Overclock.net banner
1 - 14 of 14 Posts

·
Registered
5900X, RTX 3080
Joined
·
854 Posts
Discussion Starter · #1 ·
Intro
After reading a number of articles on other websites, and a number of threads here, I thought I would try and disable GDM and manually set the CR to 1. I have already tested my RAM kit (CMK32GX4M2Z3600C18 / 3600 CL18) at its rated XMP (an old habit as it's still overclocking) and it passed memtest, TestMem, tKarhu, and HCI Memtest without an issue. The screenshot below shows the default timings and configuration for this kit.

Font Rectangle Parallel Number Screenshot


Software for Testing
BIOS/UEFI
I entered the BIOS, disabled GDM, manually set the CR to 1, and rebooted into Windows. I then ran TestMem5 with the "[email protected]" configuration and it started spitting out errors all over the place. Within a few minutes there were over 45 errors.

I rebooted, re-enabled GDM, set CR to Auto (it defaults to 1 with GDM enabled) and booted back into Windows and ram the same test - it passed.

I've seen a number of highly knowledgeable people on OCN adjusting even tertiary timings, so thought it best to ask here. Is there any advice you can offer for trying to obtain stability and performance with GDM disabled? I haven't tried GDM disabled with CR2, but that sort of defeats the purpose.
 

·
Overclocked
Joined
·
3,988 Posts
Increase SOC voltage gradually. I can run my memory with tight timings at 1v SOC and GDM enabled, but with it disabled I had to increase SOC voltage to 1.1v. I got like 0.5ns less latency with GDM off and slightly better memory scores, but I didn't like the voltage increase so I enabled it again.
 
  • Rep+
Reactions: J7SC and Piers

·
Registered
5900X, RTX 3080
Joined
·
854 Posts
Discussion Starter · #3 ·
Increase SOC voltage gradually. I can run my memory with tight timings at 1v SOC and GDM enabled, but with it disabled I had to increase SOC voltage to 1.1v. I got like 0.5ns less latency with GDM off and slightly better memory scores, but I didn't like the voltage increase so I enabled it again.
Thank you for the clear information. The SoC voltage on auto goes between 1.087V and 1.106V. I'll try your suggestion as I'm seeing if I can reduce latency. It's not terrible (and hard to measure on Windows 11 AFAIK) but could be improved. This RAM kit doesn't appear to like any changes. Normally I'd expect to be able to go from 3600 to 3800, for example, without the need to increase voltage, but when doing that I had to reset CMOS. Maybe that's an old-fashioned way of looking at RAM overclocking.

Edit: I've just noticed your signature - how does one create a signature like that? I can't seem to work it out.
 

·
Tank destroyer and a god
Joined
·
2,834 Posts
On my current setup I did not went into much RAM tuning, since the system was rock-solid and performing beyond all my expectations...

The only tuning of RAM i did was to set CR1 and disable Gear Down Mode. Originally it was for performance purposes, yet I noticed that few unexpected BSODs completely vanished afterwards. I did not looked into it for more details, and it does not make much sense.
 

·
Registered
Joined
·
4,690 Posts
Yes, it's possible. It's very CPU / IMC dependant as well. My old B0 5900X could not do 1T GDM Off above 3600. This B2 one does it at 3866 with no real extra voltage needed. It does usually need adding of AddrCmdSetup (56-60) to work properly.
 

·
Registered
5900X, RTX 3080
Joined
·
854 Posts
Discussion Starter · #6 ·
Yes, it's possible. It's very CPU / IMC dependant as well. My old B0 5900X could not do 1T GDM Off above 3600. This B2 one does it at 3866 with no real extra voltage needed. It does usually need adding of AddrCmdSetup (56-60) to work properly.
I know my B0 is capable of running DDR4 4000 at 1:1 (GDM on, tested with a Kingston 2*16 GB single rank kit that was three times the price of the RAM I chose) so I'm not that concerned by the IMC. I probably should do it manually and start with a small SoC voltage bump as suggested earlier.
 

·
Registered
Joined
·
4,690 Posts
I know my B0 is capable of running DDR4 4000 at 1:1 (GDM on, tested with a Kingston 2*16 GB single rank kit that was three times the price of the RAM I chose) so I'm not that concerned by the IMC. I probably should do it manually and start with a small SoC voltage bump as suggested earlier.
Did you test 1T with AddrCmdSetup 56 already or? Without it there's almost no chance of it being stable or even booting.
 
  • Rep+
Reactions: Piers

·
Registered
5900X, RTX 3080
Joined
·
854 Posts
Discussion Starter · #8 ·
Did you test 1T with AddrCmdSetup 56 already or? Without it there's almost no chance of it being stable or even booting.
Nope. I wasn't aware of the need to change AddrCmdSetup before you mentioned it. So, would you suggest trying:

  • enable XMP
  • disable GDM
  • change AddrCmdSetup to 56? I've seen some say 61
  • change command rate to 1
 

·
Registered
Joined
·
4,690 Posts
Nope. I wasn't aware of the need to change AddrCmdSetup before you mentioned it. So, would you suggest trying:

  • enable XMP
  • disable GDM
  • change AddrCmdSetup to 56? I've seen some say 61
  • change command rate to 1
Yeah. Exactly. 56 is the general minimum required. Mine does fine with 56 up to 3866 1T. Most people accept a range from 56-60 or a bit above.

I do not daily it however since my combination of board, CPU and DIMM's only train tPHYRDL on a even number on 1T GDM Off so I need to run 3866C14 1T and that takes way too much vDIMM. I just run 3866C15 2T GDM Off with nice and low voltages. Latency is almost the same as tPHYRDL matches now and at 3866C15 1T mismatched tPHYRDL isn't faster.
 
  • Rep+
Reactions: Piers

·
Registered
5900X, RTX 3080
Joined
·
854 Posts
Discussion Starter · #10 ·
Yeah. Exactly. 56 is the general minimum required. Mine does fine with 56 up to 3866 1T. Most people accept a range from 56-60 or a bit above.

I do not daily it however since my combination of board, CPU and DIMM's only train tPHYRDL on a even number on 1T GDM Off so I need to run 3866C14 1T and that takes way too much vDIMM. I just run 3866C15 2T GDM Off with nice and low voltages. Latency is almost the same as tPHYRDL matches now and at 3866C15 1T mismatched tPHYRDL isn't faster.
OK, just tried two combinations and tested for 10 minutes (TestMem seems to produce errors very quickly if unstable - for daily use I would test with 5 applications over a week):

First: hundreds of errors in TestMem5 (w/ anta777 Extreme1) within 10 minutes
  • XMP enabled (3600 - CL18 - 1.35V)
  • SoC 1.1V
  • GDM disabled
  • 1T
  • AddrCmdSetup of 56
Second: zero errors in TestMem5 (w/ anta777 Extreme1) within 10 minutes
  • XMP enabled (3600 - CL18 - 1.35V)
  • SoC 1.1V
  • GDM disabled
  • 2T
 

·
Registered
Joined
·
4,690 Posts
OK, just tried two combinations and tested for 10 minutes (TestMem seems to produce errors very quickly if unstable - for daily use I would test with 5 applications over a week):

First: hundreds of errors in TestMem5 (w/ anta777 Extreme1) within 10 minutes
  • XMP enabled (3600 - CL18 - 1.35V)
  • SoC 1.1V
  • GDM disabled
  • 1T
  • AddrCmdSetup of 56
Second: zero errors in TestMem5 (w/ anta777 Extreme1) within 10 minutes
  • XMP enabled (3600 - CL18 - 1.35V)
  • SoC 1.1V
  • GDM disabled
  • 2T
Aight so it's unstable. However, it booted 1T with AddrCmdSetup 56 so that is already a small win. Now try again with as an only change vDIMM 1.45v in stead of 1.35v. See how it responds. If it still errors the same way and amount it wasn't vDIMM holding it back and must be something else like AddrCmdSetup must be 58 or 60, wrong RTT's, wrong resistances..
 
  • Rep+
Reactions: Piers

·
Registered
5900X, RTX 3080
Joined
·
854 Posts
Discussion Starter · #13 ·
Aight so it's unstable. However, it booted 1T with AddrCmdSetup 56 so that is already a small win. Now try again with as an only change vDIMM 1.45v in stead of 1.35v. See how it responds. If it still errors the same way and amount it wasn't vDIMM holding it back and must be something else like AddrCmdSetup must be 58 or 60, wrong RTT's, wrong resistances..
What does the AddrCmdSetup value do in relation to stabilising RAM clocks? For example, a value of 56 vs 60.

Will try bumping voltage to 1.45V to see if that helps and will report back. It won't be immediate as I'm using the PC to serve 900GB of files to a work (S)FTP so can't reboot atm.
 

·
Registered
Joined
·
4,690 Posts
What does the AddrCmdSetup value do in relation to stabilising RAM clocks? For example, a value of 56 vs 60.

Will try bumping voltage to 1.45V to see if that helps and will report back. It won't be immediate as I'm using the PC to serve 900GB of files to a work (S)FTP so can't reboot atm.
Honestly I don't know. I just learned this from Veii and some others in the AMD topic and all I know is that it works but not how it works. There was a post yesterday as well from someone running DR B-Die at 3800C13 1T GDM Off stable. He was using 59 for AddrCmdSetup and the other 2 as well.
 
  • Rep+
Reactions: Piers
1 - 14 of 14 Posts
Top