1usmus ClockTuner for Ryzen

[porina]

Who's going to give this a try when it releases later today? I make that 5pm UK, noon Eastern US, 9am Pacific if I haven't messed up my time zones.

 

I was and remain sceptical as to what it can do since Linus hyped it in the video below. However I'll give it a chance. Maybe I misunderstand what it does, but as much as the potential performance increases would be nice, I'll be focusing on stability. Unlikely to start testing until tomorrow though.

 

 

 

[Guest]

Doesnt support the processors I have, so unfortunately wont be able to try this. 

 

Since the app is only focused on seemingly a single generation of Ryzen chips, it doesnt look too appealing from here.

 

Hopefully it works great for those who do have chips that are supported and gives that extra "free" performance above and beyond the Ryzen cpu capability with current software and bios overclocking!!

[porina]

11 minutes ago, ShrimpBrime said:

Since the app is only focused on seemingly a single generation of Ryzen chips, it doesnt look too appealing from here.

It covers Zen 2 desktop CPUs which would probably account for the majority of performance minded Ryzen owners. If you're still on older gen than that, you're choosing value over performance.

[Skiiwee29]

I was definitely intrigued by this when Linus got early access on it. I will follow it I think for a bit before deciding if its worth it or not for me and my 3700x. 

[Guest]

2 hours ago, porina said:

It covers Zen 2 desktop CPUs which would probably account for the majority of performance minded Ryzen owners. If you're still on older gen than that, you're choosing value over performance.

My 2700x flagship processor was 350$. 

 

It's a whopping 10% slower at stock then Zen 2 equivalent. 

 

I gotta say it's very decent performance to the dollar.... if I bought it today.

 

...... 

[SFFDesigns]

I will try it and probably keep it

 

1usmus is well known and I bet this utility will be pretty cool. Per CCX OC will definitely be useful and PBO puts a lot of voltage through the chip!

[porina]

2 hours ago, Hymenopus_Coronatus said:

Per CCX OC will definitely be useful and PBO puts a lot of voltage through the chip!

That's my concern. AMD don't put voltage through a chip for the fun of it. It is to ensure it is stable under any possible workload. Reducing that may lead to instability in some situations, unless AMD really are over doing it, and there is some shaving that can be done.

[Mister Woof]

3 hours ago, ShrimpBrime said:

My 2700x flagship processor was 350$. 

 

It's a whopping 10% slower at stock then Zen 2 equivalent. 

 

I gotta say it's very decent performance to the dollar.... if I bought it today.

 

...... 

i regret not picking up the $159 Ryzen 7 2700s when they were a thing

[SFFDesigns]

I just used CTR and tweaked the profile, 4.3GHz all core @ 1.29V right now. I'm going to try to push my stronger CCX (CCX 1) later

[Zektbach]

15 hours ago, Hymenopus_Coronatus said:

I just used CTR and tweaked the profile, 4.3GHz all core @ 1.29V right now. I'm going to try to push my stronger CCX (CCX 1) later

How? Mine keeps stopping at overclock when I press start

[Deli]

Interesting. Let's see what it can do.

[porina]

Still haven't had time to try it... not looking good for this week. I look forward to seeing results of those that have tried though.

[Vals89]

Tried it on my r5 3600 and on the first run with the recommended voltage of 1.275, it settled with CCX1@4225mhz, CCX2@4200mhz. Compared to my manual OC of 4.3Ghz @ 1.325 (SET) or the default CPU settings, it looks like a straight undervolt. I did another run with CTR, but this time I set the voltage to 1.325 and it got me the same OC results as my manual OC. The tool achieved the same OC results as I did manually, but it took only 15 mins instead of a day of testing.

[Deli]

2 hours ago, Vals89 said:

Tried it on my r5 3600 and on the first run with the recommended voltage of 1.275, it settled with CCX1@4225mhz, CCX2@4200mhz. Compared to my manual OC of 4.3Ghz @ 1.325 (SET) or the default CPU settings, it looks like a straight undervolt. I did another run with CTR, but this time I set the voltage to 1.325 and it got me the same OC results as my manual OC. The tool achieved the same OC results as I did manually, but it took only 15 mins instead of a day of testing.

 

15 min?

 

It took forever to first do the diagnostic. Then it took another forever to do the first run. The default reference voltage is 1.250v for me.

 

Now doing the second run. Up the reference voltage to 1.300v.....takes hours.

[Vals89]

6 minutes ago, Deli said:

15 min?

 

It took forever to first do the diagnostic. Then it took another forever to do the first run. The default reference voltage is 1.250v for me.

 

Now doing the second run. Up the reference voltage to 1.300v.....takes hours.

It takes less time in my case, because the 3600 has 2 CCX. During the OC process I noticed that the tool increases/decreases clock speed in 25mhz increments for each CCX, then tests and if the test fails, the tool backs up by 25mhz and tests again. With more CCX's the testing until failure should take more time. Also, the reference clock speed is important, because if you set a speed close to the total limit of the chip, it will take less time for the tool to reach stable OC.

It might have been a bit misleading on my part, because I already knew the limit of my chip and set the reference speed to a similar one (reference speed after diagnostic was close though).

[Deli]

1 minute ago, Vals89 said:

It takes less time in my case, because the 3600 has 2 CCX. During the OC process I noticed that the tool increases/decreases clock speed in 25mhz increments for each CCX, then tests and if the test fails, the tool backs up by 25mhz and tests again. With more CCX's the testing until failure should take more time. Also, the reference clock speed is important, because if you set a speed close to the total limit of the chip, it will take less time for the tool to reach stable OC.

It might have been a bit misleading on my part, because I already knew the limit of my chip and set the reference speed to a similar one (reference speed after diagnostic was close though).

First time first run. Didn't mess with the settings too much. The default reference frequency is 4.075GHz, I should start higher like 4.3Ghz.

 

Mine is also 3600. Also I need to increase the LLC  from Turbo to Extreme in my gigabyte board, because it keeps said too much Vdroop.

 

Now it's stress testing 4.55GHz. Probably will crash very soon.

[Deli]

8 minutes ago, Vals89 said:

It takes less time in my case, because the 3600 has 2 CCX. During the OC process I noticed that the tool increases/decreases clock speed in 25mhz increments for each CCX, then tests and if the test fails, the tool backs up by 25mhz and tests again. With more CCX's the testing until failure should take more time. Also, the reference clock speed is important, because if you set a speed close to the total limit of the chip, it will take less time for the tool to reach stable OC.

It might have been a bit misleading on my part, because I already knew the limit of my chip and set the reference speed to a similar one (reference speed after diagnostic was close though).

CCX 2 crashes at 4.525GHz. Probably will run OCCT afterwards to be sure the OC is stable.

[Vals89]

Just now, Deli said:

CCX 2 crashes at 4.525GHz. Probably will run OCCT afterwards to be sure the OC is stable.

Yeah, it's obviously not polished enough, it is a beta after all. Multiple runs with the same reference clock speed and voltage give me different results. First run - I got 4.3GHz on both CCX, second run I got CCX1 at 4.275Ghz and CCX2 at 4.3Ghz, which points at inadequate stress testing. Also, although it detects that my CCX2 contains better quality cores, it doesn't OC CCX2 higher and even sometimes clocks it lower as it did during the default reference voltage setting (my screenshot above). 

Additional stability testing is needed in my opinion. 

[Deli]

1 minute ago, Vals89 said:

Yeah, it's obviously not polished enough, it is a beta after all. Multiple runs with the same reference clock speed and voltage give me different results. First run - I got 4.3GHz on both CCX, second run I got CCX1 at 4.275Ghz and CCX2 at 4.3Ghz, which points at inadequate stress testing. Also, although it detects that my CCX2 contains better quality cores, it doesn't OC CCX2 higher and even sometimes clocks it lower as it did during the default reference voltage setting (my screenshot above). 

Additional stability testing is needed in my opinion. 

It's Beta version. Still great and many thanks to the creator of this app, 1usmus.

[SFFDesigns]

I'm at 4.275 GHz right now at 1.275V on a 3700X. I'm going to try 1.3V or even 1.325V since thats safe and push further

[Deli]

5 minutes ago, Hymenopus_Coronatus said:

I'm at 4.275 GHz right now at 1.275V on a 3700X. I'm going to try 1.3V or even 1.325V since thats safe and push further

One thing strikes me when running CTR. The load the app puts on the CPU is much lighter than OCCT and AIDA64. When the CPU did 4.5GHz at only 70C, I think,"Wow, my CPU runs so cool." Only when I apply the OC and then run OCCT. It instantly spikes pass 80C and climbing. So, although the CPU can do 4.5GHz with safe voltage. The cooling just can't keep up in real stress test.

[SFFDesigns]

23 minutes ago, Deli said:

One thing strikes me when running CTR. The load the app puts on the CPU is much lighter than OCCT and AIDA64. When the CPU did 4.5GHz at only 70C, I think,"Wow, my CPU runs so cool." Only when I apply the OC and then run OCCT. It instantly spikes pass 80C and climbing. So, although the CPU can do 4.5GHz with safe voltage. The cooling just can't keep up in real stress test.

I think it runs Prime 95 (at least thats included with the app).

 

My OC is on a stock cooler and mine doesn't break 80 degrees. But you are right, it's just testing stability.

[Zektbach]

1 hour ago, Hymenopus_Coronatus said:

I'm at 4.275 GHz right now at 1.275V on a 3700X. I'm going to try 1.3V or even 1.325V since thats safe and push further

Is that voltage still considered safe on 3700X?

Because I’m really scared when I set my voltage above 1.3V 
 

[Vals89]

25 minutes ago, Zektbach said:

Is that voltage still considered safe on 3700X?

Because I’m really scared when I set my voltage above 1.3V 
 

do note that the voltage in CTR is VID, the actual voltage under heavy load is lower than that, even at high LLC. My voltage drops to 1.290-1.300v under load at 1.325 VID. 

Look at core voltage (SVI2 TFN) readings for the actual voltage while under load. 

There is a lot of discussion on what voltage is safe for Zen 2, but the method in the thread below looks the safest. 

 

[SFFDesigns]

4 hours ago, Zektbach said:

 

Because I’m really scared when I set my voltage above 1.3V 

1.3V is fine, nothing will kill your chip!  1.325V is the max I'd use though