AGESA 1.2.0.8

[BatRastard]

Yeah, I'm thinking those with the instability probably have 4 DIMMs of 4000mhz CL14 and got their poor old memory controller working harder than an ugly stripper.

 

[INSTG8R]

It let me use Curve Optimizer for my 5800X3D so I’m very happy with the results and allowed me to OC my 3600 RAM to 3800.

 

[iuliug]

Can the 5800X3D run 4 x16 GB (single sided) at 3600mhz regardless of silicon quality - Micron E-dies for ex? Thanks.

 

[Veseleil]

Can the 5800X3D run 4 x16 GB (single sided) at 3600mhz regardless of silicon quality - Micron E-dies for ex? Thanks.

I'm also curious, but with 4x8GB, and expect at least 3800MHz. But as people said, X3D tolerate slower RAM speeds and worse timings for a decent latency numbers. Anyway I'll share my experience when I get it. Maybe next month.
And BTW, I'm on 1206, and don't plan to go for 1208 if everything goes fine.

 

[Celadoniac]

Does 1.2.0.8 have any effect on 5600G? Or is this only for 5800X3D?

 

[Mussels]

Can the 5800X3D run 4 x16 GB (single sided) at 3600mhz regardless of silicon quality - Micron E-dies for ex? Thanks.

I'm running 2x32GB (four ranks) at 3800, so yes it should be fine.

You'll just need to raise the SoC voltage, i need 1.15v on all my Zen3 based CPU's, 3D or not.

Does 1.2.0.8 have any effect on 5600G? Or is this only for 5800X3D?

It's for all AM4 CPUs, but that doesnt mean it has any major changes to other CPUs, sometimes its just as simple as different default settings

 

[iuliug]

I'm running 2x32GB (four ranks) at 3800, so yes it should be fine.

You'll just need to raise the SoC voltage, i need 1.15v on all my Zen3 based CPU's, 3D or not.


It's for all AM4 CPUs, but that doesnt mean it has any major changes to other CPUs, sometimes its just as simple as different default settings

Hi, what Ram are you using ? Ty

 

[Mussels]

Hi, what Ram are you using ? Ty

It's in my specs, a corsair vengeance 2x32GB 3600 kit

 

[iuliug]

It's in my specs, a corsair vengeance 2x32GB 3600 kit

Ah right, ty - at xmp 3600 (default) is the SOC voltage lower? Is the manual 266mhz OC worth it?

 

[Mussels]

Ah right, ty - at xmp 3600 (default) is the SOC voltage lower? Is the manual 266mhz OC worth it?

SoC voltage doesnt change with XMP, it must be done manually.

I've seen default ranges of 0.9v to 0.975v depending on the motherboard, i don't believe the CPU Controls it.
*Some* boards may change the value when XMP is enabled or some threshold is reached (RAM above a certain speed) but no one documents that sort of thing - some of those 'Auto OC' features do adjust voltages like this, but often far more than needed.

I overclocked my PC just because i could - I had the time to spare. Undervolting (same performance for less power/heat) was more important.
The gains are minimal from 3600 to 3866, but remembering that Zen3 CPU's only officially support 3200 at stock voltages, going from 3200 to 3800+ was a large benefit to feeding my GPU.
(I'm running a 2x2R setup with two dual rank 32GB sticks, and two dummy lighting modules)

Edit: TPU's most recent DDR4 review shows the type of gains and what to expect, with a 2x16 setup

Acer Predator Vesta RGB DDR4-3600 2x 16 GB Review - AMD Performance Results | TechPowerUp

Notice how the *all* the worst results are 2x8 (two single ranks) regardless of their speeds and latencies and that 3200 CL14 and 3600CL14 outperform everything else, despite speeds all the way up to 5066

Maintaining 1:1 ratio (1900 IF/3800 RAM for example) and four ranks is a big performance boost, then its about the best latencies you can get. In 32GB modules, CL18 was the best available for me so I did what i could to regain that loss, with the higher speed.

Edit: updated Aida64 to throw my my own results at 3800 so you can compare to the review ones (I have a 5800x3D that does make it less sensitive to slower timings)

 

[INSTG8R]

SoC voltage doesnt change with XMP, it must be done manually.

I've seen default ranges of 0.9v to 0.975v depending on the motherboard, i don't believe the CPU Controls it.
*Some* boards may change the value when XMP is enabled or some threshold is reached (RAM above a certain speed) but no one documents that sort of thing - some of those 'Auto OC' features do adjust voltages like this, but often far more than needed.

I overclocked my PC just because i could - I had the time to spare. Undervolting (same performance for less power/heat) was more important.
The gains are minimal from 3600 to 3866, but remembering that Zen3 CPU's only officially support 3200 at stock voltages, going from 3200 to 3800+ was a large benefit to feeding my GPU.
(I'm running a 2x2R setup with two dual rank 32GB sticks, and two dummy lighting modules)
View attachment 293662


Edit: TPU's most recent DDR4 review shows the type of gains and what to expect, with a 2x16 setup

Acer Predator Vesta RGB DDR4-3600 2x 16 GB Review - AMD Performance Results | TechPowerUp

Notice how the *all* the worst results are 2x8 (two single ranks) regardless of their speeds and latencies and that 3200 CL14 and 3600CL14 outperform everything else, despite speeds all the way up to 5066

Maintaining 1:1 ratio (1900 IF/3800 RAM for example) and four ranks is a big performance boost, then its about the best latencies you can get. In 32GB modules, CL18 was the best available for me so I did what i could to regain that loss, with the higher speed.
View attachment 293663


Edit: updated Aida64 to throw my my own results at 3800 so you can compare to the review ones (I have a 5800x3D that does make it less sensitive to slower timings)View attachment 293669View attachment 293670

That score isn't so bad! I feel pretty good about mine now but you're totally right about it not making a huge difference. That was a "dirty run" I easily could have closed more apps that had like 7GB in use

 

[freeagent]

I use 1.15v SOC for 1933 running 4x8 GB 1:1

I didn't have SOC LLC enabled, that's why it droops so hard

But it still ok

 

[Veseleil]

I use 1.15v SOC for 1933 running 4x8 GB 1:1

View attachment 293687

I didn't have SOC LLC enabled, that's why it droops so hard

But it still ok

Which memory? By the timings I can assume it's B-die.

 

[freeagent]

Which memory? By the timings I can assume it's B-die.

Trident Z 3200C14 Royal, and 3200C14 Black and White

 

[Veseleil]

Trident Z 3200C14 Royal, and 3200C14 Black and White

Nevermind, read it in your specs, OFC the royals are the Samsung B-die... My bad.

 

[freeagent]

Nevermind, read it in your specs, OFC the royals are the Samsung B-die... My bad.

It’s all good my man, both pairs are B-Die

 

[Mussels]

That score isn't so bad! I feel pretty good about mine now but you're totally right about it not making a huge difference. That was a "dirty run" I easily could have closed more apps that had like 7GB in use
View attachment 293682

Your numbers are incredibly close to mine - odds are i could lower my secondary timings slightly, but i'm spending enough time gaming lately that crashes would drive me mental.
TRFC defaulted SUPER high on my RAM (~1100) but dropped to 712 @ 3866 or 704 @3800 and reduced latency up a heap. It seems to go up based on ranks or capacity, as its far worse than lower capacity kits but doesnt lose the performance you'd expect from it being that much higher.

Forgot my zentimings screenshot -SoC is set to 1.15v, but droops. From what i've seen this is the opposite problem on AM5, where they're kicking SoC voltage up and melting things.

1.15v SoC seems to be the golden standard to try for as it adds very little heat (1-2W) vs 1.10v, while solving issues for some people and allowing higher clocks for others

I've found very little variance between BIOS and motherboard changes - their defaults/auto settings mine change but manual settings have been more or less identical between them all
(One board wouldnt go above 3200 no matter the CPU, one CPU wont go above 3200 no matter the RAM, but the good CPU's work one every other board at 3600 with 1.15v)





The only variance with my setups is too low SoC voltage behaves differently. On the 5800x3D I get USB dropouts, The 3700x gets PCI-E dropouts that trigger a GPU driver crash, or black-screen freeze the PC.
The 5800x can have either one or both, depending on how its setup - if the CPU temps hit that 80c margin with +200 I can get the PCI-E dropouts, while PBO disabled it suffers the USB dropouts.

I have found that higher CPU temps can decrease SoC stability - setting a thermal limit of 75c stopped the black screen crashes on the 5800x, while not really altering performance at all - it only hit those temps for milliseconds at a time with those low thread boosts in low-high load transitions.

Freebies result is showing the gains from C14 RAM - they exist and are measurable but that small gain comes at a heck of a cost

Vs two of these

I wouldnt wanna go from $300 to $840 for 10% more RAM performance (at best) - I got my x3D for that price difference

 

[Ferrum Master]

Why your RAM Write speeds are so piss poor?

 

[freeagent]

Why your RAM Write speeds are so piss poor?

View attachment 293806View attachment 293807

1 CCD.

 

[INSTG8R]

Your numbers are incredibly close to mine - odds are i could lower my secondary timings slightly, but i'm spending enough time gaming lately that crashes would drive me mental.
TRFC defaulted SUPER high on my RAM (~1100) but dropped to 712 @ 3866 or 704 @3800 and reduced latency up a heap. It seems to go up based on ranks or capacity, as its far worse than lower capacity kits but doesnt lose the performance you'd expect from it being that much higher.

Forgot my zentimings screenshot -SoC is set to 1.15v, but droops. From what i've seen this is the opposite problem on AM5, where they're kicking SoC voltage up and melting things.
View attachment 293798


1.15v SoC seems to be the golden standard to try for as it adds very little heat (1-2W) vs 1.10v, while solving issues for some people and allowing higher clocks for others

I've found very little variance between BIOS and motherboard changes - their defaults/auto settings mine change but manual settings have been more or less identical between them all
(One board wouldnt go above 3200 no matter the CPU, one CPU wont go above 3200 no matter the RAM, but the good CPU's work one every other board at 3600 with 1.15v)





The only variance with my setups is too low SoC voltage behaves differently. On the 5800x3D I get USB dropouts, The 3700x gets PCI-E dropouts that trigger a GPU driver crash, or black-screen freeze the PC.
The 5800x can have either one or both, depending on how its setup - if the CPU temps hit that 80c margin with +200 I can get the PCI-E dropouts, while PBO disabled it suffers the USB dropouts.

I have found that higher CPU temps can decrease SoC stability - setting a thermal limit of 75c stopped the black screen crashes on the 5800x, while not really altering performance at all - it only hit those temps for milliseconds at a time with those low thread boosts in low-high load transitions.

Freebies result is showing the gains from C14 RAM - they exist and are measurable but that small gain comes at a heck of a costView attachment 293796

Vs two of these
View attachment 293797

I wouldnt wanna go from $300 to $840 for 10% more RAM performance (at best) - I got my x3D for that price difference

Yeah I refuse to go down the "Timings Rabbit Hole" All I have done is changed the default CL16 to CL18 and I am done....My SoC is .962V and I have looked through my BIOS and anything SoC related I have adjusted hasn't changed it. I ran 1.12V on my previous GB Board with the same "OC" with my 5600X
I mean I JUST game on my PC, I am well past my "chasing numbers" days unless it's FPS...

 

[Mussels]

Why your RAM Write speeds are so piss poor?

View attachment 293806View attachment 293807

ram writes are doubled on dual CCX chips, so the 5900x and 5950x are always twice as fast as the other ryzen 5000 CPUs in that one test. It doesnt seem to help gaming.

Tried C16 (semi-stable) and C14 (bootlooped even at 3200) on this RAM, it's a nogo.
3866 is happy times, however. I'll see if higher is stable, now that winter is here i can justify the PC heating up the room while i screw around with testing.
Should probably move over to the Zen Garden thread, even if this is all related to the new AGESA (and overclocking on it)

(The tiny 0.2ns latency increase is likely some annoying background task, the read/write went up appropriately)

Yeah I refuse to go down the "Timings Rabbit Hole" All I have done is changed the default CL16 to CL18 and I am done....My SoC is .962V and I have looked through my BIOS and anything SoC related I have adjusted hasn't changed it. I ran 1.12V on my previous GB Board with the same "OC" with my 5600X
I mean I JUST game on my PC, I am well past my "chasing numbers" days unless it's FPS...

If you dont need to, you dont need to. If that's single rank RAM, you may be able to undervolt that SoC.
You have a CPU with great ST/gaming performance and nothing makes it overheat - you've got an 88W cap where mines got a 142W cap if PBO is enabled, with the same space to spread that heat out from, so you can see why a 5800x(3D) user cares more about tweaking that down.

 

[INSTG8R]

If you dont need to, you dont need to. If that's single rank RAM, you may be able to undervolt that SoC.
You have a CPU with great ST/gaming performance and nothing makes it overheat - you've got an 88W cap where mines got a 142W cap if PBO is enabled, with the same space to spread that heat out from, so you can see why a 5800x(3D) user cares more about tweaking that down

It’s dual rank as far as I know and my Mobo has ‘Kombo Strike” which is “special sauce” UV and now with CO I can run it at -30 so I’m down to a 1.24V max. I’ve tested all the combinations of the two and running both gave me the best overall results. The RAM OC is just a bonus whether it’s better or not. it’s not affecting my temps or needs any special voltage tweaking.

 

[Mussels]

It’s dual rank as far as I know and my Mobo has ‘Kombo Strike” which is “special sauce” UV and now with CO I can run it at -30 so I’m down to a 1.24V max. I’ve tested all the combinations of the two and running both gave me the best overall results. The RAM OC is just a bonus whether it’s better or not. it’s not affecting my temps or needs any special voltage tweaking.

kombo strike is 100% just PBO's undervolt curve, 1.0.0.8 gave that to all x3D users on every board
MSI just added it in prior to AMD making it available to everyone

x3D behaves differently for some reason to any other Zen 3 CPU, it's like they measure the voltage and treat it as VID - My board lets me use the curve undervolt AND a voltage offset, and you can never make it unstable - it just downclocks instead so past a certain point you lose performance

Kombo strike 3/-30 UC, and -.05v is the limit before i take any performance hit at all, but -0.61 shaves a lot of power and heat for something like a 0.1% loss (that lets me run the RAM harder)


I beat w1zzards score, so i'm happy
I didnt even need a debloated OS!

 

[Cowboystrekk]

Can the 5800X3D run 4 x16 GB (single sided) at 3600mhz regardless of silicon quality - Micron E-dies for ex? Thanks.

It will depend on motherboard. If you get a T-top MB you are more likely to achieve that.

Your numbers are incredibly close to mine - odds are i could lower my secondary timings slightly, but i'm spending enough time gaming lately that crashes would drive me mental.
TRFC defaulted SUPER high on my RAM (~1100) but dropped to 712 @ 3866 or 704 @3800 and reduced latency up a heap. It seems to go up based on ranks or capacity, as its far worse than lower capacity kits but doesnt lose the performance you'd expect from it being that much higher.

Forgot my zentimings screenshot -SoC is set to 1.15v, but droops. From what i've seen this is the opposite problem on AM5, where they're kicking SoC voltage up and melting things.
View attachment 293798


1.15v SoC seems to be the golden standard to try for as it adds very little heat (1-2W) vs 1.10v, while solving issues for some people and allowing higher clocks for others

I've found very little variance between BIOS and motherboard changes - their defaults/auto settings mine change but manual settings have been more or less identical between them all
(One board wouldnt go above 3200 no matter the CPU, one CPU wont go above 3200 no matter the RAM, but the good CPU's work one every other board at 3600 with 1.15v)





The only variance with my setups is too low SoC voltage behaves differently. On the 5800x3D I get USB dropouts, The 3700x gets PCI-E dropouts that trigger a GPU driver crash, or black-screen freeze the PC.
The 5800x can have either one or both, depending on how its setup - if the CPU temps hit that 80c margin with +200 I can get the PCI-E dropouts, while PBO disabled it suffers the USB dropouts.

I have found that higher CPU temps can decrease SoC stability - setting a thermal limit of 75c stopped the black screen crashes on the 5800x, while not really altering performance at all - it only hit those temps for milliseconds at a time with those low thread boosts in low-high load transitions.

Freebies result is showing the gains from C14 RAM - they exist and are measurable but that small gain comes at a heck of a costView attachment 293796

Vs two of these
View attachment 293797

I wouldnt wanna go from $300 to $840 for 10% more RAM performance (at best) - I got my x3D for that price difference

Getting RRD\FAW, RDWR\WRRD and RC lower would improve performance by quite a margin

 

[Mussels]

It will depend on motherboard. If you get a T-top MB you are more likely to achieve that.

This - but mostly you just need to not get one of the WORST motherboards, 99% of them can achieve 3600 with four ranks. Usually SoC voltage is all thats needed, but some rare boards just wont do it at all (my ROG STRIX B450-I just refused to do above 3200, no matter what)

Getting RRD\FAW, RDWR\WRRD and RC lower would improve performance by quite a margin

True, but it gets into that complex math that hurts my brain with how they need to be tweaked in relation to each other

No one has examples of Hynix MJR to compare with, because no one overclocks 32GB DDR4 sticks

Example from reddit:

I found one random example on reddit, i'll give these a shot and see if they work
I dont have high hopes since my tRFC needed to be over 700 to POST, but theres a chance his other tweaks allow it to work