Ryzen 7000 / Zen 4 Owner's Club

[kaliz]

14 hours ago, neurotix said:

Is the TridentZ5 that I'm getting (I convinced her ) dual or single rank?

 

Amazon.com

WWW.AMAZON.COM

 

i couldnt find the part number if you posted it here, but if its 2x16GB or 2x24GB it is single rank, and 2x32gb most likely dual rank. Just as with DDR4 single rank clocks a bit better vs dual rank and has better latency. 

[kaliz]

o yeah those are single rank 2x24GB Hynix M-die modules, very nice stuff!

[kaliz]

i screwed my OS by using 8000C32-34 the whoile day without knowing it, so now sometimes my browser hangs

[neurotix]

20 hours ago, J7SC_Orion said:

 

 per Zen sheets, mine are single rank.

 

After RAM and standard CPU, bclk+, and eclk+ profiles are all set now, I played around a bit with with Process Lasso , something @ENTERPRISE @neurotix  and Co will also surely enjoy soon. It is more complex than Microsoft's  XBox GameBar but ultimately gives better control.

 

edit:   ...did a few quick benchies; all that CPU & memory tuning is worth it:

 

 

 

Nice dude! My best is 17k.

 

https://hwbot.org/submission/5244641_

[J7SC_Orion]

8 hours ago, kaliz said:

i screwed my OS by using 8000C32-34 the whoile day without knowing it, so now sometimes my browser hangs

 

Ouch ! Complete reinstall or 'repair' ? 

[ENTERPRISE]

Still working on this RAM at 6600MHz, I have managed to get the timings down which now means consistently I get around 62ns on the Latency. I am almost finished with the timings now I feel, any lower on the values recommended by @kaliz and I start to get issues. Will post up a Zen Timings screenie when done. 

 

I have not touched the Nitro settings for now, will investigate that once timings are done to see if that can aid in latency reduction.

 

Once finished I will work on lowering the voltages  

 

On another note I did contact MSI about the missing GDM and RTL settings. I am not the only one that has complained about this. I told them that for an enthusiast board I would expect these settings to be available. I have not heard back yet but I imagine it will likely not be the response I am after.

 

[neurotix]

Hey guys, I got my kit of TridentZ5 with single rank Hynix M-Die today. I'm having problems where it's unstable (in memtestCL in the old Ryzen DRAM Calculator), I am going to try applying more voltage and see if it passes.

 

Unfortunately these are my results, using the settings @J7SC_Orion gave me:

 

 

The bandwidth on all the memory readings looks quite low to me for DDR5-8000MHz. These results are barely better than my tightly tuned B-Die at 6200MHz I was previously using.

 

@kaliz I am going to try adding more voltage to see if it will pass memtestCL and then GSAT in Linux, but can you go through and mark timings and make them a little looser to see if that gets me stable?

 

Thanks everyone.

[neurotix]

Update:

 

 

I passed memtestCL in Ryzen DRAM Calculator at 7600MHz on 24 threads with 32GB being tested and to 2400% task scope.

 

Next is stressapptest in Linux for 4 hours.

 

I'd still be interested in some looser timings from @kaliz to see if I can run 8000MHz. Unfortunately, 8000MHz and 7800MHz both errored out in this test pretty early on (within 10 minutes).

 

On another note, I use Linux as my main operating system and fold in it as well as have it act as the house server. Unfortunately, when I reboot the RAM sticks lighting turns off then turns back on with rainbow puke that doesn't match my rig's cyan/magenta theme. I guess I'll have to try learning OpenRGB as I saw it was available for Linux, I also used it on Windows and unfortunately it messed up my keyboard's custom lighting profile. So yeah, pretty unhappy that the lighting profile doesn't get saved into the sticks profile or whatever and persist across reboots. I'll figure it out though.

[J7SC_Orion]

1 hour ago, neurotix said:

Hey guys, I got my kit of TridentZ5 with single rank Hynix M-Die today. I'm having problems where it's unstable (in memtestCL in the old Ryzen DRAM Calculator), I am going to try applying more voltage and see if it passes.

 

Unfortunately these are my results, using the settings @J7SC_Orion gave me:

 

 

The bandwidth on all the memory readings looks quite low to me for DDR5-8000MHz. These results are barely better than my tightly tuned B-Die at 6200MHz I was previously using.

 

@kaliz I am going to try adding more voltage to see if it will pass memtestCL and then GSAT in Linux, but can you go through and mark timings and make them a little looser to see if that gets me stable?

 

Thanks everyone.

 

...Keeping in mind that we have different boards and also that both CPU and RAM kits can have some variation, I would start with the less-tight settings shown on the left > here   ...start with the stock RAM speeds and primaries per XMP setting, then work your way up. Also make sure that your FCLK is rock-solid; I typically lock that in after RAM tuning is done

[neurotix]

5 minutes ago, J7SC_Orion said:

 

...Keeping in mind that we have different boards and also that both CPU and RAM kits can have some variation, I would start with the less-tight settings shown on the left > here   ...start with the stock RAM speeds and primaries per XMP setting, then work your way up. Also make sure that your FCLK is rock-solid; I typically lock that in after RAM tuning is done

 

Thank you. I know 100% that my fclk is rock solid at 2100MHz. The looser timings helps too, but they don't seem that different from the others excepting the tRFCs. I'll give it a shot but I doubt it will be stable. I'll eat my words if I'm wrong, but it seems like my board and these DIMMs like 7600MHz and looking at the two AIDA's I just posted, there isn't really that much of a difference besides slightly higher latency and it seems like the bandwidth readings are actually higher in AIDA at 7600MHz vs the 8000MHz I posted.

 

Thank you again though. I'll try giving the looser timings a shot but I doubt they'll be stable at 8000.

[neurotix]

As I suspected, the looser timings @ 8000MHz failed memtestCL with tons of errors within 10 minutes. I think for 8000, I'm going to need timings that are way looser, but it may be better to stick with the tight timings @ 7600MHz. About to start GSAT on it now.

[J7SC_Orion]

5 hours ago, neurotix said:

As I suspected, the looser timings @ 8000MHz failed memtestCL with tons of errors within 10 minutes. I think for 8000, I'm going to need timings that are way looser, but it may be better to stick with the tight timings @ 7600MHz. About to start GSAT on it now.

 

What about 7800 MHz on your board, noting that the Aorus boards are usually among the better ones for RAM oc'ing. In any event, 7800 MHz was fairly easily locked in with tight timings. I only went up to 8000 after I 'learned' the particulars about my mobo, CPU and specific RAM at 7800. Also, depending on your board, bclk can also get you there, but again subject to A LOT of testing.

 

FYI, another thing to remember per earlier posts in this thread is that V_SOC adjustments work in opposite directions for higher FCLK vs higher RAM speed; you might want to try 2033 fclk and see what that brings re. top RAM speed.

[neurotix]

8 hours ago, J7SC_Orion said:

 

What about 7800 MHz on your board, noting that the Aorus boards are usually among the better ones for RAM oc'ing. In any event, 7800 MHz was fairly easily locked in with tight timings. I only went up to 8000 after I 'learned' the particulars about my mobo, CPU and specific RAM at 7800. Also, depending on your board, bclk can also get you there, but again subject to A LOT of testing.

 

FYI, another thing to remember per earlier posts in this thread is that V_SOC adjustments work in opposite directions for higher FCLK vs higher RAM speed; you might want to try 2033 fclk and see what that brings re. top RAM speed.

 

Appreciated. Repped.

 

I think the gains are so marginal, if any at all (according to AIDA anyway), on my setup that I am going to leave it at 7600 with the 'tight' timings. As shown for whatever reason my bandwidth in AIDA was low for 8000 and 7600 nearly matched or possibly outperformed it though I'd need to check them side by side.

 

7600/2100 fclk is stable in memtestCL to 2400% task scope on 24 threads, as well as 3 hours GSAT stable in Linux.

 

I think to get 7800 or 8000 stable the timings would need to be increased significantly causing higher latency and thus, 7600 with your set of tight timings would outperform it.

 

Thank you so much for all your advice and help.

[ENTERPRISE]

Hey guys, 

 

Just working on voltages. I have SOC down to 1.25. Oddly, if I deviate from 1.5v for VDDIO and VDDQ, System wont post, so they will remain as they are. I was expecting to get those a little lower, but it is what it is.

 

Manually set VDDP to 1.05v Will confirm stability and then post final results. 

 

Out of interest what is a sensible manual VDDG voltage ?

 

On another note MSI got back to me regarding the missing GDM and RTL settings in the BIOS and said they will investigate further with their technical team, will keep fingers crossed for positive outcome.

[neurotix]

59 minutes ago, ENTERPRISE said:

Hey guys, 

 

Just working on voltages. I have SOC down to 1.25. Oddly, if I deviate from 1.5v for VDDIO and VDDQ, System wont post, so they will remain as they are. I was expecting to get those a little lower, but it is what it is.

 

Manually set VDDP to 1.05v Will confirm stability and then post final results. 

 

Out of interest what is a sensible manual VDDG voltage ?

 

On another note MSI got back to me regarding the missing GDM and RTL settings in the BIOS and said they will investigate further with their technical team, will keep fingers crossed for positive outcome.

 

On Zen 3, they recommended (The_Stilt) that VDDG voltages should be 0.40v lower than VDDP.

 

On Zen 4, I have not touched VDDG but I believe @kaliz suggested keeping it below 1v. I think on Auto on my board both VDDG_IOD and VDDG_CCD are around 0.900v. I haven't touched VDDP either but I think the 0.40v under VDDP is no longer the rule.

 

You may be able to get your IF stable at 2100MHz if you look into VDDG and add voltage but I'd keep it under 1v. I know in AIDA64 Memory Read on my setup it is like 3000GB/sec higher with Fclk at 2100 versus 2000.

 

Good luck, I hope this helps. I will look into VDDG on Zen 4 further for you.

[ENTERPRISE]

16 minutes ago, neurotix said:

 

On Zen 3, they recommended (The_Stilt) that VDDG voltages should be 0.40v lower than VDDP.

 

On Zen 4, I have not touched VDDG but I believe @kaliz suggested keeping it below 1v. I think on Auto on my board both VDDG_IOD and VDDG_CCD are around 0.900v. I haven't touched VDDP either but I think the 0.40v under VDDP is no longer the rule.

 

You may be able to get your IF stable at 2100MHz if you look into VDDG and add voltage but I'd keep it under 1v. I know in AIDA64 Memory Read on my setup it is like 3000GB/sec higher with Fclk at 2100 versus 2000.

 

Good luck, I hope this helps. I will look into VDDG on Zen 4 further for you.

Thanks for the info and history lesson bud. I actually have 2100 FCLK stable currently with VDDG set to auto but I wondered if...BIG if I could push to 2133 by increasing VDDG. Will have to look into it.  

 

Let me know if you find any juicy info

[neurotix]

[J7SC_Orion]

...don't despair, gentlemen - setting up new memory usually is a big chase down a deep rabbit hole - it can get frustrating....but also be fun in a weird sort of way 

 

FYI, once I had locked in my daily setting shown before, I went back and slightly lowered several memory-related voltages (which weren't high to begin with) and also dialed back both LLCs to medium (fyi, more aggressive LLCs can help with RAM oc sometimes). This meant going through the short and long test suite again but in any case, the newer SK Hynix M-die 2x24GB kit really likes the higher speeds, up to CL36 8100+, though it will also do CL30 6000 at stock voltage (didn't even try a lower CL). 

[ENTERPRISE]

Running stability test with VDDG CCD & VDDG IO at 0.950v to check all is good, may play with these to see if I can achieve 2133 Fclk but its a big IF still.

[neurotix]

45 minutes ago, ENTERPRISE said:

Running stability test with VDDG CCD & VDDG IO at 0.950v to check all is good, may play with these to see if I can achieve 2133 Fclk but its a big IF still.

I looked into it and it should be ok and possible to run the VDDG voltages at 1.050v or possibly higher. Though maybe kaliz will correct me.

 

Good luck.

[kaliz]

VDDG's auto are 0,850v , they are safe daily up to 1,1v. Ive set mine to VDDG CCD 1,050 and VDDG IOD 0,950 just as i did with Ryzen AM4 and it works for high FCLK speeds! Safe is up to 1,15 but anything over 1,05v is out of spec.

[J7SC_Orion]

21 minutes ago, kaliz said:

VDDG's auto are 0,850v , they are safe daily up to 1,1v. Ive set mine to VDDG CCD 1,050 and VDDG IOD 0,950 just as i did with Ryzen AM4 and it works for high FCLK speeds! Safe is up to 1,15 but anything over 1,05v is out of spec.

 

For DDR5 8000/+, I've set both to 1.0 V daily but don't exceed that even with bench settings at higher bclk / eclk.

[ENTERPRISE]

Quick one chaps. Talking to MSI about the missing GDM and RTL settings and they have asked me to reference mobo models that have these settings accessible on AM5

 

Could you guys give me some of yours and I can go back to them. 

 

Thanks!

[J7SC_Orion]

3 hours ago, ENTERPRISE said:

Quick one chaps. Talking to MSI about the missing GDM and RTL settings and they have asked me to reference mobo models that have these settings accessible on AM5

 

Could you guys give me some of yours and I can go back to them. 

 

Thanks!

 

...my 670E Aorus Master has most of these

[neurotix]

Asus Strix X670E-E Gaming Wifi has them both.

 

Weird that you don't have gear down mode. That's pretty standard (and necessary) for RAM OC on Ryzen.

 

Also @J7SC_Orion you said you set VDDG voltages to 1.0v for 8000mhz. Mine were in Auto when I was trying. Think it might help stabilize me? Also what VDDP are you using?

Edited October 5, 2023 by neurotix