RAM Tuning - Help with subtimings.

[Kodehawa]

Not really, but they scale with voltage pretty well, so if you ever decide to go above 3600 don't be afraid to pump more voltage.

When you finish tertiaries, try tRRDS 4 tRRDL 4 tFAW 16 tWTRL 8

No luck at 3733, but those tertiaries worked fine. Thanks you so much btw


This are the "final" timings (after tweaking everything). I gotta run an overnight test to be sure, though.

 

[mechtech]

No luck at 3733, but those tertiaries worked fine. Thanks you so much btw


This are the "final" timings (after tweaking everything). I gotta run an overnight test to be sure, though.
View attachment 288500

pfffffffffffffffff

check out these stock bad boys - 100% default lol

 

[Kodehawa]

pfffffffffffffffff

check out these stock bad boys - 100% default lol
View attachment 288501

lmao what die is this

to be fair, the stock ones here are... awful too lol, definitely not tCL 22 awful, though, secondaries are actually worse sometimes (stock tRFC on XMP shoots to 620)

EDIT: That kit is E-Die too. Maybe you could use my timings as a guide? Even if just at 3200

 

[3x0]

This are the "final" timings (after tweaking everything). I gotta run an overnight test to be sure, though.

tRDRDSD should be 4 instead of 5 and tWRWRSD should be 6 instead of 7.

Also, check if tPHYRDL is the same 26/26 for both sticks across several reboots. When I tweaked my RAM, one in 10 reboots would result in missmatch 26/28 tPHYRDL...

 

[Kodehawa]

tRDRDSD should be 4 instead of 5 and tWRWRSD should be 6 instead of 7.

Also, check if tPHYRDL is the same 26/26 for both sticks across several reboots. When I tweaked my RAM, one in 10 reboots would result in missmatch 26/28 tPHYRDL...

Oh. ty, those were on auto.

For me it's 26/28 yeah. On all boots, though.

 

[3x0]

For me it's 26/28 yeah. On all boots, though.

A mismatch is not ideal. Try lowering ProcODT to the lowest value, or a notch above the lowest and see if it fixes the mismatch.

Unfortunately you'll have to find the sweetspot for this one, it's either ProcODT, vSOC or VDDP/G. Try with ProcODT first since that was what fixed it for me.

 

[Kodehawa]

A mismatch is not ideal. Try lowering ProcODT to the lowest value, or a notch above the lowest and see if it fixes the mismatch.

Unfortunately you'll have to find the sweetspot for this one, it's either ProcODT, vSOC or VDDP/G. Try with ProcODT first since that was what fixed it for me.

Lowering ProcODT did not help, sadly.

Seems to also happen at 3200MHz.

Should be the "final" timings now. I tried tRCDWR 16 alone and that didn't work either. Sad, because I've seen Rev. E as low as 8 on tRCDWR.

 

[3x0]

Lowering ProcODT did not help, sadly.

Seems to also happen at 3200MHz.

That's really weird. It should be doing that at all at such low speeds. Can you try and disable Memory Fast Boot in your BIOS? (I don't know how it's called in ASUS BIOS)
It should force the motherboard to retrain RAM/Mem controller instead of restoring previous "training data".

 

[Dragonsyph]

I’m messing with mine aswell, how can we tell if what we are even doing is increasing FPS? Do we like run a special benchmark before and after and compare? If so what is said benchmark?

 

[AMF]

Hello there!

I have a pair of T-Force DDR4 3200MHz RAM I managed to push to 3600MHz (and to verify it's stable, which was a pain but )
Was wondering whether any of the subtimings are suboptimal, since the automatic settings on the motherboard seem to be fairly lenient on them, but I have no framework to work with.

About main timings: CL 18-21-21-21-40 at 3600 is the least that will not produce errors over a 12h long memory test. Doubt it's possible to push that further (even doing 18-20-20-20-40 fails) at 1.4V (1.45V too, but 1.4V works just as well on that).
Taiphoon Burner does not tell me the RAM chip type of this specific RAM. Screenshot attached. If anything, I wonder if anyone knows what specific memory chips this uses.

View attachment 285983

Thanks you all!

lower your vddg iod to 1.001 for cldo vddp try .851 vddg ccd try .951

i was able to do this with cheap mem

No luck at 3733, but those tertiaries worked fine. Thanks you so much btw


This are the "final" timings (after tweaking everything). I gotta run an overnight test to be sure, though.
View attachment 288500

try 1.1 v even on vsoc on my bios i had to set to 1.118 to get even 1.1 in hwmon second pic is stock freq 1.2v 2666 xmp to 3800 with same cas 1.35 v

 

[izy]

Is it Micron E die? You can try mine see if they work:

 

[Kodehawa]

Is it Micron E die? You can try mine see if they work:
View attachment 288525

tRCDWR < 18 = no post (cpu light)
tRCDRD < 21 = post, but bsod or thousands of errors in TM5 memtest
tRTP < 18 = no post (dram light)

tRFC works, the rest look pretty similar. All at 1.45V

What does GDM Disabled do?

That's really weird. It should be doing that at all at such low speeds. Can you try and disable Memory Fast Boot in your BIOS? (I don't know how it's called in ASUS BIOS)
It should force the motherboard to retrain RAM/Mem controller instead of restoring previous "training data".

Will try.

 

[3x0]

What does GDM Disabled do?

Practically, it allows for odd numbered tCL i.e 15, 17 etc.

 

[mechtech]

lmao what die is this

to be fair, the stock ones here are... awful too lol, definitely not tCL 22 awful, though, secondaries are actually worse sometimes (stock tRFC on XMP shoots to 620)

EDIT: That kit is E-Die too. Maybe you could use my timings as a guide? Even if just at 3200

Ya micron e die. It’s a pair of dual rank 3200 from crucial in my system. Runs at 1.20v jedec 3200.

 

[izy]

tRCDWR < 18 = no post (cpu light)
tRCDRD < 21 = post, but bsod or thousands of errors in TM5 memtest
tRTP < 18 = no post (dram light)

tRFC works, the rest look pretty similar. All at 1.45V

What does GDM Disabled do?


Will try.

Did you try with exactly same settings? Try same settings with CL15 , my RAM for some reason can boot with this settings at this speed but if i increase some values it just wont boot anymore for example if i increase tRCDRD it doesnt work with some higher values, some weird things like that (was expecting only lowering things to make it bad), if im trying to use lets say 18 - 22 - 22 @ 3733 it just doesnt boot, even with CL16 was a bit sensible. My ram seems that it is working better with CL 15 / CL 17, allowing my system to boot and be stable and allowing more random values (and it was good place to start) , ive started with cl17 then went C15 then ive found this settings at CL14 that are working stable but changing any values from here stops booting or im getting TM5 errors (even if im increasing some of the values).

Edit: i just noticed that your RAM is Dual Rank and its harder to OC DR , mine works with 4 sticks too and same settings but i am not sure if Dual Rank is more or less taxing than 4 single rank sticks, try to set a lower FLCK and check TM5 for errors OR try 2T, i think 2T can help with dual rank RAM OC. (sometimes the problem can be with the memory controller and not the ram)

 

[Kodehawa]

Did you try with exactly same settings? Try same settings with CL15 , my RAM for some reason can boot with this settings at this speed but if i increase some values it just wont boot anymore for example if i increase tRCDRD it doesnt work with some higher values, some weird things like that (was expecting only lowering things to make it bad)

Ah huh. Odd. I might try the full thing then.

 

[izy]

Ah huh. Odd. I might try the full thing then.

Yeah this E die is weird on my sticks , try to OC your RAM first with half speed on memory controller to be sure its not that and for dual rank 32gb you should try 2T , i dont get big latency hit 1t vs 2t , GDM its somehow 2T anyway, when i started to OC my RAM i had problems with getting tRCDRD right as i said sometimes higher value was not working for some reason. Good luck

 

[Kodehawa]

oops, time to see what caused it now

ugh, running 1usmus is impossible, i can't get any errors before it just stops running altogether in a very short amount of time (give it 20~30 minutes)

even at stock it will eventually just stop running and say all memory is available while still "running" lol

It does this:

No matter what I do really, this was with nothing running in the background while running the test. I wonder if I need to give it less memory (anta777 which runs does 1.7GB x 16 instead of 1.8GB x 16).

 

[izy]

View attachment 288599
oops, time to see what caused it now

ugh, running 1usmus is impossible, i can't get any errors before it just stops running altogether in a very short amount of time (give it 20~30 minutes)

even at stock it will eventually just stop running and say all memory is available while still "running" lol

It does this:
View attachment 288731
No matter what I do really, this was with nothing running in the background while running the test. I wonder if I need to give it less memory (anta777 which runs does 1.7GB x 16 instead of 1.8GB x 16).

Did you try to close and reopen it after you free some RAM? Or just use the ram cleaner button from 1usmus ram calculator or what is called.

 

[Kodehawa]

Did you try to close and reopen it after you free some RAM? Or just use the ram cleaner button from 1usmus ram calculator or what is called.

Giving more RAM to Windows seemed to have done it. It was 256MB by default reserved to Windows, which is unthinkable lol. 1024MB should be the minimum you give it so it doesn't kill the workers.

 

[WhenMusicAttacks]

Just some tips:
1remember to test memory with load on the dGpu if you care about actual game stability,
2 test performance with different programs, exp when you get to a close-to-instability value, as sometimes you can have perf regression (that's mostly on d5 but had it on d4 aswell)
3 huge help to stabilize higher fclk is to undervolt the cores with curve to reduce the heat buildup inside the cpu itself

 

[The King]

Not all Micron Rev.E kits can perform the same or share the same settings. There is a binning process that makes some kits better than others.

This is why its good to find our what IC you have and what Speed grade they are rate at.

If your stock XMP is 3200 CL16-20-20-40-60 its not going to run the same settings of a higher bin IC that has XMP like 3600 CL16-18-18-38-58.

 

[Veseleil]

I have two identical kits of 2x8GB Ballistix with 3200 16-18-18-36 1.35V and can say it's a great bin, overclocks like crazy and it's stable. All of the fast timing variations and other settings of the @1usmus DRAM calculator work excelent. Talking about 2 sticks. I've tried to run 4 sticks on my daizy chain tomahawk max, and there's no way for it to work.

 

[The King]

I have two identical kits of 2x8GB Ballistix with 3200 16-18-18-36 1.35V and can say it's a great bin, overclocks like crazy and it's stable. All of the fast timing variations and other settings of the @1usmus DRAM calculator work excelent. Talking about 2 sticks. I've tried to run 4 sticks on my daizy chain tomahawk max, and there's no way for it to work.

Could be because of ZEN 2 CPU or you not giving the the dimms enough VDIMM (Voltage) or enough VSOC.
Running 4 X dimms is harder on the CPUs IMC which is part of the SOC.

I have 3200 16-18-18-18-36 1.35V (C9BJZ) it does 4000MT or 2000 FCLK on ZEN 3 with no issues @ 1.45V

Share your AIDA 64 cache and memory benchmark here

This is another Micron Kit I have Crucial Ballistix 3600 CL 16-18-18-38-58 (C9BLM) 16GB DR Micron Rev. E Running ar 4133 CL16-19-19-44-63 1:1:1. Left CCD on Auto it will only boot with 1.2 CCD. This passes at least 5 cycles of TM5 1usmus. I see you are above 2K.. very nice. I am on the fence...

www.techpowerup.com

Share your AIDA 64 cache and memory benchmark here

This is another Micron Kit I have Crucial Ballistix 3600 CL 16-18-18-38-58 (C9BLM) 16GB DR Micron Rev. E Running ar 4133 CL16-19-19-44-63 1:1:1. Left CCD on Auto it will only boot with 1.2 CCD. This passes at least 5 cycles of TM5 1usmus. I see you are above 2K.. very nice. I am on the fence...

www.techpowerup.com

 

[Veseleil]

Could be because of ZEN 2 CPU or you not giving the the dimms enough VDIMM (Voltage) or enough VSOC.
Running 4 X dimms is harder on the CPUs IMC which is part of the SOC.

I have 3200 16-18-18-18-36 1.35V (C9BJZ) it does 4000MT or 2000 FCLK on ZEN 3 with no issues @ 1.45V

Share your AIDA 64 cache and memory benchmark here

This is another Micron Kit I have Crucial Ballistix 3600 CL 16-18-18-38-58 (C9BLM) 16GB DR Micron Rev. E Running ar 4133 CL16-19-19-44-63 1:1:1. Left CCD on Auto it will only boot with 1.2 CCD. This passes at least 5 cycles of TM5 1usmus. I see you are above 2K.. very nice. I am on the fence...

www.techpowerup.com

Share your AIDA 64 cache and memory benchmark here

This is another Micron Kit I have Crucial Ballistix 3600 CL 16-18-18-38-58 (C9BLM) 16GB DR Micron Rev. E Running ar 4133 CL16-19-19-44-63 1:1:1. Left CCD on Auto it will only boot with 1.2 CCD. This passes at least 5 cycles of TM5 1usmus. I see you are above 2K.. very nice. I am on the fence...

www.techpowerup.com

Sorry to jump on a thread, but as there's talk about micron e, can't help it.
My 3600 isn't stable at anything beyond 1866 FCLK.
As I said, 2x8 is fine w/e I throw at it, but 4x8 is a problem I can't seem to solve.
Apart from trying to raise SOC and DRAM voltages, maybe try running 2T, is there anything else I'm missing from the big picture? I'm well aware not to expect anything from a daizy chain topology and 4 sticks, but is there a way to make it work without shooting above 70ns latency?