Question Is there a large enough deviation from 3600mhz here to cause any problems such as stuttering etc ?

May 1, 2023
194
7
95
So yesterday, I ended up manually setting my memory timings without proper knowledge which caused my pc to completely screw itself over, so I needed to do a CMOS reset. the reason I did that was because I am trying to figure out the cause of this microstutters that occur in very specific areas of my game, and they don't seem to be happen in any other game; yet i've asked multiple people, including some friends, to try and replicate them. they receive a bit of fps drops like me, but they don't have the stutter

one friend said it is caused by some kind of 'async' (?) because of the slight deviation from 3600mhz in my Memory Frequencies/Timings. I have XMP enabled, and FCLK set to Auto (setting it to 1800mhz made no difference), and the profile is listed as:

"DOCP- DDR4-3597 18-22-22-42-1.35v" this isn't the full 3600mhz of the my Corsair Vengeance DDR4-3600mhz ram, but I was told a few months back that it wasn't a big enough deviation for it to necessarily matter.

View: https://imgur.com/a/9f3np9M

Above are my Memory Timings as shown by CPU-Z; but here's what confuses me...the Memory tab lists a frequency of 1799.6. to get the full 3600mhz, you would need 1800mhz + 1800; yet 1799.6 + 1799.6 comes out as 3599.2? and under the SPD tab, it is listed as 1798mhz...1798mhz + 1798 comes out as 3596?

3596mhz is not 3597mhz, which is what the XMP profile says. I get that this is a -1 deviation from 3597, but it also becomes a -4 deviation from 3600, instead of a -3 deviation. is this still irrelevant? is this simply too small of a deviation to actually ever matter?

even if this deviation does not matter, why does is it not seem to add up?
 
Thats normal for ram to not be right at its full speed, happens on a lot of board, but yeah its only 4mhz off, the not even a .1 of an FPS in games, it will not mess with performance in any meaningful way. You can possibly fix this but setting BCLK to 100.1, but being on Ryzen, that can have the potential to disable PBO changing that, its not worth the hassle in my opinion.

You are literally better off making your GPU fans or CPU fan spin faster so they can boost higher or longer than 4mhz on the RAM.

Good Luck!
 
Apr 14, 2023
74
4
35
So yesterday, I ended up manually setting my memory timings without proper knowledge which caused my pc to completely screw itself over, so I needed to do a CMOS reset. the reason I did that was because I am trying to figure out the cause of this microstutters that occur in very specific areas of my game, and they don't seem to be happen in any other game; yet i've asked multiple people, including some friends, to try and replicate them. they receive a bit of fps drops like me, but they don't have the stutter

one friend said it is caused by some kind of 'async' (?) because of the slight deviation from 3600mhz in my Memory Frequencies/Timings. I have XMP enabled, and FCLK set to Auto (setting it to 1800mhz made no difference), and the profile is listed as:

"DOCP- DDR4-3597 18-22-22-42-1.35v" this isn't the full 3600mhz of the my Corsair Vengeance DDR4-3600mhz ram, but I was told a few months back that it wasn't a big enough deviation for it to necessarily matter.

View: https://imgur.com/a/9f3np9M

Above are my Memory Timings as shown by CPU-Z; but here's what confuses me...the Memory tab lists a frequency of 1799.6. to get the full 3600mhz, you would need 1800mhz + 1800; yet 1799.6 + 1799.6 comes out as 3599.2? and under the SPD tab, it is listed as 1798mhz...1798mhz + 1798 comes out as 3596?

3596mhz is not 3597mhz, which is what the XMP profile says. I get that this is a -1 deviation from 3597, but it also becomes a -4 deviation from 3600, instead of a -3 deviation. is this still irrelevant? is this simply too small of a deviation to actually ever matter?

even if this deviation does not matter, why does is it not seem to add up?
It's due to memory interleaving. AFAIK this would actually make your dual less...errory.

C18 on 3600 can be ok...when you're clocking towards 4400.

Just get it to 18 20 40 on the 1.35V and then start lowering voltage. I thinga lot of glitch is that they're frying.
 
May 1, 2023
194
7
95
It's due to memory interleaving. AFAIK this would actually make your dual less...errory.

C18 on 3600 can be ok...when you're clocking towards 4400.

Just get it to 18 20 40 on the 1.35V and then start lowering voltage. I thinga lot of glitch is that they're frying.
I don't really overclock or undervolt anything, me altering the memory timings was a stupid knee jerk response. what do you mean by that last part though? they should not be frying, since this is XMP and 1.35 is a safe voltage range, unless I'm misunderstanding your reply


also are you saying that this 'memory interleaving' thing is an issue that's causing this deviation, or an actual feature that prevents more errors?
 
Apr 14, 2023
74
4
35
It can be a way to work on 2 levels at once in the same module or fabric while avoiding fatal interactions, data corruption,...or to be able t9 perform different tasks on different frequencies. Setting a latency to +1 in relation to the other, identical module is often a way to track the activity of the twin / bidirectional bookkeeping.

If the memory timings bookkeeper marks a now 11 MT task as 10, there's a bit (0/1) to read or set,.... there's a system of managed microglitching going, possibly resulting in gaining an insane amount of bandwidth.

By choosing 3600 Mhz as Mclk, AMD SoC, IF, onboard graphics, (memcontroller!) ,... are all on 1800 Mhz.

So actually...you're smack in the middle of the best possible action: push your clock to 4000 and establish AMD uncore @2000 MHz.

In BIOS , set the command mode of your memory to 2T. Set DRAM Voltage to 1,40V. Timings to default and set tRC very royally like the sum of the first 4 timings. That should get you some stability with simple tweaks.

Up next: get that schoolbusRAM to 4000.
 
Last edited:
Apr 14, 2023
74
4
35
I don't really overclock or undervolt anything, me altering the memory timings was a stupid knee jerk response. what do you mean by that last part though? they should not be frying, since this is XMP and 1.35 is a safe voltage range, unless I'm misunderstanding your reply


also are you saying that this 'memory interleaving' thing is an issue that's causing this deviation, or an actual feature that prevents more error.
 
May 1, 2023
194
7
95
It can be a way to work on 2 levels at once in the same module or fabric while avoiding fatal interactions, data corruption,...or to be able t9 perform different tasks on different frequencies. Setting a latency to +1 in relation to the other, identical module is often a way to track the activity of the twin / bidirectional bookkeeping.

If the memory timings bookkeeper marks a now 11 MT task as 10, there's a bit (0/1) to read or set,.... there's a system of managed microglitching going, possibly resulting in gaining an insane amount of bandwidth.

By choosing 3600 Mhz as Mclk, AMD SoC, IF, onboard graphics, (memcontroller!) ,... are all on 1800 Mhz.

So actually...you're smack in the middle of the best possible action: push your clock to 4000 and establish AMD uncore @2000 MHz.

In BIOS , set the command mode of your memory to 2T. Set DRAM Voltage to 1,40V. Timings to default and set tRC very royally like the sum of the first 4 timings. That should get you some stability with simple tweaks.

Up next: get that schoolbusRAM to 4000.
interesting, I might look into this. although it isn’t absolutely necessary, right? technically speaking, I can continue running my ram as is without issue, even though I’m losing potential performance/speed gain
 
Apr 14, 2023
74
4
35
interesting, I might look into this. although it isn’t absolutely necessary, right? technically speaking, I can continue running my ram as is without issue, even though I’m losing potential performance/speed gain
You have the Samsung Lego RAM.

Each module is 2 sticks combined into one. But they put half as big chips on them. To use them at.....eh m...to be able to click things and have the screen change. more than once daily, install 2 modules to the board as a dual channel formation.

And then that is that. RAM full. Add more and you're full on back in DDR3 times. Besides they could put more than 2 sticks then. And timings were better.

This is probably because of the same thing as you could not go over 4000 MHz with RAM until now blabla....now you CAN, but only with half. Of double half.

Which you can multiplu by to get DDR4 speeds..like 3600 MHz. But less. And less less than with more RAM.

So. People who say AMD normally work for AMD, or tell you not to buy anything. It's like fake baby tablet'
 
May 1, 2023
194
7
95
You have the Samsung Lego RAM.

Each module is 2 sticks combined into one. But they put half as big chips on them. To use them at.....eh m...to be able to click things and have the screen change. more than once daily, install 2 modules to the board as a dual channel formation.

And then that is that. RAM full. Add more and you're full on back in DDR3 times. Besides they could put more than 2 sticks then. And timings were better.

This is probably because of the same thing as you could not go over 4000 MHz with RAM until now blabla....now you CAN, but only with half. Of double half.

Which you can multiplu by to get DDR4 speeds..like 3600 MHz. But less. And less less than with more RAM.

So. People who say AMD normally work for AMD, or tell you not to buy anything. It's like fake baby tablet'
Samsung Lego Ram? what does that mean

I have dual ram, I just don’t think I’m following what you’re saying
 
Apr 14, 2023
74
4
35
Samsung Lego Ram? what does that mean

I have dual ram, I just don’t think I’m following what you’re saying
I have 4 x 8 Gb in those timings. They're 1Rx8 of Hynix C or D a piece. They don't 'need voltage for this I think, at least not much.

Maybe yours is the 'Samsung B trash die.'
Good for running at 2133 or crashing.No one knew what it was. Was an XMP on the box. Crashed it every time. It couldn't do 16-18. It went directly to 22-20. if yo7 let it loose.

It's baa-aaack.
 
Last edited:
Apr 14, 2023
74
4
35
if my memory timings with XMP are fine (which I've been told they are), and the .4 deviation isn't big enough to impact my performance like the other guy said, I'll probably just leave it as it is. I've never been big on overclocking or anything
These timings are what I ended up using for the worst bin in the world. Almost no voltage needed, timings expresssed in days.

It was still glitchy though until I lowered its voltage to 1.32V and padded it from the architecture..