3600 @4.4 gained 15°C

forteh

forteh

forteh

Associate
Joined
18 Jan 2010
Posts
224
Location
Lichfield
I've recently noticed my r5 3600 running hotter than it used to. Built it in summer 2020 and got a good stable overclock of 4.4 with peak temps of 66°C in prime 95.

I'd reset my bios end of last year and forgot to make a note of my overclock settings, I found my original thread on here and reset the overclock accordingly.

https://forums.overclockers.co.uk/threads/r5-3600-overclocks.18897190/

Now in prime 95 I'm getting instability within 5 minutes or so and temps are spiking up to 82°C! I've tried bumping the voltages and LLC up and it's not helped. Could the PSU be failing to supply enough current? It's a fairly old antec 550w and admittedly it's possibly at the low end of capacity now I have a 3070 as well?

Cooling is the same, thermalright true spirit 140bw but I now have an extra 140mm fan blowing over the ram, CPU and vrms. The heatpipes are cold to touch, as if they've stopped conducting. Is this a thing?

edit: heatsink still works, if I turn the fan off it all gets toasty hot. CPU vcore doesn't budge an inch when the prime 95 worker fails.

edit: the only hardware configuration change is I've gone from a 1070 to a 3070 and gone from 16gb 3200 to 32gb 3600. Might I need more voltage/LLC on the north bridge or fabric clock?
 
Last edited:
Heatpipes will be cold / room temp at ends. To feel heat you need to touch heatpipes between cooler base and fins.

Heat from CPU moves up thru cooler base into heatpipe boiling coolant expanding it from liquid to vapor that moves toward ends of heatpipe where it condenses back into liquid in wicking on inside of heatpipe radiating heat into fins. This liquid then wicks back to cooler base where cycle starts all over.

Not sure what problem may be. Is there any dust / film on cooler fins .. on fan blades? When was system last cleaned? Are case filters / vent grills clean?
 
Heatpipes are cold right down to the base block, I would anticipate that they would have been at least warm given the CPU temperature. Turning the fan off got the heat going through them and they got hot right up to the end, but by that point it was up to 85.

The fans are clean and no filters, motherboard is fitted into an old mobo tray that I cannibalised and turned into an open wall mount case. Heatsink fins are also clear.

When I switched the thermalright from my old 2500k to the 3600 I found that there was quite a lot of concavity on the heatsink base, it appears that the intel IHS wasn't as flat as the AMD so the was a gap in the middle. I lapped the bottom of the heatsink on glass set onto a surface plate and it dropped the temps dramatically.

I don't think the heatsink has since warped as it still suctions into the IHS with minimal paste.

I have picked up a spare thermalright 140 power (15 quid on your favourite auction website) that I'll test but it's almost as though the IHS isn't conducting through to the heatsink properly. I cleaned and repasted it last night with artic silver ceramique that I've had for about 15 years and it's settled the temps a little bit it's still warmer and completely unstable in prime where as it wasn't 18 months ago.
 
Kinda strange because normally heatpipes on a hard working cooler are warm to touch between base and fins. Difference between no fan and fan shows cooler is obviously working.

"Open wall mount case" is not likely to have any problem with air entering cooler being pre-heated.

Most coolers (especially Thermalright) have slightly convex (higher in center) base. convex cooler bases started shortly after heatpipes started being used because most CPU had concave IHS surfaces. Lapping, especially when both IHS and cooler base are lapped almost always improve their mating / better heat transfer / lower temps.

I don't think base shape ever changes after assembly and coating. At least I've never seen a cooler base change shape over time.

3600 has soldered IHS so not likely that's the problem.

I assume you R&R'ed cooler cleaning base and re-seating with new TIM?

Do problems change when running stock settings vs overclock settings?

TRUE Spirit 140 Power are one of the very best coolers ever made.

Edit: Definitely a strange problem.
 
Last edited:
Well I've turned all the fans up to 100 and
vcore 1.3v LLC mode 8 - previously 1.25v mode 1
NB/soc 1.175 mode 8 - previously auto with no llc

Basically turned all the voltages up to see what it's doing.

Fresh into Windows, vcore is at 1.304 but as soon at prime starts the torture test this drops down to 1.288. I was under the impression that llc was there to prevent it dropping volts under load, stop prime 95 and it's back up to 1.304.

PSU degraded enough to not supply enough juice on vcore?
 
It's fresh on last night, arctic silver ceramique, admittedly ancient but do they have a shelf life?
With all fans running maximum it's peaking at 72°C now but obviously with much higher voltage than I normally use. Previously it had peaked at 66 after 4 hours of prime and fans were set to auto so span right down.

The more concerning thing now is why is the overclock no longer stable, even with more voltage than it ever needed before? Small FFTs error instantly on core 1 & 6, two fastest cores are 2 and 5.

edit1: reset the clock speeds and voltages to auto and it crashed prime instantly...
I've now reset the ram to completely auto and turned off xmp and it's seeming more stable initially. It's now running at 1333 20-19-19-43. The previous overclock was on 16gb 3200 ram, this time it's on 32gb 3600; perhaps my cpu doens't like 1800 fclck? Do I run the ram slower and see if I can get silly tight timings? It's ballistix 3600 cl 16-18-18-38. How do I make the fclck more stable at 3600?

edit2: with auto voltages vcore has jumped up to 1.424 peak and NB is up to 1.12, peak cpu temperature is 74°C with these silly voltages so temps should be well under control if I can get it stable back on my original voltages.

edit3: loaded the 4.4ghz overclock again but this time with ram running at 3200 with the xmp timings, temps have peaked at 68°C but prime95 has errored again...
 
Last edited:
More playing with this one...

I reset bios to optimised defaults, all stable for 10 mins with temps reasonable between 62-72 on full load boosted up to around 4.1Ghz.
Manually reset the 4.4ghz overclock leaving all voltage options on auto, it didn't error within the 10 mins or so but it pushed the vcore up to 1.524v and hit 99°C! At which point I stopped it.

Reset the manual voltage override to 1.4v and it pretty quickly gave errors on cores 1&6 again.

I did update the bios to latest version back in november but is that likely to have affected the ability of the cpu/mobo to overclock?
On further testing with the cold heatpipes, if I turn off the 140mm fan that's feeding cold air to the ram and conversely into the cpu heatsink then the heatpipes start to get warm once the cpu is over 80°C, it must simply be the airflow cooling them before the heat gets into the fin stack. It will be interesting to see if the replacement cooler behaves in the same manner.
 
Made a little headway this afternoon.
Comparing ryzen master from my screenshot from my previous thread I noticed that my current PPT, TDC and EDC maximum range values were much lower than previously. Turns out I had forgotten to enable PBO when I had updated bios and rebuilt the overclock.

Additionally the MSI LLC modes have no indication as to what each level does, turns out that 1 is highest and 8 is lowest. Whichever setting doesn't seem to fix the issue.

At 1.25 vcore LLC1 it's a no go, at 1.375 vcore LLC1 it seems much more stable but temps are soaring up to 95°C no matter the fan speed.

The only other thing it can be is the ram. When I did the original overclock, I had 2x8 teamgroup vulcanz, I upgraded to 4x8 a few months back. I switched the 4x8 teamgroup Vulcanz 3200 to 2x16 crucial ballistix 3600 last week (borrowed the ballistix from work to test the effect of lower memory latency in DCS vr). I've tested it at xmp timings at 3200 and it's still unstable, even with dram pumped up to 1.4v and the infinity fabric set to 1600mhz so it's synchronous. I'll switch the ram back during the week and see if I can reattain the 4.4ghz with the teamgroup ram.

Wtf is going on with this thing in the meantime?

edit: I should point out that I never have any instability in windows with the [email protected]. without the overclock the CPU boosts to 4.1ghz which costs 2ms CPU frametime in DCS so if rather keep the overclock running if I can, it's just going to bug me that it's not prime stable though :(
 
Last edited:
Reset BIOS to stock default completely and start from scratch. Put it on 4.2ghz see what happens, without changing anything else at all.
 
That's effectively what I've been doing, it's gone up to 4.3 @1.25v at the moment and done a quick and dirty hour of prime without any issues. I've left it running all day so will see whether our not it's looking term stable.

If I take out one of the ram sticks so only driving 16gb then the temps are almost 10°C cooler, I guess that it must be the IMC heating up with the additional load of 32gb.
 
I feel that I may have found the source of my extra temperature!
Embarrassingly so.....

I'm fairly certain that the version of prime95 I used before didn't have the avx instruction options.

Turning them off results in a 10°c drop in small ffts testing giving a peak temperature of 68°c at 4.3ghz, I would be willing to put (a small amount of) money on it being prime stable at 4.4 without avx aswell. Will be testing it tomorrow.

The additional couple of degrees could either be ambient or due to increased ram load, either way I'm happy with the temps now, idles at 32°c :)
 
Glad you got it sorted!
Realistically there is no noticeable difference between 4.3 and 4.4 so might just be worth leaving it there especially if it means lower voltages and temps. 0.05v is quite a bump for 100mhz, which is less than 2% increase.
 
Ran 8 hours of non avx prime95 blend torture test today without a hitch, 4.325ghz 1.25v 67°C peak temperature . I'll take the 75mhz reduction in clock speed for having 32gb ram installed.
 
forteh said:
Ran 8 hours of non avx prime95 blend torture test today without a hitch, 4.325ghz 1.25v 67°C peak temperature . I'll take the 75mhz reduction in clock speed for having 32gb ram installed.
Click to expand...

That poor CPU - no need to run prime 95 for 8 hours. All you've done there is give your electricity bill a boost. p95 is an unrealistic load on the CPU. Do you just play games on your computer?
 
Well I've been using prime to test stability for the last 15 years or so and I've always had stable overclocks so just stuck to what I know :)

Yes it's massively heavy load compared to anything that I'm going to use it for but I also know that it's unlikely to fall over. The whopping increase in the electricity bill will be less than 50p, I think I'm willing to make that sacrifice :D
 
I've frequently found (I'd say 2/12 overclocks) that Prime95 is stable for a few hours but finds an error in a 24 hour test.

I'd rather pay a few quid in electricity and know my computer won't bluescreen testing with an 'unrealistic' workload.
 
forteh said:
I've recently noticed my r5 3600 running hotter than it used to. Built it in summer 2020 and got a good stable overclock of 4.4 with peak temps of 66°C in prime 95.
Click to expand...

that must have been a bug, no way you got max temps of 66c in summer with an OC, a stock 3600 would be over 66c in prime
 
You must log in or register to reply here.
Back
Top Bottom