Ryzen Owners Zen Garden

[phill]

Guys, I'm wondering if you might be able to point me in the right direction, (whichever way it might be in!! lol) Following on from the purchase thread with @Mussels ( Linky to Mussels's post )

I'm trying to set my 5950X up pretty much the same as my 3950X... Although I'm stupid or its stupid or we both stupid in which case, I'm...... (I'll leave you all to fill in the blanks...) lol

So, from my limited knowledge of AMD/Zen all I have been doing before with my 3900X/3950X's are limiting them with a vcore set and setting to 3200 RAM and forgetting about it to be honest... My 3950X has been rock solid, so have both my 3900X's but they have been a little different (the two 3900X's in that one requires more volts than the other and such like, silicone lottery I'm presuming ) but when it comes to my 3950X vs my 5950X, it seems I'm completely lost and I'm really stupid!

So I've grabbed some hopefully useful info from CPU-Z, Thaiphoon reader and a few tests in the background with WCG running (which is basically what I'm trying to set it up for mostly), so any advise would be amazing.
End goal? Efficiency over anything else really but also if I can get 3600MHz to run on the RAM and at least 4GHz from this monster of a CPU, I would like that! I've not been really messing about with the voltages and settings in the bios, I've not had to with the 3900/3950X systems at all, maybe due to the 3200MHz RAM in them but they just seem to work
Voltages I can remember setting in the bios -

CPU VCore - Auto
SOC - 1.15v - tried less, things went wrong, so just left it here for the moment
DRAM - 1.35v
RAM is 3600, but set to 3333 for the moment, didn't seem to like 3400 or 3600 - think it might have been down to the fabric link if I am correct in saying?? Hopefully.....
Load line calibration set to level 2.... Hopefully that helps...

Here goes
A few standard CineBench tests - R15, R20 and R23


The Vcore tests with WCG running -



The Thiaphoon RAM results -



I've ran WCG with the 5950X with a set voltage and an auto voltage, the more I higher the voltage up and set it at that, the lower the CPU speed goes. If I leave to Auto, it sits about the 3.90GHz to 4.00GHz area... My 3950X I've included the grabs for that too, here they are...



What I'm confused about for the most part, is the amount of variance the 5950X has when it's in use with a set Vcore like the 3950X. As you can see from that screen grab, it's pretty much a solid straight line with very little movement. I'll try and grab the bios settings for the 3950X just so we can compare if needed.
Is there something I'm doing completely wrong at this point?? If there is, I'd love to find out so I can learn and put it right.... Both CPUs are under custom water, one a 420 rad (5950X for testing) and a 560 rad for the 3950X ..

Very much looking forward to any replies Both CPUs use the same motherboard, a Crosshair 8 board (ones WiFi and the other isn't - Can't imagine that would make any difference tho?? )

 

[mstenholm]

Guys, I'm wondering if you might be able to point me in the right direction, (whichever way it might be in!! lol) Following on from the purchase thread with @Mussels ( Linky to Mussels's post )

I'm trying to set my 5950X up pretty much the same as my 3950X... Although I'm stupid or its stupid or we both stupid in which case, I'm...... (I'll leave you all to fill in the blanks...) lol

So, from my limited knowledge of AMD/Zen all I have been doing before with my 3900X/3950X's are limiting them with a vcore set and setting to 3200 RAM and forgetting about it to be honest... My 3950X has been rock solid, so have both my 3900X's but they have been a little different (the two 3900X's in that one requires more volts than the other and such like, silicone lottery I'm presuming ) but when it comes to my 3950X vs my 5950X, it seems I'm completely lost and I'm really stupid!

So I've grabbed some hopefully useful info from CPU-Z, Thaiphoon reader and a few tests in the background with WCG running (which is basically what I'm trying to set it up for mostly), so any advise would be amazing.
End goal? Efficiency over anything else really but also if I can get 3600MHz to run on the RAM and at least 4GHz from this monster of a CPU, I would like that! I've not been really messing about with the voltages and settings in the bios, I've not had to with the 3900/3950X systems at all, maybe due to the 3200MHz RAM in them but they just seem to work
Voltages I can remember setting in the bios -

CPU VCore - Auto
SOC - 1.15v - tried less, things went wrong, so just left it here for the moment
DRAM - 1.35v
RAM is 3600, but set to 3333 for the moment, didn't seem to like 3400 or 3600 - think it might have been down to the fabric link if I am correct in saying?? Hopefully.....
Load line calibration set to level 2.... Hopefully that helps...

Here goes
A few standard CineBench tests - R15, R20 and R23
View attachment 228768 View attachment 228769View attachment 228770 View attachment 228771

The Vcore tests with WCG running -
View attachment 228772 View attachment 228773 View attachment 228774
View attachment 228775

The Thiaphoon RAM results -

View attachment 228776

I've ran WCG with the 5950X with a set voltage and an auto voltage, the more I higher the voltage up and set it at that, the lower the CPU speed goes. If I leave to Auto, it sits about the 3.90GHz to 4.00GHz area... My 3950X I've included the grabs for that too, here they are...

View attachment 228777

What I'm confused about for the most part, is the amount of variance the 5950X has when it's in use with a set Vcore like the 3950X. As you can see from that screen grab, it's pretty much a solid straight line with very little movement. I'll try and grab the bios settings for the 3950X just so we can compare if needed.
Is there something I'm doing completely wrong at this point?? If there is, I'd love to find out so I can learn and put it right.... Both CPUs are under custom water, one a 420 rad (5950X for testing) and a 560 rad for the 3950X ..

Very much looking forward to any replies Both CPUs use the same motherboard, a Crosshair 8 board (ones WiFi and the other isn't - Can't imagine that would make any difference tho?? )

@phill these days are not good for uploading pics…..all I see is the filenames.

 

[Taraquin]

Guys, I'm wondering if you might be able to point me in the right direction, (whichever way it might be in!! lol) Following on from the purchase thread with @Mussels ( Linky to Mussels's post )

I'm trying to set my 5950X up pretty much the same as my 3950X... Although I'm stupid or its stupid or we both stupid in which case, I'm...... (I'll leave you all to fill in the blanks...) lol

So, from my limited knowledge of AMD/Zen all I have been doing before with my 3900X/3950X's are limiting them with a vcore set and setting to 3200 RAM and forgetting about it to be honest... My 3950X has been rock solid, so have both my 3900X's but they have been a little different (the two 3900X's in that one requires more volts than the other and such like, silicone lottery I'm presuming ) but when it comes to my 3950X vs my 5950X, it seems I'm completely lost and I'm really stupid!

So I've grabbed some hopefully useful info from CPU-Z, Thaiphoon reader and a few tests in the background with WCG running (which is basically what I'm trying to set it up for mostly), so any advise would be amazing.
End goal? Efficiency over anything else really but also if I can get 3600MHz to run on the RAM and at least 4GHz from this monster of a CPU, I would like that! I've not been really messing about with the voltages and settings in the bios, I've not had to with the 3900/3950X systems at all, maybe due to the 3200MHz RAM in them but they just seem to work
Voltages I can remember setting in the bios -

CPU VCore - Auto
SOC - 1.15v - tried less, things went wrong, so just left it here for the moment
DRAM - 1.35v
RAM is 3600, but set to 3333 for the moment, didn't seem to like 3400 or 3600 - think it might have been down to the fabric link if I am correct in saying?? Hopefully.....
Load line calibration set to level 2.... Hopefully that helps...

Here goes
A few standard CineBench tests - R15, R20 and R23
View attachment 228768 View attachment 228769View attachment 228770 View attachment 228771

The Vcore tests with WCG running -
View attachment 228772 View attachment 228773 View attachment 228774
View attachment 228775

The Thiaphoon RAM results -

View attachment 228776

I've ran WCG with the 5950X with a set voltage and an auto voltage, the more I higher the voltage up and set it at that, the lower the CPU speed goes. If I leave to Auto, it sits about the 3.90GHz to 4.00GHz area... My 3950X I've included the grabs for that too, here they are...

View attachment 228777

What I'm confused about for the most part, is the amount of variance the 5950X has when it's in use with a set Vcore like the 3950X. As you can see from that screen grab, it's pretty much a solid straight line with very little movement. I'll try and grab the bios settings for the 3950X just so we can compare if needed.
Is there something I'm doing completely wrong at this point?? If there is, I'd love to find out so I can learn and put it right.... Both CPUs are under custom water, one a 420 rad (5950X for testing) and a 560 rad for the 3950X ..

Very much looking forward to any replies Both CPUs use the same motherboard, a Crosshair 8 board (ones WiFi and the other isn't - Can't imagine that would make any difference tho?? )

What do you mostly do? Gaming? Productivity? I would use curve optimizer with pbo at 0 and set a PPT limit. This way you get the best from all three world: Good single core at low voltage, good allcore similar to static allcore and very low idle/load consumption due to low temps/voltage. If you are unlucky with a few core you will probably get better allcore efficiency with CO since you can run other cores much better. A static clock is only as good as your worst core allows.

Example: 2 core do -30, 4 do -25, 1 do -20, 5 do -15, 2 do -10 and 2 do -5. The -5 cores allows for 4.2GHz@1.15V, while the 2 -30 allows for 4.6GHz@1.15V, with a static OC you lose out on performance.

 

[Flyordie]

I think AM4 in general is winding down. Latest BIOS for me is 3801, the exact same as your 2423 (just different numbering due to X570/B550), and exact same as Dark Hero which is also 3801. So if that's the best available on Asus' top dog, then either we're not being left behind or we're ALL being left behind lol

I'm not touching 3801 though. Tried once and CPU performance is horrendous across the board. Not sure what they or AMD did.

Your post made me check Gigabyte for BIOS update for my board.. and sure enough there is a new update. 11-30-2021...

 

[tabascosauz]

Your post made me check Gigabyte for BIOS update for my board.. and sure enough there is a new update. 11-30-2021...

Some were speculating that Gigabyte is the only one addressing the vulnerability because they're more "forward thinking". Others were saying that Gigabyte is the only one that needs it because only their BIOS-level security is poor compared to other vendors.

Whatever the case might be, all Asus boards are still on the same 1.2.0.3C BIOS (3801 for X570, 2423 for B550) after all this time.

Some redditor says that the new update packaging is more secure and brings it up to par with competitors:

I'm guessing the .CAP file that Asus BIOSes usually come in, refers to this new format. My B550I Aorus Pro AX hasn't made the switch yet, still on F14.

 

[Flyordie]

Some were speculating that Gigabyte is the only one addressing the vulnerability because they're more "forward thinking". Others were saying that Gigabyte is the only one that needs it because only their BIOS-level security is poor compared to other vendors.

Whatever the case might be, all Asus boards are still on the same 1.2.0.3C BIOS (3801 for X570, 2423 for B550) after all this time.

Some redditor says that the new update packaging is more secure and brings it up to par with competitors:

View attachment 228799

Well, I'll be updating to it tomorrow maybe. Depends how well I feel. lol. Currently sick. NO.. its not COVID. lol.

I am hoping they did some bug fixes too. A simple one is the voltage setting for the main CPU volt rail. Set it to manual and it defaults to 1.6V no matter what setting you set. Its stupid. Only way to manually adjust voltage is in OS with Ryzen Master now.

 

[Mussels]

Guys, I'm wondering if you might be able to point me in the right direction, (whichever way it might be in!! lol) Following on from the purchase thread with @Mussels ( Linky to Mussels's post )

I'm trying to set my 5950X up pretty much the same as my 3950X... Although I'm stupid or its stupid or we both stupid in which case, I'm...... (I'll leave you all to fill in the blanks...) lol

So, from my limited knowledge of AMD/Zen all I have been doing before with my 3900X/3950X's are limiting them with a vcore set and setting to 3200 RAM and forgetting about it to be honest... My 3950X has been rock solid, so have both my 3900X's but they have been a little different (the two 3900X's in that one requires more volts than the other and such like, silicone lottery I'm presuming ) but when it comes to my 3950X vs my 5950X, it seems I'm completely lost and I'm really stupid!

So I've grabbed some hopefully useful info from CPU-Z, Thaiphoon reader and a few tests in the background with WCG running (which is basically what I'm trying to set it up for mostly), so any advise would be amazing.
End goal? Efficiency over anything else really but also if I can get 3600MHz to run on the RAM and at least 4GHz from this monster of a CPU, I would like that! I've not been really messing about with the voltages and settings in the bios, I've not had to with the 3900/3950X systems at all, maybe due to the 3200MHz RAM in them but they just seem to work
Voltages I can remember setting in the bios -

CPU VCore - Auto
SOC - 1.15v - tried less, things went wrong, so just left it here for the moment
DRAM - 1.35v
RAM is 3600, but set to 3333 for the moment, didn't seem to like 3400 or 3600 - think it might have been down to the fabric link if I am correct in saying?? Hopefully.....
Load line calibration set to level 2.... Hopefully that helps...

Here goes
A few standard CineBench tests - R15, R20 and R23
View attachment 228768 View attachment 228769View attachment 228770 View attachment 228771

The Vcore tests with WCG running -
View attachment 228772 View attachment 228773 View attachment 228774
View attachment 228775

The Thiaphoon RAM results -

View attachment 228776

I've ran WCG with the 5950X with a set voltage and an auto voltage, the more I higher the voltage up and set it at that, the lower the CPU speed goes. If I leave to Auto, it sits about the 3.90GHz to 4.00GHz area... My 3950X I've included the grabs for that too, here they are...

View attachment 228777

What I'm confused about for the most part, is the amount of variance the 5950X has when it's in use with a set Vcore like the 3950X. As you can see from that screen grab, it's pretty much a solid straight line with very little movement. I'll try and grab the bios settings for the 3950X just so we can compare if needed.
Is there something I'm doing completely wrong at this point?? If there is, I'd love to find out so I can learn and put it right.... Both CPUs are under custom water, one a 420 rad (5950X for testing) and a 560 rad for the 3950X ..

Very much looking forward to any replies Both CPUs use the same motherboard, a Crosshair 8 board (ones WiFi and the other isn't - Can't imagine that would make any difference tho?? )

Oof, all the images are dead thanks to the TPU server emergency

It's PBO settings you need to be tweaking, combined with the curve undervolting on zen3 chips

higher voltage means more wattage used, which makes you hit the PBO limits faster for lower clocks
The curve undervolt means less voltage used, so you get higher clocks within the same PBO limits

 

[phill]

@phill these days are not good for uploading pics…..all I see is the filenames.

What do you mostly do? Gaming? Productivity? I would use curve optimizer with pbo at 0 and set a PPT limit. This way you get the best from all three world: Good single core at low voltage, good allcore similar to static allcore and very low idle/load consumption due to low temps/voltage. If you are unlucky with a few core you will probably get better allcore efficiency with CO since you can run other cores much better. A static clock is only as good as your worst core allows.

Example: 2 core do -30, 4 do -25, 1 do -20, 5 do -15, 2 do -10 and 2 do -5. The -5 cores allows for 4.2GHz@1.15V, while the 2 -30 allows for 4.6GHz@1.15V, with a static OC you lose out on performance.

Oof, all the images are dead thanks to the TPU server emergency

It's PBO settings you need to be tweaking, combined with the curve undervolting on zen3 chips

higher voltage means more wattage used, which makes you hit the PBO limits faster for lower clocks
The curve undervolt means less voltage used, so you get higher clocks within the same PBO limits

Sorry guys, they worked when I put them up (wondering if to do todays update.....)

I would love to do some gaming but I seem to spend most of my time looking through websites and crunching... Currently, set to Auto vcore, I'm pulling about 210w from the wall with all 32 threads loaded and working away, with my 1080 TI and a basic custom water setup... That's running about the 3.80GHz mark at the moment... I've never really used the PBO method, so best go and see what that's all about

I'll see if I can upload the pics elsewhere just so they are showing, hopefully that'll help a bit

EDIT - I'll hang on and wait till the images are fixed, I'll get some better ideas and knowledge then, at the moment, I simply have no idea and I'd rather not do damage the CPU Trying to keep the core clock as high as possible and the temps as low as possible So far, auto vcore (for some reason) seems to be the best. Have CB R23 looping for 10 minutes, the temp has hit 52C and stuck there, 3.86GHz so not too far off the 4.00GHz target.... Power usage, 198w full system load

 

[Mussels]

The images died because TPU's servers died and w1zzard had to restore from a backup, losing a few hours of images

The clock is pre-set, based on the voltage. If you set a static voltage, you're also locking the max clock speed.

 

[phill]

The images died because TPU's servers died and w1zzard had to restore from a backup, losing a few hours of images

The clock is pre-set, based on the voltage. If you set a static voltage, you're also locking the max clock speed.

That's never really given me any problems with the performance and setup, most of the games I did get to play I would be limited by other things than my GPU, probably me more so! My aim is to keep the CPU as cool as possible whilst maintaining the highest boost I can. Since whenever the PC is on, it's always under a 100% load, so 50C temps are my aim so when the summer hits, they might spike to about 60C if its rather warm.. That said, with my little AC unit now, it might get a bit cooler again....

 

[Taraquin]

Sorry guys, they worked when I put them up (wondering if to do todays update.....)

I would love to do some gaming but I seem to spend most of my time looking through websites and crunching... Currently, set to Auto vcore, I'm pulling about 210w from the wall with all 32 threads loaded and working away, with my 1080 TI and a basic custom water setup... That's running about the 3.80GHz mark at the moment... I've never really used the PBO method, so best go and see what that's all about

I'll see if I can upload the pics elsewhere just so they are showing, hopefully that'll help a bit

EDIT - I'll hang on and wait till the images are fixed, I'll get some better ideas and knowledge then, at the moment, I simply have no idea and I'd rather not do damage the CPU Trying to keep the core clock as high as possible and the temps as low as possible So far, auto vcore (for some reason) seems to be the best. Have CB R23 looping for 10 minutes, the temp has hit 52C and stuck there, 3.86GHz so not too far off the 4.00GHz target.... Power usage, 198w full system load

The pbo 0+curve optimizer will net you around 200MHz more allcore at same consumption so really worth it

 

[phill]

Thanks for the help What sort of Vcore could I set it to? Trying to set at like 1.0Vcore like my 3900X was a big no no lol

 

[freeagent]

For straight crunching like with WCG I think it’s better to run an all core clock, 1 speed 1 voltage.. but for everything else PBO is the way to go most of the time

I use 1.25v for 4500 and that is all load stable. 4600 comes at 1.35 and that’s good for everything except the most stressful of loads.

 

[phill]

For straight crunching like with WCG I think it’s better to run an all core clock, 1 speed 1 voltage.. but for everything else PBO is the way to go most of the time

I use 1.25v for 4500 and that is all load stable. 4600 comes at 1.35 and that’s good for everything except the most stressful of loads.

What sort of temps do you get with those @freeagent ? I'm trying to keep it low, 50's if I'm honest... Crunching is a long game in my opinion, there's no end in sight at all, just let them keep coming

 

[Taraquin]

Thanks for the help What sort of Vcore could I set it to? Trying to set at like 1.0Vcore like my 3900X was a big no no lol

For pbo+CO you dont set vcore, you modify the voltage curve through values from 0 to -30.

 

[freeagent]

What sort of temps do you get with those @freeagent ? I'm trying to keep it low, 50's if I'm honest... Crunching is a long game in my opinion, there's no end in sight at all, just let them keep coming

High 50s low 60s at 4500, and a bit spicier at 4600 with temps in the low to mid 70s..

Edit:

Whoops mid to high 70s at 4600..

 

[Mussels]

That's never really given me any problems with the performance and setup, most of the games I did get to play I would be limited by other things than my GPU, probably me more so! My aim is to keep the CPU as cool as possible whilst maintaining the highest boost I can. Since whenever the PC is on, it's always under a 100% load, so 50C temps are my aim so when the summer hits, they might spike to about 60C if its rather warm.. That said, with my little AC unit now, it might get a bit cooler again....

Then you want to adjust PBO settings, and possibly just set a max CPU temp value

60C as a goal would be laughable with a 5800x, fyi> i can do it MT, but not ST.

Heat is what we have be trained to see is bad, amperage is what kills the CPUs - so 4.7GHz all core at 60C could kill a chip while 5.05Ghz boost with low threads at 80C wont

 

[Deleted member 202104]

What sort of temps do you get with those @freeagent ? I'm trying to keep it low, 50's if I'm honest... Crunching is a long game in my opinion, there's no end in sight at all, just let them keep coming

Hey Phill - For reference I just set up a 5950x as a 24/7 WCG box. I started by setting the BIOS to the 95w Eco mode (10w less than stock by lowering PPT, EDC, and TDC). That gave me 16c/32t at around 3.4GHz on all cores @ 52c on a Scythe Mugen 5B. I figured that was too slow and cool so I brought it back to just stock. It's now 3.8 on all cores @59c. Ambient room temp is 20-21c

You'll probably gain some more MHz by attempting to optimize each core with a negative offset, but only you can decide if it's worth the time and effort.

 

[freeagent]

but only you can decide if it's worth the time and effort

For WCG? Nah not worth it. All core or bust, unless you run PBO right to the thermal edge, but even still.. not really worth it to me. Results are fairly similar..

Now with water? I would be interested in trying.. someone just has to send me a loop

But in the end I am not a pro,, I only played with it for about a month or two..

And some of you guys have big pulsing brains, unlike moi

 

[Mussels]

Ah, if you're doing WCG then yeah you're gunna want static all core, at efficient settings

 

[Mussels]

Aww damnit

Time to play with core cycler again, got a WHEA error and a crash overnight

 

[Taraquin]

Aww damnit

Time to play with core cycler again, got a WHEA error and a crash overnight
View attachment 229200

No, just adjust the negative on no 14 slightly. Usually 1 tick fixes it

 

[HD64G]

For anyone in need of higher clocks in his Zen CPU I have a suggestion to do. Since the RAM timings contribution in the system's performance increase have diminishing returns when you get too far, I would loosen the timings a bit to allow the same stable system with lower SOC voltage. This decrease allows more power to get to the cores of the CPU when the power limit isn't ultra high (stock or close to that) and thus, to reach higher core clocks. In my case when I managed to reduce the SOC voltage from 1,125v to 1,1v without losing in stability (-2 clicks in the UEFI) the cinebench R20 score raised from 2800 to 2840.

 

[Mussels]

Turned out i'd been daft, and raised the PBO values to +200Mhz as well as -20 on the curve optimiser - simply forgotten why i'd lowered it.
At stock speeds, -20 on all cores, but -14 on core 3 seems stable.

 

[Taraquin]

For anyone in need of higher clocks in his Zen CPU I have a suggestion to do. Since the RAM timings contribution in the system's performance increase have diminishing returns when you get too far, I would loosen the timings a bit to allow the same stable system with lower SOC voltage. This decrease allows more power to get to the cores of the CPU when the power limit isn't ultra high (stock or close to that) and thus, to reach higher core clocks. In my case when I managed to reduce the SOC voltage from 1,125v to 1,1v without losing in stability (-2 clicks in the UEFI) the cinebench R20 score raised from 2800 to 2840.

Lowering timings should not affect what soc voltage you need, lowering infinity fabric speed does. On my system running 38000/1900 tight or loose needs 1.06V soc and 0.98V iod. Going to 4000/2000 I need 1.11V soc and 1.03V iod no matter what timings I use In CB this translates to 1-2% score for me since clockspeed is 50-75MHz lower. In games 4000cl16 tuned is a bit better than 3800cl15 so better BW makes up for it.

Another suggestion for improving is lowering VDDG CCD and VDDP as these usually run at 1.1V each on auto which is far above required. In reality only 0.85-1.00V og CCD is required and 0.8-1.00V og VDDP is required. This steals a lot from CPU budget.