DRAM calculator for Zen 3

[dgianstefani]

Redid the cooling from Alphacool LT240 Eisbaer to 40mm EK 240 copper rad, Optimus Foundation AMD block, D5 Pump/res combo and another ram fan. Took sustained load temps from 72c max to 51c max.

I was able to push the ram a little harder too with the extra ram fan.

 

[juanyunis]

View attachment 183733View attachment 183734

Redid the cooling from Alphacool LT240 Eisbaer to 40mm EK 240 copper rad, Optimus Foundation AMD block, D5 Pump/res combo and another ram fan. Took sustained load temps from 72c max to 51c max.

I was able to push the ram a little harder too with the extra ram fan.

View attachment 183735

Impressive

 

[harm9963]

Hi guys,

I'm new here, I just read the whole thread and I see some really good people with memory settings, and I was wondering if i could get some help with my system.

I'm currently running a Ryzen 5950X with X570 Aorus Master, using PBO Advanced. I'm able to achieve some nice CPU performance, single boost clocks upto 5.1Ghz and all core clocks about 4.45 - 4.5Ghz. However I'm not an expert with memory settings and I think my system could do better. The memory sticks I have are 4x8GB 4000Mhz CL18 Teamgroup T-Force Xcalibur Special Edition and according to Thaipoon Burner they are Samsung B-die A2 bin. This is what I have achieved so far using Ryzen DRAM Calculator running the FAST profile but with gear down mode enabled.

View attachment 182936
View attachment 182937

I have not been able to be stable with 1T command rate and gear down mode disabled, if I enabled gear down or put command rate to 2T it works. What you guys think?
Note: I tried running FCLK at 2000Mhz with 4000Mhz memory but my system doesn't post, I guess is because of the strain for the memory controller with 4 sticks.

Thanks for the help.

1.5 v for sure .

 

[Trusconi]

:-o

Why FLCK is not 1900?

 

[juanyunis]

:-o

Why FLCK is not 1900?

Did you manually change it? You have to go into BIOS and AMD Overclocking menu and change infinity fabric clock to 1900

 

[brandon7171]

:-o

Why FLCK is not 1900?

I would like to see a user bench mark test on this.. Those numbers look really nice...

 

[Trusconi]

I would like to see a user bench mark test on this.. Those numbers look really nice...

i would like to have fclk equal to mclk before make test...

Did you manually change it? You have to go into BIOS and AMD Overclocking menu and change infinity fabric clock to 1900

no, i change nothing. Before was 1600, it goes automatic to 1800

 

[juanyunis]

i would like to have fclk equal to mclk before make test...


no, i change nothing. Before was 1600, it goes automatic to 1800

Go to BIOS > AMD Overclocking and change Infinity Fabric Clock (Hopefully this will update UCLK too)

:-o

Why FLCK is not 1900?

I got 90% memory quality, reaching up to 4000Mhz CL16.

 

[Trusconi]

Go to BIOS > AMD Overclocking and change Infinity Fabric Clock (Hopefully this will update UCLK too)


I got 90% memory quality, reaching up to 4000Mhz CL16.

i lost 100/150 points in CB20 with these timings...Maybe i could go 3800c15 or 4000c16?
Is possible i forgot to change somethings?

Thank you

Manu

 

[juanyunis]

i lost 100/150 points in CB20 with these timings...Maybe i could go 3800c15 or 4000c16?
Is possible i forgot to change somethings?

Thank you

Manu

In my case pushing FCLK to 2000Mhz using 4 sticks was impossible, I have seen only 1 person achieving that. I'm currently waiting for gigabyte to release a new BIOS with AGESA 1.1.9.0 (it is supposed to help achieving 2000Mhz FCLK) just to give it a try. However some users that already have that AGESA said that it brought some memory problems for them. So if you go 4000C16 you won't have a 1:1 ratio unless you are lucky enough to have 2000Mhz FCLK (if you do try it and it fails to post you might have to clear CMOS depending on the quality of the board, mine has dual BIOS).

Also for 3800CL15, you are using gear down mode (so even if you put the CAS to 15, the system will increase it to 16, gear down mode doesn't support odd numbers) you could try turning off gear down mode.

I think before you try any other timings or clock speed, try turning off gear down mode, if it works it should boost your memory speed. (Memory command rate is set to 1T which means that memory clock will run at 1900Mhz, using GDM the system will reduce internally the clock in half to provide stability, so a system with GDM is considered a 1.5T command rate)

 

[Trusconi]

In my case pushing FCLK to 2000Mhz using 4 sticks was impossible, I have seen only 1 person achieving that. I'm currently waiting for gigabyte to release a new BIOS with AGESA 1.1.9.0 (it is supposed to help achieving 2000Mhz FCLK) just to give it a try. However some users that already have that AGESA said that it brought some memory problems for them. So if you go 4000C16 you won't have a 1:1 ratio unless you are lucky enough to have 2000Mhz FCLK (if you do try it and it fails to post you might have to clear CMOS depending on the quality of the board, mine has dual BIOS).

Also for 3800CL15, you are using gear down mode (so even if you put the CAS to 15, the system will increase it to 16, gear down mode doesn't support odd numbers) you could try turning off gear down mode.

I think before you try any other timings or clock speed, try turning off gear down mode, if it works it should boost your memory speed. (Memory command rate is set to 1T which means that memory clock will run at 1900Mhz, using GDM the system will reduce internally the clock in half to provide stability, so a system with GDM is considered a 1.5T command rate)

If i turn off GDM it does not boot, maybe is better to try 3600c14?
These are the temp after 90 min of BFV with XMP profile

 

[juanyunis]

If i turn off GDM it does not boot, maybe is better to try 3600c14?
These are the temp after 90 min of BFV with XMP profile

Try it. And compare scores.

 

[Trusconi]

Brief update. I further knocked down some secondary and tertiary timings. Results below:
View attachment 174910View attachment 174911View attachment 174912

I am gonna test it first at 1.43V DRAM . 1.44V should work for sure.

This is starting to be an obsession for me so I am gonna have to quit fast before it consumes even more of my time and energy.

Interestingly enough, RAM OC has nearly no effect on Cinebench R20 or Passmark CPU bench compared to stock 3200CL14. Maybe something is bugged right now, given those 4stick setups are showing much better performance.

I thought I would post sources I found helpful.

OC Guide:

MemTestHelper/DDR4 OC Guide.md at master · integralfx/MemTestHelper

C# WPF to automate HCI MemTest. Contribute to integralfx/MemTestHelper development by creating an account on GitHub.

github.com

Timings explained:

RAM timings knowledge base.

DDR3 Memory Timings Explained | MSI HQ User-to-User FAQ

<p>An explanation of DDR3 and memory timings.</p>

forum-en.msi.com

DDR4 SDRAM - Understanding Timing Parameters - systemverilog.io

Explanation to several of the key timing parameters in DDR4 memory like CL, CWL, tREFI, tRFC, tDQSS, tCCD_S, tCCD_L and many more

www.systemverilog.io

Visualization:
https://users.ece.cmu.edu/~koopman/lectures/ece348/08_bus_memory_handouts.pdf Page 18 and further.

Book (1k pages):

Cache and Memory Hierarchy Design

An authoritative book for hardware and software designers. Caches are by far the simplest and most effective mechanism for improving computer performance. This innovative book exposes the characteristics of performance-optimal single and multi-level cache hierarchies by approaching the cache...

books.google.cz

Available via Scribd subscription: https://www.scribd.com/book/282518929/Memory-Systems-Cache-DRAM-Disk

I try with settings...but it do not boot :-(

 

[juanyunis]

I was only able to run with gdm off at 3400 CL14 but according to AIDA I lose memory bandwidth and the latency increased. So I reverted to 3800 CL16 with gdm.

I will still hunt for 3800 CL14 and wait for future AGESA to try 4000CL16 with FCLK at 2000Mhz.

 

[Zach_01]

If you like AIDA high scores you can hunt better settings. But those high-er scores do any actual good to the system?
I mean in real life apps and games...

 

[juanyunis]

@Zach_01 that is what I would like to know. I don't really care about cinebench scores or cpu z but having lower l3 cache speeds does affect any real life apps/games?

Found this in reddit. I guess that is what it makes my cinebench scores higher. But doesn't really matter that much. I think.

The extra L3 cache will make a difference in memory intensive applications that are using multiple cores. Video encoding and intense gaming applications that are fully ramping up your CPU will benefit from the extra 2MB. However, this difference is going to be small. The big difference comes from having L3 cache vs not having L3 cache.

 

[Zach_01]

Specifically Cinebench does not care about RAM performance. Its raw CPU power mostly that increases scores.
My worst and my best DRAM settings with 10~15% difference in memory bandwidth and 5ns for latency had zero difference on CB R15/20.
On game FPS (min/max/avg) from 1~3%.
If I could "play" with my CPU like you all do here, maybe I could gain some more? How much? Can't really say..

It all comes down to experiments mostly and not gaining actual worthy performance increase. Who really cares about the 2, 3... 10 or even 15 more FPS when you already have 60, 100 or 150?
What I'm saying is... That it's good to experiment and try a lot of different settings and tune the system, but its not really worth risking a part, CPU mostly, or maybe a GPU that costs a small fortune.
Its fun to do, if anyone likes to fiddle, but doing it for actual performance... it gets you in dangerous paths.

 

[JGalland]

What latency (and read/write speed) should I be aiming for with this kit : https://www.gskill.com/specification/165/166/1562839659/F4-3600C16D-32GTZR-Specification and some manual tuning of some of the timings ?
Also whats the best software to measure the above values ?

 

[dgianstefani]

Specifically Cinebench does not care about RAM performance. Its raw CPU power mostly that increases scores.
My worst and my best DRAM settings with 10~15% difference in memory bandwidth and 5ns for latency had zero difference on CB R15/20.
On game FPS (min/max/avg) from 1~3%.
If I could "play" with my CPU like you all do here, maybe I could gain some more? How much? Can't really say..

It all comes down to experiments mostly and not gaining actual worthy performance increase. Who really cares about the 2, 3... 10 or even 15 more FPS when you already have 60, 100 or 150?
What I'm saying is... That it's good to experiment and try a lot of different settings and tune the system, but its not really worth risking a part, CPU mostly, or maybe a GPU that costs a small fortune.
Its fun to do, if anyone likes to fiddle, but doing it for actual performance... it gets you in dangerous paths.

I have a 240hz 1440p Samsung G7, so FPS gains of 5% are very relevant.

 

[juanyunis]

I just found out that my IF1900 was giving me some random WHEAs (CPU bus/interconnect), everything related to the SoC in BIOS was set in auto, so I started tweaking it a little bit and was able to find correct settings that don't produce WHEAs and is stable enough, I ran Karhu RAMTest for about 4 hours and 16 mins with a coverage of 8000%, 0 errors. I also ran OCCT memory test with 0 errors.

I think my IF at 1900 is solid now. Next step for me will to try better timings if possible, any help is welcome.

 

[Makaveli]

This is a great thread.

Currently running at 3200 with Fast profile applied and getting this.

Last night I tried this and benching is fine but seeing some instability after more testing any recommendations?

 

[dgianstefani]

If you have B die you should be able to do 3600/3800 with tight timings. Even if you have to loosen a couple timings you'll still have better latency and performance.

Also B die scales with voltage and temperature. Keep it under 40c and bump voltage anywhere up to 1.7v daily to unlock tighter timings.

Personally I run 4000/14-16-14-28 on a 3800/14 kit.

 

[Makaveli]

If you have B die you should be able to do 3600/3800 with tight timings. Even if you have to loosen a couple timings you'll still have better latency and performance.

Also B die scales with voltage and temperature. Keep it under 40c and bump voltage anywhere up to 1.7v daily to unlock tighter timings.

Personally I run 4000/14-16-14-28 on a 3800/14 kit.

Is it B-die 3200/14

 

[dgianstefani]

maybe not.

 

[tabascosauz]

View attachment 186445maybe not.

Bdie finder is a resource from like goddamn 2017...it hasn't been updated like it says it has, it's pretty much dogshit at finding new kits, but apparently it's great for identifying older kits that are no longer Bdie but it claims they are, or haven't been made for the past 2 years.

3200/14 doesn't come in anything but [average] Bdie. It's just not balls-to-the-wall potential and expensive like your 3800/14 kit.