Ryzen Voltages—WHAT THE @#$%?????

[genghisquan]

OK, so when I first built this b450 + Ryzen 3600 system, the vcore was at 1.45V when it was set to auto! Obviously that's a little too high, since the recommended upper limit for vcore voltages are between 1.3-1.35V.

 

After overriding the vcore and dropping it down to a constant 1.36V, the performance & temps were as expected. However, I don't want a constant voltage and want it to be able to adjust to the CPU loads.

This is where things get weird because changing the vcore to an offset mode (that kept the load voltage at 1.36V) resulted in the worst performance in Cinebench r20; i.e, ~3250 pts. My CPU never went above base clock (3.6 GHz all cores) at this point.

I switched back to auto vcore, but now the BIOS set the vcore to just 1.1V! WHY? What does auto even do?

So after that, I went back to the override voltage, keeping it at the constant 1.36V that it was at previously...Cinebench score instantly went up 100 pts (up to 3350 pts). The CPU boosted up to 3.7-3.8 GHz (all cores) during the benchmark.

WHY? What's the difference between offset and constant voltage performance?! 

It also gets weirder because going back to a voltage offset (NEGATIVE offset this time) resulted in a vcore of 1.28V during load. Now the CPU boosted to around 3.974 GHz (all cores) during Cinebench run. Except now the score is around 3500 pts!!!

 

Throughout all of these tests, using all these different voltages, my CPU never even touched 72ºC at the highest load voltage. So I don't think the performance was being thermal-throttled.

 

WHAT THE HELL IS GOING ON WITH RYZEN VOLTAGES & PERFORMANCE?!

Please help me understand what's going on and figure out what voltage setting I should be using.

[Jurrunio]

Stock voltage can go up to 1.47V for short bursty single threaded workload, it will run at most somewhere around 1.3V at stock for the most stressful of all core workloads like Prime95 smallFFT. Basically, it's really dynamic and the 1.3V-ish recommended voltage (which is lower than what you're using now) is for all core workloads and only a thing because manual mode cant be as adaptive as auto.

 

18 minutes ago, genghisquan said:

I switched back to auto vcore, but now the BIOS set the vcore to just 1.1V! WHY? What does auto even do?

sometimes settings get stuck, whenever something like that happens just clear CMOS and try again

 

19 minutes ago, genghisquan said:

WHY? What's the difference between offset and constant voltage performance?! 

I suspect you're seeing clock stretching, basically performance lower than shown frequency. This is done to avoid shutdowns when voltage is too low (of course it can only compensate so much, you cant boot it at all if you only give it 0.8V for 3.8GHz for example)

 

20 minutes ago, genghisquan said:

It also gets weirder because going back to a voltage offset (NEGATIVE offset this time) resulted in a vcore of 1.28V during load. Now the CPU boosted to around 3.974 GHz (all cores) during Cinebench run. Except now the score is around 3500 pts!!!

Probably hitting the power and current limits (TDC EDC PPT) with stock voltage

[genghisquan]

11 minutes ago, Jurrunio said:

Stock voltage can go up to 1.47V for short bursty single threaded workload, it will run at most somewhere around 1.3V at stock for the most stressful of all core workloads like Prime95 smallFFT. Basically, it's really dynamic and the 1.3V-ish recommended voltage (which is lower than what you're using now) is for all core workloads and only a thing because manual mode cant be as adaptive as auto.

 

sometimes settings get stuck, whenever something like that happens just clear CMOS and try again

 

I suspect you're seeing clock stretching, basically performance lower than shown frequency. This is done to avoid shutdowns when voltage is too low (of course it can only compensate so much, you cant boot it at all if you only give it 0.8V for 3.8GHz for example)

 

Probably hitting the power and current limits (TDC EDC PPT) with stock voltage

So what should I be doing to keep my voltage at a good level under load but allow it to drop when idle or low load? It's confusing because I saw the vcore at 1.45V (on auto mode) when in UEFI. Why would that voltage be sustained that high when there's no load on the system?

[Master Disaster]

36 minutes ago, genghisquan said:

OK, so when I first built this b450 + Ryzen 3600 system, the vcore was at 1.45V when it was set to auto! Obviously that's a little too high, since the recommended upper limit for vcore voltages are between 1.3-1.35V.

 

After overriding the vcore and dropping it down to a constant 1.36V, the performance & temps were as expected. However, I don't want a constant voltage and want it to be able to adjust to the CPU loads.

This is where things get weird because changing the vcore to an offset mode (that kept the load voltage at 1.36V) resulted in the worst performance in Cinebench r20; i.e, ~3250 pts. My CPU never went above base clock (3.6 GHz all cores) at this point.

I switched back to auto vcore, but now the BIOS set the vcore to just 1.1V! WHY? What does auto even do?

So after that, I went back to the override voltage, keeping it at the constant 1.36V that it was at previously...Cinebench score instantly went up 100 pts (up to 3350 pts). The CPU boosted up to 3.7-3.8 GHz (all cores) during the benchmark.

WHY? What's the difference between offset and constant voltage performance?! 

It also gets weirder because going back to a voltage offset (NEGATIVE offset this time) resulted in a vcore of 1.28V during load. Now the CPU boosted to around 3.974 GHz (all cores) during Cinebench run. Except now the score is around 3500 pts!!!

 

Throughout all of these tests, using all these different voltages, my CPU never even touched 72ºC at the highest load voltage. So I don't think the performance was being thermal-throttled.

 

WHAT THE HELL IS GOING ON WITH RYZEN VOLTAGES & PERFORMANCE?!

Please help me understand what's going on and figure out what voltage setting I should be using.

Whats going on is Zen 2 is a unique architecture in the way it works (even compared to Zen 1 & Zen+) and it took a few tries for AMD to get the microcode correct for motherboards plus them basically overriding the Windows CPU scheduler almost entirely before the CPUs started operating as intended. There's 2 important things for Zen 2, make sure your board is running the latest BIOS and make sure you manually download and install the AMD chipset driver directly from AMD.com. (edit - You must also be running Windows 10 1903 or newer)

 

Even with these 2 things installed you might see the chip/board doing things you'd consider weird from any other CPU (like using lots of voltage for what seems like a trivial task), that's normal.

 

AMD also recommend you only monitor Zen 2 CPUs using Ryzen Master, other software can give false readings or even artificially increase the workload on the CPU by polling it for a readout faster than the CPU is expecting.

[Jurrunio]

1 minute ago, genghisquan said:

It's confusing because I saw the vcore at 1.45V (on auto mode) when in UEFI. Why would that voltage be sustained that high when there's no load on the system?

Remember that I said "high voltage for single core workload"? You cant get more single threaded and light than moving the mouse around.

 

5 minutes ago, genghisquan said:

So what should I be doing to keep my voltage at a good level under load but allow it to drop when idle or low load?

Ryzen's power management focuses on turning off parts of the core rather than lowering voltage to save power (like Intel does), so auto settings pretty much doesnt lower voltage. I recall Zen and Zen+ capable of dropping voltage when literally doing nothing (background tasks disabled as many as possible, no mouse movements) so maybe Zen 2 can do that as well.

[5x5]

7 minutes ago, genghisquan said:

So what should I be doing to keep my voltage at a good level under load but allow it to drop when idle or low load? It's confusing because I saw the vcore at 1.45V (on auto mode) when in UEFI. Why would that voltage be sustained that high when there's no load on the system?

Leave it on auto and stop touching things you don't understand and things which you approach with panic and not rationale

[genghisquan]

5 minutes ago, 5x5 said:

Leave it on auto and stop touching things you don't understand and things which you approach with panic and not rationale

Oh, thanks. I'll just stop messing around with performance parts and tuning. You're such a big help. LOL.

[Master Disaster]

8 minutes ago, Jurrunio said:

Ryzen's power management focuses on turning off parts of the core rather than lowering voltage to save power (like Intel does), so auto settings pretty much doesnt lower voltage. I recall Zen and Zen+ capable of dropping voltage when literally doing nothing (background tasks disabled as many as possible, no mouse movements) so maybe Zen 2 can do that as well.

To add to this, nor does it need to lower voltage. If its only using 2 cores for light workloads then it can happily use 1.4v, remember 2 cores using 1.4v is still much less heat output than 8 cores using 1.3v.

 

7 minutes ago, 5x5 said:

Leave it on auto and stop touching things you don't understand and things which you approach with panic and not rationale

Basically this.

[5x5]

1 minute ago, genghisquan said:

Oh, thanks. I'll just stop messing around with performance parts and tuning. You're such a big help. LOL.

Would you prefer to break things and cry on forums about how little you understood and yet still messed with everything? The Engineers who created this system are several times more intelligent than all of us combined so please, don't act like you know more about tuning that the person who made the bloody things

[genghisquan]

Just now, 5x5 said:

Would you prefer to break things and cry on forums about how little you understood and yet still messed with everything? The Engineers who created this system are several times more intelligent than all of us combined so please, don't act like you know more about tuning that the person who made the bloody things

I'm asking for help on forums to try to learn and understand how my system works. I never claimed to be smarter than you or the engineers at AMD. Where are you getting that notion?

[Master Disaster]

25 minutes ago, genghisquan said:

Oh, thanks. I'll just stop messing around with performance parts and tuning. You're such a big help. LOL.

While the way @5x5 said it wasn't the way I'd have said it, what he said was true. Update the BIOS, install the latest chipset driver and leave it alone. Trying to manually override AMDs settings will hurt performance more than help it (thats a lesson I learned the hard way).

 

Edit - Also avoid using NZXTs CAM software

 

[genghisquan]

2 minutes ago, Master Disaster said:

While the way @5x5 said it wasn't the way I'd have said it, what he said was true. Update the BIOS, install the latest chipset driver and leave it alone. Trying to manually override AMDs settings will hurt performance more than help it (thats a lesson I learned the hard way).

Thanks for your replies and recommendation. I'm using current BIOS and chipset driver versions. I'll try auto settings.

[Damascus]

19 minutes ago, 5x5 said:

Would you prefer to break things and cry on forums about how little you understood and yet still messed with everything? The Engineers who created this system are several times more intelligent than all of us combined so please, don't act like you know more about tuning that the person who made the bloody things

Or, and just hear me out here, motherboard manufacturers constantly fuck up the default voltage settings!  Exhibit A being early gen asus x99 boards pumping 2v through cpus and killing loads of them on (you guessed it) stock settings.  How about all the z390 boards with entirely different Tjmax, voltage and allowed power draws?  Or maybe the time every manufacturer ever did their own little fucky thing with voltage controls and auto boost weirdness.  

 

Lowering vcore isn't going to hurt a CPU, worst case scenario OP gets some black screens, resets the bios and ups the score a little bit to correct for going too low

[Damascus]

To actually respond to your questions-

https://www.gamersnexus.net/guides/3494-amd-ryzen-3000-undervolting-offset-override

 

Gamers nexus did a pretty interesting piece on this a while back, worth a read/watch 

[Master Disaster]

14 minutes ago, Damascus said:

Or, and just hear me out here, motherboard manufacturers constantly fuck up the default voltage settings!  Exhibit A being early gen asus x99 boards pumping 2v through cpus and killing loads of them on (you guessed it) stock settings.  How about all the z390 boards with entirely different Tjmax, voltage and allowed power draws?  Or maybe the time every manufacturer ever did their own little fucky thing with voltage because controls and auto boost weirdness.  

 

Lowering score isn't going to hurt a CPU, worst case scenario OP gets some black screens, resets the bios and ups the score a little bit to correct for going too low

You're comparing Intel to AMD. Unlike Intel, AMD provide solid baseline numbers to board manufacturers for things like Max Safe Voltage, TJMax and TDP.

 

I bought my 3800X on release day (I had to wait a week for it) and I tried all the "fix your Zen 2" and the "these settings will make your Zen 2 faster"  guides that were doing the rounds. Some did nothing, some seemed to help with voltage while actually hurting performance and some made things much worse. None of them helped.

 

I played around with manual overclocking and discovered this tended to hurt single core performance which in turn made gaming slower. By manually OCing you stop your fast cores from boosting which ends with a net loss in tasks that need single threaded performance.

 

Once AMD released AGESA 1.0.0.4B and the corresponding Chipset Driver I noticed a significant increase in SC performance, the "fast cores" would boost much higher than before (up to 4.5Ghz) and overall performance increased while temps decreased.

[GoldenLag]

18 minutes ago, Damascus said:

Lowering vcore isn't going to hurt a CPU, worst case scenario OP gets some black screens, resets the bios and ups the score a little bit to correct for going too low

It will when you do what OP did......

 

Setting it to 1.36v. which at this point is considered "lethal" voltage. In the sense it will relativly quickly degrade your CPU

[genghisquan]

1 minute ago, Damascus said:

To actually respond to your questions-

https://www.gamersnexus.net/guides/3494-amd-ryzen-3000-undervolting-offset-override#!/ccomment-comment=10012222

 

Gamers nexus did a pretty interesting piece on this a while back, worth a read/watch 

Haha, I actually remembered this and rewatched it when @Jurrunio mentioned clock stretching. As shown in GN's video, performance dropped significantly from the 1.0V vcore while the GHz were the reported as normal. That's why my situation confused me because my results were the inverse; i.e, slightly lowered vcore actually increased my all core frequency and raised my cinebench scores by 250 pts.

 

And yeah, I'm trying settings on auto right now. I was just initially afraid that those vcore spikes were a little too high. Given Zen 2 being so new, I'm hearing so many conflicting information regarding all its settings.

[genghisquan]

5 minutes ago, GoldenLag said:

It will when you do what OP did......

 

Setting it to 1.36v. which at this point is considered "lethal" voltage. In the sense it will relativly quickly degrade your CPU

Right. I tried to lower it from 1.36V because I saw from LTT's Ryzen overclocking video that it's the upper limit for vcore. Prior to tonight, I didn't know that it's normal for burst voltages to reach 1.4V! 

[Mihle]

My 3600 do 1,47v even idle in Windows in HWinfo, while it does 1,364v under all core load. Only thing I have done is put on PBO and AutoOC.

 

What's weird to me is that one place some people said 1,35v is the max safe, but other places some AMD representative or forum person or whatever said higher than that is also safe?

[genghisquan]

1 minute ago, Mihle said:

My 3600 do 1,47v even idle in Windows in HWinfo, while it does 1,364v under all core load. Only thing I have done is put on PBO and AutoOC.

 

What's weird to me is that one place some people said 1,35v is the max safe, but other places some AMD representative or forum person or whatever said higher than that is also safe?

Thanks for your input (as another 3600 owner). Yeah, I'm seeing the same ~1.36V under all-core load. And like you, I'm also reading/hearing a lot of conflicting info regarding Ryzen voltages...that's why I tried to undervolt in the first place.

[GoldenLag]

5 minutes ago, genghisquan said:

Prior to tonight, I didn't know that it's normal for burst voltages to reach 1.4V! 

Yeah its normal behavior for low current loads. 

 

There have been iterations of this since zen 1. And people keep trying to "fix" it. Even tho its not actually an issue. 

 

7 minutes ago, genghisquan said:

Right. I tried to lower it from 1.36V because I saw from LTT's Ryzen overclocking video that it's the upper limit for vcore

Yeah the upper limit has been lowered and lowered since day 1. Currently 1.325v is the upper limit. At which point. Outside of allcore performance, it makes very little sence to overclock. 

2 minutes ago, Mihle said:

AMD representative or forum person or whatever said higher than that is also safe?

Probably in relation to what AMD runs themselves, which leads to a lot of confusion. 

 

Id rather keep an eye on what people have run, and experienced degrading with. Which is anything above 1.325v from what i can tell. 

 

Its not a definitive line and it varies ln a CPU to CPU basis, but its better to assume you lost the silicon lottery, rather than you have won it. 

[genghisquan]

3 minutes ago, GoldenLag said:

Yeah its normal behavior for low current loads. 

 

There have been iterations of this since zen 1. And people keep trying to "fix" it. Even tho its not actually an issue. 

 

Yeah the upper limit has been lowered and lowered since day 1. Currently 1.325v is the upper limit. At which point. Outside of allcore performance, it makes very little sence to overclock. 

Probably in relation to what AMD runs themselves, which leads to a lot of confusion. 

 

Id rather keep an eye on what people have run, and experienced degrading with. Which is anything above 1.325v from what i can tell. 

 

Its not a definitive line and it varies ln a CPU to CPU basis, but its better to assume you lost the silicon lottery, rather than you have won it. 

God damn it. Now I'm worried that the auto setting is pumping too high of a vcore during all-core load!

[Mihle]

So both of us should try to get the voltage lower?

[genghisquan]

Just now, Mihle said:

So both of us should try to get the voltage lower?

I don't even know what's the proper method for voltage control at this point. People recommend Ryzen Master, but others say it's too complicated / unnecessary. Albeit, Cinebench is an extremely stressful program, and I don't think I ever hit my CPU that hard during gaming / Adobe app usage.

[GoldenLag]

3 minutes ago, Mihle said:

So both of us should try to get the voltage lower?

Either you run stock (or PBO). And not touch any vcore related settings. 

 

Or stay at or below 1.325v (when overclocking), is what i would suggest.