Do you have the ability to monitor the wattage being used? Is it really operating within a 65w envelope? I had a 3700x in my 2U server, and had to drop down to the 3600x as I was seeing some pretty high numbers regardless of what was advertised.
If true, this would be a very interesting CPU, and I would likely pick one up.
You do know that TDP has nothing to do with power draw these days anymore, right?
Just want to see the heat output under regular load, but yes you bring up a good point. I know I can already assume that if the 3700X generated too much heat, that the Pro 3900 is likely out as well. But maybe with the reduced clocks, one can dream 
Multiply AMDās TDP value with a factor of about 1.3 and youāll get the amount of watts a CPU is allowed to draw permanently as long as temperature and VRMs play along.
This value (PPT) is 88 W with the 3900 PRO, just like the 3700X.
The Eco Mode is nothing more than lowering the PPT value, you can also do this manually with Precision Boost Overdrive 2 on the X470D4U.
Increasing PPT > 88 W doesnāt have any effect on the X470D4U (in theory more multi-core performance (CB R15 MC around 2900), just like a regular 3900X at 105 TDP), I donāt know if this is a fixed limitation by the PRO SKU or just a bug of ASRock Rackās BIOS.
Lowering PPT on the other hand works as expected, although the 3900 PRO with its default package power of 88 W is pretty close to the siliconās sweet spot; 75 W leads to about 10 % lower CB R15 MC scores, 65 W about 30 % less.
Single-Core performance (CB R15 SC: 205) isnāt really affected by this (the 3900 PROās maximum single core boost is 4.35 GHz), with PBO I saw 4.45 GHz in HWiNFO shortly.
But during all these tests the CPU package power never exceeded 88 Watts (HWiNFO displays 90 Watts, I guess the motherboardās software sensors arenāt the most exact kind).
If you could use negative offset voltage you might be able to squeeze more performance out of the 3900 PRO
Thanks for the detailed response. That answers my question.
Iām happy to report success - I got the 2x16GB Samsung M378A2K43CB1-CTD DIMMs today. They work just fine with the R5 2600 both on the Asus B450M-A & the ASRock X470D4U. 
As a side note - Iāve had the board running continuously for about 2 weeks, and save for time installing & swapping components itās been running 24/7 with a 2200G, and in the last hour an R5 2600.
I havenāt seen the CPU_PROCHOT error or any throttling at all yet & I too have the issue with not being able to read the SPD memory information in the OS with two sets of DIMMs. IPMI has been great throughout aswell even when the system was locking up with the Crucial DIMMs & 2600.
For anyone who doesnāt want to read that much:
You can achieve the same as āEco Modeā by enabling Precision Boost Overdrive in the āAMD Overclockingā menu of the X470D4U.
The key is PPT (Package Power Tracking), a value in Watts that limits the total power that goes to the CPU socket.
Default PPT values by comparison to TDP:
AMD 65 W TDP (3600, 3700X, 3900 PRO) = 88 W PPT
AMD 105 W TDP (3900X, 3950X) = 142 W PPT
Donāt know the PPT for 95 W TDP (3600X), never had that SKU go through my hands.
Can anyone check if you can manually actually increase the PPT above what your SKU defaults to?
As mentioned in my previous posting, with the 3900 PRO it seems locked to 88 W and I would like to know if this a general limitation of that model or if the ASRock Rack X470D4U is just broken in this regard, too.
Thanks!
Could you make a screenshot of HWiNFO?
Posting was a brain fart.
I think you mis-read my post. 
I canāt read the SPD memory details either in the OS. Iāve had the same issue with the new Samsung M378A2K43CB1-CTD DIMMs & the Crucial DIMMs I had with the 2200G.
This is with BMC Firmware 1.90 & BIOS Version 3.30 (Newest of both).
Iām running the 3600X in my 4U2-2T board right now, where can I pull that info for you?
Just install the Ryzen Master Software, after launching you see various stats for the CPU, including current PPT, default AMD values if you havenāt toyed around with PBO.
isnt that windoze only?
Yes, I donāt know about Linux hardware monitoring tools. But you can indirectly find out your current PPT by making a āsimpleā multi-core load for a minute where the CPU isnāt throttled by temperature.
In HWiNFO this value is āCPU Package Powerā.
Yeah, will take me a while in that case. This is my unraid server, so to do this would involve loading windows temporarily (since a VM likely wonāt do).
Another observation - in the AMD PBS section of the BIOS, thereās an option to split the PCIe x16 slot into 1x16, 2x8 or 4x4.
This was completely absent for me with the 2200G, but itās there with the 2600. Kind of makes sense I guess, given I think the 2200G only has 8 lanes to play with.
I just got this board yesterday, and Iāll admit Iām rather new to all of this and I probably am lacking a lot of knowledge everyone else has, but Iāve been having issues with this board. Iām just trying to do some sanity checking right now because I was buying this board with a 2700x to replace a proxmox box I have running couple VMs and one with GPU passthrough. I was reading GPU pass through is broken for everyone on the current agesa version but Iām trying to figure out if that is whatās causing the issues or if I have another problem. Iāve used the gpu before on proxmox that Iām trying to pass through without issues, even rebooting and shutting off the VM without getting caught by a reset bug, but when I attempt to pass through the gpu and start the vm with it the VM never starts, and after that proxmox gets stuck shutting down so I have no choice but to do a hard shutdown.
Iām also pretty new to the IPMI stuff and Iām not sure if the errors that I was getting were something to just ignore or if something is faulty. I got the bus degraded a couple of times when I first added some PCI-e cards and it seems if I have any card in at all Iām getting a system firmware progress error. I have no idea what either of those mean.
i just wish rack had some full ATX or even Eatx boards for am4, especially with more SATA on the mobo like asrock used to do
I wish upon a star that ASRock Rack would actually fix bugs reported close to nine months ago - maybe it can become a real motherboard then
The only thing (they) fixed seems to be the CPU_PROCHOT throttling with the BMC/IPMI update 1.90.
The other stuff was pre-packaged by AMD (AGESA) and slapped on the existing BIOS. 
BTW: Is there any way to PCIe-passthrough a device in ESXi that is connected via the X470 chipset? Iāve failed so far (Intel I210 NICs, onboard SATA, onboard USB). CPU-PCIe-connected stuff in the slots works fine.
Whats the best pcie card to add 4 more m.2 nvme ssds for the X470D4U mobo? It seems like its better to add a 10G eth card separately vs buying the the X470D4U2-2T because of the way the PCI is linked. Has anyone tried getting both to fit in a 1u case?
I just picked up an IO crest 2x U.2 pcie bifurication adapter on amazon.
I havent used it yetā¦ but might need it in the future.