5600X, Aorus Extreme, 4,000 MHz RAM works with two but not one or four sticks

Tanquen

[H]ard|Gawd
Joined
Feb 15, 2005
Messages
1,226
When I put one stick in it defaults to the 1333 MHz and ignores the XMP profile.

When I put two sticks in it runs at 2,000 megahertz using the XMP profile but it sets the Infinity fabric at 1800 and I don't know how to change that yet. I'd rather have the memory and the Infinity fabric match at 18 or 1900.

When I put all four sticks in it again ignores the XMP profile and runs the memory and fabric clock at 1333 megahertz.

Is there a good guide on the best way to manually configure the RAM settings for the four sticks to run at hopefully 1900 megahertz?
 
IF is under AMD overclocking you just set the MHz you want it to run assuming it can post that high.

A good place to start would be to use Thaiphoon burner to find out what chips are on your RAM and see if settings suggested by Ryzen DRAM calc can help stability.
 
I found a Corsair guide for a Aorus X570 and can get past 1333MHz but only to 1667. The Fabric Clock is at 1800. So is it better to leave them out of sync or make them match?

The guide say that with 4 stick that are 1 Rank 4000MHz you should get 3600MHz and if 2 Rank then only 2933MHz.

If two sticks can run at 4000MHz I was hoping 4 could do 3800MHz. :(

Looks like the chips are Hynix but it don't say if MFR or AFR or CJR and so on.
Looks like it M-Die but I don't know what to set the DRAM calc at for it.
 
Last edited:
With 2 sticks I can set the get the RAM and FCLK at 1900 but the game benchmark I'm running had a lower score than with both at 1800. ???
 
Cinebench

Score RAM FClk CPU
4682 2000 1800 4.7 OC
4672 1800 1800 4.7 OC
4634 1600 1800 4.7 OC
4578 1800 1800 4.6 OC
4336 1900 1900 4.6 OC
4335 1800 1800 Stock
4311 2000 1800 Stock
4069 2000 1900 Stock

I also did some games and they mostly followed the above. I wonder if I can tighten RAM timings when not running the KIT at 2000?

Cinebench doesn't care about RAM speed just CPU speed so if increasing RAM speed is dropping performance then your clock speed is likely dropping due to the MB setting a higher SOC\VDDG and thermal or power throttling.
Watch Ryzen master while it runs so you can see temps and power limits at 1800 vs 1900 and check for any v change.
The BIOS are immature though so it is possible something else is going on.
How are does your single threaded Cinebench respond?

Most games will be GPU bottlnecked so if you could run AIDA64 and 3DMark TimeSpy and show the scores along with a screenshot of ZenTimings & HWiNFO64 (sensors only) then I may be able to see what is going on.
 
Thanks for the response.

I know you say Cinebench doesn't care about RAM speed but the score changes by enough to tell that if I lower the RAM speed it gets worse, if I raise the RAM speed it gets better until I make the RAM and the FClk 1900. And again the games follow, if I lower the RAM clock my frame rate drops and if I raise it they go up unless I use 1900 for the RAM and f clock. The game benchmarks I'm running are at 1950x1080 so the (RTX3080) GPU's not really a bottleneck as much. Even when the GPU is a bottleneck, faster RAM and CPU can squeak out a few more frames per second.

Okay so that's a lot of stuff. when you say clock speed is likely dropping do you mean the CPU clock speed? In Ryzen master the peak speed with a single thread on Cinebench just bounces around from 1300 to 2000 and that doesn't seem right but when the multi-threaded test is done it says I'm getting pretty much the same score as a similar 5600x system listed in the app.

Temp seems kinda high at 50°C? Is that normal for Ryzen? The idle temp is about 34°C.

What version of Aida64 should I run? Extreme or engineer etc.

I hate having to load all these little apps I'm sure I'm going to get a virus. :)

HWiNFO64 has pages and pages with the sensor only status, is there an export file you want or just the first page or something?

Untitled1.png
Untitled2.png
Untitled3.png
 
So you want to run MCLK FCLK and UCLK all at the same speed for best performance.
So drop the RAM to 3800 for 1900 FCLK or 3600 for 1800FCLK.
I will give cinbench a run at some different RAM speeds to show you what difference you should see.
 
Yes, I have set the RAM and FClk to match. Again the worst score is when RAM and FClk are at 1900.

Where is UCLK in Ryzen Master?

Here are the Time Spy runs I just ran:

Time Spy Hardware Speeds

Total GPU CPU RAM FClk CPU

15356 17640 8858 1800 1800 4.7
14957 17228 8562 2000 1800 4.7
14753 17120 8274 1900 1900 4.7
 
Last edited:
Here is 1900-1900-1900

Untitled.jpg


Here is 1800-1800-1800
ZenTimings 1.2.1 11_18_2020 3_07_25 PM.png
 
Last edited:
So timings change a little which would offset the gains from higher speed a bit but there is no obvious reason there for a drop in Cinebench score.

With AIDA64 click tools Cache and memory benchmark.
 
Temp seems kinda high at 50°C? Is that normal for Ryzen?
With a custom loop my 5800X hits over 70c with multithreaded Cinebench and some are having trouble keeping them under 90c on air but the 5600X does run a bit cooler.

when you say clock speed is likely dropping do you mean the CPU clock speed?
That is what I was wondering but it seems not.

As for tweaking RAM timings for the lower frequency. You certainly can do this and end up with higher performance than just setting 4000 1:1 and ignoring timings.
Ryzen DRAM calculator is the best starting point.
https://www.techpowerup.com/download/ryzen-dram-calculator/

And this guide is kinda the bible for RAM tweakers that want to squeeze a little more out.
https://github.com/integralfx/MemTestHelper/blob/master/DDR4 OC Guide.md#finding-the-maximum-frequency
 
I was just going to tweak some of the simpler timings just to bump it up a little bit. Setting the RAM different speeds does change the benchmarks a little but I'm a little disappointed. For the last many years and many builds I'd given up on messing with RAM after reading reviews and whatnot it just made small changes in benchmarks mainly nothing really noticeable. But there was a lot of talk about how Ryzen really needs really fast RAM and then AMD was saying that 4,000 megahertz was the new sweet spot for the 5000 series. But it doesn't seem to be making a huge difference.

I'm also not sure about the forced overclock, it makes a difference (a small difference) in stuff like cinebench but I'm just forcing all the cores to run at what one or two cores would turbo to and I think that's what 99% of the apps would do and I'm wasting a bit of power I think forcing them all to run at 4.6 or 4.7 or 4.8.

I was disappointed in the 3080 as well. I messed around with afterburner for a little bit and I can crank up the memory like a thousand megahertz and max out the little voltage bump and the benchmarks don't really change.

I'm not sure if I should keep this kit or get a smaller one that has better cas latency? I'm tempted to just get the two stick version so I can run it 1800 x 3. (MCLK FCLK and UCLK) but then it's the old, do you get a 3600 kit with better latency or a 400 megahertz kit and the hopes that 1900 or 2000 Infinity fabric becomes a thing but then on top of that mine's slower when I run it 1900 Infinity fabric. But maybe while nothing's crashing there's something not quite right when it's running at 1900 that's causing the slow down. ????

I don't know why can't I just buy the part with the right numbers and stick it in and have it work???! :(
 
With a custom loop my 5800X hits over 70c with multithreaded Cinebench and some are having trouble keeping them under 90c on air but the 5600X does run a bit cooler.


That is what I was wondering but it seems not.

As for tweaking RAM timings for the lower frequency. You certainly can do this and end up with higher performance than just setting 4000 1:1 and ignoring timings.
Ryzen DRAM calculator is the best starting point.
https://www.techpowerup.com/download/ryzen-dram-calculator/

And this guide is kinda the bible for RAM tweakers that want to squeeze a little more out.
https://github.com/integralfx/MemTestHelper/blob/master/DDR4 OC Guide.md#finding-the-maximum-frequency
I don't think I can use the Ryzen DRAM calculator as:
Looks like the chips are Hynix but it don't say if MFR or AFR or CJR and so on.
Looks like its M-Die but I don't know what to set the DRAM calculator at for it. Many say not supported.
 
Those high timings are sure killing your RAM performance.
Try set the first 3 timings suggested by Ryzen DRAM calc and if that works try the rest as there is a lot to be gained by those other settings then look at AIDA64 and 3dmark scores again.
If it doesn't work get ready to reset the BIOS.
tmp.jpg
 
I was able to import the XMP into DRAM Calculator. What settings did you use for, Rank, MTS, number DIMM Modules and so on?

DRAM Calculator for Ryzen™ 1.7.3 by 1usmus 11_19_2020 9_32_12 AM.png
 
I use single rank and A0/B0 then calculated safe as a good starting point.
Rest of the settings are entered are the same as what you plugged in.

TCL 14 is a interesting recommendation.
 
Good start, latency has come down a lot and read speed has improved a bit still plenty of room to move you should be able to get latency under 60.
Dropping tRFC, tFAW, tWR, tCWL and there associated timings should help a lot.
 
I lowered a few more but things got slower. Will not boot if I try and lower tRCDWR. Are all timings, the lower the better?

ZenTimings 1.2.1 11_19_2020 12_16_30 PM.png
AIDA64 Cache & Memory Benchmark  [ TRIAL VERSION ] 11_19_2020 12_17_37 PM.png
 
I lowered a few more but things got slower. Will not boot if I try and lower tRCDWR. Are all timings, the lower the better?
You can go too low with some.

From this guide
https://github.com/integralfx/MemTestHelper/blob/master/DDR4 OC Guide.md#finding-the-maximum-frequency

Minimum tFAW can be is tRRDS * 4.
tRAS = tCL + tRCD(RD) + 2
tRC = tRP + tRAS + x1. Setting tRAS lower than this can incur a performance penalty.
1Your RAM might not be able to do tRP + tRAS, hence the x. Setting x to 8, i.e. tRP + tRAS + 8 should be a pretty safe gamble.

tRCD is split into tRCDRD (read) and tRCDWR (write). Usually, tRCDWR can go lower than tRCDRD, but I haven't noticed any performance improvements from lowering tRCDWR. It's best to keep them the same.

Also AIDA64 is very sensitive to background programs so running windows in Diagnostic mode can realy improve consistency.
 
Last edited:
I guess I'm ok where I'm at. I have a nice bump over running the RAM at stock 1300 and a little better than the XMP profile. Would like to know what is up with having the 1900 FCLK being slower but even if it worked there is not much of a gain.

Thanks for all the help.

Not sure an OC makes a real world difference. It looks like the CPU is stable at 4.7 and the temps don't seem any worse.

As for the RAM I think I need, to just get a 2 stick kit. Having 128GB sounds cool but it's double the price and slower as you need 2 sticks. I thought it would be cool to set up a RAM drive for transcoding but don't really need it.
 
Back
Top