3900x issue

[Mchart]

And thankfully, virtually no current, or we'd all be dead

Either way, Ryzen Balanced power profile shouldn't be running at nearly 1.5v under a no load condition

Set the balanced profile minimum to 5% and see if that helps.

 

[tangoseal]

And thankfully, virtually no current, or we'd all be dead

Either way, Ryzen Balanced power profile shouldn't be running at nearly 1.5v under a no load condition

I actually agree with you. Why not try windows balanced and see what happens.

 

[funkydmunky]

I actually agree with you. Why not try windows balanced and see what happens.

Any Ryzen profile bork's my system. I have a monitor off then a sleep. A Ryzen profile will either not enact the monitor off, or will put it into a "sleep" that isn't a sleep. Meaning the monitor will go off but the system keeps running. Then when I go to wake it, it will do a re-boot to the desktop. Also on any profile hibernate is not listed.
I have tried every combination. Right now I am running a non Ryzen profile with only the monitor off.\
I hope they fix this soon. Not sure who is to blame. Problem became an issue after I got my RTX card. If anyone has info on that please share.

 

[Deleted member 82943]

I have a Gigabyte Aorus x570 master. It was cooking my chip but using neg offset I now have it at 1.28v full load. Idle in 30s full load at 68ish. Room is AC cooled. Custom loop cooled cpu and gpu.

williacm

Your AIO probably evaporated most of the coolant out. AIOs just like custom loops lose liquid volume over time. Or your pump was weak or your microfins in the AIO block were clogged up with biogunk.

I dont think die placement has any affect on thermal transfer to be honest. Once I tempered my voltages my thermals were as normal as any typical cpu.

Unrelated but I didn’t know this happened. Sheesh. Big Noctua coolers for me from now on or a custom loop where I can add more coolant and a new pump as needed.

 

[tangoseal]

So I got my 3900x down to 1.21v full load. And its still dropping. I am tired of rebooting tonight and need to go to sleep. Lower V then reboot. Success over and over. I have a feeling I can get this thing to like 1.05v full load haha I am going to keep shooting for it.

No matter what I do though I cant get more than a 194 in single thread Cinebench when revieewers out there were showing 204s etc... I am not sure what I am doing wrong. Unless they were manually overclocking and that is why the ST was higher.

 

[RamonGTP]

So I got my 3900x down to 1.21v full load. And its still dropping. I am tired of rebooting tonight and need to go to sleep. Lower V then reboot. Success over and over. I have a feeling I can get this thing to like 1.05v full load haha I am going to keep shooting for it.

No matter what I do though I cant get more than a 194 in single thread Cinebench when revieewers out there were showing 204s etc... I am not sure what I am doing wrong. Unless they were manually overclocking and that is why the ST was higher.

Are you still boosting as high with the lower voltage?

 

[Dan_D]

So I got my 3900x down to 1.21v full load. And its still dropping. I am tired of rebooting tonight and need to go to sleep. Lower V then reboot. Success over and over. I have a feeling I can get this thing to like 1.05v full load haha I am going to keep shooting for it.

No matter what I do though I cant get more than a 194 in single thread Cinebench when revieewers out there were showing 204s etc... I am not sure what I am doing wrong. Unless they were manually overclocking and that is why the ST was higher.

I provided both stock (PB2) results and a manual overclocked results using Cinebench R20 in multi-threaded and single-threaded workloads. I got results between 495 and 516 depending on configuration and BIOS (AGESA code) revision. Since lowering your voltage, what are you boost clocks like? Boost clocks were generally around 4.4GHz the first time around and with AGESA 1.0.0.3 patch AB, I saw 4.573GHz. This is pretty damn close to the 4.6GHz advertised boost clock speeds.

 

[RamonGTP]

I provided both stock (PB2) results and a manual overclocked results using Cinebench R20 in multi-threaded and single-threaded workloads. I got results between 495 and 516 depending on configuration and BIOS (AGESA code) revision. Since lowering your voltage, what are you boost clocks like? Boost clocks were generally around 4.4GHz the first time around and with AGESA 1.0.0.3 patch AB, I saw 4.573GHz. This is pretty damn close to the 4.6GHz advertised boost clock speeds.

That 4.573GHz might be shy of 4.6GHz because of the base clock. CPUZ reports my base clock at 99.80 instead of 100

 

[cbutters]

So I got my 3900x down to 1.21v full load. And its still dropping. I am tired of rebooting tonight and need to go to sleep. Lower V then reboot. Success over and over. I have a feeling I can get this thing to like 1.05v full load haha I am going to keep shooting for it.

No matter what I do though I cant get more than a 194 in single thread Cinebench when revieewers out there were showing 204s etc... I am not sure what I am doing wrong. Unless they were manually overclocking and that is why the ST was higher.

Keep in mind that zen 2 works a bit differently from what I understand; you might lower voltage and your clocks might look still ok; but your performance will be dropping. This is due to "clock stretchers". Be sure to benchmark as you drop voltage to see when your performance starts to drop off as well.

Smarter people's words:

"Its not wise to undervolt Matisse, unless you enable the OC-Mode (fixed frequency).

That's because the CPU is constantly monitoring its supply voltage. When it sees the supply voltage below the desired level, it starts increasing the VID request from the VRM controller. When the VID request increases above the maximum voltage allowed by FIT, the clock stretchers will kick in. Your performance will decrease, despite the change is not visible in the frequencies diplayed by the usual monitoring software.

For example on the 3900X sample I have, the default VID request during CB20 NT is 1.3125V. With the offset set to -25mV the VID request raises to 1.33125V or so. At -75mV the VID request is already 1.36875V and increasing the negative offset any further will trip the stretchers.

There is an undervolting margin available, but it is tiny. The exact behavior will depend on the silicon specimen and the workload.

Its also the very reason why this guy thinks he is running at > 4.2GHz with just 1.00V..."​

 

[Dan_D]

That 4.573GHz might be shy of 4.6GHz because of the base clock. CPUZ reports my base clock at 99.80 instead of 100

This is true. MSI has basically always run its motherboards shy of a 100MHz base clock or even FSB back in the day. It may also be the external clockgen that MSI uses. I'm not sure off hand what each manufacturer uses, so it might not be that, but rather a choice MSI has consciously made while designing their boards.

 

[RamonGTP]

Keep in mind that zen 2 works a bit differently from what I understand; you might lower voltage and your clocks might look still ok; but your performance will be dropping. This is due to "clock stretchers". Be sure to benchmark as you drop voltage to see when your performance starts to drop off as well.

Smarter people's words:

"Its not wise to undervolt Matisse, unless you enable the OC-Mode (fixed frequency).

That's because the CPU is constantly monitoring its supply voltage. When it sees the supply voltage below the desired level, it starts increasing the VID request from the VRM controller. When the VID request increases above the maximum voltage allowed by FIT, the clock stretchers will kick in. Your performance will decrease, despite the change is not visible in the frequencies diplayed by the usual monitoring software.

For example on the 3900X sample I have, the default VID request during CB20 NT is 1.3125V. With the offset set to -25mV the VID request raises to 1.33125V or so. At -75mV the VID request is already 1.36875V and increasing the negative offset any further will trip the stretchers.

There is an undervolting margin available, but it is tiny. The exact behavior will depend on the silicon specimen and the workload.

Its also the very reason why this guy thinks he is running at > 4.2GHz with just 1.00V..."​

I've actually seen this direct correlation between voltage reduction and performance drop in my own testing with Cinebench. It's not a whole lot, but it is there.

 

[CMDR1337]

I have a Gigabyte Aorus x570 master. It was cooking my chip but using neg offset I now have it at 1.28v full load. Idle in 30s full load at 68ish. Room is AC cooled. Custom loop cooled cpu and gpu.

williacm

Your AIO probably evaporated most of the coolant out. AIOs just like custom loops lose liquid volume over time. Or your pump was weak or your microfins in the AIO block were clogged up with biogunk.

I dont think die placement has any affect on thermal transfer to be honest. Once I tempered my voltages my thermals were as normal as any typical cpu.

Wait! so the problem is happening with X570 boards as well? i thought it was only b450 and x470 etc boards that were not working correctly without a Bios update?
NUTs! so that means i will have to be super careful when i start my build. I am not a Novice but far from a PRO building systems.

 

[Mchart]

I have no idea why people are lowering their voltages, these things will use the voltages they need, you are only hurting performance. The idle voltage 'issue' is only an issue in that you are using more power than you need at idle. The CPU is fine taking higher voltages when there isn't load.

I've done more testing as to what makes idle voltages stay high. The choice of power plan within windows doesn't really do anything.

I've discovered many, many, different applications keep these CPU's from going into idle state. This seems like a problem that AMD will need to fix with Microsoft, to be honest, or at least update the chipset drivers.

For instance, Steam, even when not in the foreground prevents my 3800x from dropping to idle voltages. I've discovered pretty much any application, even if it's using a minuscule amount of compute, will cause this.

 

[Algrim]

No matter what I do though I cant get more than a 194 in single thread Cinebench when revieewers out there were showing 204s etc...

WTH? I get a straight 200 single core on a 32-nm Westmere. (This is Cinebench R20; adding context because maybe R15 is a lower number.)

 

[Mchart]

If he is lowering the voltage his single core results will suffer tremendously.

 

[Algrim]

If he is lowering the voltage his single core results will suffer tremendously.

I just edited my post as R15 and R20 are very different in the numbers they display.

 

[tangoseal]

I'm doing it for research. I'm aware that low voltage might hurt ST perf but on 7nm do we know for certain? If so I'm not sure to what extent. Also I want to see how low 7nm can be stable under full load for my own enthusiast motivations. Once I find the floor ill raise it back to see where the sweet spot is.

we've never had 7nm before. So even assumptions about voltage could be wrong. I'm finding much higher boost clocks at low voltage and higher MT perf. but st is hurting even at higher clocks. My chip boosts to 4.4ghz at 1.20v full loaded. I'm exploring LLC right now.

 

[cbutters]

I'm doing it for research. I'm aware that low voltage might hurt ST perf but on 7nm do we know for certain? If so I'm not sure to what extent. Also I want to see how low 7nm can be stable under full load for my own enthusiast motivations. Once I find the floor ill raise it back to see where the sweet spot is.

we've never had 7nm before. So even assumptions about voltage could be wrong. I'm finding much higher boost clocks at low voltage and higher MT perf. but st is hurting even at higher clocks. My chip boosts to 4.4ghz at 1.20v full loaded. I'm exploring LLC right now.

No thats great to test the undervolting limits; however we're just saying you might not find a traditional floor where the processor stops working; because the CPU is very aware of its power situation and will adjust itself and stretch clocks out if it doesn't have what it needs; doing so without crashing; although I guess if you set it to 0.2V it probably won't even boot at all. I guess testing this in itself would be a fun research project. Let us know.

 

[RamonGTP]

No thats great to test the undervolting limits; however we're just saying you might not find a traditional floor where the processor stops working; because the CPU is very aware of its power situation and will adjust itself and stretch clocks out if it doesn't have what it needs; doing so without crashing; although I guess if you set it to 0.2V it probably won't even boot at all. I guess testing this in itself would be a fun research project. Let us know.

Review outlets have all the fun with samples they can test on without compromising their personal systems. Now that my 3770k is a secondary system, i'm only now really working on pushing it to see what it can do.