Safe cpu voltage offset for 5800x3D?

[Origami Cactus]

At default my 5800x3D ran at 90.4C, 4000ghz in cinebench, 1.22->1.25V, not good.

In my Asus x470 beta bios I can freely overclock the 5800x3D, so I chose CPU core Voltage offset, and set it to "-", and -0.075. This is all I did, everything else is set to auto.

Now in cinebech the temp is 82c, 4.25ghz, 1.138->1.144V. Score in R23 is 14134.

I know about the weird PBO2-tuner program you can install in windows and set per core voltage offsets etc, but I would like to keep it in BIOS, no windows programs.

1) Is running at 1.144V instead of 1.25V safe?

2) Could I technically run at 1.1, maybe even 1.0V, and would the CPU be smart enough to just drop the clocks a bit to keep up? Or would it be unstable?

3) If there is anything else in BIOS I could tweak to get more performance or lower temps, also let me know

 

Pic during test:

Spoiler

[Guest]

90c at 4ghz? You should be closer to 4.8ghz all core.

 

The cooling solution is.....?

[Origami Cactus]

1 minute ago, Guest 5150 said:

90c at 4ghz? You should be closer to 4.8ghz all core.

 

The cooling solution is.....?

It is the 5800X3D, not the normal one, so max clockspeeds are closer to 4.6ghz singlecore, 4.4ghz all core from what I have seen. It is due to CPU cores not making contact with the heatspreader anymore, there is a layer of silicon and cache between the 2 now, making the cpu much hotter.

 

The cooler is NZXT H1 included 140mm rad, with fan running at 1600rpm, at 2400rpm the noise is not bearable.

[Guest]

Just now, Origami Cactus said:

It is the 5800X3D, not the normal one, so max clockspeeds are closer to 4.6ghz singlecore, 4.4ghz all core from what I have seen. It is due to CPU cores not making contact with the heatspreader anymore, there is a layer of silicon and cache between the 2 now, making the cpu much hotter.

 

The cooler is NZXT H1 included 140mm rad, with fan running at 1600rpm, at 2400rpm the noise is not bearable.

140mm rad!!!!! (Mean 240???)

I want to say get a bigger cooler for the hottest chip on the market. 

 

Under volt will be a personal experience. Go as low as you can, high LLC to prevent load droop.

[Origami Cactus]

Just now, Guest 5150 said:

140mm rad!!!!! (Mean 240???)

I want to say get a bigger cooler for the hottest chip on the market. 

 

Under volt will be a personal experience. Go as low as you can, high LLC to prevent load droop.

yeah, 140mm. My previous 3700x was 90c with a 92mm one, so I went all out for the 5800x3D lol.

Quite happy with 80c at load tbh, not going to change the case to fit a larger rad.

About the LLC; I have no idea what it does. Seems to increase voltage to prevent droop, but I am after the lowest voltage, so seems counter intuitive, especially as I don't have stability problems, so no droop for now.

[Guest]

40 minutes ago, Origami Cactus said:

yeah, 140mm. My previous 3700x was 90c with a 92mm one, so I went all out for the 5800x3D lol.

Quite happy with 80c at load tbh, not going to change the case to fit a larger rad.

About the LLC; I have no idea what it does. Seems to increase voltage to prevent droop, but I am after the lowest voltage, so seems counter intuitive, especially as I don't have stability problems, so no droop for now.

Well the way transistors work.....

 

You have millions. Some do not operate. Even with a default cpu.

A low volt hang up is the result of too many transistors that are not operational.

Some are used for per core, Some are used to handle memory and Some are used for IO.

 

So the lower the vcore, the less transistors operate. And as a result you can produce lower cpu scores during certain benchmarks to demonstrate that.

 

Vcore leakage is a real thing. The higher the temp, the more transistors leak. 

 

So it can be a hand to hand thing, lower temps lower ccore. You just gotta find that middle ground.

 

I would try base default frequency with vcore on auto and give it a load. See what it settles at and increase the frequency based on that voltage.