5800X Boost set to 5050Mhz all cores with AMD Curve Optimizer

[MikeS3000]

Glad I'm not the only one worried about degradation. I've basically been on the same Bios since I got the CPU 10 days ago. I booted at -5 on best 2 cores and -15 on other cores on 5800x. I found this wasnt's totally stable so I have been running -2 on best and -10 on others. This passed CB20 and CB23 single core without error and OCCT large. Slowly single core in CB20 and CB23 has been failing and I keep increasing voltage to best cores. I was down to -2 and thought I was stable. I had about 5 failures in a row on CB this morning. I will test again tonight but this doesn't make sense.

 

[Dreamic]

.

 

[MikeS3000]

All we can do is speculate. Now I did have pretty chilly temps this morning in my office around 17 to 18c. I wonder if the CPU tried to hit higher single core boost frequencies than previous tests and it just wasn't stable. Usually I boost the 2 best cores around 4900 to 4950 in Ryzen Master. I have a +200 boost override setup. I'll retest tonight with warmer temps.

 

[gerardfraser]

A CPU will last 10 years ,lol you gals/guys can not be serious,so soon after release on lower CPU voltages. If it was an Intel CPU would you say the same.F This I am going to sell my 5800X and buy an Intel 9900K there cheap and I will not have to worry about degradation .

 

[MikeS3000]

Don't know the reason, all I can do is report my findings. This 5800x applies a heck of a lot more all-core voltage than my 3900x with PBO and all the overclocking goodies turned on. I know AMD says this is by design, but still can't explain why I could complete CB20 and CB23 10 days ago at -4 or -5 and now -2 is failing repeatedly.

 

[Dean Guile]

Anyone wanting to find something to make a reboot happen is to test system stability, as there's nothing worse than thinking your finally stable, then after 3 days a reboot out of nowhere and back to bios.

I found this out by pure accident and have no idea why it causes a reboot, but Utorrent, I have 6 torrents currently seeding only, but everytime I start utorrent within 10-15mins a reboot has happened every time on an unstable setup.
I found a correlation also between having the client on the desktop or minimized to taskbar, if it's minimised doesn't seem to trigger so much, so what I do is open utorrent, whip it over to my 2nd monitor and hope i'm now stable, which after 30mins at the moment I am.

 

[gerardfraser]

@MikeS3000 I do not think you lying.
I put my 5800X up for sale and got 9900K and cheap Z390 at my local computer store.Always wanted to try 9900K for fun.

 

[Dean Guile]

@MikeS3000 I do not think you lying.
I put my 5800X up for sale and got 9900K and cheap Z390 at my local computer store.Always wanted to try 9900K for fun.

Ha funny! Imagine someone ACTUALLY doing this, anyway lets get back on topic.,

 

[MikeS3000]

I'm going to blame Gigabyte's bios and hopefully my 5800x can return to running more undervolt on later revisions. Not a good sign that we are on BIOS f31o. What happens when they hit "z"? Probably that new AGESA 1.1.0.8 will drop any day now. Zen2 had teething issues so no surprise that zen3 has some hiccups.

 

[alec]

Don't know the reason, all I can do is report my findings. This 5800x applies a heck of a lot more all-core voltage than my 3900x with PBO and all the overclocking goodies turned on. I know AMD says this is by design, but still can't explain why I could complete CB20 and CB23 10 days ago at -4 or -5 and now -2 is failing repeatedly.

I suggest resetting bios and starting from scratch. Gave another try at curve optimizer after last bios update, so far much better results than before.
Difference is I didn't mess with BCLK at all this time. Gigabyte bios gets into weird states sometimes, even after you revert all changes.

Btw bclk 102 has very similar effect (performance/volt/temp wise) as Curve optimizer -10 combined with PBO +100MHz. You could try if that works for you, just note the memory clock and fclck both go up as well.

 

[DarkBahamut]

Thing is degradation doesn't make much sense since we're literally undervolting, it's supposed to do the opposite of degradation so idk why my PC is starting to reboot because it's not enough juice or something

The thing is, curve optimiser isn't really undervolting. Not how most users would expect it, anyway.

It's true that for a given frequency a negative curve optimiser will result in a lower voltage, but peak absolute voltages are unchanged. On my 5950X. At stock, 1T peak boost vCore hits 1.488v. With CO -20 it hits, you guessed it, 1.488v. The same is true of multi thread loads too, the exact same voltages are seen.

Precision boost uses the voltage headroom to further increase the peak/sustained boost clocks, which results in the same absolute voltages before and after CO. So for 99% of workloads, you'll never see an 'undervolt' (less voltage). You'll just see more clockspeed. So it's far more accurate to call curve optimiser an overclock, and not really an undervolt at all.

The idle crashes are on C6 exit transitions because CO has this 'smart' algorithm that tugs extra voltage out at low load, but then the CPU can't cope as the voltage doesn't recover in the ~1-2ms when a load becomes heavy.

 

[Dreamic]

.

 

[woppy101]

Finally got my cores boosting a bit better, still can’t break the 5ghz mark constantly though, but with all cores loaded with real bench I get 4.4ghz over all cores, also got my 3200c14 memory running at 3800 c16

 

[DarkBahamut]

Guess my 5800x is a trash overclocker then, who knows if -5 All Core is even stable only time will tell. Thought I was good after -10 for weeks. Can only post at 1867 FCLK as well, hoping a new BIOS might help whatever century MSI feels like getting around to that for the B550 Tomahawk, sure taking their time doing nothing.

In reality mine is no better either. While I can run Cinebench R23 and even video encode on -20, that isn't stable for gaming. Even -5 results in random reboots while gaming, so I personally have curve optimiser disabled now. I'm also stuck at 1867 FCLK too (no post at 1900 ). I don't think there is much lost here really. Even at -20 I only gained 3% single thread and 5% multi thread in Cinebench so the gains are hardly earth shattering. Not big enough that I'd consider risking system stability for, anyway!

Thinking back on the discussions of degradation, it should be noted that negative CO values will result in higher degradation in the longer term. As load voltages remain unchanged but clocks go up, you also see increased power and temps. At -20 in CB R23 MT I observed an additional 20 W load power consumption and an extra 5 C CPU temps. As load voltages aren't exactly low, there can be little argument that the increased currents and temps seen will eventually wear the chip more than it would otherwise. If it will be within a meaningful length of time.. who knows. But for anyone concerned about chip longevity, curve optimiser isn't here to help you.

 

[OC2000]

I've noticed Curve Optimizer puts a lot of stress on the Imc. 1900 FCLK becomes unstable when CO is used. Im testing it at the moment with +75, 10x scaler and CO -5 / -10. I managed a stable run of Single CPU on R20 with +50 -5/-10 for a 650 score, but Im not sure about the long term stability of it.
100+ crashes the system instantly, yet with RAM set to 3600 I can run at +200

 

[alec]

additional 20 W load power consumption

if PBO limits for EDC and PPT are kept default, this shouldn't happen. Then it is not CO that is contributing to (mythical) degradation, its the power limit increase.

 

[alec]

I can run at +200

Are you sure it is a stable mode? And is there any measurable gain over +100?

 

[OC2000]

Are you sure it is a stable mode? And is there any measurable gain over +100?

I think it wouldn’t matter if you set it to +1000 MHz

with PBO on and CO off it maxes out at 5075 on a single core. Introducing CO raises the core over 5100 and the limits then apply as some cores can’t handle 5100.
I’m working out now how to keep the limit raised for the cores that can handle over 5100 and keep those that can’t below.

 

[DarkBahamut]

if PBO limits for EDC and PPT are kept default, this shouldn't happen. Then it is not CO that is contributing to (mythical) degradation, its the power limit increase.

The PPT limit on this board with PBO enabled is 500 W, it's not power limited in either situation. The PPT/EDC/TDC are identical with and without CO. The difference is without CO it runs at roughly ~4.4 GHz @ 1.25v, after CO -20 it runs at roughly 4.55 GHz @ 1.25v. That extra clockspeed is not free. The CPU is running against what it feels it's safe voltage is so just uses CO to ramp clocks hard which in turn increase power and thermals.

There's nothing about it's operation that is mysterious, the results are exactly as expected. If a lower power limit was engaged (say PPT = 200 W) then CO would result in reduced load voltages as it would have to offset the increased clocks with a lower vcore. But when unbounded by power limits CO will always result in higher power as it just pushes the clocks with no reduction in peak voltage.

 

[alec]

PPT limit on this board with PBO enabled is 500 W

Ah. I'm not interested in (limited by stock power limits) multicore perf, only single core and gaming style loads. So EDC and PPT stay stock for me. Multicore clocks improve a little with CO, but power stays same.