• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Question Ryzen 2600 OC @ 4.1GHz - 1.35v to 1.256v @ 70c in Prime95

VforV

Member
Oct 19, 2019
33
1
11
Hey guys, since 2 days now I've been playing with my new setup:

CPU: Ryzen 2600 BOX
MB: MSI B450M MORTAR MAX
RAM: Crucial Ballistix Sport LT 16GB Kit (8GBx2) DDR4 3000 CL15
CPU Cooler: ARCTIC AC Freezer 34 eSports DUO White

I started with CPU OC and I'm still yet to find the limits. My memory is running XMP2 @ 2933MHz CL16 (not great, but for now I leave it like that until I finish my CPU OC).

So I got to this point today: my Ryzen 2600 OC @ 4.1GHz - 1.35v + LLC4 and after 10 min of Prime95 Small FFTs @ 70c temps, it get as low as 1.256v on all cores. I plan on doing longer stress tests, but I'm asking: is that low voltage safe for this OC?

I have Core Boost, PBO, C-State, Spread Spectrum and Cool' n' Quiet all Disabled and Relaxed EDC Throttling Enabled.

I gonna try next 4.15GHz and if it works then 4.2GHz @ same voltage, but I would not want to increase my voltage much more. I like it how it is now, cool and quiet when not stress testing. So if the trade off is too much I'll stay happy at 4.1GHz.

What do you guys suggest to do next with my CPU OC?
 

DrMrLordX

Lifer
Apr 27, 2000
14,437
3,420
136
Low voltage will not hurt your CPU, per se. Voltage spikes when undergoing a reduction in load can be a killer though.

Also 4.1 GHz is not gad bad for a vanilla R5 2600.
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
14,437
3,420
136
Can you explain a little more about what you said in the 1st sentence?

Also, what does "gad" mean?
Oops gad = bad.

Also . . . let's say you are running Prime95, and your voltage is set to 1.35v. vdroop kicks in and voltage drops to 1.256v (ouch) during the benchmark. When you stop Prime95, current draw drops, causing voltage to swing back to the normal value, which causes an upward spike that lasts for maybe a few nanoseconds. Voltage spikes may also be known as overshoot:


Motherboards with strong VRMs at a high switching frequency are less likely to give you vdroop and overshoot/voltage spikes (or the spikes will be shorter in duration).
 

VforV

Member
Oct 19, 2019
33
1
11
Oops gad = bad.

Also . . . let's say you are running Prime95, and your voltage is set to 1.35v. vdroop kicks in and voltage drops to 1.256v (ouch) during the benchmark. When you stop Prime95, current draw drops, causing voltage to swing back to the normal value, which causes an upward spike that lasts for maybe a few nanoseconds. Voltage spikes may also be known as overshoot:


Motherboards with strong VRMs at a high switching frequency are less likely to give you vdroop and overshoot/voltage spikes (or the spikes will be shorter in duration).
Thanks for that explanation, now I get it.

So in my case (4 phase VRM MB and no option to select high switching frequency) going from 1.256v up to 1.35v can that be dangerous? Since 1.35v it's a pretty safe voltage, even for an OC, I mean for a Ryzen 2600...

Anyway, I settled for 4.15GHz OC at the same settings as the 4.1GHz OC and after the 1st batch of tests: Cinebench + Intel BurnTest (10min) + Prime95 (10min) it seems it's stable at max 70c, like the 4.1GHz OC which is funny.

I'll do some extended tests tomorrow, but so far I'm quite happy with this result. I did manage a 4.2GHz OC, but @ 1.375v and LLC1, it was passing Cinebench + Intel BurnTest (5min), but it was crashing testing for 10min or Prime95 and the temps and cooler noise was much higher too. I don't want to increase the voltage beyond that value, it's not worth it for +50MHz.
 

DrMrLordX

Lifer
Apr 27, 2000
14,437
3,420
136
@VforV

It depends on the strength of the voltage spike. Some people have complained of Pinnacle Ridge breaking down at 1.4v and higher over time. If you do a ton of heavy-current work and are constantly starting and stopping the workload, and if the spikes manage to go over 1.4v (which they probably will, with vdroop like that), then I could see a problem. If current draw is lower, vdroop and spikes will be less pronounced. It is also generally not a problem for people who run 100% high-current workloads 24/7 since the spikes only occur when you end the workload.
 

killster1

Diamond Member
Mar 15, 2007
4,950
177
106
well unless all you do is stress test i think it will be fine, but what are your plans for the computer? why do you need to overclock? will it help your render time a little?
 

VforV

Member
Oct 19, 2019
33
1
11
@VforV

It depends on the strength of the voltage spike. Some people have complained of Pinnacle Ridge breaking down at 1.4v and higher over time. If you do a ton of heavy-current work and are constantly starting and stopping the workload, and if the spikes manage to go over 1.4v (which they probably will, with vdroop like that), then I could see a problem. If current draw is lower, vdroop and spikes will be less pronounced. It is also generally not a problem for people who run 100% high-current workloads 24/7 since the spikes only occur when you end the workload.
Alright, so after 5 days of OC-ing I think I got to the best and optimum OC values for both CPU and RAM, which are:

CPU: Ryzen 2600 OC'ed @ 4.15GHz [Vcore in BIOS set to 1.35v manual, shows 1.368v-1.352v (in BIOS)] + LLC4 + AMD Cool' n' Quiet + some other settings.

RAM: 3000MHz CL15 (only recognized as 2933MHz CL16 for XMP2) OC'ed to 3600MHz CL16 with lots of custom sub-timings @ 1.40v and SOC 1.1v.

Now, I still have some issues with understanding the values reported by HWiNFO64 for Core VID, Core Voltage and Vcore, because all of them are different and I don't know which one is the most important one.

Because of this when I use LLC4 it's stable in Prime95, but lower or higher it's not. LLC3 crashed and PC restarted in less than 10 min in Prime95. LLC4 I ran it 30 min, no issues.

I've attached some screenshots with LLC3 and LLC4 settings and corresponding values when stress testing:
LLC3a
LLC3a.jpg

LLC3b
LLC3b.jpg

LLC4a
LLC4a.jpg

LLC4b
LLC4b.jpg

LLC4c
LLC4c.jpg

So which of these 3 values do I need to look out for when adjusting LLC? And why are they so different, because I think I was looking at the wrong one, Core VID which has the biggest min/max and also is very different than Vcore value. Core VID drops when on full load and Vcore raises, so I'm confused...

edit: I'm still on original BIOS which is ComboPI 1.0.0.3ab and I don't see a reason to go to the newest one ComboPI1.0.0.4 Patch B. I don't think it will get me better OC results for my CPU and RAM.

well unless all you do is stress test i think it will be fine, but what are your plans for the computer? why do you need to overclock? will it help your render time a little?
Gaming is the most demanding thing I do on my PC, but I need to OC because my GTX 1060 is old and can barely keep up with the new games at min 60 FPS in 1080p... so every little bit of performance matters to me. OC'ing the CPU in some games can make the difference between 55 min FPS and 60+ min FPS, so smoother gameplay. In some games it can make zero difference also, but it all depends on the game and how well it is optimized for Ryzen.

I did not test a lot of games yet, but for example, here are some scores from what I tested:

1080p (quality preset) // Stock CPU & RAM // OC'ed CPU & RAM

Battlefield V // Low=116 (min FPS), Ultra=61 (min FPS) // Low=121, Ultra=65
Rage 2 // Low=100 Ultra=67 // Low=100, Ultra=69
Forza Horizon 4 // Very Low =156 (min.GPU/CPU=157.2/140.9/194.3) // Very Low =165 (min.GPU/CPU=158.6/151.9/215.5)
WoT enCore RT // Minimum=111408 // Minimum=121536

In most of the cases you can see on Ultra that the GPU is the bottleneck, as FH4 nicely reports that in the test results page, but in the Low settings when I take out of the equation the GTX 1060, you can see some pretty big increases in performance when OC'ed, which shows me that if next year I plan to buy a new GPU it will actually help when my CPU is OC'ed vs stock even more.

To me all this OC'ing and testing shows that there is so much untapped potential in these Ryzen CPUs and the more devs optimize for them the better thy get. Can't wait for PS5 and new Xbox games that will work on PC too, those should run much better than what we have now.
 
  • Like
Reactions: lightmanek

VforV

Member
Oct 19, 2019
33
1
11
@VforV

Stick to CPU-z for monitoring your CPU voltage. HWiNFO64 can be a pain in that department.
Well, that simplifies things a lot... CPUz has only one value for CPU voltage which is named Core Voltage and the corresponding value in HWiNFO64 is actually named Vcore (Core Voltage shows different values in HWiNFO64) and they both report the same. Now I know what to look out for, thanks.

Also what's your opinion on CPU Over Voltage Protection, Under Voltage Protection and Over Current Protection? Should I manually set them to max in BIOS or leave them on auto?

And as I side note, I find it funny that after you OC Ryzen you can still use half the Cool' n' Quiet function, as in it downclocks the cores when not of full load, but it does not drop the voltage. But I guess it's still better than running them 24/7 and max OC, right?
 

DrMrLordX

Lifer
Apr 27, 2000
14,437
3,420
136
Also what's your opinion on CPU Over Voltage Protection, Under Voltage Protection and Over Current Protection? Should I manually set them to max in BIOS or leave them on auto?
I would leave those on auto. You're not doing anything that would require you to choose custom settings.

And as I side note, I find it funny that after you OC Ryzen you can still use half the Cool' n' Quiet function, as in it downclocks the cores when not of full load, but it does not drop the voltage. But I guess it's still better than running them 24/7 and max OC, right?
Probably better overall, yes. It's the same way on Matisse, though Matisse can lower the voltage too. With Summit Ridge (like my 1800x), you got max clocks and max voltage all the time, which was a bit silly.
 

VforV

Member
Oct 19, 2019
33
1
11
@DrMrLordX

Thanks and also now I have another dilemma (sorry for dragging this OC stuff so long):

With my current OC (see above post) I can pass any benchmark and stress test except IntelBurnTest Very High preset... I mean Standard and High are successful also Prime95 30min is fine and stable, but when I try Very High on IntelBurnTest it makes my PC reset after the 1st pass or during it...

If I reset my CPU to stock while keeping the RAM OC'ed I can pass Very High, so it's clear that something about the CPU OC is the issue, but considering I can pass all those other stress tests, do I really need to worry, since I only game on this PC?

I'll try raising the Vcore to 1.375v or 1.4v, just for the sake of testing, but even if it's stable at those voltages I don't like em, I'm not gonna use them so high. :/
 

VforV

Member
Oct 19, 2019
33
1
11
Well now I really think it is final, I can't squeeze any more performance or efficiency.

I can get stable 4.2GHz @ 1.4v set in BIOS with LLC3 which under stress test raises the max Vcore to almost 1.42v, so it's a big NO for me.

The last tweak I did was to get the 4.15GHz OC Vcore in BIOS down to 1.3375v with LLC3. So basically I have 2 options to chose from now:

A. 1.35v with LLC4 in BIOS = in Windows under stress test it gets from 1.352v up to 1.368v.
B. 1.3375v with LLC3 in BIOS = in Windows under stress test it gets from 1.344v up to 1.368v.

So I think option B is the better one right? Since the idle voltage is a little lower. The load voltage has to be at least 1.368v (thus the LLC change to match the other one) otherwise it's not stable under stress tests.

So yeah, my work is done now and I'm pretty happy. I'll post one more benchmark score that I really like:

Cinebench R15 with CPU+RAM @ stock > MT = 1295
Cinebench R15 with CPU+RAM @ Ryzen 2600 @ 4.15GHz & RAM @ 3600MHz CL16 > MT=1436

This shows that if game devs optimize for Ryzen to the fullest you can get at least 10% more performance from OC'ing the CPU & RAM and this is with the Ryzen 2600, the 3000 series is even better with RAM OC'ing and those CPUs (at least the X ones) don't need manual OC since they do that alone, programmed by AMD.

Thanks everyone for the help, in this thread and the other one too. And a special Thanks to
DrMrLordX, I really appreciated all your help.
 

ASK THE COMMUNITY