AM5 boot times improve RADICALLY with memory context restore enabled

[Space Lynx]

I was dealing with an MSI B650 Tomahawk with latest BIOS. Enabling MCR sorted the RAM testing/training cycle issues after a stable boot. At least three dozen boots with zero issues before delivering to the customer. They've not reported any issues and this was handed to them months ago.

ah I just noticed the OP has 64gb of ram, hence the extra time. makes more sense now. I missed that initially.

I have to say this whole boot thing was a weird design choice for AMD. It doesn't really bother me, just strategically seems like a strange move, not everyone is as tech savvy as those of on this site, lot of people want to hit a button and want it to load just as fast as they were used to in the past.

 

[mb194dc]

It's absolutely shocking am5 was released in this state and the long boot times are still going on. Surely there was a better design solution with memory training?

 

[Space Lynx]

It's absolutely shocking am5 was released in this state and the long boot times are still going on. Surely there was a better design solution with memory training?

I still don't even understand why it had to change from stable XMP set it and forget it like we are so used to.

I mean ram timings make so little difference in gaming, oh you gained 10 fps going from 180 to 190 fps in one game out of ten, congrats I guess...

 

[LethaR]

Wow guys thanks for all the replies, didnt think this thread was gonna get all this attention at first. I've been reading all of your comments, im not sure if "power down mode" is enabled or not in my bios, didnt change it at all, just enabled MCR, so i guess it is enabled. Did this just yesterday, so i cant tell if its going to be stable 24/7 or not, as i said ill post here if anything weird happens.

Have to clarify my situation though. I've been having an Asus B650-A Gaming for the last year, with the same 64gb of ram and the EXPO profile enabled. Never enabled MCR in that board, because the boot times were "decent" (around 50 seconds, which is still awful for a pc with a pcie 4.0 7500mb/s m.2 ssd...) but i could deal with it and didnt care that much. But with this msi board it was just unacceptable, over 250 seconds is just too f*cking much. At first i thought it was gonna take that long only on the first boot after building the pc, but the 250 seconds were still there after 3-4 boots.

Just if anyone wants to know, enabling MCR, your XMP or EXPO profile, setting UCLK DIV1 MODE to 1:1 and using the curve optimizer (PBO (Precision Boost Overdrive) -> Curve Optimizer set to All Cores -> All Core Curve Optimizer Sign set to Negative -> All Core Curve Optimizer Magnitude set between 5 and 30 (set mine at 20)) is the way to go when dealing with AM5 bios settings if you just want a pretty easy way to optimize your ryzen 7000 build.

 

[Assimilator]

It's absolutely shocking am5 was released in this state and the long boot times are still going on. Surely there was a better design solution with memory training?

This is unfortunately standard for a new AMD CPU platform. First generation is always rubbish, second generation has most of the bigger bugs ironed out, third gen is what first gen should have been. This is one of the reasons I never buy 1st-gen AMD, consider 2nd-gen, but generally wait until 3rd-gen.

 

[chrcoluk]

I think this would be driving me nuts personally.

Ite feels "broken" to me if it needs to train for over a minute to be stable on every power up.

Does this training occur also when waking up from sleep or just from full power off state?

This is unfortunately standard for a new AMD CPU platform. First generation is always rubbish, second generation has most of the bigger bugs ironed out, third gen is what first gen should have been. This is one of the reasons I never buy 1st-gen AMD, consider 2nd-gen, but generally wait until 3rd-gen.

Yeah wont surprise me if by the end of AM5 this will be completely gone as an issue.

 

[Panther_Seraphin]

I think you forgot the ASRock memory slot sticker at AM5 launch.

View attachment 346791

Can testify that the 4x16 setup was every bit of that 200 secs on boot on the early BIOS if not even more!!! ¬_¬

 

[RJARRRPCGP]

I think this would be driving me nuts personally.

Ite feels "broken" to me if it needs to train for over a minute to be stable on every power up.


Yeah wont surprise me if by the end of AM5 this will be completely gone as an issue.

Buggy-ness= Reminds me of when I first went to socket 462 in June, 2001, was getting crashes at default!

 

[Kapone33]

To be fair even on AM4 4x16 could be problematic with RAM. I have always seen that basically everyone recommends 2 sticks for best stability/performance.

 

[jarablue]

I have done this and absolutely love it. I have 0 problems with it enabled. And it cut my post time down to like 3 seconds from 30 seconds.

Works perfectly fine for ME. If it doesn't work for you....don't do it. Not everyone who enables this has issues. One persons opinion on his rig does not mean everyone will have that persons issues.

 

[ir_cow]

Works perfectly fine for ME. If it doesn't work for you....don't do it. Not everyone who enables this has issues.

It is odd that not everyone has this problem. My tests were controlled too with different ram and motherboards. So what gives?

All I can do is tell people who thinks it's fine to run memtest5 for at least the default 3 loops. If its passes congrats! But if you get random BSODs down the road, probably just disabling it will fix the problem.

 

[chrcoluk]

To be fair even on AM4 4x16 could be problematic with RAM. I have always seen that basically everyone recommends 2 sticks for best stability/performance.

It wasnt problem free, I got used to downclocking RAM, but I didnt/dont have minute long post times.

 

[Space Lynx]

It is odd that not everyone has this problem. My tests were controlled too with different ram and motherboards. So what gives?

All I can do is tell people who thinks it's fine to run memtest5 for at least the default 3 loops. If its passes congrats! But if you get random BSODs down the road, probably just disabling it will fix the problem.

testmem5 is all I can find when I Google that? do you just mean memtest86?

 

[AusWolf]

@W1zzard Do you know of any logic/reason why this would not have been enabled by default by AMD through a BIOS update in the AGESA releases? Or by motherboard manufacturers in general?

What are we missing here? I feel like Lisa Su is smart enough to not make people have long boots if she didn't have to.

Because it may or may not work depending on your system config and settings. I've had it work and then not work with different BIOS versions on the same system.

 

[ir_cow]

testmem5 is all I can find when I Google that? do you just mean memtest86?

https://www.overclock.net/threads/memory-testing-with-testmem5-tm5-with-custom-configs.1751608/ . First thing to come up on Google.

 

[Space Lynx]

https://www.overclock.net/threads/memory-testing-with-testmem5-tm5-with-custom-configs.1751608/ . First thing to come up on Google.

thanks I am going to run 5 cycles now just to make sure my stock experience stable. (expo on and that's it)

 

[Devon68]

I just timed my boot time and it's between 20-25 seconds. No idea about that MCR setting being enabled.

 

[Space Lynx]

I just timed my boot time and it's between 20-25 seconds. No idea about that MCR setting being enabled.

hey what's the difference between my Mobo and your Mobo? I have b650 non-e steel legend, but it has gen 5 GPU slot and gen5 storage... I was under the impression that was what the e meant? I wonder why mine doesn't have e but still support gen 5 both ways

edit nm I see now, you have two more phases in power delivery than me and a front 2x2 gen 2 type c port for a case. I suppose that doesn't matter really, mine has an updated audio codec though, interesting. strange they made the same Mobo with such small changes.

 

[jarablue]

I have a asus prime x670-p wifi board and teamgroup ud5 6000 memory. Works awesome here. Shaves 30 seconds off posting.

 

[lolek86]

Why on earth should they require each other?

I dunno. Its probably simple. In order to not loose ddr5 speed trained at last boot you and restore it next time (Memory context restore) you cant totally cut off power for memory (power down)

 

[tabascosauz]

I dunno. Its probably simple. In order to not loose ddr5 speed trained at last boot you and restore it next time (Memory context restore) you cant totally cut off power for memory (power down)

I'm pretty sure PDM is relevant only when the PC is on. It puts DRAM in a lower volts/power state until performance is needed, but at the very onset of the workload can cause extra latency due to the wake up.

MCR is pretty sketchy overall though. Sometimes even without obvious instability it just won't train properly, then without a hitch the next reboot.

 

[R-T-B]

I dunno. Its probably simple. In order to not loose ddr5 speed trained at last boot you and restore it next time (Memory context restore) you cant totally cut off power for memory (power down)

But it REQUIRES PowerDown, not the other way around.

 

[lolek86]

I'm pretty sure PDM is relevant only when the PC is on. It puts DRAM in a lower volts/power state until performance is needed, but at the very onset of the workload can cause extra latency due to the wake up.

MCR is pretty sketchy overall though. Sometimes even without obvious instability it just won't train properly, then without a hitch the next reboot.

But it REQUIRES PowerDown, not the other way around.

It might also be doing other things in off state. Why it would be required to not loose ddr5 training settings?

In my experience on asus strix B650-A during last year, not even once encountered any random retraining/reboots or anything like that looking MCR ON related and my sticks are pushed rather hard. Surprised people still report issues about that.

 

[tabascosauz]

It might also be doing other things in off state. Why it would be required to not loose ddr5 training settings?

In my experience on asus strix B650-A during last year, not even once encountered any random retraining/reboots or anything like that looking MCR ON related and my sticks are pushed rather hard. Surprised people still report issues about that.

Sounds like you didn't push them hard enough then I saw it a few times at 6400 but not any lower

idk honestly, given how agesa has had 5+ years of basically being held together by twine, I guess nothing is impossible

Last year people said it was a bug, not intended behaviour, and that AGESA 1008 would fix it. I guess they all went on vacation and forgot cause it's still here in AGESA 1100

 

[Space Lynx]

@GerKNG is your B650E steel legend having memory context enabled by default?

I just checked and in my BIOS it actually is enabled by default... that's why my boot times are most instantaneous... huh strange, if it causes so many problems, why would AsRock have it on by default? (stock bios when I looked)