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

[UK_SPAWN]

ok so messing with the CPU/FCLK and failed to post. ok normal.... just leave the board and it will retry then reset the CMOS if it fails to boot a few times right?

Well my B550 PRO doesnt seem to want to do it. it also wont clear the CMOS with the jumper (turned off the PSU waited for a bit, cleared power from board pressing power button, shorted jumper for 5-20 seconds etc)

The only thing that clears it is to take out the GPU, remove the battery and then short the battery terminals on the board then reinstall the battery and the GPU.... obviously this is incredibly infuriating and inefficient when you may need to clear the bios 10-20 times in a row when messing with memory/overclocking etc...

Anyone know how to make it auto-clear the cmos after X boot loops/boot failures etc?

Thanks

 

[stooeh]

I tried setting 1.25V core override as per gerard's vid above on my 5800X and it worked as in showed expected boost clocks and did exhibit less performance likely due to clock stretching? (about -400 on multi and -30 on single in R20)

I am using the X570 tomahawk as well though it seems on mine if I enable the PBO feature via OC menu on left button it does nothing and need to enable it within the motherboard settings on top left button. Though I get a BSOD when booting windows using the settings in AMD presentation posted earlier

 

[UK_SPAWN]

Just another thing regarding this curve optimiser.

Just had an email back from AMD stating using curve optimiser with LLC isnt optimal/reccomended. kind of makes sense as if you are reducing load voltages with curve optimiser then LLC is tring to maintain a higher load voltage... kind of fighting eachother

 

[gerardfraser]

Just another thing regarding this curve optimiser.

Just had an email back from AMD stating using curve optimiser with LLC isnt optimal/reccomended. kind of makes sense as if you are reducing load voltages with curve optimiser then LLC is tring to maintain a higher load voltage... kind of fighting eachother

Why would anyone use LLC with the Curve optimizer.That be crazy talk . I would unplug your reset switch on the case and plug it into the JBAT! so you do not have to remove battery. I am guessing you were not hitting the JBAT! correct,I have messed it up before after I was 100% sure I jump it.

I tried setting 1.25V core override as per gerard's vid above on my 5800X and it worked as in showed expected boost clocks and did exhibit less performance likely due to clock stretching? (about -400 on multi and -30 on single in R20)

I am using the X570 tomahawk as well though it seems on mine if I enable the PBO feature via OC menu on left button it does nothing and need to enable it within the motherboard settings on top left button. Though I get a BSOD when booting windows using the settings in AMD presentation posted earlier

Yes the 1.25v was just a random number I hit,just to show boost will work with CPU override. It will defiantly clock stretch at that voltage. It could prove useful for those 5800X 90°C temperatures, weird but could work.

 

[stooeh]

I have not had a proper mess with memory timings yet as resetting the CMOS is a right pain! Might have to connect that reset switch to it hah

Haven't figured out this curve thing yet - best 2 cores with a negative offset and it BSODs but leaving at 0 and sticking a -25 on the rest and able to finish a CB R20 run

6122 MT, 633 ST

 

[Alyjen]

@gerardfraser question. How exactly do you set up your RAM for these tests? I checked some of the videos, and posts on other forums, and it seems you don't really OC FCLK/ram that much as it's not benefiting 4k much.
I ask because CPU SoC Power in your recent videos is like ~6W, while for higher FCLKs it's more of 25W and it draws from the same limit as CPU

 

[ubasti]

Thanks for all the info on this thread, I finally got round to trying a few things yesterday. It seems that on my 5900 the best core will only go -8 in curve optimiser and the rest -5 on my X570 MAG Tomahawk. Still, it was able to boost over 5000 on a single thread, up from 5950, with some small speed gains and temperature reductions on multithread. I will try the core override next as well.

I will probably try pushing the negatives in curve optimiser again when the new AGESA is out in Decemeber. After watching the AMD guide linked above, it sounds like they threw this on certain motherboards almost as a beta for users to discover, so we might see further refinements.

 

[ubasti]

Thanks for all the info on this thread, I finally got round to trying a few things yesterday. It seems that on my 5900 the best core will only go -8 in curve optimiser and the rest -5 on my X570 MAG Tomahawk. Still, it was able to boost over 5000 on a single thread, up from 5950, with some small speed gains and temperature reductions on multithread. I will try the core override next as well.

I will probably try pushing the negatives in curve optimiser again when the new AGESA is out in Decemeber. After watching the AMD guide linked above, it sounds like they threw this on certain motherboards almost as a beta for users to discover, so we might see further refinements.

Sorry meant up from 4950mhz of course!

 

[Guest2]

I tried the 1.250 V override and PBO left on Auto.

Clock stretching seems to happen as all core shows boost to 4725Mhz, single core to 4850Mhz however CPU-Z bench scores are 616 single and 5851 multi. Cinebench Multi is 5208, way less than the ~6050 with PBO just left enabled and all core boost to 4550Mhz.
Temps are lower though, max 65c

1.325 V override and PBO auto results in 4500Mhz all core boost, 4850Mhz single core and correct cores - ~6000 multi core score in R20 and 81c max temp

 

[Guest2]

I have not had a proper mess with memory timings yet as resetting the CMOS is a right pain! Might have to connect that reset switch to it hah

Haven't figured out this curve thing yet - best 2 cores with a negative offset and it BSODs but leaving at 0 and sticking a -25 on the rest and able to finish a CB R20 run

6122 MT, 633 ST

I thought it's a "Good idea to have negative offset on the two best cores" not negative on all of them except the best cores

https://www.anandtech.com/show/1626...ptive-undervolting-for-ryzen-5000-coming-soon

https://images.anandtech.com/doci/16267/AMD Ryzen 5000 Series - Precision Boost Overdrive 2-page-011.jpg

@gerardfraser on your BIOS settings, do you have DigitALL Power all left on auto? This includes settings for LLC and CPU over voltage and under voltage protection.

edit. I have re-read psot 264 and 265 above. Exactly the same issues i'm having. Gerard, which LLC setting do you use as it can't be straight disabled?

 

[alec]

BTW for me many benchmarks and games like BFV and COD 2019 where stable with curve optimizer -15 but Minecraft crashed with WHEA all the way down to 0 offset.

I guess BF V and CoD load multiple threads, so CPU doesn't boost too high. So it doesn't hit modes where curve made it unstable.
This is the downside to overclocking with functioning boost. You never know which core at what load could be unstable at single thread. And it is tricky and time consuming to test all cores properly due to Windws scheduler preferring two fastest cores for most workloads.

 

[alec]

single core stress tests

Stress tests again, put too much continuous stress on a core for it to hit the fastest speed/voltage combinations. And probably the wrong core.
Games are more transient loads, perhaps spread over many threads so scheduler spreads them out as well.

Try https://krakenbenchmark.mozilla.org/
It helps me find unstable modes much quicker.
Possibly run this in two browser windows simultaneously

 

[Guest2]

I don't think I will bother to do any more testing with this 5950X sample, it's utter shaite. More testing will only give me more frustration. Hopefully I will get my new 5950x sample on Wednesday this week, will get back to you then with the results.

What you you doing with your current 5950X?

 

[gerardfraser]

@gerardfraser on your BIOS settings, do you have DigitALL Power all left on auto? This includes settings for LLC and CPU over voltage and under voltage protection.

edit. I have re-read psot 264 and 265 above. Exactly the same issues i'm having. Gerard, which LLC setting do you use as it can't be straight disabled?

Like I said a few times everything is auto voltages except the BIOS settings I showed in Post #1. I done all auto to keep it simple and not confuse people even more.

Also to not get clock stretching you have enough CPU voltage in override mode. That value would be around 1.33v for everyone without any voltage drops to avoid clock stretching in Override mode. There is no way around this or any kind of magic to not have clock stretching in this override mode on low voltages,just saying it some more.

I don't think I will bother to do any more testing with this 5950X sample, it's utter shaite. More testing will only give me more frustration. Hopefully I will get my new 5950x sample on Wednesday this week, will get back to you then with the results.

This is the correct move,I had a 3800XT last month that would crash in Idle ,I returned it. You buy a product if it does not work return it,do not keep buying gear to try to fix it.

@gerardfraser question. How exactly do you set up your RAM for these tests? I checked some of the videos, and posts on other forums, and it seems you don't really OC FCLK/ram that much as it's not benefiting 4k much.
I ask because CPU SoC Power in your recent videos is like ~6W, while for higher FCLKs it's more of 25W and it draws from the same limit as CPU

You are exactly correct ,I did not overclock Ram for any of the test,trying to keep the test simple as possible. ALL test are done at XMP of my currently installed Rank 2 DDR4 CL19 3600Mhz.

Here is a 3800X with DDR4 3866Mhz (1933Mhz FCLK)You can not get any higher on Ryzen 3 FCLK, If you want to check power

Spoiler

 

[ubasti]

That sucks.

With my 5800X I installed Ryzen master, it ranked cores 1 (core 0) and core 7 (core 6) the highest.
I went into BIOS and set PBO t oadvanced, 150Mhz boost, motherboard PBO limits and -15 on cores 0 and 6.
Tested Cinebench R20, instant crash. Pretty sure its a BIOS thing on the MEG Unify. I put money on it doing better on the Tomahawk. Will wait for a new none beta BIOS

With my 5900x on the X570 Tomawhawk when I tried -10 on the 2 best cores it could barely load windows without crashing, I had to bring it down. You may well do better on the Tomahawk, as you say, but it's still not guaranteed for all samples - as you point out it's in many ways a ridiculous lottery that seems to just be accepted as part of life.

 

[Guest2]

I also tried -15 on all cores except strong cores (0 & 6) which I set -5, LLC on mode 4, +200Mhz offset

This held out in R20, but crashed on the menu in a game.

I would be interested swapping chips with Gerard / anyone who seems to have a 'good one' if only to see if the different motherboard has any affect. Of course, this is not possible (well it is but isnt going to happen)

I dont know how AMD / Intel can get away selling chips that are really good / really crap for the same price. Surely just test each one to make sure it can hit X without failing. Basic quality control

 

[UK_SPAWN]

Surely just test each one to make sure it can hit X without failing. Basic quality control

they do, the min and max boost it says on the 'tin' (box) - anything over that is overclock and thats not covered by any of that.

 

[Alyjen]

Like I said a few times everything is auto voltages except the BIOS settings I showed in Post #1. I done all auto to keep it simple and not confuse people even more.

You are exactly correct ,I did not overclock Ram for any of the test,trying to keep the test simple as possible. ALL test are done at XMP of my currently installed Rank 2 DDR4 CL19 3600Mhz.

Here is a 3800X with DDR4 3866Mhz (1933Mhz FCLK)You can not get any higher on Ryzen 3 FCLK, If you want to check power

Spoiler

Thanks, exactly what I wanted to check! Sill power pumped into your SoC is noticeably lower than what I see in my case and may other examples. Maybe that's the difference between various motherboard brands, or it's side effect of your tuning.

 

[gerardfraser]

Thanks, exactly what I wanted to check! Sill power pumped into your SoC is noticeably lower than what I see in my case and may other examples. Maybe that's the difference between various motherboard brands, or it's side effect of your tuning.

Well you can check these if you want. I have owned 6/7 Ryzen CPU's X470/X570 boards blah blah blah. Here is a couple DDR4 Ram clocks

2600X @ 4300Mhz DDR4 3466Mhz MSI X470

Spoiler

3600X @ 4300Mhz DDR4 CL16 4200Mhz MSI X470

Spoiler

 

[Alyjen]

hmm I wonder if the fact that I have pcie AC WLAN addapter, pcie sound card and extra m.2 drive (and quite a few USB devices) contributes to this high SoC power draw, it's closer to 25W like on some of your tests, your 5800X is way lower