• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Is this why Ryzen is having problems with higher RAM frequencies?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

trents

Senior Member
Joined
Dec 27, 2008
Saw this in a NewEgg review:

"The issue is the memory clock and Infinity Fabric that connects the two CCX units in the CPU are tied together and both run at 1T command rate - impossible to change - whereas most RAM is specified for 2T command rate."

Anything to it?
 

Woomack

Benching Team Leader
Joined
Jan 2, 2005
RAM is specified to run at CR 2/3 because of compatibility with memory controllers on various platforms. There is no DDR4 memory which has CR1 in SPD/XMP.
Ryzen memory controller runs at CR1 and there are no issues with memory kits because of that. The only issue is that memory controller in Ryzen can't handle higher frequencies when memory capacity is higher or when memory has more ranks than 1. AMD said that clear but users don't read specification and later wonder why their 64GB 3200 kit can't pass 2133.

Single rank 2 modules will make ~3200 ( at auto probably something like 2400-2666 )
Single rank 4 modules will make ~2666 ( at auto probably something like 1866-2133 )
Dual rank 2 modules will make ~2666 ( at auto probably something like 1866 )
Dual rank 4 modules will make ~2133 ( at auto probably something like 1866 )

In my tests on ASUS X370-Pro max was like:
2x8GB single rank = 3200
4x8GB single rank = 2666
2x16GB dual rank = 2666
 
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
RAM is specified to run at CR 2/3 because of compatibility with memory controllers on various platforms. There is no DDR4 memory which has CR1 in SPD/XMP.
Ryzen memory controller runs at CR1 and there are no issues with memory kits because of that. The only issue is that memory controller in Ryzen can't handle higher frequencies when memory capacity is higher or when memory has more ranks than 1. AMD said that clear but users don't read specification and later wonder why their 64GB 3200 kit can't pass 2133.

Single rank 2 modules will make ~3200 ( at auto probably something like 2400-2666 )
Single rank 4 modules will make ~2666 ( at auto probably something like 1866-2133 )
Dual rank 2 modules will make ~2666 ( at auto probably something like 1866 )
Dual rank 4 modules will make ~2133 ( at auto probably something like 1866 )

In my tests on ASUS X370-Pro max was like:
2x8GB single rank = 3200
4x8GB single rank = 2666
2x16GB dual rank = 2666

"At auto"? You mean not using the XMP setting in bios?
 

Johan45

Benching Team Leader Super Moderator
Joined
Dec 19, 2012
"At auto"? You mean not using the XMP setting in bios?

I can't use XMP in any of my testing. I have 3 kits and all of them are too fast so it leaves me only manual settings. I find 32-3300 is pretty easy for my CPUs. I have had it over 3600 but it was too unstable for me.

aida 3300 CL14.JPG
 

Bluefalcon13

Member
Joined
Jan 1, 2008
Location
Colorado Springs, CO
I can use XMP on my kit, but it is only a 3000 kit. I do have to fiddle with voltages a bit and it seems to have stopped my 'overclock failed' errors on boot, for the most part. Sometimes my system get a hair up it's rear but it's fairly infrequently. I am honestly considering running auto on my ram (2133) until the May update cause it is impacting being able to remotely startup my rig if there is a power outage, which tanks my PPD for that day.
 

Johan45

Benching Team Leader Super Moderator
Joined
Dec 19, 2012
There's some new BIOS for CHVI surfacing right now with the new AGESA code. One of them also added support for a wide array of ram kits. So you may get relief sooner than May

Test BIOS 0082

Same as 0081 but with a DRAM compability patch for the below part numbers

CMK16GX4M4A2133C13
AHU08GGB13CGT7G
PV416G240C5QK
F4-2400C15Q-32GRR
TCD44G2400C14BK
F4-2666C16Q2-64GRB
AVD4UZ126661504G
BLT8G4D26AFTA.16FAD
IMA41GU6MFR8N-C F0
MD16GK4D4266615AXR
HX430C15PB2K4/16
HX430C15PBK4/32
AX4U3333W4G16-QGZ
GAM4DBLBM2133D15IE041C
TC48G24S817
SP004GBLFU213N02
78.C1GM3.AF10B
F4-3200C16D-8GVKB

MTA4ATF51264AZ-2G6B1
MTA8ATF1G64AZ-2G3B1
MTA16ATF2G64AZ-2G6B1
HMA41GU7AFR8N-TF T0
HMA451U7AFR8N-TF T0
HMA41GU6AFR8N-UH N0
M378A2K43BB1-CTD
M378A1K43BB1-CRC
M378A1G43EB1-CRC
 

ratbuddy

Member
Joined
Aug 24, 2007
RAM is specified to run at CR 2/3 because of compatibility with memory controllers on various platforms. There is no DDR4 memory which has CR1 in SPD/XMP.
Ryzen memory controller runs at CR1 and there are no issues with memory kits because of that. The only issue is that memory controller in Ryzen can't handle higher frequencies when memory capacity is higher or when memory has more ranks than 1. AMD said that clear but users don't read specification and later wonder why their 64GB 3200 kit can't pass 2133.

Single rank 2 modules will make ~3200 ( at auto probably something like 2400-2666 )
Single rank 4 modules will make ~2666 ( at auto probably something like 1866-2133 )
Dual rank 2 modules will make ~2666 ( at auto probably something like 1866 )
Dual rank 4 modules will make ~2133 ( at auto probably something like 1866 )

In my tests on ASUS X370-Pro max was like:
2x8GB single rank = 3200
4x8GB single rank = 2666
2x16GB dual rank = 2666

Same mobo, I run 4x16GB dual rank. Auto chose 2133, and I was able to set it to 2400 and passed overnight memtest just fine. I guess it's a YMMV thing.

edit: Oh, I couldn't even find the XMP selection in BIOS, but I found the dropdown to change auto to 2400 and that seemed to work.
 

ATMINSIDE

Sim Racing Aficionado Co-Owner
Joined
Jun 28, 2012
I have had no issues with DDR4-3200 on multiple board and CPU combinations in 2x8GB dual channel.
 

Woomack

Benching Team Leader
Joined
Jan 2, 2005
On some motherboards it's called XMP, on some DOCP. In theory XMP is Intel profile technology so it can't be called the same on AMD but exactly the same profiles are usually working on both platforms. If memory is not working at any profiles then memory has incompatible profile ( usually single timings are too tight/loose ). On Ryzen you can't correct any sub timings so then you have to set auto/manual settings and adjust what is possible manually. Every memory has timing table based on SPD profile. It's usually ( or at least should be ) compatible with JEDEC. Usually sub timings in SPD profile are more relaxed but often 2133 SPD timings can be set manually at 3600+ and will work too ( on Ryzen can't be set as there are no sub timings options in BIOS or are not working ).
BIOS is not changing much. It adds compatibility to memory kits which have some weird profiles or tight sub timings. However I can't see any difference in memory capacity support. On two motherboards ( 3-4 BIOS releases each ) were the same results on 2x8, 4x8 and 2x16GB.
 

Woomack

Benching Team Leader
Joined
Jan 2, 2005
There are only couple of timings. Programable is what you see in AIDA64 SPD table so maybe 12-13 timings. Full list of timings depends on platform but is much longer. Still Command Rate depends on platform, memory is only suggesting it should be 2/3. The same some main timings on some platforms never run as in XMP. Sometimes what you see in software or in BIOS isn't real. I mean that you set CL11 but it's running at CL12 in real. This is on Ryzen too. CL11/13 go one digit up so in real run as CL12/14. However at CL15 run as CL15. There are also other things like some timings affect real value of other timings.
SPD is typical setting which is usually compatible with JEDEC. Exception are memory kits like Kingston Fury/Impact where SPD=XMP so you don't have to enable XMP and can run memory above standard frequency on motherboards without memory settings in BIOS ( mainly laptops ).

SPD has nothing to do with XMP. These are 2 separated profiles. You could call them SPD1 and SPD2 and it wouldn't change much. In real what matters is timing table which specification depends on used IC. It's still compatible with JEDEC settings so if you won't use XMP then raising memory frequency will cause much more relaxed timings at auto settings.

and really ... again wikipedia, there are better sources that that
 

gupsterg

Registered
Joined
Feb 25, 2017
Location
Lurking over a keyboard
I had 2x R7 1700, same country but differing batches, 1st UA 1706PGT vs 2nd UA 1709PGT. So UEFI defaults with x264 loading CPU, 3.2GHz ACB shows VCORE: ~1.089V SOC: ~0.838V vs VCORE: ~1.119V SOC: ~0.893V. Both OC to 3.8GHz ACB with +137mV vs +162mV.

UEFI v1002 was used on C6H with F4-3200C14D-16GTZ, same all other hardware.

1st only capable of using strap 2933MHz 14-14-14-34-1T, 1.35V VBOOT & VDIMM, SOC is 0.900V in UEFI (~0.901V on DMM). Increasing VCORE to +181mV and SOC to 1.050V yields no 3200MHz, may that be [Auto] timings or manually set. I get Q-Code 15 and sometimes 15 turns to 0d, the Q-LED for RAM is also on at that time.

2nd is using strap 3200MHz 14-14-14-34-1T, 1.35V VBOOT & VDIMM, SOC is 0.950V in UEFI (~0.966V on DMM).

As I no longer have the 1st CPU I can't test it with the new UEFI for C6H which has AGESA 1004a. From what other members on another forum have experienced some are still struggling to gain 3200MHz on single sided 2x8GB configs. I have also noted several posts by members where they do the same as I did above and they gain higher speed RAM.

Either some IMC/DF due to immature FW just "hiccup" with high speed RAM or there is variability between the CPUs.

Even "X" CPU does not mean guaranteed high speed RAM IMO from posts I have read, one such example link.
 
Last edited:

Johan45

Benching Team Leader Super Moderator
Joined
Dec 19, 2012
Welcome aboard mate, at least here you can keep up to the threads.
 

wingman99

Member
Joined
Dec 10, 2003
There are only couple of timings. Programable is what you see in AIDA64 SPD table so maybe 12-13 timings. Full list of timings depends on platform but is much longer. Still Command Rate depends on platform, memory is only suggesting it should be 2/3. The same some main timings on some platforms never run as in XMP. Sometimes what you see in software or in BIOS isn't real. I mean that you set CL11 but it's running at CL12 in real. This is on Ryzen too. CL11/13 go one digit up so in real run as CL12/14. However at CL15 run as CL15. There are also other things like some timings affect real value of other timings.
SPD is typical setting which is usually compatible with JEDEC. Exception are memory kits like Kingston Fury/Impact where SPD=XMP so you don't have to enable XMP and can run memory above standard frequency on motherboards without memory settings in BIOS ( mainly laptops ).

SPD has nothing to do with XMP. These are 2 separated profiles. You could call them SPD1 and SPD2 and it wouldn't change much. In real what matters is timing table which specification depends on used IC. It's still compatible with JEDEC settings so if you won't use XMP then raising memory frequency will cause much more relaxed timings at auto settings.

and really ... again wikipedia, there are better sources that that

Serial presence detect (SPD) is a standardized way to automatically access information about a memory module. Earlier 72-pin SIMMs included five pins that provided five bits of parallel presence detect (PPD) data, but the 168-pin DIMM standard changed to a serial presence detect to encode much more information.

SPD is specified by JEDEC. SPD holds all the memory data including JEDEC timings, subtimings and MXP timings, subtimings. When the PC boots up the BIOS access all the SPD memory data it needs, including XMP data, if Bios is set to use that data.

JEDEC (Joint Electron Device Engineering Council)
SPD (serial presence detect), it is all the memory data in a chip on the memory stick.
XMP A similar, Intel-developed JEDEC SPD extension. This uses bytes 176–255, which are unallocated by JEDEC, to encode higher-performance memory timings
https://en.wikipedia.org/wiki/Serial_presence_detect#XMP

Woomack where do you acquire your information from? Links please.
 

Johan45

Benching Team Leader Super Moderator
Joined
Dec 19, 2012
I can say the new BIOS releases for the CHVI with updated AGESA and memory work has helped. I managed 36xx Cl14 last night fairly easily The extra speed, higher FSB and tight timings did require a few extra voltage tweaks to stabilize it.
VDIMM 1.6v
VTT 0.9v
VDPP 1.2v
PLL 2.0v

3600 cl14.JPG

Gave a pretty good bump to the physics score. One thing I have to mention is I'm not sure if it's the whole platform or just using an NVMe drive but my card is giving me all kinds of issue. Won't recognize it as ASUS so GPUTweak won't open Wattman says it's overclocking but GPUz is calling it a liar. Will have to try an SSD and see if that helps.


3d phy 22673.JPG