Curve Optimizer maximum offset?

[GotNoRice]

I'm just curious what the maximum Curve Optimizer offset is supposed to be?

With my 9800X3D and Gigabyte Aorus Pro X870E, I have it set as -60 (all core) in the BIOS (anything -61 or more returns "invalid input range"). But Ryzen Master reports Curve Optimizer as still being set as -50. I've rebooted multiple times to confirm that it's still set at -60 in the BIOS, but I'm not sure which one is correct. Up until -50, both the BIOS and Ryzen Master always reported the same value.

Searching the internet, I found one post from about two years ago from a supposed AMD employee claiming that the Maximum in a then current AGESA update was -50 for Zen 4 CPUs. Whether that is/was true, or whether that still applies to the current AGESA, Zen 5, and/or X870E, I don't know. I've also found posts from owners of other motherboards (including other brands) that claim their boards also allow values up to -60.

I'm not sure how else to confirm what the Curve Optimizer offset is actually set at other than Ryzen Master, to know if it's actually at -60 (as set in the BIOS) or -50 (as reported by Ryzen Master).

 

[tabascosauz]

I don't think I've heard anything I'd regard as definitive proof that beyond -50 is possible yet. Reddit is a no-go for being authoritative.

I'm aware that some board vendors set the limit at -60, but iirc they didn't wait until the ceiling was officially raised in Raphael either.

In any case, -50 is still a hefty bar for most CPUs to actually attain with true stability under torture (long testing with HYDRA). Make sure you test and that clocks and performance actually increase - it's not just about running some OCCT and calling the win just cause it doesn't crash usually.

 

[GotNoRice]

In any case, -50 is still a hefty bar for most CPUs to actually attain with true stability under torture (long testing with HYDRA). Make sure you test and that clocks and performance actually increase - it's not just about running some OCCT and calling the win just cause it doesn't crash usually.

Absolutely. It took me a good 6 months or so to fully dial-in my 5800X3D. I've been using Core-cycler, along with a mix of all-core loads, as well as regular gaming and normal usage to test stability. With my 9800X3D, I've been taking it slow, increasing my offset and then waiting a few days before increasing it more. I thought that I was doing good when I got to -30. I'm blown-away that I've even got to the point where the max limit would potentially be an issue. Part of me almost wonders if Curve Optimizer is working correctly, since it seems almost too good to be true. My 5800X3D was very picky and took a lot of work to really dial-in, whereas this CPU has not been picky at all so far.

I'm currently able to maintain an all-core boost speed of over 5.5Ghz at 1.155v during all-core workloads, using curve optimizer and a 102 BCLK. Everything seems stable, in fact, I have not had even one single BSOD or crash/freeze since my mobo/CPU/RAM upgrade. If this holds up and remains stable, this would even take the crown from my 2500k as the best CPU I've ever owned IMO.

 

[Launcestonian]

What apps are you using to test your OC settings?

 

[GotNoRice]

What apps are you using to test your OC settings?

Mostly CoreCycler and just using the computer like normal, since a large Curve Optimizer offset usually causes instability at idle and during lightly-threaded loads, but I've also used Cinebench R23 and Prime95 to test All-Core loads.

 

[tabascosauz]

Mostly CoreCycler and just using the computer like normal, since a large Curve Optimizer offset usually causes instability at idle and during lightly-threaded loads, but I've also used Cinebench R23 and Prime95 to test All-Core loads.

I use corecycler on AM4, pretty reliable there, but not on AM5.

By the standards of that test suite my 7800X3D can run all day at -35. By Hydra's standards, it gets humbled quite a bit. Hence the remark about actual stability.

 

[GotNoRice]

I use corecycler on AM4, pretty reliable there, but not on AM5.

By the standards of that test suite my 7800X3D can run all day at -35. By Hydra's standards, it gets humbled quite a bit. Hence the remark about actual stability.

I'll investigate the program and give it a shot. But at the moment I'm in a different place than I was at with my 5800X3D. With my 5800X3D I was getting random occasional freezes, and was doing everything I could to identify the problem core(s) and exactly what settings I needed to make it work. This time it's different. Everything is totally stable during normal usage. There is a point where, if I can go weeks if not months of actual normal usage without a single BSOD or freeze, then I don't really care if it's stable or not when running an artificial stability test. But since that much time has not elapsed yet, I'll still give it a try.

 

[tabascosauz]

I'll investigate the program and give it a shot. But at the moment I'm in a different place than I was at with my 5800X3D. With my 5800X3D I was getting random occasional freezes, and was doing everything I could to identify the problem core(s) and exactly what settings I needed to make it work. This time it's different. Everything is totally stable during normal usage. There is a point where, if I can go weeks if not months of actual normal usage without a single BSOD or freeze, then I don't really care if it's stable or not when running an artificial stability test. But since that much time has not elapsed yet, I'll still give it a try.

Don't get me wrong, there are AM5 samples that are truly golden up to -50 on every core. Just a thought, to be sure with your testing.

Hydra isn't freeware, so just be aware of that. Good enough is still good enough - just be aware that Zen 4/5 does not behave quite the same way as predecessors. Solely SSE max freq is not the whole stability question anymore, AVX is worth a look too.

 

[Zach_01]

I'm just curious what the maximum Curve Optimizer offset is supposed to be?

With my 9800X3D and Gigabyte Aorus Pro X870E, I have it set as -60 (all core) in the BIOS (anything -61 or more returns "invalid input range"). But Ryzen Master reports Curve Optimizer as still being set as -50. I've rebooted multiple times to confirm that it's still set at -60 in the BIOS, but I'm not sure which one is correct. Up until -50, both the BIOS and Ryzen Master always reported the same value.

Searching the internet, I found one post from about two years ago from a supposed AMD employee claiming that the Maximum in a then current AGESA update was -50 for Zen 4 CPUs. Whether that is/was true, or whether that still applies to the current AGESA, Zen 5, and/or X870E, I don't know. I've also found posts from owners of other motherboards (including other brands) that claim their boards also allow values up to -60.

I'm not sure how else to confirm what the Curve Optimizer offset is actually set at other than Ryzen Master, to know if it's actually at -60 (as set in the BIOS) or -50 (as reported by Ryzen Master).

+/-30 steps for 5000series (90~150mV)
+/-50 steps for 7000/9000series (150~250mV)

Never heard anything beyond -50steps. And very few chips can actually be fully stable past -35~40
And I’m under the impression that AM5 CPUs initially was up to +/-40 steps(?).

Do you notice anything different in CPU behavior (speed, temp, V) past -50? Or even going from -40 to -50?

 

[GotNoRice]

+/-30 steps for 5000series (90~150mV)
+/-50 steps for 7000/9000series (150~250mV)

Never heard anything beyond -50steps. And very few chips can actually be fully stable past -35~40
And I’m under the impression that AM5 CPUs initially was up to +/-40 steps(?).

Yeah it's frustratingly difficult to find any kind of official information about it.

I found a ~2 year old reddit post from a supposed "Verified AMD Employee" referencing a change to a max of -50:

https://www.reddit.com/r/Amd/comments/12i30it/ryzen_7000_maximum_curve_optimiser_value/

We extended the Curve Optimizer range with the release of the 7000x3d CPUs. I think that should be AGESA 1005 and higher.

We found some AM5 CPUs could benefit from a bigger range (I think Amit suggested it), so we implemented it.

But I've also seen references from people with ASUS AM5 boards about also being able to set Curve Optimizer offsets up to -60 (I have a Gigabyte board). So it seems a bit of an odd coincidence that multiple different brands would have -60 as the maximum offset if -50 is the actual limit. I'm certainly not claiming that as proof either, it just makes things more confusing.

Do you notice anything different in CPU behavior (speed, temp, V) past -50? Or even going from -40 to -50?

I have not been taking extremely granular measurements at each step. I just assumed that I would find my unstable all-core limit quickly and it would mainly be a battle of finding my weakest core (since that's how it played out with my 5800X3D). But instead, I never reached the point of instability.

The biggest difference overall is that earlier in the process it was only boosting to ~4.8 during all-core loads. Then it got to the point that it was boosting to ~5.3Ghz during all-core loads. Now at -60(as set in the BIOS) it maintains ~5.5Ghz all-core boost at 1.155v during all-core loads, which is the same speed it boosts to during lighter loads. Temps are usually in the high 70's when running something like CinebenchR23 MultiCore, using my 5-year old Cooler Master ML360R AIO with 6 fans in push-pull and a somewhat aggressive fan curve.

 

[Ruru]

I have a 5800X so not comparable, but Ryzen Master says that all are stable at -30, but after few crashes, I ended up with -20 all-core.

 

[tabascosauz]

But I've also seen references from people with ASUS AM5 boards about also being able to set Curve Optimizer offsets up to -60 (I have a Gigabyte board). So it seems a bit of an odd coincidence that multiple different brands would have -60 as the maximum offset if -50 is the actual limit. I'm certainly not claiming that as proof either, it just makes things more confusing.

The biggest difference overall is that earlier in the process it was only boosting to ~4.8 during all-core loads. Then it got to the point that it was boosting to ~5.3Ghz during all-core loads. Now at -60(as set in the BIOS) it maintains ~5.5Ghz all-core boost at 1.155v during all-core loads, which is the same speed it boosts to during lighter loads. Temps are usually in the high 70's when running something like CinebenchR23 MultiCore, using my 5-year old Cooler Master ML360R AIO with 6 fans in push-pull and a somewhat aggressive fan curve.

Do you notice an actual difference between -50 and -60 though? Corroborated by HWInfo metrics under load. I assume the 4.8 all-core is referring to stock 0 CO.

Often times both board vendors and AGESA can slip things in ahead of time, which aren't functional yet so make no sense until officially released.

And do make sure that any clock increase corresponds to an actual score improvement - if not, usually a pretty clear sign of instability.

It does seem like Granite Ridge stock settings are hiding a lot of oomph that is otherwise expected from a move to mature N4 though, so it could also just be a golden goose at -60! Higher (lower?) stable CO is always a good thing - means yields are both improving and also holding more in reserve for stability, both of which are always good.

 

[GotNoRice]

Do you notice an actual difference between -50 and -60 though? Corroborated by HWInfo metrics under load. I assume the 4.8 all-core is referring to stock 0 CO.

No I don't notice a difference between -50 and -60. It was already maintaining maximum all-core boost at -50.

 

[tabascosauz]

No I don't notice a difference between -50 and -60. It was already maintaining maximum all-core boost at -50.

Exact same Vcore?

If so, then it doesn't sound like -60 is actually doing anything (yet).

 

[GotNoRice]

Exact same Vcore?

I don't know. I was mostly testing for stability while monitoring boost clocks. Maybe at some point I'll go back into the BIOS, set it to -50, and see if the Vcore changes at all.

 

[Zach_01]

I don't know. I was mostly testing for stability while monitoring boost clocks. Maybe at some point I'll go back into the BIOS, set it to -50, and see if the Vcore changes at all.

I bet its not easy task to come into a solid conclusion.
Need to keep tracking speed, voltage (SVI3 TFN), temp as well but only with exactly the same cooling conditions (CPU cooler speed, case fan speed, ambient temp) at least.
Maybe CPU PPT and EDC too

For sure, 5.3~5.5GHz all 8 cores and under 80°C... sounds like a really good piece of silicon.
Did it get up to 150~160W? (CPU PPT)

 

[GotNoRice]

Did it get up to 150~160W? (CPU PPT)

In CinebenchR23 (Multi Core), CPU PPT shows about 151W