Curve Optimizer on 5900X - Better than manual OC!

[mrdoubtfull]

What's up??

 

Copying this over from my post on reddit. Figured it might be of interest to someone, maybe not.. Whatever  

 

I have a 5900X and I'm currently playing with the "Curve Optimizer" and am getting the best results I've seen on my CPU!

 

I'm getting 646 single core and over 8900 muti in r20!

 

A little context..

 

I thought I was getting the best performance I could without a manual overclock by using custom PBO settings and setting the curve optimizer to negative 10 all core. 

 

I couldn't get lower than negative 10 to work for all core. I even tried doing per-core and at first I tried increasing on what I thought were my best cores (through HWInfo).. I tried -15 on a few different ones at different times and it just didn't work for me.

 

I was tired of trying things and gave up for a while..

 

With a manual 4.7GHz OC I was looking at 8900-9100 (mostly below 9000) in r20 but my single core score would suffer, dropping to ~617 from ~635. I even tried per CCX (or CCD?) overclocking and it just wasn't stable, 4.7GHz all-core seemed to be the best manual OC I could get.

 

Before, I was using the curve optimizer and only PBO and was seeing ~8500 multi and - 625 single which was easy to reproduce.

 

With custom PBO + negative 10 all-core curve optimizer I was just barely getting over 8700 multi and ~635 single (642 was max I've EVER seen and couldn't reproduce it). 

 

After adjusting the curve optimizer per core, along with PBO, I was able to get 8950 multi and 646 single in r20. Over 640 is now common.

 

That's nearly as good as a manual OC, I can literally get the same multi core score now AND get higher single core score! That's exactly what I've been working towards!

 

ALL-CORE OVERCLOCKING IS DEAD!

 

I was hitting 5GHz here and there but now I'm hitting over that just ever so slightly AND more often. I'm seeing 5150MHz on one core, 5075MHz on another and 5050MHz on the third one that reaches 5GHz. I've actually seen two of them hit 5150MHz but not sure if was at the same time (probably not).

 

I'm seeing 4850MHz and 4775MHz often on several cores (up to 6) while playing Apex when it use to be 4400-4675MHz. I'm also seeing 5GHz while playing when I've never seen that before while playing Apex. What this translates to actual FPS I have no fucking clue. 

 

What did I do? 

 

After seeing the thread about PBO2 with the video (didn't watch it) and seeing AMD_Roberts replies, it regained my interest, so I went back to the BIOS to try a couple things..

 

Instead of sticking to one number for curve optimizer for all core, or just trying to set what I thought were the best cores to lower numbers, I actually started from the bottom, raising each number except for the top 5 (because I tried higher numbers before and failed) and did a lot of testing. I could get the bottom number to 70 but that didn't help much more than 40.

 

Currently, this is what's getting the results above (curve optimizer per-core):

1. Negative 10

2. Negative 10

3. Negative 10

4. Negative 10

5. Negative 10

6. Negative 35

7. Negative 35

8. Negative 35

9. Negative 35

10. Negative 35

11. Negative 35

12. Negative 35

Seems simple right? Maybe some of you already figured this out, but for those who want to push your CPU a little more, do some testing. Try my numbers. BUT also try testing other numbers and combinations.. Each CPU may work differently.

 

I did eventually try higher numbers again in the top 5 but it resulted in crashes or lower r20 scores. I tried positive numbers here and there, I tried keeping a few cores the same numbers, I tried decreasing in increments from bottom to top. For instance, I tried (from the bottom) 70, 60, 50, 40, 35, 30, 25, 15, 10, 10, 10, 10.. I saw a decrease in performance but it still allowed me to run tests without crashing. This is all kind of weird and new to me.

 

My PBO settings are manually set to:

• Precision Boost Overdrive: ADVANCED

• PBO Limits: MOTHERBOARD

• PBO Scalar: Manual, 5X

• MAX CPU PBO: 200MHz

• Platform Thermal Throttle Limit: 255 (max)

On MSI you can just set to 'Enhance Level 2' to get this.

 

SOMETHING IMPORTANT TO NOTE: before I was playing heavily with the curve optimizer, I found that OCing the RAM, whether increasing frequency or tightening the timings, DECREASED PERFORMANCE (on top of the WHEA errors obviously)!!!! That's why I'm running XMP ONLY at the moment. But after playing with the curve, I might try tweaking it again, but may have to wait for a BIOS update. Not sure.

 

Here's my specs (RAM is running XMP only!):

• 5900X

• B550 Gaming Carbon

• GTX 1080 TI

• 2x16GB Crucial Ballistix RGB 3600 Cl16 Dual Rank

• RM850

[mrdoubtfull]

 Looking at a couple reviews, I noticed they had scores similar to this, at like default, but that's not what me or  A LOT of others got at default OR with PBO!

 

So maybe this can help someone.. Or not..

[Letgomyleghoe]

can you do a bench in R23?

[mrdoubtfull]

2 minutes ago, Letgomyleghoe said:

can you do a bench in R23?

No.

 

 

JK. give me a minute

[mrdoubtfull]

8 minutes ago, Letgomyleghoe said:

can you do a bench in R23?

First time using.. Should I turn trest duration off or leave it at 10 min?

[Letgomyleghoe]

7 minutes ago, mrdoubtfull said:

First time using.. Should I turn trest duration off or leave it at 10 min?

either way is fine

[mrdoubtfull]

11 minutes ago, Letgomyleghoe said:

either way is fine

Haven't seen much info on r23, any good?

 

[Letgomyleghoe]

41 minutes ago, mrdoubtfull said:

Haven't seen much info on r23, any good?

just a tad over a normal 5900x bench 

https://www.cpu-monkey.com/en/cpu-amd_ryzen_9_5900x-1748

[x10kuliai]

Well, i dont get anywhere near the results reviewers get on their 5900x processors. At stock with C R23 i get around 20500 on stock, 21500 with PBO enabled. And with these settings mrdoubtfull has shared with us i get 22500 and its running stable it seems, so thanks for your findings. Maybe you have tweaked it a little more since posing this and got more perfomance? 

[mrdoubtfull]

On 12/31/2020 at 1:26 AM, x10kuliai said:

Well, i dont get anywhere near the results reviewers get on their 5900x processors. At stock with C R23 i get around 20500 on stock, 21500 with PBO enabled. And with these settings mrdoubtfull has shared with us i get 22500 and its running stable it seems, so thanks for your findings. Maybe you have tweaked it a little more since posing this and got more perfomance? 

Glad it helped you! I haven't bothered with it much more since, I actually backed off slightly in favor of stability. 

 

Latest BIOS on my MSI B550 decreases performance and boost clocks, BIOS version definitely plays a role as well. I reverted to the previous one. 

[blockbusta]

Are your temps really high when doing that overclock and do you offset your voltage?

 

Can't seem to get past 20500 with PBO and curve optimizer.

 

Currently stable OC for me is:

Vcore voltage, Offset, -0.125 volts

PBO Advanced

Scalar 10x

200 MHz

PBO limit Auto

 

Curve Optimizer 

All cores

Negative 14

 

4.6 Ghz average and some boost to 5.15/5.05 Ghz on Cinebench with temps near 75C

 

I've gotten other OC that were a bit higher doing per core (negative 14 on the fastest cores from Ryzen Master and negative 24 on all others) but temps with cinebench were up to 90C with Noctua Dh15 so I went down to more stable. 

 

Think im doing something wrong haha. 

 

[x10kuliai]

4 hours ago, blockbusta said:

Are your temps really high when doing that overclock and do you offset your voltage?

 

Can't seem to get past 20500 with PBO and curve optimizer.

 

Currently stable OC for me is:

Vcore voltage, Offset, -0.125 volts

PBO Advanced

Scalar 10x

200 MHz

PBO limit Auto

 

Curve Optimizer 

All cores

Negative 14

 

4.6 Ghz average and some boost to 5.15/5.05 Ghz on Cinebench with temps near 75C

 

I've gotten other OC that were a bit higher doing per core (negative 14 on the fastest cores from Ryzen Master and negative 24 on all others) but temps with cinebench were up to 90C with Noctua Dh15 so I went down to more stable. 

 

Think im doing something wrong haha. 

 

Hey there. I just hit the same settings that MRdoubtfull posted and it worked got CINEBENCH score of ~22500 from 1st try. sometimes it goes to 22300,223500,22400.... Temps while running CINE for me was MAX 82-83C.  Correct me if i'm wrong, but i think Ryzen will start throttle down if it goes above 85C for you. So that might be the difference. Curve only allows for freq to go up if temp allows it.   I am using ARCTIC liquid freezer II 280mm AIO, that is one of bigger AIOs you can get not considering 360mm ones. So if its struggling that much  its no surprise to me that NHD15 will struggle more as liquid cooler takes time to absorb all the heat around 3min and CINE ends quicker, so AIO can hold higher clock on very short burst task like this. If i ran CINE on the loop for 10-15in i might not boost as high too maybe.

 

[blockbusta]

7 hours ago, x10kuliai said:

Hey there. I just hit the same settings that MRdoubtfull posted and it worked got CINEBENCH score of ~22500 from 1st try. sometimes it goes to 22300,223500,22400.... Temps while running CINE for me was MAX 82-83C.  Correct me if i'm wrong, but i think Ryzen will start throttle down if it goes above 85C for you. So that might be the difference. Curve only allows for freq to go up if temp allows it.   I am using ARCTIC liquid freezer II 280mm AIO, that is one of bigger AIOs you can get not considering 360mm ones. So if its struggling that much  its no surprise to me that NHD15 will struggle more as liquid cooler takes time to absorb all the heat around 3min and CINE ends quicker, so AIO can hold higher clock on very short burst task like this. If i ran CINE on the loop for 10-15in i might not boost as high too maybe.

 

Interesting...

 

I have the asus x570-e gaming motherboard so can only do up to 30 on the curve optimizer.

 

Seems like I got a stable overclock doing -10 on the fastest cores and -20 on the rest with similar cinebench results but seems like clocks are faster...

 

I wonder if its my memory since I have 4 sticks of 16 gb gskill ram. Also, had to make the voltage offset to -0.11875 though to get temperatures where I like them. I guess that's my real question with regards to the memory and cpu voltage, since I've seen people overclocking and getting some good results but for me I always have to offset the voltage otherwise the temps are way too high on Auto.

[x10kuliai]

3 hours ago, blockbusta said:

Interesting...

 

I have the asus x570-e gaming motherboard so can only do up to 30 on the curve optimizer.

 

Seems like I got a stable overclock doing -10 on the fastest cores and -20 on the rest with similar cinebench results but seems like clocks are faster...

 

I wonder if its my memory since I have 4 sticks of 16 gb gskill ram. Also, had to make the voltage offset to -0.11875 though to get temperatures where I like them. I guess that's my real question with regards to the memory and cpu voltage, since I've seen people overclocking and getting some good results but for me I always have to offset the voltage otherwise the temps are way too high on Auto.

Well not all motherboards are the same, X570 MAG Tomahawk is one of cheapest X570 boards and it has VRMs on par with 400-500dol boards and i think that might affect it to a degree, but it also might be silicon lottery was worse for your chip as asus x570 should be good for overclock. 

[blockbusta]

I just noticed the x570 strix from Asus doesn't have agesa 1180 yet haha.

 

Maybe i'll revisit this when the bios firmware is more ironed out...

 

Thanks for the suggestions!

[x10kuliai]

4 hours ago, blockbusta said:

I just noticed the x570 strix from Asus doesn't have agesa 1180 yet haha.

 

Maybe i'll revisit this when the bios firmware is more ironed out...

 

Thanks for the suggestions!

yeah mine does not have 1.1.8.0 too, it is said to come out early this month so i am eagerly waiting for it, it should make curve optimiser work better also add support for nvidia direct memory similar to AMD SAM. 

[blockbusta]

On 1/3/2021 at 6:03 AM, x10kuliai said:

yeah mine does not have 1.1.8.0 too, it is said to come out early this month so i am eagerly waiting for it, it should make curve optimiser work better also add support for nvidia direct memory similar to AMD SAM. 

Hey, is a 280mm AIO really much better than Noctua Dh15? I thought its only a couple degrees difference that's why I always stuck with big air coolers. Also less hassling with wires...

 

Was looking at the new phanteks line that just came out as they look pretty nice! 

[x10kuliai]

29 minutes ago, blockbusta said:

Hey, is a 280mm AIO really much better than Noctua Dh15? I thought its only a couple degrees difference that's why I always stuck with big air coolers. Also less hassling with wires...

 

Was looking at the new phanteks line that just came out as they look pretty nice! 

Not sure, ARCTIC Liquid freezer II 240,280,360 AIOs are considered one of the best AIO's around, even though they are on the cheap side considering. Not sure if it will last. Gamers nexus gave it real good review and said its generally very good product. It should be better than NHD15 only by 2-4C i think. Its just better on short tasks like CINEBENCH i think. After a while if i ran it on a loop for lets say 1hour i think my AIO would get hotter too. 

[mrdoubtfull]

8 hours ago, blockbusta said:

Hey, is a 280mm AIO really much better than Noctua Dh15? I thought its only a couple degrees difference that's why I always stuck with big air coolers. Also less hassling with wires...

 

Was looking at the new phanteks line that just came out as they look pretty nice! 

I've never used a good air cooler as I went straight to AIOs and custom loops when I started building about 3 years ago. But right now I have a cheap $70 240 Cooler Master AIO (until I do another custom loop when I get 68(9)00 XT) and I'm barely hitting 70C while gaming for hours, mostly Warzone, MW, and Apex. I'm honestly surprised how cool it's keeping my 5900X, but I also think I got pretty decent silicon. 

 

I'll check my temps while running r20/r23 and report back, but like the other guy (girl?) said, AIOs take longer to reach equilibrium, so gaming is really the only good way to measure temps (unless you do other intensive things like video editing/streaming). I forgot what the user had but I think they had the same cooler as you and they said they were getting 10-15C+ higher temps than me in Warzone. 

 

To your earlier truly: I don't undervolt and the only time I tried it locked my frequency so I haven't tried since, but might give it a go when the next BIOS comes out for my board. 

 

 

[x10kuliai]

6 hours ago, mrdoubtfull said:

I've never used a good air cooler as I went straight to AIOs and custom loops when I started building about 3 years ago. But right now I have a cheap $70 240 Cooler Master AIO (until I do another custom loop when I get 68(9)00 XT) and I'm barely hitting 70C while gaming for hours, mostly Warzone, MW, and Apex. I'm honestly surprised how cool it's keeping my 5900X, but I also think I got pretty decent silicon. 

 

I'll check my temps while running r20/r23 and report back, but like the other guy (girl?) said, AIOs take longer to reach equilibrium, so gaming is really the only good way to measure temps (unless you do other intensive things like video editing/streaming). I forgot what the user had but I think they had the same cooler as you and they said they were getting 10-15C+ higher temps than me in Warzone. 

 

To your earlier truly: I don't undervolt and the only time I tried it locked my frequency so I haven't tried since, but might give it a go when the next BIOS comes out for my board. 

 

 

Forgive me if I'm wrong, but you are using curve optimizer at Negative values and that's essentially the same as undervolting just in a smarter way :). Curve optimiser lets you lower voltage curve to get higher overclock at the end of the curve and less heat/power consumption at the beginning of the curve. Its a very interesting technology and as you said earlier its future of overclocking (atleast on ryzen). When it works as advertised you should get lower voltages with higher boosts if your temps allow for it and that is a WIN/WIN situation no matter how you look at it. 

[blockbusta]

Looks like I was getting destroyed by clock stretching from the Asus motherboard with Fmax enabled.

 

Just turned that off and did a little bit less voltage offset to -.125 to get temps lower and got a score of 22046! 

 

I guess the hint was to look at the core clock vs effective clock when cinebench was running and now there isn't such a difference between the two.

 

Still think though since the Asus x570 only has Agesa 1.1.0.0, I'll probably wait until a more ironed out bios. 

[x10kuliai]

5 hours ago, blockbusta said:

Looks like I was getting destroyed by clock stretching from the Asus motherboard with Fmax enabled.

 

Just turned that off and did a little bit less voltage offset to -.125 to get temps lower and got a score of 22046! 

 

I guess the hint was to look at the core clock vs effective clock when cinebench was running and now there isn't such a difference between the two.

 

Still think though since the Asus x570 only has Agesa 1.1.0.0, I'll probably wait until a more ironed out bios. 

Great to hear you have some progress. Yeah AGESA 1.1.8.0 will fix some of things with new features so just have to wait. MSI just posted new BIOS with AGESA 1190 already yesterday they jumped over 1180. I updated and all the curve optimizer features seem to run fine now. One strange thing is with AGESA1100 was able to set negative 35 value and now it wont let me set below negative 30. So i did it like this:
 

1. Negative 10

2. Negative 10

3. Negative 10

4. Negative 10

5. Negative 10

6. Negative 30

7. Negative 30

8. Negative 30

9. Negative 30

10. Negative 30

11. Negative 30

12. Negative 30

And am getting little lower score in CNr23 instead of 22400-22500 getting now 22200, maybe should try playing with these settings some more. RYZEN Clock calculator 2.0 should come out with support for 5xxxx ryzen so hopefully that will help us get easier stable OC to get the most our chips can handle. They said next month they will release 2.1 version and it will do per core OC automatically, that will be so sick . Will be able to get max out of your processor with just one click if it works as they promised.  

[blockbusta]

Ahh I see. Yea I was kind of confused how your older bios let you go past 30 when the amd videos was showing the curve optimizer had a max of 30 counts.  My current asus bios 1.1.0.0 patch C can only go to 30.

 

Yea cant wait. This is already loads better than my i7-6700k lol! 
 

Any reason why you did the first 5 cores as -10 vs doing the 4 fastest cores (ryzen master circle/Star)? 
 

I see some people doing that instead though idk which methodology is better. 

[x10kuliai]

27 minutes ago, blockbusta said:

Ahh I see. Yea I was kind of confused how your older bios let you go past 30 when the amd videos was showing the curve optimizer had a max of 30 counts.  My current asus bios 1.1.0.0 patch C can only go to 30.

 

Yea cant wait. This is already loads better than my i7-6700k lol! 
 

Any reason why you did the first 5 cores as -10 vs doing the 4 fastest cores (ryzen master circle/Star)? 
 

I see some people doing that instead though idk which methodology is better. 

 

 

Its as on the 1st post, these numbers are what mrdoubtfull thought to be best cores. You are right its better to set higher value on best cores. I will try to check best cores (since every CPU is different on that regard) on my ryzen master and change this order. 

On 11/25/2020 at 5:49 AM, mrdoubtfull said:

Instead of sticking to one number for curve optimizer for all core, or just trying to set what I thought were the best cores to lower numbers, I actually started from the bottom, raising each number except for the top 5 (because I tried higher numbers before and failed) and did a lot of testing. I could get the bottom number to 70 but that didn't help much more than 40.

 

[blockbusta]

There was a beta bios with Agesa 1190 for the x570 Asus board so I thought of trying it out.

 

Seems like the curve optimizer is much better with 1190 and no whea errors or random reboots. 

 

I made the voltage auto and LLC auto this time around to see the difference. Seems like for my CPU, I can only decrease the 4 fastest cores (circle/star on Ryzen master) to -10 and then -25 on the rest as every other combination I tried would not boot or windows crash soon after. 

 

The scores I got were pretty good I think. 

 

Hopefully they have a new dram calculator for 5000 series so I can try overclocking that too. Just kept that at XMP settings for now.