pushed. Not sure if your polling data will catch that or not though. Perhaps that's the catch. Again, no-load voltage spike wouldn't really worry me either.

Thanks so much Scott.

Yes, I expected a load and unload impact, certainly not such a visible sustained load difference. Quite weird!