CPU running cooler after overclocking?

[GrannyDane]

I've recently switched from i7-4790 to Ryzen 5 3600, and since it's an unlocked CPU, I figured that I want to try overclocking it lightly. With no knowledge or previous experience, I started by raising the base clock to match the boost clock of the CPU and lock it there, which was a shift from 3.6 GHz base to 4.2 GHz. No other settings touched, just raised the base clock on all cores and disabled the clock boost.

 

The performance did get better with about 400 higher Cinebench score, but when I stress tested it using AIDA64 Extreme to see if it's stable, I noticed that the CPU temperature was noticeably lower than it was before.

 

Here's the result after 10 minutes of stress testing before overclocking:

 

And here's the result after:

 

The max CPU temperature is 16 degrees lower and the max core voltage is 0.3 volts lower after overclocking, which seems quite bizarre and counterintuitive. It is also worth noting that the CPU voltage was quite inconsistent before overclocking, ranging from 0.948 to 1.320 volts throughout, whereas it was almost constant after overclocking at 1.044 to 1.092 volts.

 

Any idea why this might be happening?

 

Edit: If you need my PC spec, please refer to my profile page.

Edited August 21, 2020 by GrannyDane
Specs matter, right?

[Guest]

I have some issues with the evidence this cpu is golden as could be loaded AVX instructions at 4.2ghz running 1.080v

 

1st issue are the screen shots. I cant see the multiplier of the processor. There's just no frequency information at all except you saying 4.2ghz.

 

Can you provide screen shots of OCCT with AVX 2 enabled all threads please? 

And maybe monitor with HWInfo64 so we can see everything that's going on. 

 

But yeah!! Other than that.... you have the top golden sample!! 

 

 

[Mark Kaine]

And now we wait...

 

 

If this works I'm doing it too!   How many fps could I gain by this on average?  

I suppose like 5 max? 

[GrannyDane]

Sorry for updating it late guys, I played MHW all day yesterday to see if the system stayed stable with the OC. (I know it's not a particularly CPU-demanding title, but I gotta have some fun, right?) I couldn't find any problem in the game.

 

16 hours ago, ShrimpBrime said:

Can you provide screen shots of OCCT with AVX 2 enabled all threads please? 

And maybe monitor with HWInfo64 so we can see everything that's going on. 

So I ran the test you've suggested, although since I'm not familiar with either of the software, I don't know if the screenshots I took does enough job for you. Let me know if you need anything else.

 

First I ran it with the OC enabled, HWInfo running on the side:

 

As you can see, it gave me lots of errors with the OC. I did a quick Googling about this and found out that there's some issue between HWInfo and OCCT on 1st gen Ryzen, and although mine isn't a 1st gen, I figured that it's still worth a shot.

 

I ran the second test with AIDA64 graph on the side instead (no stress test running):

It gave me no error up until about 4:30 mark, but then it started to produce errors and crashed not long after. Maybe because It was right after 10 minutes of stress testing?

 

I rolled back the settings (with a reboot) and ran the stress test again, and:

and when I re-ran it with HWInfo:

No errors detected even with HWInfo.

 

So there's the 'gotcha' I guess. The next thing I can do is probably try increasing the voltage with OC since it was running at less than 1.1V while non-OC ran at nearly 1.3V anyway.

[Guest]

It's running 4ghz at 1.250v. 

So the chip is throttled when loaded and not at 4.2ghz.

 

Looks like each test run was 00 seconds long. Why are all the counters at 0? 

 

Havent gotten anyone yet. So far I havent seen anything at 4.2ghz, besides idle clocks. At load looks like the frequency drops to accommodate the low vcore 

[GrannyDane]

24 minutes ago, ShrimpBrime said:

Looks like each test run was 00 seconds long. Why are all the counters at 0? 

The first few tests that says 10:00 were running on infinite time and I manually canceled it at 10 minute mark (that's why it says 10:00), and the rest of them that says 0:00 is because I figured that I can actually make it run for a set number of minutes so I took the screenshot once the countdown stopped.

2 hours ago, GrannyDane said:

Those two images are from same test. You can see that the second image says "Timed" for "00:10:00" on the top left.

 

24 minutes ago, ShrimpBrime said:

Havent gotten anyone yet. So far I havent seen anything at 4.2ghz, besides idle clocks. At load looks like the frequency drops to accommodate the low vcore 

 

2 hours ago, GrannyDane said:

First I ran it with the OC enabled, HWInfo running on the side:

Well, the minimum clock says 4,192 MHz on the first image under "Core 0 Clock" and so on. Isn't that enough?

[Guest]

5 minutes ago, GrannyDane said:

The first few tests that says 10:00 were running on infinite time and I manually canceled it at 10 minute mark (that's why it says 10:00), and the rest of them that says 0:00 is because I figured that I can actually make it run for a set number of minutes so I took the screenshot once the countdown stopped.

Those two images are from same test. You can see that the second image says "Timed" for "00:10:00" on the top left.

 

 

Well, the minimum clock says 4,192 MHz on the first image under "Core 0 Clock" and so on. Isn't that enough?

Yeah that's good. Got a lot of errors. 

 

Does a small increase in vcore eliminate the errors? Something like 1.125v or something?

[GrannyDane]

So I tinkered with the voltage for about 2 hours, and settled on a mere +0.012V (I could probably try to lower it, but I'm pretty tired after sitting in front of computer doing nothing for 2 hours) with no errors:

I also retried after setting the voltage to +0.000 V (instead of 'Auto' that it was set to) and it started to spit out errors after less than a minute.

 

Bit hotter than it was before, but still cooler than the stock settings! The title still holds true!

[GrannyDane]

Also now the saving button is missing on OCCT for some reason… Maybe related to the pop-ups that tell me to make $1 donation at the start of the stress test?

[Guest]

1 minute ago, GrannyDane said:

Also now the saving button is missing on OCCT for some reason… Maybe related to the pop-ups that tell me to make $1 donation at the start of the stress test?

Yeah that's possible, or the system isnt actually stable, the latter part probably true.

 

2 hours testing all this isnt enough. I'd think 2 hours spent on a single setting to test stability would be ideal. 

 

 

[GrannyDane]

8 minutes ago, ShrimpBrime said:

2 hours testing all this isnt enough. I'd think 2 hours spent on a single setting to test stability would be ideal.

Yeah, I guess I'll leave it to run while I'm not using the computer. Right now I want to play some more MHW…

[Guest]

5 minutes ago, GrannyDane said:

Yeah, I guess I'll leave it to run while I'm not using the computer. Right now I want to play some more MHW…

Good chip though! 

 

Nice low leakage. As long as the temps are low, itll leak less. Nice work.

[GrannyDane]

I let OCCT run for a while and came back to take a look, and it had something like 3000 errors at 44 minute mark. Raised the voltage to +0.024 V, and:

All seems alright to me. I mean, I could let it go for longer, but I'm pretty sure it won't give me a real problem anyway considering how many cases of 'OCCT spitting out errors but having no issue for half a year,' people on the web seem to have. The temperature doesn't seem any different either, so that's pretty much all I could wish for.

[Guest]

7 minutes ago, GrannyDane said:

I let OCCT run for a while and came back to take a look, and it had something like 3000 errors at 44 minute mark. Raised the voltage to +0.024 V, and:

 

 

All seems alright to me. I mean, I could let it go for longer, but I'm pretty sure it won't give me a real problem anyway considering how many cases of 'OCCT spitting out errors but having no issue for half a year,' people on the web seem to have. The temperature doesn't seem any different either, so that's pretty much all I could wish for.

Hell yeah. Looks good from my seat