i9-7980XE OC problems

Magogan · January 10, 2022

Hi,

I got a used i9-7980XE and I tried to OC it.

I set different turbo limits for 1, 2, 3, ..., 17, 18 cores, how can I test whether this is stable? Do I just have to leave it running for multiple days? Can't get it to hit more than 3.9 GHz with Prime95 with 3 cores and without AVX-512, even though I set the 3 and 4 core turbo to 4.3 GHz.

It seems to run stable at 3.9 GHz on all cores at 100% load but it randomly crashed (BSOD) once after 2 days at maybe 5% load. I also had to undervolt it by -0.02 V because the PC would shut down instantly due to too much current. The PSU (beQuiet Dark Power 1000W) can supply 32 A on the CPU rail, so I have no idea why it would have a problem with 280-ish watts. I don't want much more than that anyway due to cooling, but not triggering the over current protection would be nice.

Somerandomtechyboi · January 10, 2022

I think ocp is usually adjustable or even disable in the bios, but you can use elmor evc2 if the board is being a prick with a garbage ocp limit thats wayyyy too low and there isnt a way to disable it via bios

ofc an 18 core is a nuclear reactor so you gotta have good cooling to oc it, have you delidded it yet? cause the stock tim is trash and iirc you can get upto 20c lower temps with liquid metal

Jurrunio · January 10, 2022

46 minutes ago, Magogan said:

I set different turbo limits for 1, 2, 3, ..., 17, 18 cores, how can I test whether this is stable? Do I just have to leave it running for multiple days?

This is X299, you can let different cores run at different frequencies. I assume you're refering to this? You'll have to test each core independently, at reduced cooling fan speeds so it will get hot as if you're running all the cores.

If you're talking about the "how many cores are under work" kind, just don't. Not worth the time and effort to control 18 cores with the same core voltage. Let them run at the same clocks.

50 minutes ago, Magogan said:

It seems to run stable at 3.9 GHz on all cores at 100% load but it randomly crashed (BSOD) once after 2 days at maybe 5% load. I also had to undervolt it by -0.02 V because the PC would shut down instantly due to too much current.

Current limits are usually ignored on motherboards when you overclock it, the exception are usually awful boards that will blow up if the current limit isn't in place. What board is this?

And reduced voltage (what undervolt means) hurts stability at higher frequencies, this is a sign -0.02V is too low for more than 3.9GHz. Stock is just 3.4GHz on all cores.

Magogan · January 10, 2022

The motherboard is a Gigabyte UD4. Forgot to mention that.

4 hours ago, Somerandomtechyboi said:

I think ocp is usually adjustable or even disable in the bios, but you can use elmor evc2 if the board is being a prick with a garbage ocp limit thats wayyyy too low and there isnt a way to disable it via bios

I think the OCP is from the PSU, not the motherboard. But I'm not sure and I have no idea how to test it. Well, I could disable it in the PSU, but that seems like a fire hazard.

4 hours ago, Somerandomtechyboi said:

ofc an 18 core is a nuclear reactor so you gotta have good cooling to oc it, have you delidded it yet? cause the stock tim is trash and iirc you can get upto 20c lower temps with liquid metal

Bought it as "delided" (sic!) on eBay. Cooling seems to be fine with a Noctua NH-D14. Got some thermal throttling in Cinebench R23 but it's still way above stock. It's not gonna run at 100% load for very long anyway, I'm just compiling some code which takes about a minute at a time.

4 hours ago, Jurrunio said:

If you're talking about the "how many cores are under work" kind, just don't. Not worth the time and effort to control 18 cores with the same core voltage. Let them run at the same clocks.

That's what I did. The only alternative is limiting it to 3.9 GHz no matter what the load is. That's below stock though, which is 4.4 GHz for 2 cores.

4 hours ago, Jurrunio said:

And reduced voltage (what undervolt means) hurts stability at higher frequencies, this is a sign -0.02V is too low for more than 3.9GHz. Stock is just 3.4GHz on all cores.

I've reduced the turbo OC for 3-12 cores by 0.1 GHz so we'll see I guess. If that doesn't work I somehow need to test single cores or some combinations of cores. 4.4 GHz seems to be fine because every core has hit it at some point without crashing, though it's for a very short time, so it's hard to tell. I may even reduce that to 4.3 GHz because it's running at that speed for maybe 1% of the time so it shouldn't matter much.

I could try to increase the offset to -0.01 V but that might trigger the OCP. It's literally a couple Watt difference between stock and -0.02 V, I'm very close to the limit apparently.

Jurrunio · January 10, 2022

4 hours ago, Magogan said:

That's what I did. The only alternative is limiting it to 3.9 GHz no matter what the load is. That's below stock though, which is 4.4 GHz for 2 cores.

It could also be a VCCIN (CPU input voltage) problem, that defaults to 1.8V but I'd keep it at 2V because it lets the VRM run cooler and doesnt hurt the CPU (and is required when overclocking)