DRAM calculator for Zen 3

[Ampeg]

Hey hoping I can piggy back off this thread.

I've been trying to lower my timings and I'm relatively new to all of this. Can anyone suggest what timings to go after next?

I have the dimm voltage set to 1.45v in BIOS.
Sticks I'm using are Gskill F4-3800C14D-16GTZN.

Thanks in advance, any help/advice is appreciated!

EDIT: I have done 3 successful passes of Memtest86+ with these timings / voltage.

 

[Zach_01]

Hey hoping I can piggy back off this thread.

I've been trying to lower my timings and I'm relatively new to all of this. Can anyone suggest what timings to go after next?

View attachment 174925
View attachment 174928
View attachment 174927

I have the dimm voltage set to 1.45v in BIOS.
Sticks I'm using are Gskill F4-3800C14D-16GTZN.

Thanks in advance, any help/advice is appreciated!

EDIT: I have done 3 successful passes of Memtest86+ with these timings / voltage.

Those are nice sticks!

tRRDS, tFAW, tWR... and also how is your GearDownMode?
A ZenTimings screenshot will be helpful to show all DRAM settings.
tFAW must always be x4~6 of tRRDS.

 

[Ampeg]

Those are nice sticks!

tRRDS, tFAW, tWR... and also how is your GearDownMode?
A ZenTimings screenshot will be helpful to show all DRAM settings.
tFAW must always be x4~6 of tRRDS.

Hey thanks for the response.

I tried to launch ZenTimings and it doesn't pop up. I can see it try to start in task manager, but then it just disappears.

It shows my GearDownMode is disabled.. Is that something I should enable? I'll try and look at that in a bit to see if I can find it.

EDIT: Checked in the BIOS. GearDownMode is greyed out set to [Auto] and I'm unable to change it.

 

[Zach_01]

Hey thanks for the response.

I tried to launch ZenTimings and it doesn't pop up. I can see it try to start in task manager, but then it just disappears.

It shows my GearDownMode is disabled.. Is that something I should enable? I'll try and look at that in a bit to see if I can find it.

EDIT: Checked in the BIOS. GearDownMode is greyed out set to [Auto] and I'm unable to change it.

Don't know whats that with ZenTimings...

GearDownMode when enabled it loosens some timings despite the settings. ("odd" numbers to next "even").
Some BIOSs when you set manually the CR(Cmd) to 1T they gray out GDM. You must first alter CR to auto, adjust GDM and then set CR to 1T again. Know that if now GDM is enabled and you disabled it, it may cause instability, but if its still stable it may decrease some latencies further.

I have it enabled because my sticks not so good...

 

[Ampeg]

Don't know whats that with ZenTimings...

GearDownMode when enabled it loosens some timings despite the settings. ("odd" numbers to next "even").
Some BIOSs when you set manually the CR(Cmd) to 1T they gray out GDM. You must first alter CR to auto, adjust GDM and then set CR to 1T again. Know that if now GDM is enabled and you disabled it, it may cause instability, but if its still stable it may decrease some latencies further.

I have it enabled because my sticks not so good...

Yeah, for some reason ZenTimings isn't running on Zen3 from what I can tell.

Here are my new results and my current timings. (Let me know if you'd like to see any others)

Thanks again for all of your help. It's greatly appreciated.

 

[harm9963]

Hey hoping I can piggy back off this thread.

I've been trying to lower my timings and I'm relatively new to all of this. Can anyone suggest what timings to go after next?

View attachment 174925
View attachment 174928
View attachment 174927

I have the dimm voltage set to 1.45v in BIOS.
Sticks I'm using are Gskill F4-3800C14D-16GTZN.

Thanks in advance, any help/advice is appreciated!

EDIT: I have done 3 successful passes of Memtest86+ with these timings / voltage.

Try MEMbench -default test , report temps please.

 

[Ampeg]

Try MEMbench -default test , report temps please.

Ran the test and had to back off some timings to not throw an error.

Ran through it without an error now, but any idea why my time is so awful? 312s vs 196s?

Is this anything to do possibly with the MSI RAM/BIOS issues they're having?

EDIT: Dimm voltage set to 1.45v in BIOS, shows as 1.472v in HWInfo

Thanks for your help.

 

[thesmokingman]

Are you watching for WHEA errors too?

 

[Zach_01]

Ran the test and had to back off some timings to not throw an error.

Ran through it without an error now, but any idea why my time is so awful? 312s vs 196s?

Is this anything to do possibly with the MSI RAM/BIOS issues they're having?

EDIT: Dimm voltage set to 1.45v in BIOS, shows as 1.472v in HWInfo

Thanks for your help.

View attachment 174982

Dont pay to much attention to "Best result" Nobody catches that one...



PS:
5.05GHz? Nice...!!

 

[harm9963]

Ran the test and had to back off some timings to not throw an error.

Ran through it without an error now, but any idea why my time is so awful? 312s vs 196s?

Is this anything to do possibly with the MSI RAM/BIOS issues they're having?

EDIT: Dimm voltage set to 1.45v in BIOS, shows as 1.472v in HWInfo

Thanks for your help.

View attachment 174982

What are your MB spec, have the latest bios and chipset.

 

[Zach_01]

What are your MB spec, have the latest bios and chipset.

If you're asking what board he has, its MSI X570 Tomahawk

 

[harm9963]

What are your MB spec, have the latest bios and chipset.

Updated AMD AGESA ComboAm4v2PI 1.1.0.0 Patch C , 11-4-2020

MAG X570 TOMAHAWK WIFI | RETURN TO HONOR

Military style with Pre-installed IO shielding, tuned for better performance by Core Boost, DDR4 Boost, M.2 Shield Frozr, Wi-Fi 6, Frozr Heatsink Design, Lightning Gen4

us.msi.com

https://www.amd.com/en/support/chipsets/amd-socket-am4/x570 -
Revision Number
2.10.13.408

TechPowerUp

MemTest64 is a free system memory stability and error detection software for all Intel and AMD processors. It runs on 64 bit Windows without installation.

www.techpowerup.com

 

[northvisit]

I think your membench time 312 looks quite big (and also read speed 42,2 GB/s a bit low). I have slower memory, but only 6-core CPU and smaller time 212 (default). I dont know if results depend at all on core count but I assume it should be the same. This is a memory test. Do you have some other program running in background consuming CPU load? There is some bios issues with MSI (and other as well) that could still be a problem as you suggested but I can not confirm if this is the case. Some complain also multicore(!!) boost clocks. How does CB20 results look? Are these about as expected?

 

[Nekyno]

I have created another thread as I have encountered issues with manual adjustment to SoC and cLDO VDDP, VDDG voltages.

Long story short, when OCing memory I have to set the stock values manually otherwise they go sky high. Once I do, SoC power draw goes 1.6x up. While keeping the same PPT, that means CPU Power runs 3W lower on the stock RAM.

Ryzen 5600x - SoC Power +60% when VDDCR SoC, CLDO VDDP/G are set manually.

When I was tuning memory I noticed that SoC Voltage has increased by a huge margin. I thought about it and went ahead and tested 2 configurations to confirm my hypothesis that manual voltage setting causes SoC Power draw to go up. 1. Stock config, memory set to XMP. SOC - Auto VDDP - Auto VDDG...

www.techpowerup.com

 

[Zach_01]

I think your membench time 312 looks quite big (and also read speed 42,2 GB/s a bit low). I have slower memory, but only 6-core CPU and smaller time 212 (default). I dont know if results depend at all on core count but I assume it should be the same. This is a memory test. Do you have some other program running in background consuming CPU load? There is some bios issues with MSI (and other as well) that could still be a problem as you suggested but I can not confirm if this is the case. Some complain also multicore(!!) boost clocks. How does CB20 results look? Are these about as expected?
View attachment 175001View attachment 175002

Not comparable, his with yours and mine...
Its 12 threads vs 16threads (task scope 180 vs 240)

 

[Ampeg]

I did some more tinkering last night. I can't explain why Aida and memtest read speeds are 10,000 MB/s different?

14-16-16-16-36 @ 1.45v in BIOS.

14-14-14-14-28 @ 1.5v in BIOS

When I leave it on auto and XMP it shows up as 1.52v as well, so I assume it's a safe voltage?
I'm not sure which to leave it on for now.. although I'll probably tinker with it some more at some point especially when MSI fixes their BIOS/RAM issue.

EDIT: Here's my Cinebench R20 result as requested above. (Not sure if this is where it should be compared to others with the 5800X?)

 

[Zach_01]

You cannot compare AIDA64 with MEMbench. Different software, different measuring method. You can only compare different settings on the same software.

 

[tabascosauz]

@Ampeg some things here - your RRDS/RRDL/FAW and WTRS/WTRL/WR are looser than a bucket of fishing worms, tRFC is stupidly high for 1T B-die. At the very least, you should be able to do 4/6/16, 4/12/12 and 350 or so without adding voltage to what you're already using now.

There are some minute differences in performance between different core count SKUs, but membench results are generally roughly comparable to see where your memory performance lies. In this case, 277-300+sec is pretty damn terrible for 3800CL14.

AIDA doesn't care too much about secondaries, but membench certainly does if your timings are that loose.

After you properly spend time on the secondaries, I wouldn't be surprised if this all gets better once a more mature AGESA firmware hits the shelves. Ryzen 5000 does appear to behave a little differently on the memory side than Ryzen 3000, more so than AMD initially let on. From the custom latency of 3ns, it's pretty obvious that DRAM calc hasn't been updated to reflect Zen 3's 8-core CCD. So get your timings in order now, then wait till the software and firmware are updated, then try again.

Here's some old 4Gb E-die that behaves similarly to B-die:



And stop posting your Cinebench results. It's not a memory benchmark or stability test and gains nothing from faster memory.

 

[Ampeg]

@tabascosauz Thanks for the info. Helps a lot!

I re-ran MEMbench with the settings you suggested and it still threw a single error @ 1.5v set in BIOS. (Tested from 1.45v to 1.5v)

Which timings would you suggest I back off on to find out what's stable?
Additionally, do you know if over 40C is too hot? The sticks are rated to run at 1.5v, but I'm not 100% sure if I need them to stay cooler than that.

Screenshot below of the test.

 

[tabascosauz]

@Ampeg for starters, turn Geardown on. It costs you about 3sec on Easy and a bit more on Default, but the stability boost is massive. If you really want the 3 seconds, you can try to pump more voltage or mess with Clkdrv strengths, but I recommend neither at this point in time. There's plenty of room to tighten timings from what they are right now, after you get rid of the instability.

Try that, then run membench. Better yet, download HCI Memtest and run a stability test for about an hour - the free version limits you to 3GB of RAM per instance, so just keep opening instances of HCI until you can fill about all of your RAM save for 2GB or so left for Windows, and test all of the instances concurrently. Like, membench literally includes HCI.exe and uses that in its extremely short and light "stability" test, so you may as well use the actual program instead of a half-assed quickie in membench.

HCI catches Geardown off instability really quickly, so if it ends up being stable in HCI, you'll know what was up with the instability.

40C is cool as a cucumber. Start thinking about more airflow if you hit 50C in HCI.

 

[northvisit]

Looks your results are already much better and about in good values for B-die. I would not be worried about the results. Maybe coming bios/agesa releases will improve RAM performance. Many other people seem to have problems with RAM (with zen 3) and also multicore boosts in MSI motherboards in other discussions (mainly reddit), but our results seem pretty good. I asked CB20 results just to check there is nothing strange with the system (actually poor bios). Your single core clock and also multicore (CB20 result) look just fine and actually good. CB20 test does not test RAM speeds at all but it is one method to validate CPU performance to see whether there is some problem which could affect also in (other) RAM tests.

 

[Ampeg]

@tabascosauz Turning GDM on did the trick! I was able to lower voltage to 1.45v and didn't lose any time compared to my other run with GDM disabled.

Would you recommend still running HCI memtest with my current settings? Or would you say MEMbench is sufficient at this point?

Thanks again for your help!

@northvisit Interesting that you mention the MSI all core boost issue, do you have a link to a thread on that? I noticed that mine is 100-150mhz lower than my friend with a 5800X + Asus board. I had just chalked it up to silicon lottery. Appreciate your help / advice as well!

 

[northvisit]

There is some discussion here;

https://www.reddit.com/r/MSI_Gaming/comments/jp3f47

There is also discussions about RAM issues as well. No idea how common issue these are (or not),

Personally I believe you dont have same issue at least in that extent. Your CB20 score was over 5900 points. Some tests (hardware unboxed etc) show only slightly bigger numbers. In the results (CB20 and membench) all background tasks affect negatively so have you checked active background tasks? Also motherboard vendor settings (default limits etc) and CPU silicon lottery affect as you said. Also cooling have some impacts. At least somewhere was tested golden 5950x which showed better results due lower power consumption. So all 16 cores had higher boost clocks (and temps) than others as the used 142W power limit did not limit the performance. Clocks could be kept higher in 25-32 thread loads. Have you compared something like 12-14 thread clocks? If this is power limiting issue then PBO setting for PPT (or TDP) can help. I have at least those settings in my MSI B450 board in overclocking setting. In my case (chip) I can not get practically any performance improvement so not worth to PBO OC for me. Maybe minor max thread boosts but as said it is minimal. I preferred only set a small (0.075mV) negative offset for lower temps (lower fan noise) and power consumption as well smaller vCore voltages. It just feels a bit better to have below max 1,4V for a CPU even close 1,5 is not a problem in zen2. It is not constant but maximun short term voltage. Manual OC with constant voltages over 1,25-1.35V may be problematic for long tern (silicon decration).

 

[tabascosauz]

@Ampeg good results to see. I'm surprised you didn't lose any on account of Geardown. At this point, 4/6/16 - 4/12/12 offers a respectable performance baseline if you want to leave it there, or if you want to go further the github Memtesthelper guide offers some further info if you want to push for 4/4/16 - 4/10/10 or further on your own. Personally, I'd leave it as it is and enjoy.

Still worth doing HCI overnight or at least to 1000% completion on each instance (percentage in the bottom of the window) to catch intermittent errors, and doing a half hour or an hour of Prime95 LargeFFT to heat things up.

At this point I'm not sure if overnight HCI is worth it, since AGESA is (if reports believed) quite unstable right now. I do 200% if I don't have time, 400% if I want to use it for the rest of the week, 5000% if I want it completely stable.

@northvisit small offset up to 0.075V with LLC (0.05V without LLC) can give you lower load temps while sacrificing little or no multi threaded performance, but can negatively affect single thread scores. Back when I first got my 3700X I used -0.075V and the MT benchmarks looked great, but games suffered. Not to mention HWInfo's "1.4-1.5V" Vcore is not only completely harmless, but highly misleading if you ever take a look at what the CPU is actually doing in idle through Ryzen Master.

 

[Space Lynx]

@Ampeg good results to see. I'm surprised you didn't lose any on account of Geardown. At this point, 4/6/16 - 4/12/12 offers a respectable performance baseline if you want to leave it there, or if you want to go further the github Memtesthelper guide offers some further info if you want to push for 4/4/16 - 4/10/10 or further on your own. Personally, I'd leave it as it is and enjoy.

Still worth doing HCI overnight or at least to 1000% completion on each instance (percentage in the bottom of the window) to catch intermittent errors, and doing a half hour or an hour of Prime95 LargeFFT to heat things up.

At this point I'm not sure if overnight HCI is worth it, since AGESA is (if reports believed) quite unstable right now. I do 200% if I don't have time, 400% if I want to use it for the rest of the week, 5000% if I want it completely stable.

@northvisit small offset up to 0.075V with LLC (0.05V without LLC) can give you lower load temps while sacrificing little or no multi threaded performance, but can negatively affect single thread scores. Back when I first got my 3700X I used -0.075V and the MT benchmarks looked great, but games suffered. Not to mention HWInfo's "1.4-1.5V" Vcore is not only completely harmless, but highly misleading if you ever take a look at what the CPU is actually doing in idle through Ryzen Master.

just do light oc right now - wait for new BIOS AMD already said one is coming.