Asrock X570D4U-2L2T Build

[NateS]

I could never get it to work in either m.2 slot; tried a couple different brands of adapter and switched the bios settings for the m.2 from nvme to sata but made no difference; these adapters aren't passing sata through or there's some special board setting that's missing. I've never seen the adapter Asrock mentioned for sale anywhere either.

It won't be a bios setting, and since your particular adapter is designed for NVMe, it won't be able to do it without hardware modifications. M.2 ports determine what mode they run in based on particular pins being tied to either VCC or ground, so any passive adapter that can do either mode would need to have jumpers or dip switches, and I don't know of any that do.

Theoretically, if you're good with a soldering iron, you could cut a trace and add a jumper on the adapter to modify an NVMe adapter into a SATA one. Personally I'm not sure if it's worth the effort since you're only getting one more SATA port out of it, but it's an option. See this document for which pin you'd need to modify if you want to try going that route: https://www.congatec.com/fileadmin/...Pinout_Descriptions_and_Reference_Designs.pdf

 

[homelab]

No worries, glad it's working well.

Yea I spent so much time looking at cooler options and configurations, but in the end I settled with the Dynatron A24 and it's built in fan in all 3 of my servers. I did consider push/pull and I can't remember for sure but I don't think it would fit without some bodging.

In the end I was quite impressed with the Dynatron fan to the point where I think it will offer better cooling power due to it's thickness, shroud and pressure so I stuck with that and it's working well.

I do have a custom fan curve setup though, that's definitely advisable if you want it quiet as the Dynatron fan will act as if it's in a datacenter by default and run at very high RPM. My small server rack lives in a cupboard so some noise is fine, but not super loud!

Also for anyone interested, AGESA ComboAM4v2 1.2.0.2 is currently being developed for the board and is available in their BETA bios section now. Once that's updated, I think ClockTuner would be worth another shot to see if we can get those improved efficiencies!

Thanks for the warning.

The server rack is unfortunately in the home office so it has to be quiet.

I've already been told the 2x 80mm & 1x 120mm noctuas running at about 1000RPM when in the BIOS was way too loud so it could be a fun.

would you be willing to share your fan curve & max temps?

 

[petersnows]

Server Bare Bone 1U4LW-X570/2L2T 699€

ASRock Rack > 1U4LW-X570/2L2T

www.asrockrack.com

MotherBoard X570D4U-2L2T (included on the server barebone)

ASRock Rack > X570D4U-2L2T

www.asrockrack.com

CPU: AMD Ryzen 9 5900X 12-Core 3.7GHz c/ Turbo 4.8GHz 70MB SktAM4 453€
Ram: 4x 32Gb KSM32ED8/32ME (4x155 = 620€)
Disk: 1TB WD SN850 138€
OS: Windows 2019 server (used some of the windows 10 drivers (installed win 10 and got the drivers backup))

Total : 1911€ + VAT

So far works great.
Fans are loud, but the server is in the server room.

 

[petersnows]

Server Bare Bone 1U4LW-X570/2L2T 699€

ASRock Rack > 1U4LW-X570/2L2T

www.asrockrack.com

MotherBoard X570D4U-2L2T (included on the server barebone)

ASRock Rack > X570D4U-2L2T

www.asrockrack.com

CPU: AMD Ryzen 9 5900X 12-Core 3.7GHz c/ Turbo 4.8GHz 70MB SktAM4 453€
Ram: 4x 32Gb KSM32ED8/32ME (4x155 = 620€)
Disk: 1TB WD SN850 138€
OS: Windows 2019 server (used some of the windows 10 drivers (installed win 10 and got the drivers backup))

Total : 1911€ + VAT

So far works great.
Fans are loud, but the server is in the server room.

 

[Shadowsong]

Thanks for the warning.

The server rack is unfortunately in the home office so it has to be quiet.

I've already been told the 2x 80mm & 1x 120mm noctuas running at about 1000RPM when in the BIOS was way too loud so it could be a fun.

would you be willing to share your fan curve & max temps?

Sorry for the delay, if your still interested this is what I currently have 3 servers fan curve set to which is quiet enough for me. But I expect It could be pushed further.



As for max temps, it varies a considerable amount depending on if I have the AC turned on and server load. Also going from Balanced to High Performance adds significantly more heat due to the clock speeds, however I did not see the performance benefit even for my game servers. So I leave it at balanced and let it clock up to max on the cores that It needs to.

 

[t3k]

Guys, as @Shadowsong told us Bios 1.40 is available since the 12th of August ! We have now Agesa 1.2.0.2, and Performance Boost Override and Curve Optimizer with advanced settings are available in the Bios ! \o/

So I updated and told myself it was a good time to start playing with CPU fine tuning to try to max out the watt / performance ratio...

I followed two ways : the first one is PBO + CO fine tuning, and the second one, CTR 2.1 RC5

PBO + CO fine tuning :

More information here : https://www.reddit.com/r/overclocking/comments/octeuh
The main advantage is that you still have boost enabled for one threaded applications with this configuration, so you can achieve between 4800 and 5100 Mhz boost depending on your proc and cooling system

It's much longer to fine tune than CTR which makes it all automatic for you, here are my results :

Fine tuning PBO and CO following the video tutorial :

PPT 200​

TDC 150​

EDC 150​

Curve Optimizer set to -20 on all core​

I didn't take the time to test which core was failing at -25, as the watt consumption with these parameters didn't satisfy me for my usage (335 Watts on Cinebench R20 benchmark)

CTR 2.1 RC5 :

This soft is so easy to use ! In a matter of seconds it is able to tell you the optimized frequencies for each CCX depending on the voltage you use for each profile. If your workload works on WIndows, you can let the software start at boot and automatically switching profiles depending on your workload, allowing you to boost for monothreaded applications or optimizing efficiency when all your cores are working on multithreading apps.

My workload is containerization on Linux, so I will just use the P1 multithreaded profiles and put the values in BIOS.

Here are the synthetic results :

Watts measurements are taken from the computer plug with this tool (https://www.amazon.fr/54365-Compteu...1&keywords=mesure+watt&qid=1632214282&sr=8-16)

For CTR results, Voltage and CCX frequency are calculated by the soft

Configuration	Voltage (mA)	CCX1 Frequency (Mhz)	CCX2 Frequency (Mhz)	Cinebench R20 Score	Watt on 100% CPU
Stock config PBO enabled no advanced config	(max voltage seen with HWINFO) 1387	4400	4400	11057	340
Stock config PBO with advanced config	forgot to check sorry	forgot to check sorry	forgot to check sorry	11300	335
Stock config PBO disabled	1125	3930	3930	10115	203
CTR Undervolt 1	1000	3900	3825	9982	176
CTR Undervolt 2	1050	4100	4000	10445	196
CTR Undervolt 3	1100	4250	4150	10840	215
CTR small overclock 1	1150	4425	4300	11250	238
CTR overclock 2	1200	4550	4400	11535	250
CTR overclock 3	1250	4625	4475	11740	290

The next try at 1300 mA, 4700 Mhz CCX1 - 4550 Mhz CCX2 was not successfull, error popped out from Cinebench at first run.

Looking at the number, I decided that overclock 1 was a good basis for working on final values and putting them in BIOS. Sadly the Bios does not give an option to put different values for each CCX, so you've got to put the same Frequency for both.

At 1150 mA, 4425 Mhz was not stable in Cinebench, error popped out fast enough, like at the second or third run. First run still had 11311 points

At 1150 mA, 4375 Mhz, error popped out around the 15th pass. 11230 points though

At 1150 mA, 4350 Mhz, no errors after half an hour of cinebench ! \o/ 11114 points on Cinebench R20 and 28649 on Cinebench R23, Watts are around 230 on full load, temperatures don't go over 70°, I'm happy with these numbers IMO it's still a lot better than stock with or without PBO activated !

If you want to optimize your RIG and your OS is windows, update to the last BIOS and use CTR, for me it is a no brainer. If your OS is Linux, make your choice between boost with PBO or stable frequencies for multi threading !

Happy under/overclocking

 

[Shadowsong]

@t3k Thanks for sharing the results, i've still yet to do mine as was waiting for the BIOS to come out of beta which it now is. Are those configurations provided by the software after the diagnostic or just what you put together for testing? I've followed Clocktuner for a long time but yet to use it as these are my only ryzen systems.

I'm likely going to try and run stock settings with an undervolt if possible, stability is still the priority followed by efficiency.

 

[t3k]

CTR will create profiles P1 and P2 with different voltage values, but you can modify them and click on calculate and it will give you two new frequencies values for CCX1 and 2.

Here is the youtube video that 1usmus did to explain the process :

If undervolt is your priority I think you should be able to lower voltage at least 100 mV and get stable and more efficient usage, if you get a good CPU

Btw mine was rated gold by CTR I hope you will have similar luck (or better) at the silicon lottery !

 

[Shadowsong]

CTR will create profiles P1 and P2 with different voltage values, but you can modify them and click on calculate and it will give you two new frequencies values for CCX1 and 2.

Here is the youtube video that 1usmus did to explain the process :

If undervolt is your priority I think you should be able to lower voltage at least 100 mV and get stable and more efficient usage, if you get a good CPU

Btw mine was rated gold by CTR I hope you will have similar luck (or better) at the silicon lottery !

Thanks that video was ideal! Just checked the software and it seems to run fine on Server 2019 so I will get one of my servers upgraded and tuned, then test out the other 2 once i'm comfortable with the process. Looks very straight forward though, just needs the time to run the diagnostic to evaluate the system.

 

[Shadowsong]

@t3k Not sure if you have looked into it yet, but I was also considering using Hydra which is the version of the tool dedicated to the 5000 series - HYDRA 1.0A PRO COMING SOON!!! | Yuri Bubliy on Patreon

 

[t3k]

@t3k Not sure if you have looked into it yet, but I was also considering using Hydra which is the version of the tool dedicated to the 5000 series - HYDRA 1.0A PRO COMING SOON!!! | Yuri Bubliy on Patreon

Version of which tool ? CTR ? 2.1 RC5 is completely compatible with 5000 series

Well maybe you will have better results with this tool, I'm eager to read what underclock you will achieve !

I'm fine with my numbers haha

 

[Shadowsong]

Version of which tool ? CTR ? 2.1 RC5 is completely compatible with 5000 series

Well maybe you will have better results with this tool, I'm eager to read what underclock you will achieve !

I'm fine with my numbers haha

Yea it's a version of ClockTuner dedicated for the 5000 series, though the normal CTR is also compatible of course I will let you know the results I get, just need to get my current workloads moved off one of the servers and then I will take it down and test this out!

 

[Shadowsong]

Finally started testing on one of my servers, tried CRT and also tried Hydra both of which worked pretty well. Though Hydra definitely did a better job with the underclock as there is a nice preset for it.

However while the tuning has given me some useful info on voltages for certain clock speeds, it doesn't seem to have helped overall. The main issue i'm finding is that the CPU boost on the 5950X is just crazy and adds imo far to much voltage creating excess heat. This is less of an issue in balanced power mode, but in high performance it is insane. Roughly a 20 degree difference in temps on the same workload.

I need high frequencies for the workload I have, even though the load overall is quite low. So the only option I can come up with so far is to set a manual voltage and frequency in BIOS with the CPU boost turned off which gives me better performance over the Balanced plan (3.4Ghz vs 4Ghz+) but with good temps.

Has anyone else experienced this sort of situation? I think ultimately it comes from the 5950X not being a server chip but still would hope for an easier way to manage it then manually setting voltage and frequency!

 

[Shadowsong]

@t3k Did you end up seting your voltage/frequency in the BIOS in the end and if so what did you end up going with? Looks like I will need to do the same if I want a constant high frequency with sensible temps!

 

[deepred]

Has anyone else experienced this sort of situation? I think ultimately it comes from the 5950X not being a server chip but still would hope for an easier way to manage it then manually setting voltage and frequency!

I am using 5950X in Proxmox 7.0 with stock settings so I can't comment on Windows power plans. With Noctua L9x65 5950X is barely kept cool in a range between 85-90 Celsius when turboing for some time. The CPU frequency fluctuates around 4.1-4.4 GHz at these times. I am a bit space constrained so I use a small mATX case but case fans help a lot. The CPU voltage gets up to (uncomfortable) 1.46V according to IPMI. Please keep in mind that I don't have AVX intensive workloads, just a bunch of service containers and development VMs. I guess in prolonged AVX-heavy tasks it should get only worse. I don't know if that's a normal behavior or some BIOS/AGESA bug. CPU's 1.46V would be an instant death for my 14nm LGA-2011-v3 CPUs.
I tried enabling Eco Mode in order to lower temperatures and voltages but the CPU became quite unstable. The symptoms looked like RAM issues, e.g. random reboots and freezes in normal scenarios, even though Memtest86+ was OK. Currently CPU is at stock settings and I just increased case fans speed so CPU cooler always has enough fresh air.

@All, what's the maximum DDR4 frequency you're getting with 4x dual-rank 32GB ECC (128GB in total)? According to the manual it should work on 2666 and indeed it's fine and stable. But if I increase it even slightly, say, to 2933 or even 3200 it becomes unstable. Sometimes it would work for a half-day then freeze on reboot or in the middle of some VM installation. I am using non-QVL Micron MTA18ASF4G72AZ-3G2B1 modules. I checked compatibility on Micron site and while they don't have Asrock Rack as a vendor these modules were marked as 100% compatible with workstation-class X570 motherboards like ASUS Pro WS X570-ACE that are known to have a working ECC support. Thank you in advance.

 

[Shadowsong]

I am using 5950X in Proxmox 7.0 with stock settings so I can't comment on Windows power plans. With Noctua L9x65 5950X is barely kept cool in a range between 85-90 Celsius when turboing for some time. The CPU frequency fluctuates around 4.1-4.4 GHz at these times. I am a bit space constrained so I use a small mATX case but case fans help a lot. The CPU voltage gets up to (uncomfortable) 1.46V according to IPMI. Please keep in mind that I don't have AVX intensive workloads, just a bunch of service containers and development VMs. I guess in prolonged AVX-heavy tasks it should get only worse. I don't know if that's a normal behavior or some BIOS/AGESA bug. CPU's 1.46V would be an instant death for my 14nm LGA-2011-v3 CPUs.
I tried enabling Eco Mode in order to lower temperatures and voltages but the CPU became quite unstable. The symptoms looked like RAM issues, e.g. random reboots and freezes in normal scenarios, even though Memtest86+ was OK. Currently CPU is at stock settings and I just increased case fans speed so CPU cooler always has enough fresh air.

@All, what's the maximum DDR4 frequency you're getting with 4x dual-rank 32GB ECC (128GB in total)? According to the manual it should work on 2666 and indeed it's fine and stable. But if I increase it even slightly, say, to 2933 or even 3200 it becomes unstable. Sometimes it would work for a half-day then freeze on reboot or in the middle of some VM installation. I am using non-QVL Micron MTA18ASF4G72AZ-3G2B1 modules. I checked compatibility on Micron site and while they don't have Asrock Rack as a vendor these modules were marked as 100% compatible with workstation-class X570 motherboards like ASUS Pro WS X570-ACE that are known to have a working ECC support. Thank you in advance.

Glad I’m not the only one, currently for my 3 servers the plan is to run a static clock at 4225Mhz and 1180mv set in the BIOS. Keeps It super responsive and runs significantly cooler! Could even run the 4.4Ghz and still be far cooler than stock with boost….

As for memory I run 32GB modules at 3200Mhz but I am running the Kingston 3200Mhz rated chips.Not had any issues with all 128GB at that speed, also got the infinity fabric set to 1600Mhz as well to match.

 

[Borromini]

My server has the 3700X capped at 65W, my workstation I think I manually configured a voltage offset after testing (been a while now though).

@EffrafaxOfWug Would you mind sharing how you capped it? I have looked in the BIOS but apart from switching power profile under the Advanced > AMD CBS submenu, and disabling CPS (Core Performance Boost) I haven't found any such option in the BIOS. Running Linux on the board.

 

[Shadowsong]

@EffrafaxOfWug Would you mind sharing how you capped it? I have looked in the BIOS but apart from switching power profile under the Advanced > AMD CBS submenu, and disabling CPS (Core Performance Boost) I haven't found any such option in the BIOS. Running Linux on the board.

You can use ECO mode to cap at 65w, but the option is a bit buggy as it doesn’t always show. For me I had to reset the BIOS to defaults then go back in to the AMD overclocking area and it was listed.

But from my testing it still made huge voltages (same highs as stock) but the temps were a bit more manageable. I’d suggest a static clock if you want something slightly better than just disabling core performance boost.

 

[Borromini]

Thanks!

 

[llowrey]

I've been chasing stability issues with this board. I have a 5950x and (2) nemix 32GB ECC DIMMs but had experienced stability issues with a different pair of 16GB ECC DIMMs as well. I'm starting to wonder if it's the board or the CPU. Anyway, I don't want to hijack with that but thought y'all would be interested to see ECC errors as reported by linux (Fedora 35, kernel 5.14).

[307854.154669] mce: [Hardware Error]: Machine check events logged
[307854.160652] [Hardware Error]: Deferred error, no action required.
[307854.166850] [Hardware Error]: CPU:0 (19:21:0) MC18_STATUS[Over|-|MiscV|AddrV|-|-|SyndV|UECC|Deferred|-|Scrub]: 0xdc2031000000011b
[307854.178649] [Hardware Error]: Error Addr: 0x000000044e60c340
[307854.184406] [Hardware Error]: IPID: 0x0000009600150f00, Syndrome: 0x565e00100b800003
[307854.192251] [Hardware Error]: Unified Memory Controller Ext. Error Code: 0, DRAM ECC error.
[307854.200714] EDAC MC0: 1 UE Cannot decode normalized address on mc#0csrow#3channel#1 (csrow:3 channel:1 page:0x0 offset:0x0 grain:64)
[307854.212728] [Hardware Error]: cache level: L3/GEN, tx: GEN, mem-tx: RD
[307854.220958] mce: [Hardware Error]: Machine check events logged
[307854.226878] [Hardware Error]: Deferred error, no action required.
[307854.233061] [Hardware Error]: CPU:0 (19:21:0) MC18_STATUS[Over|-|MiscV|AddrV|-|-|SyndV|UECC|Deferred|-|Scrub]: 0xdc2031000000011b
[307854.244802] [Hardware Error]: Error Addr: 0x000000044e60d840
[307854.250556] [Hardware Error]: IPID: 0x0000009600150f00, Syndrome: 0x25ca01010b800003
[307854.258387] [Hardware Error]: Unified Memory Controller Ext. Error Code: 0, DRAM ECC error.
[307854.266830] EDAC MC0: 1 UE Cannot decode normalized address on mc#0csrow#3channel#1 (csrow:3 channel:1 page:0x0 offset:0x0 grain:64)
[307854.278828] [Hardware Error]: cache level: L3/GEN, tx: GEN, mem-tx: RD
[307854.285443] [Hardware Error]: Deferred error, no action required.
[307854.291623] [Hardware Error]: CPU:0 (19:21:0) MC18_STATUS[Over|-|MiscV|AddrV|-|-|SyndV|UECC|Deferred|-|Scrub]: 0xdc2031000000011b
[307854.303362] [Hardware Error]: Error Addr: 0x000000044e6bdc00
[307854.309111] [Hardware Error]: IPID: 0x0000009600150f00, Syndrome: 0x9d1e00000b800003
[307854.316938] [Hardware Error]: Unified Memory Controller Ext. Error Code: 0, DRAM ECC error.
[307854.325375] EDAC MC0: 1 UE Cannot decode normalized address on mc#0csrow#3channel#1 (csrow:3 channel:1 page:0x0 offset:0x0 grain:64)
[307854.337372] [Hardware Error]: cache level: L3/GEN, tx: GEN, mem-tx: RD

# edac-util -v
mc0: 0 Uncorrected Errors with no DIMM info
mc0: 0 Corrected Errors with no DIMM info
mc0: csrow2: 0 Uncorrected Errors
mc0: csrow2: mc#0csrow#2channel#0: 0 Corrected Errors
mc0: csrow2: mc#0csrow#2channel#1: 0 Corrected Errors
mc0: csrow3: 3 Uncorrected Errors
mc0: csrow3: mc#0csrow#3channel#0: 0 Corrected Errors
mc0: csrow3: mc#0csrow#3channel#1: 0 Corrected Errors

I've been running AMD desktop CPUs with ECC RAM for a long time but this is the first time I've actually seen errors.