F5F Stay Refreshed Power Users Overclocking Ram timings help.

Ram timings help.

Ram timings help.

Pages (3): Previous 1 2 3
M
mckahla
Member
56
12-02-2016, 02:25 PM
#21
No, it won't. At least don't take it for granted, since it's not a guaranteed outcome. I'm currently handling another thread with someone who has a Ryzen setup but XMP is active on a kit using CL15 timings. When XMP is enabled, it doesn't adjust the timings to rounded values, which might be contributing to his performance problems. We suspect this change in CAS latency (CL) to 16 is affecting tRCD, tRP, and tRAS, even though we usually see Ryzen-compatible kits with all timings set to even numbers.
M
mckahla
12-02-2016, 02:25 PM #21

No, it won't. At least don't take it for granted, since it's not a guaranteed outcome. I'm currently handling another thread with someone who has a Ryzen setup but XMP is active on a kit using CL15 timings. When XMP is enabled, it doesn't adjust the timings to rounded values, which might be contributing to his performance problems. We suspect this change in CAS latency (CL) to 16 is affecting tRCD, tRP, and tRAS, even though we usually see Ryzen-compatible kits with all timings set to even numbers.

H
Haz3lNut
Member
97
12-02-2016, 05:43 PM
#22
if gdm is active, it adjusts the value by rounding, even when showing cl15 or similar, because gdm's method involves halving speeds for tCL and tCWL while also halving digits, which prevents odd numbers from working. For example, a 3600MT/s CL15 with gdm on would run tCL and tCWL at half speed (1800MT/s) and round the digit to 8. If gdm is off, it maintains the original command rate without rounding.
H
Haz3lNut
12-02-2016, 05:43 PM #22

if gdm is active, it adjusts the value by rounding, even when showing cl15 or similar, because gdm's method involves halving speeds for tCL and tCWL while also halving digits, which prevents odd numbers from working. For example, a 3600MT/s CL15 with gdm on would run tCL and tCWL at half speed (1800MT/s) and round the digit to 8. If gdm is off, it maintains the original command rate without rounding.

S
Streiyn
Posting Freak
768
12-08-2016, 10:31 PM
#23
Gear down mode was enabled by default due to frequent memory issues on Ryzen systems. In recent years, most of the boards I've worked with had GDM turned off by default, and many top experts suggest keeping it disabled while using a 1T command rate if possible, 2T if not, and only enabling Gear down when necessary because it helps stabilize memory operations but reduces speed and increases latency.
S
Streiyn
12-08-2016, 10:31 PM #23

Gear down mode was enabled by default due to frequent memory issues on Ryzen systems. In recent years, most of the boards I've worked with had GDM turned off by default, and many top experts suggest keeping it disabled while using a 1T command rate if possible, 2T if not, and only enabling Gear down when necessary because it helps stabilize memory operations but reduces speed and increases latency.

S
Summits
Junior Member
38
12-09-2016, 08:27 AM
#24
I only got notifications for the first 2 replies since my last post. I didn't realise there was more activity until yesterday. I don't know how to reply to multiple people at once. So I'll combine everything into one post.
I cannot get TM5 Absolute to run at all still. Each time I try there's a popup for each of the workers saying that it's crashed. Extreme still runs ok. Although there's a lot of fluctuation in the sec/gb reading while it's running. A few tests it's in single digits and a few others it's almost 180sec/gb. Would it happen to be staying in the cpu cache for the quicker numbers and hitting into the page file when it gets that slow?
I don't know if it works that way or not. But someone said it was 4 ranks 4x8gb that I had. Wouldn't it be 8 ranks? I'm using 4 dual rank sticks. 4x16gb.
Also see what you said a bit about the different kits. The say they are made in the same month and have the same item number. But on one set it has 04213X8810B and the other has 0421308810B on it. I don't know what the difference between the X and 0 are though.
I've been making changes everyday following the guides and advice posted. I'll post a pic at the end of this to see how it's going and for anymore tips/advice as to how much more to proceed. Oddly from before I started the post the changes are going much easier for some reason. Only one change this time that I've tried hasn't worked. Trying to drop tCWL at all ended up in having to reset the CMOS to get the computer to post again. I'm not sure what tWRRD should be at all and I haven't tried messing with tRFC/tREFI much yet. How are things looking otherwise. Some of you were talking about GDM disabled. Should that be attempted next/soon?
So far this passes 4 passes of memtest86, 3 passes of TM5 extreme, and 14.5 hours of prime95 set the way the guide says to without any errors. Temps of the inner 2 dimms still hit about 50c though. The outer 2 are about 45-47c. I can try to see if it'll still be stable at a lower voltage to try to drop temps if you still think that's a good idea. Is it odd that the temps didn't really increase with the tighter timings? Still holding about mids 30s idle and mid 40s gaming too.
Is it likely to be able to increase the frequency above the xmp rating? Or is that too much to ask from this mobo/cpu combo? With just the one kit in there (either one). I could get it to boot easy into 3800. But didn't test it much or try higher frequencies. Mostly had it at 16s instead of 14s. With all 4 in there it won't do above 3600 even with looser timings. Unless I'm missing something.
Here's what I got so far:
https://imgur.com/a/fcZujPA
View: https://imgur.com/a/fcZujPA
S
Summits
12-09-2016, 08:27 AM #24

I only got notifications for the first 2 replies since my last post. I didn't realise there was more activity until yesterday. I don't know how to reply to multiple people at once. So I'll combine everything into one post.
I cannot get TM5 Absolute to run at all still. Each time I try there's a popup for each of the workers saying that it's crashed. Extreme still runs ok. Although there's a lot of fluctuation in the sec/gb reading while it's running. A few tests it's in single digits and a few others it's almost 180sec/gb. Would it happen to be staying in the cpu cache for the quicker numbers and hitting into the page file when it gets that slow?
I don't know if it works that way or not. But someone said it was 4 ranks 4x8gb that I had. Wouldn't it be 8 ranks? I'm using 4 dual rank sticks. 4x16gb.
Also see what you said a bit about the different kits. The say they are made in the same month and have the same item number. But on one set it has 04213X8810B and the other has 0421308810B on it. I don't know what the difference between the X and 0 are though.
I've been making changes everyday following the guides and advice posted. I'll post a pic at the end of this to see how it's going and for anymore tips/advice as to how much more to proceed. Oddly from before I started the post the changes are going much easier for some reason. Only one change this time that I've tried hasn't worked. Trying to drop tCWL at all ended up in having to reset the CMOS to get the computer to post again. I'm not sure what tWRRD should be at all and I haven't tried messing with tRFC/tREFI much yet. How are things looking otherwise. Some of you were talking about GDM disabled. Should that be attempted next/soon?
So far this passes 4 passes of memtest86, 3 passes of TM5 extreme, and 14.5 hours of prime95 set the way the guide says to without any errors. Temps of the inner 2 dimms still hit about 50c though. The outer 2 are about 45-47c. I can try to see if it'll still be stable at a lower voltage to try to drop temps if you still think that's a good idea. Is it odd that the temps didn't really increase with the tighter timings? Still holding about mids 30s idle and mid 40s gaming too.
Is it likely to be able to increase the frequency above the xmp rating? Or is that too much to ask from this mobo/cpu combo? With just the one kit in there (either one). I could get it to boot easy into 3800. But didn't test it much or try higher frequencies. Mostly had it at 16s instead of 14s. With all 4 in there it won't do above 3600 even with looser timings. Unless I'm missing something.
Here's what I got so far:
https://imgur.com/a/fcZujPA
View: https://imgur.com/a/fcZujPA

A
Athame_
Senior Member
734
12-09-2016, 03:48 PM
#25
The system can be pushed to its limits, maintaining stability with tCL and tRP if needed. The following timings perform well compared to stock XMP settings. The main concern is whether it remains stable during regular use and gaming, especially considering the heat generated by the GPU into the RAM sticks. You might want to adjust the AddrCmdDrvStren setting to 24, set RTT at 560/630, keep procODT steady or switch to 48ohm, and experiment with RttPark values like 1/2/3, 240ohm, 120ohm, or 80ohm. Ensure RttNom is around 7 or 34.3 ohms and RttWr at 3 or 80ohm. Adjust tCL to tRC, then relax tCL further, and modify other timings accordingly. Gradually increase the frequency.

Incorporating RAM changes often requires these adjustments (loosen, identify maximum, then decide between tight timings or higher frequencies with slightly looser settings for stability). For higher frequencies, Zen 3 limits are mainly on the I/O die, typically reaching around 3833 for a 1:1:1 configuration (UCLK and FCLK at equal ratios), sometimes up to 3900-4000—commonly referred to as the golden bin.
A
Athame_
12-09-2016, 03:48 PM #25

The system can be pushed to its limits, maintaining stability with tCL and tRP if needed. The following timings perform well compared to stock XMP settings. The main concern is whether it remains stable during regular use and gaming, especially considering the heat generated by the GPU into the RAM sticks. You might want to adjust the AddrCmdDrvStren setting to 24, set RTT at 560/630, keep procODT steady or switch to 48ohm, and experiment with RttPark values like 1/2/3, 240ohm, 120ohm, or 80ohm. Ensure RttNom is around 7 or 34.3 ohms and RttWr at 3 or 80ohm. Adjust tCL to tRC, then relax tCL further, and modify other timings accordingly. Gradually increase the frequency.

Incorporating RAM changes often requires these adjustments (loosen, identify maximum, then decide between tight timings or higher frequencies with slightly looser settings for stability). For higher frequencies, Zen 3 limits are mainly on the I/O die, typically reaching around 3833 for a 1:1:1 configuration (UCLK and FCLK at equal ratios), sometimes up to 3900-4000—commonly referred to as the golden bin.

Pages (3): Previous 1 2 3