Ryzen 5 3600 @ 4.6 GHz ... where to go now?

[Ben333]

With a very small vcore bump doing > 4.6 GHz. Was able to validate a CPU-Z screenshot at 4.7 on stock volts (though my MB may be auto-bumping it..?) but R20 immediately crashed that. So, I scaled back a bit and went up a bit on voltage. Then I scaled the volts back till it became unstable. Then increased it by two incremental notches. So far I've been able to run Cinebench four times without a hitch. Gained almost 600 points on there vs stock! Pretty awesome IMO.

Very comfortable temps on my 2x120 cooler master closed loop / aio system.

Should I push it? Do my voltage readings look normal? Am I doing anything wrong? Last OC was on the 775 platform.

From poking around google... seems like maybe I got lucky with this chip? Or maybe I'm finding old results.

Feedback, tips appreciated.

3600 @ 4624 MHz Stable enough to run R20 four times --> https://valid.x86.fr/1ss78b 4156 In CB R20, 20pts shy of a 600pt gain over stock.
3600 @ 4702 MHz Validated but unstable --> https://valid.x86.fr/ikl2pc (I'm sure with more voltage it is totally do-able)





Don't know how far I want to push this. Need this system \ chip to last a long time!

 

[bigtallanddopey]

Have you stress tested this overclock in more demanding software than cinebench?

If so you have one hell of a chip. I don’t think I have heard of anyone hitting 4.6/4.7 all cores on normal cooling. Most people would hit 4.3 and top out at that for all core overclocks.

 

[Ben333]

I'll hit it with 12 threads prime 95. Here's the CPU-z benchmark @ 4.6 GHz:

 

[EarthDog]

4.6 or 3.6 GHz? I've seen both in this thread.

4.6 is surprising... very few will hit that and be stable. I'd like to see this tested under something more stressful than on 15s second benchmarks as well.

4.3-4.4 GHz is typically the end of the line.

 

[bigtallanddopey]

4.6 or 3.6 GHz? I've seen both in this thread.

4.6 is surprising... very few will hit that and be stable. I'd like to see this tested under something more stressful than on 15s second benchmarks as well.

4.3-4.4 GHz is typically the end of the line.

Yeh I was getting a little confused. The screen shots show 4.6 in cinebench. Need that stress test done. If it passes that, this is probably the best 3600 out there.

 

[Ben333]

Fixed one mention of "3.6" let me know if I missed any haha. 4.6 is new to me

Been running Prime95 blend the last 6 or 7 minutes, and will let it go a few hours. So far so good.



- - - Auto-Merged Double Post - - -

Ok, so after 9 minutes 40 seconds, P95 threw some errors and dropped to 50% usage. No crashes though. What do you guys recommend? Bump the vcore up by a notch or two?

 

[EarthDog]

There's 3.6 and 3.7 in the first post still.

Blend is nice for memory... what about Small FFT for the CPU? What are temps?

EDIT: Since you failed ~10 mins in on blend, I highly doubt it will pass small FFT for the CPU longer. Before making recommendations, we need to know where your temps are at your current settings?

Are you saying you did this with voltage to auto, right? Yes, the board will change the voltage if you raise the multiplier while it is on auto.

 

[Ben333]

rofl... how did I miss that!

I'm doing the "Max heat" right now (Small FFT). "Smallest FFT" at 4.6 would work for a bit, but error out on 2 - 3 threads. I'm down to 4364 MHz on stock volts or 1.312 according to CPUZ. This has been going over 5 minutes and temps are leveling out with a max of 92C. Too hot for my liking.

So, "prime stable" it was not. But I bet I could do some solid benching or gaming with a really nice OC. Would love to get this on chilled water... but not happening. For a couple years at least.

- - - Auto-Merged Double Post - - -

OK, so now holding 4442 MHz at 1.312V, 93C, SmallFFT x12 threads running, no errors so far. I'm done pushing it on this cooling

But whatever... 4.44 GHz is still not bad. And I think when I get my real custom WC loop up and going, this baby will be running happily at 4.5 / 4.6 GHz.

2nd edit:

4468.31 MHz is the best I can do with no errors for a 12 thread Small FFT torture test. 94C max temp. So, call that my "stable max" till I get better cooling and can push more volts. Btw, it is doing that 4.46 GHz with just 1.312V

 

[EarthDog]

4.3 GHz give or take 100 MHz is about right.

94C is too hot (to me) for stress testing. Keep that bad boy around 90C peak.

I don't know if these short runs are stable, but stable is in the eye of the beholder anyway. I highly doubt you'd do any gaming (for long) at 4.6 Ghz. Benching short bursts... sure! But yeah, stability for MY USES, is running AIDA64 stress test (CPU, FPU, Cache checked off) for a few hours. This, for my system and needs (gaming, light photo/video editing), works well.

I wouldn't bet that 1.3V is for 4.6 GHz, though it could be (that would be the lowest I've personally seen on these chips). These applications sometimes don't read vCore properly. I'd try to MANUALLY set vcore and see if there is a change in software.


(I changed the 3.6/3.7 to 4.6/4.7 in the first post)

 

[Ben333]

Doh Ty for changing that, been busy tweaking this thing and forgot. Been running at 4468 for several minutes with Small FFTs going. Used software to drop the voltage all the way down to 1.232 V. Much cooler now. And I agree with you, I don't leave an overclock on a system unless it is truly stable... I know I get furious when any work gets lost. So I usually get stable, then dial it back a tad for safety. And... while typing this, oh gee, an error haha.

So gonna bump the volts back up a tad. But For the most part, we are done here for today. The last couple hours have been quite fun, it has been a while.

 

[Ben333]

So, good news I suppose.Just finished a 20 minute 12 thread SmallFFT load.

4442.33 MHz (42.75*103.91) @ 1.232 V (CPU-z) and max temperature of 79C. Idle is 26C after ending the stress, and may drop as the fluid cools.

Awesome



This seems like the most sensible place to leave it. Very low voltage, and a nice little boost. Gotta get the custom loop started

 

[trents]

I'm not sure 20 minutes of stress testing, even Prime95, is sufficient. I would go at least an hour. But maybe you should wait for that until the custom loop is built.

 

[EarthDog]

I'm not sure 20 minutes of stress testing, even Prime95, is sufficient. I would go at least an hour. But maybe you should wait for that until the custom loop is built.

Agreed... 20 min stress testing is nothing. It has to go for at least a few hours for me to start calling it stable.

 

[trents]

Yeah, we often get posts about unusually high overclocks that turn out to be pretty pedestrian when some serious stress testing is done.

 

[Ben333]

What I really meant was more of, "off to a good start". I'd do 12-24 hours before calling it "100%" stable. But under normal conditions that level of stress is never reached.