Ryzen Owners Zen Garden

[tabascosauz]

Yeah, I get it. Overthinking has nothing to do with this, if that was the case then about 90% of comments on computer enthusiasts forums would be classed as that.
If AMD claim the 'sweet spot' for Zen 4 is 6000MT/S & yet if the end user does not touch the FCLK setting & leave it on auto, it boots with 2000 FCLK so, there's your 1:3 ratio right there.
I pointed out even booting with 5600MT/S & FCLK setting on auto, its still boots with 2000 FCLK as I indicated further back in this thread & was wondering why this is, yet a net search reveals many conflicting theories about this phenomena.
Latency can be reduced with higher FCLK setting, but how high depends on the silicon lottery luck with any given SKU.

Seems like a shaky conclusion to correlate the board's 5800 auto rules to the existence of some sort of invisible divider. All boards BIOSes are very different from one another. Unless the 5800 Fabric dip occurs with the majority of boards on the same AGESA, on Hynix/Micron/SS kits, there's just too many variable there to draw conclusions.

 

[dont whant to set it"']

2000MHz IF clock, say ratio 2;
3000MHz RAM clock, say ratio 3(6000MT because Double data rate);
=2 to 3 IF-MEM
Sorted.

 

[AusWolf]

Yeah, I get it. Overthinking has nothing to do with this, if that was the case then about 90% of comments on computer enthusiasts forums would be classed as that.
If AMD claim the 'sweet spot' for Zen 4 is 6000MT/S & yet if the end user does not touch the FCLK setting & leave it on auto, it boots with 2000 FCLK so, there's your 1:3 ratio right there.
I pointed out even booting with 5600MT/S & FCLK setting on auto, its still boots with 2000 FCLK as I indicated further back in this thread & was wondering why this is, yet a net search reveals many conflicting theories about this phenomena.
Latency can be reduced with higher FCLK setting, but how high depends on the silicon lottery luck with any given SKU.

And that's your overthinking right there!

Let me tell you something, Neo...

The IF runs at 2000 MHz. End of.

2000MHz IF clock, say ratio 2;
3000MHz RAM clock, say ratio 3(6000MT because Double data rate);
=2 to 3 IF-MEM
Sorted.

6000 MHz RAM sets UCLK at 3000 and FCLK at 2000.

4800 MHz RAM sets UCLK at 2400 and FCLK at 2000.

So no. See my post above.

 

[Launcestonian]

And that's your overthinking right there!

Let me tell you something, Neo...
View attachment 312124
The IF runs at 2000 MHz. End of.


6000 MHz RAM sets UCLK at 3000 and FCLK at 2000.

4800 MHz RAM sets UCLK at 2400 and FCLK at 2000.

So no. See my post above.

No, 5800MT/S boots with 1933 FCLK on my Gigabyte system with everything on auto & XMP profile enabled.

Seems like a shaky conclusion to correlate the board's 5800 auto rules to the existence of some sort of invisible divider. All boards BIOSes are very different from one another. Unless the 5800 Fabric dip occurs with the majority of boards on the same AGESA, on Hynix/Micron/SS kits, there's just too many variable there to draw conclusions.

My conclusions are based on what I see before my eyes. As you can see from my screenshot for Zen timings, its agesa 1.0.0.7c, when another version comes out I'll compare again.

 

[AusWolf]

No, 5800MT/S boots with 1933 FCLK on my Gigabyte system with everything on auto & XMP profile enabled.


My conclusions are based on what I see before my eyes. As you can see from my screenshot for Zen timings, its agesa 1.0.0.7c, when another version comes out I'll compare again.

That is your motherboard acting weird, in my opinion. The IF should run at 2000 MHz, regardless of RAM speed.

 

[Mussels]

Seems like a shaky conclusion to correlate the board's 5800 auto rules to the existence of some sort of invisible divider. All boards BIOSes are very different from one another. Unless the 5800 Fabric dip occurs with the majority of boards on the same AGESA, on Hynix/Micron/SS kits, there's just too many variable there to draw conclusions.

The boards all do have auto rules, but they vary between brand, series, and BIOS.

MSI for example would dip the UCLK? to 850? 950? for stability while asus and aorus boosted the SoC voltage instead, until GN and HUB called them out on it (sacrificing performance for stability) - they changed 'gears' far earlier than all the other brands

No, 5800MT/S boots with 1933 FCLK on my Gigabyte system with everything on auto & XMP profile enabled.


My conclusions are based on what I see before my eyes. As you can see from my screenshot for Zen timings, its agesa 1.0.0.7c, when another version comes out I'll compare again.

5800 does math nicely into 1933 despite my brain hating that math (it's 3x)

The rest comes down to what your BIOS chose to do, as the defaults have been varying wildly as each brand tries to get ram stability up without following ASUS down the exploding CPU path

 

[ColdVergil]

Hello there everyone. Awesome Mussels has guided me here in order to set my ram properly optimised. I'm not too knowledgeable in this part haha. These are my specs:

 

[Space Lynx]

Hello there everyone. Awesome Mussels has guided me here in order to set my ram properly optimised. I'm not too knowledgeable in this part haha. These are my specs:

View attachment 312630

turn on xmp/echo and enjoy gaming. that's my advice.

 

[Mussels]

Zentimings has the information we'd need on the RAM and how its running now
ZenTimings v1.2.9 Download | TechPowerUp

CPU-Z is great, but is more useful in showing the settings within the RAM, rather than what it's running at now - having both is good for comparisons.

I know from the other thread you're running two mixed kits, one dual rank one single - not an issue on it's own. Your speeds are measuring slightly off, because you've got either spread spectrum enabled which drops to 99.9MHz to reduce EMI, or memory isolation enabled in windows (this turns on if you have virtualisation enabled in the BIOS) - it's a measuring error, and not an issue or performance loss.


Your RAM kits match perfectly at 3200 16-16-18-36 and 1.35v, which is quite lucky with mixed kits and likely why it's not given you any issues.


Going by the slot values you've got the ram set up like:
SR-SR-DR-DR
Technically that's got 3 ranks per channel and may not be optimal, going SR-DR-SR-DR could be better
(Or if we called them kit 1 and 2, you've got 1-1-2-2 instead of 1-2-1-2)

I'm not sure which rank setup is faster, but usually you want each stick paired with their identical stick.
Aida64 could be used as a before and after benchmark for you here, just remember you can double click each value instead of waiting for the whole lot to finish
Trial version wont show every field, but the top row are what matter the most here - they vary between CPU types so don't stress too much about how different they are, but if latency is over 70ns we'd want to know why. Oh and quit everything before running the test - it's sensitive to other tasks running.

Zentimings shows us every timing the PC is currently running at, not just what the RAM says it should be at - letting us know if auto settings are going off the rails
A big one is making sure M/F/U clk are all identical in the top right to avoid massive latency penalties, with VSOC (SoC voltage) being a key voltage for stability as some boards will have auto from 0.9v to 1.10v - and with 6 ranks you may need that 1.10v+ to solve infinity fabric issues if you experience any (Black screen crashes/GPU driver crashes, usb disconnects, and others)

TRFC on this RAM kit ran at over 1200 on auto at 3866, but dropped to 712 stable manually - that shaved about 5ns off my setup on it's own. (2x32GB, dual rank sticks). Other users here are better than I at knowing what values work, but i've got notes on previous advice they've given once we see what yours is at, if you care to tune it in

 

[TheLostSwede]

Upping the FCLK didn't make as much difference as I expected, but hey, there's a difference...

 

[ColdVergil]

Zentimings has the information we'd need on the RAM and how its running now
ZenTimings v1.2.9 Download | TechPowerUp

CPU-Z is great, but is more useful in showing the settings within the RAM, rather than what it's running at now - having both is good for comparisons.

I know from the other thread you're running two mixed kits, one dual rank one single - not an issue on it's own. Your speeds are measuring slightly off, because you've got either spread spectrum enabled which drops to 99.9MHz to reduce EMI, or memory isolation enabled in windows (this turns on if you have virtualisation enabled in the BIOS) - it's a measuring error, and not an issue or performance loss.


Your RAM kits match perfectly at 3200 16-16-18-36 and 1.35v, which is quite lucky with mixed kits and likely why it's not given you any issues.


Going by the slot values you've got the ram set up like:
SR-SR-DR-DR
Technically that's got 3 ranks per channel and may not be optimal, going SR-DR-SR-DR could be better
(Or if we called them kit 1 and 2, you've got 1-1-2-2 instead of 1-2-1-2)

I'm not sure which rank setup is faster, but usually you want each stick paired with their identical stick.
Aida64 could be used as a before and after benchmark for you here, just remember you can double click each value instead of waiting for the whole lot to finish
Trial version wont show every field, but the top row are what matter the most here - they vary between CPU types so don't stress too much about how different they are, but if latency is over 70ns we'd want to know why. Oh and quit everything before running the test - it's sensitive to other tasks running.
View attachment 312638



Zentimings shows us every timing the PC is currently running at, not just what the RAM says it should be at - letting us know if auto settings are going off the rails
A big one is making sure M/F/U clk are all identical in the top right to avoid massive latency penalties, with VSOC (SoC voltage) being a key voltage for stability as some boards will have auto from 0.9v to 1.10v - and with 6 ranks you may need that 1.10v+ to solve infinity fabric issues if you experience any (Black screen crashes/GPU driver crashes, usb disconnects, and others)
View attachment 312639
TRFC on this RAM kit ran at over 1200 on auto at 3866, but dropped to 712 stable manually - that shaved about 5ns off my setup on it's own. (2x32GB, dual rank sticks). Other users here are better than I at knowing what values work, but i've got notes on previous advice they've given once we see what yours is at, if you care to tune it in

Alright, cool stuff. I think I can borrow Aida from a friend haha so when I get home I'll swap the ram to what you said and will also do the benchmark. Oh and will try with Zentimmings.

Zentimings has the information we'd need on the RAM and how its running now
ZenTimings v1.2.9 Download | TechPowerUp

CPU-Z is great, but is more useful in showing the settings within the RAM, rather than what it's running at now - having both is good for comparisons.

I know from the other thread you're running two mixed kits, one dual rank one single - not an issue on it's own. Your speeds are measuring slightly off, because you've got either spread spectrum enabled which drops to 99.9MHz to reduce EMI, or memory isolation enabled in windows (this turns on if you have virtualisation enabled in the BIOS) - it's a measuring error, and not an issue or performance loss.


Your RAM kits match perfectly at 3200 16-16-18-36 and 1.35v, which is quite lucky with mixed kits and likely why it's not given you any issues.


Going by the slot values you've got the ram set up like:
SR-SR-DR-DR
Technically that's got 3 ranks per channel and may not be optimal, going SR-DR-SR-DR could be better
(Or if we called them kit 1 and 2, you've got 1-1-2-2 instead of 1-2-1-2)

I'm not sure which rank setup is faster, but usually you want each stick paired with their identical stick.
Aida64 could be used as a before and after benchmark for you here, just remember you can double click each value instead of waiting for the whole lot to finish
Trial version wont show every field, but the top row are what matter the most here - they vary between CPU types so don't stress too much about how different they are, but if latency is over 70ns we'd want to know why. Oh and quit everything before running the test - it's sensitive to other tasks running.
View attachment 312638



Zentimings shows us every timing the PC is currently running at, not just what the RAM says it should be at - letting us know if auto settings are going off the rails
A big one is making sure M/F/U clk are all identical in the top right to avoid massive latency penalties, with VSOC (SoC voltage) being a key voltage for stability as some boards will have auto from 0.9v to 1.10v - and with 6 ranks you may need that 1.10v+ to solve infinity fabric issues if you experience any (Black screen crashes/GPU driver crashes, usb disconnects, and others)
View attachment 312639
TRFC on this RAM kit ran at over 1200 on auto at 3866, but dropped to 712 stable manually - that shaved about 5ns off my setup on it's own. (2x32GB, dual rank sticks). Other users here are better than I at knowing what values work, but i've got notes on previous advice they've given once we see what yours is at, if you care to tune it in

Okay I came back and did as you told me, swapped out the ram, did the Aida benchmark and installed Zentimings. Here is what I got:

 

[HD64G]

Anyone in knowledge of existence of a similar table to the one below for DDR5? A friend will build a Zen4 system in the next few days and will try to help him config it to a stable and fast RAM setup.

 

[Mussels]

turn on xmp/echo and enjoy gaming. that's my advice.

Not enough if your RAM is past a threshold on each platform
On AM4 that was 3600, on AM5 it's 5600 IIRC

Beyond that they change gears/go out of sync and you run slower than jedec

Anyone in knowledge of existence of a similar table to the one below for DDR5? A friend will build a Zen4 system in the next few days and will try to help him config it to a stable and fast RAM setup.
View attachment 312737

That list itself isn't exactly super accurate, it's a loose guideline but not correct for a lot of DDR4 these days

 

[AusWolf]

Not enough if your RAM is past a threshold on each platform
On AM4 that was 3600, on AM5 it's 5600 IIRC

Beyond that they change gears/go out of sync and you run slower than jedec

6000 works just fine with both my 7700X and 7800X3D in gear 1.

Edit: it all depends on the number of ranks and RAM sticks you use.

 

[ColdVergil]

Alright, cool stuff. I think I can borrow Aida from a friend haha so when I get home I'll swap the ram to what you said and will also do the benchmark. Oh and will try with Zentimmings.


Okay I came back and did as you told me, swapped out the ram, did the Aida benchmark and installed Zentimings. Here is what I got:View attachment 312735

Alright, cool stuff. I think I can borrow Aida from a friend haha so when I get home I'll swap the ram to what you said and will also do the benchmark. Oh and will try with Zentimmings.


Okay I came back and did as you told me, swapped out the ram, did the Aida benchmark and installed Zentimings. Here is what I got:View attachment 312735

Good afternoon everyone. Was wondering if this looked alright and what could I do to improve them.

 

[HD64G]

That list itself isn't exactly super accurate, it's a loose guideline but not correct for a lot of DDR4 these days

Thanks for lettings me know about that, any info-link you happen to own that correct the wrongs or any info on AM5/DDR5 trfc timings to test apart from those posted in the thread? Thanks in advance!

 

[Mussels]

6000 works just fine with both my 7700X and 7800X3D in gear 1.

Edit: it all depends on the number of ranks and RAM sticks you use.

It varies between boards and BIOS
It "works" for you - but what defaults is it choosing?
Already forgotten from the prior page(s) but is it doing the full 3000/2000/2000 (or whatever the order was) or is it downclocking? Having three seperate settings is where things differ as it's not 1:1 like prior gen, now its 1:2:1 or 2:1:1 or whatever order they're in
(Got the flu, brain fart on what the values are)

Good afternoon everyone. Was wondering if this looked alright and what could I do to improve them.

Replies can be slow here, and with the edited post people don't get notified of changes
You didn't need a screenshot of every single RAM stick, Zentimings doesnt change as you rotate through them - each stick has the same active settings.

Your ram has identical timings which is good, but they aren't the same RAM at all - you've got it going samsung/hynix/samsung/hynix which is best for dual channel stability.

Set your DRAM voltage to 1.4v (vdimm) and SoC voltage (vsoc) to 1.10v and see if 3600 is stable - it may be as simple as that. Since infinity fabric speeds up with the RAM, that's going to speed up more than just RAM benchmarks and generally is easy to achieve on zen3 CPUs. Those voltages shouldn't produce any extra heat either, as it's a small increase.

 

[AusWolf]

It varies between boards and BIOS
It "works" for you - but what defaults is it choosing?
Already forgotten from the prior page(s) but is it doing the full 3000/2000/2000 (or whatever the order was) or is it downclocking? Having three seperate settings is where things differ as it's not 1:1 like prior gen, now its 1:2:1 or 2:1:1 or whatever order they're in
(Got the flu, brain fart on what the values are)

Yep, it does the full 3000 RAM, 3000 UCLK, 2000 FCLK thing on Auto - it doesn't downclock at all. I think this is why AMD said 6000 is the sweet spot - it generally works with two (especially single-rank) sticks.

 

[Mussels]

Yep, it does the full 3000 RAM, 3000 UCLK, 2000 FCLK thing on Auto - it doesn't downclock at all. I think this is why AMD said 6000 is the sweet spot - it generally works with two (especially single-rank) sticks.

That's damn well perfect then.

The GN? HUB? videos when x3D's were exploding had some of the boards doing 3000/1850/950 sort of madness, no two boards did the same.


Hah, found it in my YT history

Primary timings tended to be the same, but UCLK and secondaries vary all over the places.
I'll need to find the one that ran Uclk at <1000
(Still cant find that really low one, it's out there somewhere)


Found one of em, x670e this time

You could be
3000/2000/3000
3000/2000/1500
trc? 96 to 146 - gee i wonder why stability can be an issue with some of these boards

 

[AusWolf]

That's damn well perfect then.

The GN? HUB? videos when x3D's were exploding had some of the boards doing 3000/1850/950 sort of madness, no two boards did the same.


Hah, found it in my YT history
View attachment 313067


Primary timings tended to be the same, but UCLK and secondaries vary all over the places.
I'll need to find the one that ran Uclk at <1000
(Still cant find that really low one, it's out there somewhere)


Found one of em, x670e this time
View attachment 313080

You could be
3000/2000/3000
3000/2000/1500
trc? 96 to 146 - gee i wonder why stability can be an issue with some of these boards

And one of them is the MSi Pro X670-P Wifi? Could my MSi Pro B650M-A Wifi be better than that?

 

[TheLostSwede]

Anyone in knowledge of existence of a similar table to the one below for DDR5? A friend will build a Zen4 system in the next few days and will try to help him config it to a stable and fast RAM setup.
View attachment 312737

[Sammelthread] - Ryzen DDR5 RAM OC Thread

Ryzen DDR5 RAM OC Thread Original Thread von @ro0ki , übernommen von mir @Reous Alle aufgelisteten Werte sind Empfehlungen und Informationen meinerseits. Jeder muss selbst entscheiden was er wie einstellt oder wie viel Spannung er bereit ist auf die Komponenten zu geben. Letztes Update...

www.hardwareluxx.de

The latest version of Taiphoon Burner appears to handle DDR5 just fine now as well

สล็อตเว็บตรง อันดับ 1 ในประเทศไทย Slot ทรูวอเลทที่ดีที่สุด 2024

เว็บตรงสล็อต ไม่ผ่านเอเย่นต์ รองรับทรูวอเลทไม่มีขั้นต่ำ สล็อตเว็บตรง ที่มาแรงที่สุดในตอนนี้ รวมสล็อตชั้นนำ ครบทุกค่ายจบในที่เดียวเท่านั้น

www.softnology.biz

 

[tabascosauz]

[Sammelthread] - Ryzen DDR5 RAM OC Thread

Ryzen DDR5 RAM OC Thread Original Thread von @ro0ki , übernommen von mir @Reous Alle aufgelisteten Werte sind Empfehlungen und Informationen meinerseits. Jeder muss selbst entscheiden was er wie einstellt oder wie viel Spannung er bereit ist auf die Komponenten zu geben. Letztes Update...

www.hardwareluxx.de

The latest version of Taiphoon Burner appears to handle DDR5 just fine now as well

สล็อตเว็บตรง อันดับ 1 ในประเทศไทย Slot ทรูวอเลทที่ดีที่สุด 2024

เว็บตรงสล็อต ไม่ผ่านเอเย่นต์ รองรับทรูวอเลทไม่มีขั้นต่ำ สล็อตเว็บตรง ที่มาแรงที่สุดในตอนนี้ รวมสล็อตชั้นนำ ครบทุกค่ายจบในที่เดียวเท่านั้น

www.softnology.biz

Good to see reous still going at it

A-die is legit nuts, never stops impressing

 

[HD64G]

TheLostSwede​

 

[ColdVergil]

It varies between boards and BIOS
It "works" for you - but what defaults is it choosing?
Already forgotten from the prior page(s) but is it doing the full 3000/2000/2000 (or whatever the order was) or is it downclocking? Having three seperate settings is where things differ as it's not 1:1 like prior gen, now its 1:2:1 or 2:1:1 or whatever order they're in
(Got the flu, brain fart on what the values are)

Replies can be slow here, and with the edited post people don't get notified of changes
You didn't need a screenshot of every single RAM stick, Zentimings doesnt change as you rotate through them - each stick has the same active settings.

Your ram has identical timings which is good, but they aren't the same RAM at all - you've got it going samsung/hynix/samsung/hynix which is best for dual channel stability.

Set your DRAM voltage to 1.4v (vdimm) and SoC voltage (vsoc) to 1.10v and see if 3600 is stable - it may be as simple as that. Since infinity fabric speeds up with the RAM, that's going to speed up more than just RAM benchmarks and generally is easy to achieve on zen3 CPUs. Those voltages shouldn't produce any extra heat either, as it's a small increase.

Oki doke! I don't know what happened tho, I did two posts.

It varies between boards and BIOS
It "works" for you - but what defaults is it choosing?
Already forgotten from the prior page(s) but is it doing the full 3000/2000/2000 (or whatever the order was) or is it downclocking? Having three seperate settings is where things differ as it's not 1:1 like prior gen, now its 1:2:1 or 2:1:1 or whatever order they're in
(Got the flu, brain fart on what the values are)

Replies can be slow here, and with the edited post people don't get notified of changes
You didn't need a screenshot of every single RAM stick, Zentimings doesnt change as you rotate through them - each stick has the same active settings.

Your ram has identical timings which is good, but they aren't the same RAM at all - you've got it going samsung/hynix/samsung/hynix which is best for dual channel stability.

Set your DRAM voltage to 1.4v (vdimm) and SoC voltage (vsoc) to 1.10v and see if 3600 is stable - it may be as simple as that. Since infinity fabric speeds up with the RAM, that's going to speed up more than just RAM benchmarks and generally is easy to achieve on zen3 CPUs. Those voltages shouldn't produce any extra heat either, as it's a small increase.

Okay I try as suggested but unfortunately I couldn't get it to post. Not to mention the gigabyte bios settings are kinda confusing or redundant. Or perhaps i'm too dumb lol.

 

[Mussels]

And one of them is the MSi Pro X670-P Wifi? Could my MSi Pro B650M-A Wifi be better than that?

That's my major frustration is that there's barely even any trend within the same brand

It's like each 'series' of boards is made independently of the rest and very rarely does anything get ported sideways/backwards from one series to another - the newer ones may have all the fixes or just new issues. The gigabyte boards had super tight secondary timings that caused their boot issues, for example - they only worked right on whatever RAM they initially tested with (Hynix, maybe?)

Oki doke! I don't know what happened tho, I did two posts.

Okay I try as suggested but unfortunately I couldn't get it to post. Not to mention the gigabyte bios settings are kinda confusing or redundant. Or perhaps i'm too dumb lol.

get it working at 3200 and verify the settings stuck in zentimings before raising the speed, and then save it as a bios profile. Work your way up from 3200 and see when problems start.