What is your max infinity fabric on ryzen 5000? September 2021 edition

[Dar1usElite]

Looks good, try my suggestions 1 by 1 if stable. If unstable set voltage a bit higher. Lits of errors try 1.45V, very few errors try 1.42V.

So i downloaded testmem5 and it had a config called 1usmus_v3, it only ran for 3 cycles (unsure if i can change that) but no errors

 

[Taraquin]

So i downloaded testmem5 and it had a config called 1usmus_v3, it only ran for 3 cycles (unsure if i can change that) but no errors

In the configfile change cycles to 20, edit it with wordpad. Most say 20 or 25 is enough. If you pass 10 rounds you are usually 99% stable in my xp.

 

[Valantar]

Yeah I can't say I've ever seen HCI CPU usage so low even in the free version. I never ran as many instances as CPU threads with HCI free. HCI Pro isn't much different, it just gives you the option to set # of threads, as well as capacity per thread (leaving it on Auto is fine), so I set it to CPU threads and it loads to 100% (well, "100%" as I explained earlier). Just make sure Low Priority Threads is not checked.

Obviously no user input needed to run HCI overnight, but if you're already through 4 runs of TM5 don't fret too much about HCI. Daily stable is generally regarded as 600-700%, but I just have some adverse experiences between 1000-2000% so OCD.

View attachment 224325

Sounds odd - I don't think free has an option for en/disabling low priority threads though. I got it going before I left home just now, but I forgot to uncheck that, so we'll see where it's at when I get back. I got through something like 7 runs of TM5 1usmus before this, so it seems stable - but so did my previous settings after a 2hr run of TM5 anta777.

If you only had 4 errors in HCL overnight then 1.46-1.47V is probably enough. If I pass 20 TM5 usmus1 I call it rock solid Seems your bin is very similar to mine, I need 1.47V for about the same timings at 3800, but 1T.

1T with GDM enabled or disabled? If it passes the current testing I'll make some passes at 1.46 and/or 1.47 over the coming days, as well as tuning the settings you mentioned above. This has long since gotten to a point of (very) diminishing returns, so I'm really just looking for stability and any left-over low hanging fruit now. After that it's on to CO testing to identify which of my cores aren't handling -30.

In the configfile change cycles to 20, edit it with wordpad. Most say 20 or 25 is enough. If you pass 10 rounds you are usually 99% stable in my xp.

Lol, I really should have looked into that rather than just restarting it every three cycles. I guess I got through 21 cycles with my current settings then

 

[Dar1usElite]

In the configfile change cycles to 20, edit it with wordpad. Most say 20 or 25 is enough. If you pass 10 rounds you are usually 99% stable in my xp.

This is how its going so far, does it look right to you? namely ram usage...?

 

[schidddy]

@tabascosauz
aah shit i already felt like i bought shit deciding on that board. Yeah its AN Aourus Elite Rev1.0
Buying a new one ist not the problem but i just set up my watercooling system nice and clean, got all air out and running fine so far except the RAM. When jumping up to 3866mhz on the RAM i already felt like running against a hardwall and on my experience it should do at least a boot if the ram is really having trouble with that low freq. So it is probably as u guys say, the Problem is the board. Now the question is, is it really worth that business changing the board to get to 4000mhz RAM Speed or should i just tighten the timings as @rares495 said.
@Taraquin thanks i will try that but this evening, besides of that, do you see any indifferences or settings that might hurt each other on the timings?

 

[tabascosauz]

@tabascosauz
aah shit i already felt like i bought shit deciding on that board. Yeah its AN Aourus Elite Rev1.0
Buying a new one ist not the problem but i just set up my watercooling system nice and clean, got all air out and running fine so far except the RAM. When jumping up to 3866mhz on the RAM i already felt like running against a hardwall and on my experience it should do at least a boot if the ram is really having trouble with that low freq. So it is probably as u guys say, the Problem is the board. Now the question is, is it really worth that business changing the board to get to 4000mhz RAM Speed or should i just tighten the timings as @rares495 said.
@Taraquin thanks i will try that but this evening, besides of that, do you see any indifferences or settings that might hurt each other on the timings?

If board works fine for you, hang onto it.

Don't count on doing 4000 on 5900X. Especially on dual rank. Some 5900X won't even boot 4000, all the others will be an endless stream of hundreds or thousands of WHEA errors (some turn off the WHEA reporting service and pretend all is well, kinda dumb solution).

I can boot 4000 but it's flooded with WHEA. 3866 is fine but the WHEAs come trickling in no matter what. 3800 is stable at 1.106V ish.

Don't get sucked into the hype validations with your daily rig. You have to remember dual rank is free performance in itself, you've already got a leg up on all the single rank users - they need that extra freq to even have a chance at catching up completely.

 

[Taraquin]

Valantar: If it passes 20 rounds ca all it a day 1T without gdm, but it's a bit tricky, stick with 2T I advice.

Dar1usElite: Looks good, test 10 rounds then tweak.

@tabascosauz
aah shit i already felt like i bought shit deciding on that board. Yeah its AN Aourus Elite Rev1.0
Buying a new one ist not the problem but i just set up my watercooling system nice and clean, got all air out and running fine so far except the RAM. When jumping up to 3866mhz on the RAM i already felt like running against a hardwall and on my experience it should do at least a boot if the ram is really having trouble with that low freq. So it is probably as u guys say, the Problem is the board. Now the question is, is it really worth that business changing the board to get to 4000mhz RAM Speed or should i just tighten the timings as @rares495 said.
@Taraquin thanks i will try that but this evening, besides of that, do you see any indifferences or settings that might hurt each other on the timings?

Wr 16, rtp 8, wtrs 4, wtrl 10 or 8, rfc 288 or 280, wrwrscl 4, rdrdscl 4.

 

[purecain]

Just watched a video on ProcODT and the recommended setting is between 40 and 60ohms. Mine was set to 33ohms on auto. I've changed that to 53.3 and hopefully this should give me the memory stability I needed. I'm going to retry the settings. I added the soc voltage of 1.1v and can now achieve 3733mhz so 1866mhz IF. I can not get 1900mhz IF to boot though... still working on it.

 

[Taraquin]

Just watcxhed a video on ProcODT and the recommended setting is between 40 and 60ohms. Mine was set to 33ohms on auto. I've changed that to 53.3 and hopefully this should give me the memory stability I needed. I'm going to retry the settings.

On ryzen 5000 single rank 28 to 37, dual rank 34 to 48. Ryzen 1 and 2 needed much higher. 53 is not good on 3000 and 5000.

 

[schidddy]

okay @tabascosauz i guess i stay with the board, do you guys think i run into problems clocking the CPU to 4.8 to 4.9ghz if needed? i know it as an inferior VRM design but i am actually more likely to overclock efficiently then bruteforce method. Now im actually very fine with manual oc all@46 at 1.16v under load
@Taraquin ill try that this evening, thank you. Do you think i am in range with the RAM Volts with that values or do i need to set up? Its set to 1.45v hw info showing 1.47v

greetings

 

[Dar1usElite]

Dar1usElite: Looks good, test 10 rounds then tweak.

it was upto like 7 or something and i went out and came back and the system had rebooted, upped dram voltage to 1.45, it error'd straight away, assuming this was an anomally because i hadnt ran testmem in admin mode, i restarted it and its been fine

 

[tabascosauz]

okay @tabascosauz i guess i stay with the board, do you guys think i run into problems clocking the CPU to 4.8 to 4.9ghz if needed? i know it as an inferior VRM design but i am actually more likely to overclock efficiently then bruteforce method. Now im actually very fine with manual oc all@46 at 1.16v under load
@Taraquin ill try that this evening, thank you. Do you think i am in range with the RAM Volts with that values or do i need to set up? Its set to 1.45v hw info showing 1.47v

greetings

It's not an inferior VRM design. It's the Aorus Pro with a cheaper heatsink and a 4-layer PCB, you could argue that the newer Vishay 50A parts make the Elite a stronger VRM than the Aorus Pro with those dated IR3553s. Granted, ISL69138 is probably not a splendid choice for a controller, but 12 x 50A will handle any OC 5900X just fine. In any case it has a VRM temp sensor so you can always monitor

 

[Valantar]

okay @tabascosauz i guess i stay with the board, do you guys think i run into problems clocking the CPU to 4.8 to 4.9ghz if needed? i know it as an inferior VRM design but i am actually more likely to overclock efficiently then bruteforce method. Now im actually very fine with manual oc all@46 at 1.16v under load
@Taraquin ill try that this evening, thank you. Do you think i am in range with the RAM Volts with that values or do i need to set up? Its set to 1.45v hw info showing 1.47v

greetings

Is all-core OCing on Zen3 really the best idea, performance wise? It's been a while since I was reading up on this, but back then people were reporting better results with a combination of increasing PBO limits and negative curve optimizer to allow for higher boosting (both ST and MT), rather than setting a single-all-core clock (losing ST performance but possibly gaining MT). Of course the methods needed to optimize this are very different and a bit more complicated, but it's still doable.

 

[schidddy]

@tabascosauz sure sorry u can argue like that, my reference where those boards with 16 Phase VRM Designs with PowIRStages, i think ill be happy with also because of the price i paid, got it "new" for 110 bucks

Is all-core OCing on Zen3 really the best idea, performance wise? It's been a while since I was reading up on this, but back then people were reporting better results with a combination of increasing PBO limits and negative curve optimizer to allow for higher boosting (both ST and MT), rather than setting a single-all-core clock (losing ST performance but possibly gaining MT). Of course the methods needed to optimize this are very different and a bit more complicated, but it's still doable.

i did do PBO overclocking, but no matter what i did, when PBO clocked all core to 4.65ghz at PTT210 TDC 145 and EDC 190 the voltages where up to 1.35 to 1.4v taking up to 200 watt powerpackage even i had CurveOptimizer already at -20
If i set the offset voltage too high the cores startet to drift in the clock speeds. i played already with different TDC and EDC Values.
Now i am at 4.6ghz allcore at 1.16v and with just 130watts to max 145
Sure i dont have single core boost now but im okay with that not seeing like 1.5v boost volts just for 10percent more single core performance. ALso because i most likely play in 4k resolutions. (i know ram tuning doesnt make sense at those resolutions but i just like tuning and getting some extra for free. CPU wise ill clock a bit higher later ons but starting on the RAM first)

 

[tabascosauz]

@tabascosauz sure sorry u can argue like that, my reference where those boards with 16 Phase VRM Designs with PowIRStages, i think ill be happy with also because of the price i paid, got it "new" for 110 bucks

Some of those VRMs have good performance in other metrics that make them "good". In terms of raw capacity, all of them are ridiculously overkill x 10.

Even simple 8 phase Vcore VRM with 50A DrMOS and a good heatsink will handle a mildly OC'd 5950X. It's not Intel. You will run out of cooling for your CPU before you run out of VRM.

 

[Valantar]

@tabascosauz sure sorry u can argue like that, my reference where those boards with 16 Phase VRM Designs with PowIRStages, i think ill be happy with also because of the price i paid, got it "new" for 110 bucks


i did do PBO overclocking, but no matter what i did, when PBO clocked all core to 4.65ghz at PTT210 TDC 145 and EDC 190 the voltages where up to 1.35 to 1.4v taking up to 200 watt powerpackage even i had CurveOptimizer already at -20
If i set the offset voltage too high the cores startet to drift in the clock speeds. i played already with different TDC and EDC Values.
Now i am at 4.6ghz allcore at 1.16v and with just 130watts to max 145
Sure i dont have single core boost now but im okay with that not seeing like 1.5v boost volts just for 10percent more single core performance. ALso because i most likely play in 4k resolutions. (i know ram tuning doesnt make sense at those resolutions but i just like tuning and getting some extra for free. CPU wise ill clock a bit higher later ons but starting on the RAM first)

If you want to keep overall power consumption down you can just limit PPT, TDC and EDC, and the CPU should regulate itself within those parameters for optimal performance. It's kind of a given that it will use as much power as it is allowed to when boosting as long as thermals and power delivery can keep up. Also, remember that software-reported voltages for Zen2/3 when boosting are in no way equal to setting a similar voltage manually - the chip regulates the voltages dynamically as needed, and there is a dedicated silicon protection portion of the chip to ensure that these settings don't exceed what AMD considers safe, even when using PBO (it balances thermals, power draw and voltages to minimize leakage and electromigration). Pretty much every reliable source I've seen cover this essentially says "let the chip regulate its own voltages, it will look high but that's normal and safe". If I were you, I would attempt those previous settings with lowered PPT/TDC/EDC and compare them to your current OC results. For my 5800X, I'm currently running -10 all-core CO and +100 PBO at 120/95/120, and I'm seeing 4.95GHz short ST bursts, 4.85+ sustained ST, and 4.7-4.8 sustained MT depending on the load (power viruses might drop to 4.4-4.6, but I don't consider those relevant for real-world performance). Given that 2-CCD CPUs have much lower thermal density than the 5800X you should be able to boost higher at lower power levels as you'll have less leakage current, but of course you'll likely need higher total power limits than me simply due to the higher core count.

 

[schidddy]

@Valantar sure i did that but when i decreased the different PTT TDC and EDC values, i just saw the frequencies dropping and using less power, but the voltages where still at that high ranges. my MT scores are basically the same manual compared to PBO at 210/145/190 and boosting to allcore 4.65ghz but at a power consumption off rougly 210 watts compared 4.6ghz manual oc. I tried lowering all those values, also different to each other but i just saw decreasing frequencies still at very high power consumption. Thats because PBO uses much higher Voltage tables to be sure everyone are getting those frequencies. PBO is some kind of bruteforce attack for me i know i cant trust sensored voltages on Zen3 but i can trust the power/consumption readings at load. R23 gives mit 23k points with manual @46 and pbo gives me like 300 points more on MT. I get 210watts on MT and PBO and 140watts at manual
At ST pbo gives me roughly 10% more performance thats quiet cool but as i said at voltages i dont like to see even AMD says its actually save haha.
Also temperatures are alot lower 60degrees on manual oc then 81 at PBO on full load
Edit, if i sett PTT to something like 150 or 160, my allcore freq drops to 4.3ghz something like that giving me lower r23score and still higher power consumption then manual 46x

 

[Dar1usElite]

@Taraquin no errors until the 10th run where i got 2,unsure where to go from here

@ta

 

[Valantar]

@Valantar sure i did that but when i decreased the different PTT TDC and EDC values, i just saw the frequencies dropping and using less power, but the voltages where still at that high ranges. my MT scores are basically the same manual compared to PBO at 210/145/190 and boosting to allcore 4.65ghz but at a power consumption off rougly 210 watts compared 4.6ghz manual oc. I tried lowering all those values, also different to each other but i just saw decreasing frequencies still at very high power consumption. Thats because PBO uses much higher Voltage tables to be sure everyone are getting those frequencies. PBO is some kind of bruteforce attack for me i know i cant trust sensored voltages on Zen3 but i can trust the power/consumption readings at load. R23 gives mit 23k points with manual @46 and pbo gives me like 300 points more on MT. I get 210watts on MT and PBO and 140watts at manual
At ST pbo gives me roughly 10% more performance thats quiet cool but as i said at voltages i dont like to see even AMD says its actually save haha.
Also temperatures are alot lower 60degrees on manual oc then 81 at PBO on full load
Edit, if i sett PTT to something like 150 or 160, my allcore freq drops to 4.3ghz something like that giving me lower r23score and still higher power consumption then manual 46x

Wow, that's weird. I've seen reports of manual OCs being more efficient at MT (though typically at relatively low clocks), but never this extreme. Sounds like your chip is indeed for some reason going quite overboard on its PBO voltages - makes me wonder whether this is determined by something hard-coded into the chip, its microcode, the motherboard/VRM, or some combination of these. It does indeed sound like your case is one where manual OC makes much more sense.

 

[Ibizadr]

Anyone here With gskill f4-3600c16d-16gtz.? I'm struggle with this ram kit. With 2x8gb with xmp enable give me errors on occt and memtest, so I put only one at time for testing rams and nothing appear to be bad, when I put the second stick throw errors, system runs fine without whea errors or reboots. The cause of this its my mobo? I'm with last agesa 1.2.0.3c and the previous kit at 3200 runs fine at 3333 without errors and this kit give errors even with xmp enable. Can somebody have a answer for this weird thing?

 

[Deleted member 215115]

@Taraquin no errors until the 10th run where i got 2,unsure where to go from here
View attachment 224352@ta

DDR4/5 Helper by Veii

docs.google.com

There's a TM5 Error Description sheet in this document. Focus on the 0 error first.

Anyone here With gskill f4-3600c16d-16gtz.? I'm struggle with this ram kit. With 2x8gb with xmp enable give me errors on occt and memtest, so I put only one at time for testing rams and nothing appear to be bad, when I put the second stick throw errors, system runs fine without whea errors or reboots. The cause of this its my mobo? I'm with last agesa 1.2.0.3c and the previous kit at 3200 runs fine at 3333 without errors and this kit give errors even with xmp enable. Can somebody have a answer for this weird thing?

It is not the motherboard. Can you post a screenshot of the latest ZenTimings with both modules inserted and XMP on?

ZenTimings

ZenTimings is a simple and lightweight app for monitoring memory timings on Ryzen platform.

zentimings.protonrom.com

 

[Ibizadr]

DDR4/5 Helper by Veii

docs.google.com

There's a TM5 Error Description sheet in this document. Focus on the 0 error first.


It is not the motherboard. Can you post a screenshot of the latest ZenTimings with both modules inserted and XMP on?

ZenTimings

ZenTimings is a simple and lightweight app for monitoring memory timings on Ryzen platform.

zentimings.protonrom.com

This is what a have now only change voltage trfc and trc. if you want original xmp i can give you too. i raise voltage to see if my errors passed aways but no lucky with more vdimm voltage

 

[Deleted member 215115]

This is what a have now only change voltage trfc and trc. if you want original xmp i can give you too. i raise voltage to see if my errors passed aways but no lucky with more vdimm voltage

Timings are ok. Try setting VSOC to 1.1V manually, VDDG IOD to 1.05V and VDDG CCD to 1V. If there's a setting for CLDO VDDP, raise that to 1.1V

 

[Ibizadr]

Timings are ok. Try setting VSOC to 1.1V manually, VDDG IOD to 1.05V and VDDG CCD to 1V. If there's a setting for CLDO VDDP, raise that to 1.1V

I try that and I will do some tests to see if it works

Timings are ok. Try setting VSOC to 1.1V manually, VDDG IOD to 1.05V and VDDG CCD to 1V. If there's a setting for CLDO VDDP, raise that to 1.1V

After 1min of occt memory test throw errors

 

[Taraquin]

okay @tabascosauz i guess i stay with the board, do you guys think i run into problems clocking the CPU to 4.8 to 4.9ghz if needed? i know it as an inferior VRM design but i am actually more likely to overclock efficiently then bruteforce method. Now im actually very fine with manual oc all@46 at 1.16v under load
@Taraquin ill try that this evening, thank you. Do you think i am in range with the RAM Volts with that values or do i need to set up? Its set to 1.45v hw info showing 1.47v

greetings

1.45V is good for my suggestions

@Taraquin no errors until the 10th run where i got 2,unsure where to go from here
View attachment 224352@ta

Error 0 is usually too low voltage on ram. Try upping by 0.02V, usually fixes it. 15 can be several things, tRFC maybe, but that can often be improved by a bit more volt.

Anyone here With gskill f4-3600c16d-16gtz.? I'm struggle with this ram kit. With 2x8gb with xmp enable give me errors on occt and memtest, so I put only one at time for testing rams and nothing appear to be bad, when I put the second stick throw errors, system runs fine without whea errors or reboots. The cause of this its my mobo? I'm with last agesa 1.2.0.3c and the previous kit at 3200 runs fine at 3333 without errors and this kit give errors even with xmp enable. Can somebody have a answer for this weird thing?

Is the ram in the A2 and B2-slot? Usually this is a problem uf A1-B1 is used as they are the worst slots.