Question 1 error after 1000% coverage of RAM OC to 2933MHz from 2400MHz

Jul 17, 2021
13
0
10
0
I have 2 sticks of 8GB each Corsair Vengeance LPX DDR4 2400mhz ram that I bought in 2018. The memory uses Hynix m-die, which as far as I know doesn't OC well. However with Typhoon burner and DRAM calculator, I was able to OC it to 2933MHz and even 3000MHz but I stuck to the former. On first day I ran MemTest for 4 processes of MemTest with 3000MB each ending at 350% coverage and 2 processes with 1000MB ending at 900% coverage. My CPU was also OCed to 3.7GHz, a Ryzen 5 1600 running with stock cooler and hitting as high temps as 85C based on HwInfo.


On next day I took the same test but longer and with every setting being the same. This time, on 950% coverage of 1 1000MB memtest process it got 1 error. I never faced any crashes during my gaming session on 1st day after OC. Is it anything do with hot CPU temps? Or is my OC unstable. Oh, and I used safe option in DRAM calculator
 

mamasan2000

Distinguished
AFAIK, HCI Memtest up to 400% I think it was can find 95% of all errors in RAM. With higher percentages, you are likely to find a really rare RAM error.
Crashes with unstable RAM can happen within 5 minutes or once a week. Depends on how unstable it is.
You could check if it is the CPU temp by removing the OC on it. The memory controller is on the CPU.
It could also be particles from space so who knows. Particles from space can make a bit switch in RAM. ECC is supposed to help against that, among other things, but ECC is mainly used on server RAM.
 
Jul 17, 2021
13
0
10
0
AFAIK, HCI Memtest up to 400% I think it was can find 95% of all errors in RAM. With higher percentages, you are likely to find a really rare RAM error.
Crashes with unstable RAM can happen within 5 minutes or once a week. Depends on how unstable it is.
You could check if it is the CPU temp by removing the OC on it. The memory controller is on the CPU.
The single error may be because of the CPU controller and it's heat. I will test it again with both CPU OC and RAM OC once I get a better cooler tomorrow.

It could also be particles from space so who knows. Particles from space can make a bit switch in RAM. ECC is supposed to help against that, among other things, but ECC is mainly used on server RAM.
Can you elaborate on this, like how, when and why it happens ?

With stock CPU and RAM at 2933MHz and all RAM settings at Auto, test runs fine, I tested up to 1030% across 8 MemTest processes with 1GB memory each. However yesterday with stock CPU and RAM at 2933MHz and all RAM settings at Auto, Windows didn't even boot. I didn't even need to use DRAM Calculator today.
 

mamasan2000

Distinguished
The single error may be because of the CPU controller and it's heat. I will test it again with both CPU OC and RAM OC once I get a better cooler tomorrow.



Can you elaborate on this, like how, when and why it happens ?

With stock CPU and RAM at 2933MHz and all RAM settings at Auto, test runs fine, I tested up to 1030% across 8 MemTest processes with 1GB memory each. However yesterday with stock CPU and RAM at 2933MHz and all RAM settings at Auto, Windows didn't even boot. I didn't even need to use DRAM Calculator today.
Cosmic rays https://science.slashdot.org/story/17/02/19/2330251/serious-computer-glitches-can-be-caused-by-cosmic-rays Tons of them passing earth every second, even passing through your body.

When you say Windows didn't boot, did you get past POST? Cold-booting RAM can be problematic, some motherboards deal with it by restarting twice, you might have noticed this. Second problem can be, the mobo tries to boot with suppling RAM only 1.2 volts and unless you run JEDEC-standards, it's not gonna POST. Again, this might lead to bootloop or just an outright BIOS error "Press F1 to enter BIOS and fix the error". Some BIOSes have the option to choose your RAM boot voltage, that's great if you have that.
 
Jul 17, 2021
13
0
10
0
Cosmic rays https://science.slashdot.org/story/17/02/19/2330251/serious-computer-glitches-can-be-caused-by-cosmic-rays Tons of them passing earth every second, even passing through your body.

When you say Windows didn't boot, did you get past POST? Cold-booting RAM can be problematic, some motherboards deal with it by restarting twice, you might have noticed this. Second problem can be, the mobo tries to boot with suppling RAM only 1.2 volts and unless you run JEDEC-standards, it's not gonna POST. Again, this might lead to bootloop or just an outright BIOS error "Press F1 to enter BIOS and fix the error". Some BIOSes have the option to choose your RAM boot voltage, that's great if you have that.
The Windows passed POST, while loading Windows it crashed. It stayed there and I decided to press the reset button and use default RAM speeds with 2400MHz XMP. I have the option to change RAM voltage but didn't notice any boot voltage. I am using B450 Aorus M
 

mamasan2000

Distinguished
The Windows passed POST, while loading Windows it crashed. It stayed there and I decided to press the reset button and use default RAM speeds with 2400MHz XMP. I have the option to change RAM voltage but didn't notice any boot voltage. I am using B450 Aorus M
POST = Power On Self Test. It's what the motherboard does when you turn on system. After it passes, motherboard gives control to an OS you have installed, Linux, Windows etc. Via EFI.
Not all motherboards have boot dram voltage. Mine doesn't either, Asus X470 Prime Pro.
Have you updated BIOS recently? If you do, make sure it has support for your CPU. Some mobos have a ROM chip so small, they only support latest 5000-series if you update IIRC.
Asus uses large ROM chips, 128 megs or more so it's not a problem for me. But if you have a 32 meg chip, that's a problem.

I had my system boot twice everytime while running XMP @ 3000 Mhz but I can't remember if it was on my old B350 mobo or this new Asus in the beginning (it's on latest BIOS now). I don't have that issue any more.

If windows crashes while loading, it's very unstable RAM or corrupt systemfiles. Remove RAM OC, do the whole 'sfc /scannow' etc business. https://support.microsoft.com/en-us/topic/use-the-system-file-checker-tool-to-repair-missing-or-corrupted-system-files-79aa86cb-ca52-166a-92a3-966e85d4094e
 
Jul 17, 2021
13
0
10
0
POST = Power On Self Test. It's what the motherboard does when you turn on system. After it passes, motherboard gives control to an OS you have installed, Linux, Windows etc. Via EFI.
Not all motherboards have boot dram voltage. Mine doesn't either, Asus X470 Prime Pro.
Have you updated BIOS recently? If you do, make sure it has support for your CPU. Some mobos have a ROM chip so small, they only support latest 5000-series if you update IIRC.
Asus uses large ROM chips, 128 megs or more so it's not a problem for me. But if you have a 32 meg chip, that's a problem.

I had my system boot twice everytime while running XMP @ 3000 Mhz but I can't remember if it was on my old B350 mobo or this new Asus in the beginning (it's on latest BIOS now). I don't have that issue any more.

If windows crashes while loading, it's very unstable RAM or corrupt systemfiles. Remove RAM OC, do the whole 'sfc /scannow' etc business. https://support.microsoft.com/en-us/topic/use-the-system-file-checker-tool-to-repair-missing-or-corrupted-system-files-79aa86cb-ca52-166a-92a3-966e85d4094e
PC crashes while loading Windows, it passed the BIOS screen and crashed while loading. I didn't update my BIOS as I didn't need to. Its running the first version of it, F1, and it supports Ryzen Zen and Zen+ with this version.

Windows isn't corrupted because it works otherwise. And the file that crashed was related to DirectX and krnl probably stands for kernel. Files are OK because I don't have any issues. Even now I am typing this with RAM at 2933MHz and Auto settings in BIOS and CPU at 3,7GHz, no issues whatsoever in normal tasks.
 
Jul 17, 2021
13
0
10
0
POST = Power On Self Test. It's what the motherboard does when you turn on system. After it passes, motherboard gives control to an OS you have installed, Linux, Windows etc. Via EFI.
Not all motherboards have boot dram voltage. Mine doesn't either, Asus X470 Prime Pro.
Have you updated BIOS recently? If you do, make sure it has support for your CPU. Some mobos have a ROM chip so small, they only support latest 5000-series if you update IIRC.
Asus uses large ROM chips, 128 megs or more so it's not a problem for me. But if you have a 32 meg chip, that's a problem.

I had my system boot twice everytime while running XMP @ 3000 Mhz but I can't remember if it was on my old B350 mobo or this new Asus in the beginning (it's on latest BIOS now). I don't have that issue any more.

If windows crashes while loading, it's very unstable RAM or corrupt systemfiles. Remove RAM OC, do the whole 'sfc /scannow' etc business. https://support.microsoft.com/en-us/topic/use-the-system-file-checker-tool-to-repair-missing-or-corrupted-system-files-79aa86cb-ca52-166a-92a3-966e85d4094e
New update, I turned on XMP, SVM(i need virtualization for VMware) and Windows gets in a boot loop. When I turn off XMP, 2933mhz works completely fine and Windows boots.
 

mamasan2000

Distinguished
New update, I turned on XMP, SVM(i need virtualization for VMware) and Windows gets in a boot loop. When I turn off XMP, 2933mhz works completely fine and Windows boots.
XMP is 2400 Mhz? How is it crashing at that setting?
sfc /scannow is a quick test to see if you have corrupted systemfiles. You don't notice the corruption immeadiately. Might as well run it.
 
Jul 17, 2021
13
0
10
0
XMP is 2400 Mhz? How is it crashing at that setting?
sfc /scannow is a quick test to see if you have corrupted systemfiles. You don't notice the corruption immeadiately. Might as well run it.
Memory is rated 2400MHz, so XMP settings are 2400MHz. I mean that's why it's probably crashing with XMP enabled at 2933MHz. No errors with sfc /scannow. It shouldn't be a problem with Windows naturally when only 1 condition causes the problem. When Windows crashes with XMP at 2933MHz, the error is KERNEL_MODE_EXCEPTION_NOT_HANDLED
 

mamasan2000

Distinguished
Memory is rated 2400MHz, so XMP settings are 2400MHz. I mean that's why it's probably crashing with XMP enabled at 2933MHz. No errors with sfc /scannow. It shouldn't be a problem with Windows naturally when only 1 condition causes the problem. When Windows crashes with XMP at 2933MHz, the error is KERNEL_MODE_EXCEPTION_NOT_HANDLED
I think the XMP thing is down to motherboard implementation, I only have Auto or XMP. So unless I want faster RAM than 2133 I have to enable XMP. Maybe you have three choices, I don't know. But my timings are for the most part manually set. You could check them over so they aren't ridiculously low. Or one setting or so is set by you manually but never raised, could be an issue as well. Check all 40 of them or however many they are.

From this pic, I focus on primary and secondary. View: https://imgur.com/Ie4LVtI

The only tertiary I care about is the 2 SCLs. tCKE I can't lower, wont boot.

BTW, what is Hynix M-die? Only Samsung has M-die. https://www.reddit.com/r/overclocking/wiki/ram/ddr4
 
Last edited:
Jul 17, 2021
13
0
10
0
I think the XMP thing is down to motherboard implementation, I only have Auto or XMP. So unless I want faster RAM than 2133 I have to enable XMP. Maybe you have three choices, I don't know. But my timings are for the most part manually set. You could check them over so they aren't ridiculously low. Or one setting or so is set by you manually but never raised, could be an issue as well. Check all 40 of them or however many they are.

From this pic, I focus on primary and secondary. View: https://imgur.com/Ie4LVtI

The only tertiary I care about is the 2 SCLs. tCKE I can't lower, wont boot.

BTW, what is Hynix M-die? Only Samsung has M-die. https://www.reddit.com/r/overclocking/wiki/ram/ddr4
My motherboard only gives Disabled and Profile 1 in XMP settings, ie only 2 options, also probably because 2400MHz is next to 2133 so it's just one profile. Moreover by default, XMP is disabled but RAM is set to 2400MHz, checked in BIOS and Windows. AFAIK somewhere I read anything over 2133MHz with JEDEC specifications is overclock, so my RAM is being overclocked by default and also without XMP settings.


The Primary settings are completely same. The scls are at 1. However with both Ryzen DRAM calculator and motherboard defaults, tCKE is 7.

XMP gives me problem with 2933MHz, which I understand I set for only 2400MHz.
XMP off boots fine and 2 HCl tests over 1000% gives 0 errors over multiple processes.

I am in a dilemma whether to go with DRAM calc settings or Auto, since Auto gives best stability.

Hynix M die is checked from Typhoon Burner, I don't know what this is. Seeing M, I chose the settings for Hynix MFR in DRAM calculator
 
Jul 17, 2021
13
0
10
0
Further testing, under Profile1 in XMP, only tRRDS changes to 5 from 6 in Auto default. I change setting to Profile1 XMP and change tRRDS back to 6. While 2800MHz works with XMP ON and tRRDS at 6(Auto default settings), the Windows boots completely fine. As soon as I change it to 2933 MHz, Windows crashes at a high-resolution BSOD(ykwim) with errors CRITICAL PROCESS DIED and SYSTEM SERVICE EXCEPTION.

Looks like when Windows runs at Auto settings at 2933MHz, it overrides the default settings of Auto. If this theory is correct, then my one and only Profile1 settings aren't meant for 2933MHz. It also means that it is using somewhat near settings to DRAM calculator.
 

mamasan2000

Distinguished
My motherboard only gives Disabled and Profile 1 in XMP settings, ie only 2 options, also probably because 2400MHz is next to 2133 so it's just one profile. Moreover by default, XMP is disabled but RAM is set to 2400MHz, checked in BIOS and Windows. AFAIK somewhere I read anything over 2133MHz with JEDEC specifications is overclock, so my RAM is being overclocked by default and also without XMP settings.


The Primary settings are completely same. The scls are at 1. However with both Ryzen DRAM calculator and motherboard defaults, tCKE is 7.

XMP gives me problem with 2933MHz, which I understand I set for only 2400MHz.
XMP off boots fine and 2 HCl tests over 1000% gives 0 errors over multiple processes.

I am in a dilemma whether to go with DRAM calc settings or Auto, since Auto gives best stability.

Hynix M die is checked from Typhoon Burner, I don't know what this is. Seeing M, I chose the settings for Hynix MFR in DRAM calculator
SCLs provide a decent amount of performance but also instability if set too low. Values somewhere between 2-5 for over 2133 Mhz speeds. I have Hynix AFR and some cheap Samsung sticks, yes, mixed sticks. At XMP @ 3000 Mhz, I have to set SCLs to 4. I've tried 3, doesnt work, either I get BSOD or memory test fails.

tRFC around 320 ns if you use Ryzen DRAM calculator. Go to 'Additional calculators', insert 2933 and 320. This should give you a tRFC of 469.
tRFC around 320-350 ns should work. Samsung B-die is crazy, it can run at half that, 120-180 nanoseconds.

Jedec has more than 1 standard. https://www.guru3d.com/news-story/ryzen-3000-zen-2-cpus-get-jedec-3200-mhz-specification.html
And if you look at the table here: https://en.wikipedia.org/wiki/DDR4_SDRAM#JEDEC_standard_DDR4_module 2133, 2400, 2666 Mhz etc. And timings. What you might notice is the timings are VERY loose. But it should be guaranteed to work.
But yes, when people mention the JEDEC standard, they usually mean 2133 Mhz speed.

Auto always gives best stability. So it's a question of, do you want stability or performance? You can have both but you have to spend more time to get it. And to squeeze the last bit of perf from your RAM, you will have to spend increasing amount of time.
Some timings can give you 3% more RAM performance, others 0.1%.
Here is 1 picture: https://images.idgesg.net/images/article/2017/07/ryzen-ddr4-memory-100729659-orig.jpg Subtimings auto vs tuned vs maxed
 
Last edited:

mamasan2000

Distinguished
Further testing, under Profile1 in XMP, only tRRDS changes to 5 from 6 in Auto default. I change setting to Profile1 XMP and change tRRDS back to 6. While 2800MHz works with XMP ON and tRRDS at 6(Auto default settings), the Windows boots completely fine. As soon as I change it to 2933 MHz, Windows crashes at a high-resolution BSOD(ykwim) with errors CRITICAL PROCESS DIED and SYSTEM SERVICE EXCEPTION.

Looks like when Windows runs at Auto settings at 2933MHz, it overrides the default settings of Auto. If this theory is correct, then my one and only Profile1 settings aren't meant for 2933MHz. It also means that it is using somewhat near settings to DRAM calculator.
Something to think about when you are on the edge of stability and what Mhz your sticks can do, the motherboards RAM test and setting of timings = very fast. It can't spend minutes or hours testing every setting, it needs to test everything within seconds. Otherwise, every boot up of your system and everyone elses would take minutes or hours. That would be unacceptable.
So you get what you get under Auto at the bleeding edge. Timings should be loose enough to work but who knows. So you have to test RAM for errors.
 
Jul 17, 2021
13
0
10
0
SCLs provide a decent amount of performance but also instability if set too low. Values somewhere between 2-5 for over 2133 Mhz speeds. I have Hynix AFR and some cheap Samsung sticks, yes, mixed sticks. At XMP @ 3000 Mhz, I have to set SCLs to 4. I've tried 3, doesnt work, either I get BSOD or memory test fails.

tRFC around 320 ns if you use Ryzen DRAM calculator. Go to 'Additional calculators', insert 2933 and 320. This should give you a tRFC of 469.
tRFC around 320-350 ns should work. Samsung B-die is crazy, it can run at half that, 120-180 nanoseconds.

Jedec has more than 1 standard. https://www.guru3d.com/news-story/ryzen-3000-zen-2-cpus-get-jedec-3200-mhz-specification.html
And if you look at the table here: https://en.wikipedia.org/wiki/DDR4_SDRAM#JEDEC_standard_DDR4_module 2133, 2400, 2666 Mhz etc. And timings. What you might notice is the timings are VERY loose. But it should be guaranteed to work.
But yes, when people mention the JEDEC standard, they usually mean 2133 Mhz speed.

Auto always gives best stability. So it's a question of, do you want stability or performance? You can have both but you have to spend more time to get it. And to squeeze the last bit of perf from your RAM, you will have to spend increasing amount of time.
Some timings can give you 3% more RAM performance, others 0.1%.
Here is 1 picture: https://images.idgesg.net/images/article/2017/07/ryzen-ddr4-memory-100729659-orig.jpg Subtimings auto vs tuned vs maxed
At present I think it's not worth for me to squeeze the last performance. Most games that I play don't go over 120fps so it shouldn't be noticeable .

While my motherboard provides correct settings at 2933MHz, do you think it can do the same at 3200MHz RAM? If I buy and put 2 sticks of 2933MHz RAM which I would upgrade to 3200MHz will motherboard be able to put the correct settings? I was thinking about this because my motherboard supports Ryzen Zen+ with the 2018 BIOS, and also this CPU uses 2933MHz JEDEC standard, and anything above is JEDEC OC.
 
Jul 17, 2021
13
0
10
0
Something to think about when you are on the edge of stability and what Mhz your sticks can do, the motherboards RAM test and setting of timings = very fast. It can't spend minutes or hours testing every setting, it needs to test everything within seconds. Otherwise, every boot up of your system and everyone elses would take minutes or hours. That would be unacceptable.
So you get what you get under Auto at the bleeding edge. Timings should be loose enough to work but who knows. So you have to test RAM for errors.
If it's om bleeding edge, under what conditions should it fail to configure properly? If the BIOS has the universal JEDEC standard embedded, in which cases should the preset fail? Shouldn't all JEDEC compatible RAM work at JEDEC setting, given that the DRAM chip can supply more voltage and resistance when required?
 

mamasan2000

Distinguished
At present I think it's not worth for me to squeeze the last performance. Most games that I play don't go over 120fps so it shouldn't be noticeable .

While my motherboard provides correct settings at 2933MHz, do you think it can do the same at 3200MHz RAM? If I buy and put 2 sticks of 2933MHz RAM which I would upgrade to 3200MHz will motherboard be able to put the correct settings? I was thinking about this because my motherboard supports Ryzen Zen+ with the 2018 BIOS, and also this CPU uses 2933MHz JEDEC standard, and anything above is JEDEC OC.
Will 3200 Mhz work? Maybe, if you buy 3000 Mhz RAM but no guarantees. First gen Ryzen tends to top at 3200-3333 Mhz for most people. The question is, is it worth the money? For maybe 1-2% performance?
 

mamasan2000

Distinguished
If it's om bleeding edge, under what conditions should it fail to configure properly? If the BIOS has the universal JEDEC standard embedded, in which cases should the preset fail? Shouldn't all JEDEC compatible RAM work at JEDEC setting, given that the DRAM chip can supply more voltage and resistance when required?
When you put in too low values, when computer no longer boots, all you get is a blackscreen. That is when motherboard can't figure out timings. Set tRFC to 100 for example.

As far as I know, BIOS has zero presets for JEDEC. It's either auto (quick basic defaults, usually 2133 Mhz etc) or it's XMP and those timings are coded into your RAM sticks. So BIOS reads from RAM what the timings should be, at least primary.
JEDEC compatible RAM should work at JEDEC standards, yes. But as soon as you mention XMP or manual OC, you are no longer within JEDEC spec/standard. Think of JEDEC as something for servers, stability, something that needs 99.99% uptime. They don't OC hardware in a datacenter. Uptime is too important versus a few percents in performance.
 
Jul 17, 2021
13
0
10
0
When you put in too low values, when computer no longer boots, all you get is a blackscreen. That is when motherboard can't figure out timings. Set tRFC to 100 for example.
Yea that has happened

As far as I know, BIOS has zero presets for JEDEC. It's either auto (quick basic defaults, usually 2133 Mhz etc) or it's XMP and those timings are coded into your RAM sticks. So BIOS reads from RAM what the timings should be, at least primary.
JEDEC compatible RAM should work at JEDEC standards, yes. But as soon as you mention XMP or manual OC, you are no longer within JEDEC spec/standard. Think of JEDEC as something for servers, stability, something that needs 99.99% uptime. They don't OC hardware in a datacenter. Uptime is too important versus a few percents in performance.
How does the motherboard determine what settings it should choose on Auto without XMP? Like 2933MHz shouldn't work with "the usual 2133MHz preset". If I understand correctly it's the same case with motherboards that don't support XMP, they just put default values of JEDEC standards as indicated in the Wikipedia article, but how does it determine the other settings like tRFC or tRRDS? How accurate are they when not using XMP?
 

mamasan2000

Distinguished
Yea that has happened



How does the motherboard determine what settings it should choose on Auto without XMP? Like 2933MHz shouldn't work with "the usual 2133MHz preset". If I understand correctly it's the same case with motherboards that don't support XMP, they just put default values of JEDEC standards as indicated in the Wikipedia article, but how does it determine the other settings like tRFC or tRRDS? How accurate are they when not using XMP?
I don't know how many timings are programmed into the RAM sticks. Could be most of them?
Are there motherboards that don't support XMP? I don't know of a consumer board that doesn't.
Motherboards are supposed to find timings that are safe and stable, how exactly that happens, I don't know.

Usually for the lower secondary timings, tFAW for example, usually it can go lower. It's tRRDS * 4 I think. So if tRRDS is 6, then tFAW 24 should work. mine is 30. But my setup is a special case, since I have Hynix AFR and Samsung cheapo RAM, so characteristics of both is different so both RAM kits have to work at whatever timings I set. tWTRS, tWTRL and tWR at 4-12-12 should work for most but I have 4-12-20.
tRFC at 480 = 320 ns. This my mobo sets to 520 or even higher. SCLs are 5 by default, I have set them to 4.
Values under 10 you can test to drop by 1. And test if that works. tCWL same as tCL. tRTP at 12 tends to work.
You will figure it out the more you play with it. I played with Hynix AFR for probably 6 months, across BIOS updates. And at a certain BIOS upgrade, I could no longer run 3333 Mhz stable. Had to drop to 3200 Mhz. So BIOS version matters too. Then I got 16 gigs more of the same brand and timings, Corsair Vengeance 3000 Mhz. But new sticks are Samsung, sadly. I'm just glad they work together, no guarantee of that. They DID NOT work together when I just enabled XMP. I had to play with timings, Dram voltage at 1.37. ProcODT 53 Ohm worked nicely with Hynix AFR but with both sticks I run 48 Ohm. On first gen Ryzen 48-60 Ohm is recommended. DrvStrength at 20-20-20-20 or 24-24-24-24 or a combination, as you could probably tell from Ryzen Dram Calc.

When it comes to custom RAM timings and clocks, you gotta test test test! For this I use Testmem5. Because its quick and catches most errors.
https://www.techpowerup.com/review/amd-ryzen-memory-tweaking-overclocking-guide/9.html Scroll all the way to the bottom, get Testmem5 and below it the Preset. Extract both files to same folder. That test should take 5-6 minutes with 16 gigs. It takes 10+ minutes with 32 gigs. If that passes, I know I'm close to stable. That is when I start testing with HCI Memtest.
Also, worth reading the whole guide/article, written by 1usmus, the author of Ryzen Dram Calc.

EDIT: IF you open up Thaiphoon Burner, go to EEPROM, click Read SMBus on one of your sticks then click on Report. Scroll to the bottom. I guess those are the timings that are on the sticks.
 
Last edited:
Jul 17, 2021
13
0
10
0
I don't know how many timings are programmed into the RAM sticks. Could be most of them?
Are there motherboards that don't support XMP? I don't know of a consumer board that doesn't.
Motherboards are supposed to find timings that are safe and stable, how exactly that happens, I don't know.

Usually for the lower secondary timings, tFAW for example, usually it can go lower. It's tRRDS * 4 I think. So if tRRDS is 6, then tFAW 24 should work. mine is 30. But my setup is a special case, since I have Hynix AFR and Samsung cheapo RAM, so characteristics of both is different so both RAM kits have to work at whatever timings I set. tWTRS, tWTRL and tWR at 4-12-12 should work for most but I have 4-12-20.
tRFC at 480 = 320 ns. This my mobo sets to 520 or even higher. SCLs are 5 by default, I have set them to 4.
Values under 10 you can test to drop by 1. And test if that works. tCWL same as tCL. tRTP at 12 tends to work.
You will figure it out the more you play with it. I played with Hynix AFR for probably 6 months, across BIOS updates. And at a certain BIOS upgrade, I could no longer run 3333 Mhz stable. Had to drop to 3200 Mhz. So BIOS version matters too. Then I got 16 gigs more of the same brand and timings, Corsair Vengeance 3000 Mhz. But new sticks are Samsung, sadly. I'm just glad they work together, no guarantee of that. They DID NOT work together when I just enabled XMP. I had to play with timings, Dram voltage at 1.37. ProcODT 53 Ohm worked nicely with Hynix AFR but with both sticks I run 48 Ohm. On first gen Ryzen 48-60 Ohm is recommended. DrvStrength at 20-20-20-20 or 24-24-24-24 or a combination, as you could probably tell from Ryzen Dram Calc.

When it comes to custom RAM timings and clocks, you gotta test test test! For this I use Testmem5. Because its quick and catches most errors.
https://www.techpowerup.com/review/amd-ryzen-memory-tweaking-overclocking-guide/9.html Scroll all the way to the bottom, get Testmem5 and below it the Preset. Extract both files to same folder. That test should take 5-6 minutes with 16 gigs. It takes 10+ minutes with 32 gigs. If that passes, I know I'm close to stable. That is when I start testing with HCI Memtest.
Also, worth reading the whole guide/article, written by 1usmus, the author of Ryzen Dram Calc.

EDIT: IF you open up Thaiphoon Burner, go to EEPROM, click Read SMBus on one of your sticks then click on Report. Scroll to the bottom. I guess those are the timings that are on the sticks.

How many cycles of TestMem5 should I run? Some users have recommended to use anta777's preset. It goes to test 12 and back to 2, Am I missing something?

Edit: With the default preset, it says no errors were found.
 
Last edited:

mamasan2000

Distinguished
How many cycles of TestMem5 should I run? Some users have recommended to use anta777's preset. It goes to test 12 and back to 2, Am I missing something?

Edit: With the default preset, it says no errors were found.
With the preset from the web, it should do 3 cycles.
<empty> errors found. That's good. Still gotta run HCI Memtest.
Testmem while you are dialing in each timing, Memtest when you think you are done with dialing in.

Doesn't anta-preset take forever? Some preset took an hour or two. Too long. I don't have 2 years to test if my RAM is stable.
 
Jul 17, 2021
13
0
10
0
With the preset from the web, it should do 3 cycles.
<empty> errors found. That's good. Still gotta run HCI Memtest.
Testmem while you are dialing in each timing, Memtest when you think you are done with dialing in.

Doesn't anta-preset take forever? Some preset took an hour or two. Too long. I don't have 2 years to test if my RAM is stable.
Everything seems fine with Auto OC while tested on TestMem5 and MemTest, neither did I ever face crash with this or the calculator which still got an error. Until I get any crash I think I can use this preset without problems.
 

mamasan2000

Distinguished
Everything seems fine with Auto OC while tested on TestMem5 and MemTest, neither did I ever face crash with this or the calculator which still got an error. Until I get any crash I think I can use this preset without problems.
Yup. Every time I set stuff to Auto, I get this nagging voice in my head that says: "You could try lowering the value, though. DO IT!" =). And 2 weeks later I'm doing exactly that. It's easy to never be satisfied. Have to decide whats good enough.
 

ASK THE COMMUNITY

TRENDING THREADS