it turned out that my issue was the PSU, not the board.
Nice that you found the problem in your case
That is an opinion. Not a very good one. You too are referencing a list made by some dude who has no credibility.
Yes and that is also an opinion or did you prove that this list is wrong? It's easy to request proof from others but not providing any by yourself.
Actually I didn't claim, that one of the posted boards is not able to run a Ryzen 3900X. I only gave some advice which of these boards, in question their VRMs, is more or less capable to run a Ryzen 3900X.
Sure every board which has the Ryzen 3900X on its support list will run this CPU, no question at all, but the other question is, on which of these boards will it run better or let's say does it has more headroom. And regarding this point everyone can prove this statement.
1, Regarding number of phases:
The VRM must be capable of handling the current load. That can be done with any of those solutions out there. You can do it via fake phases, that is every phase has its components twice, or even three times like Asus did in their latest X570 boards, or you can do it via doublers or do it with real phases. All of these solutions make the VRM capable of handling more current and not getting too hot. But current handling is only one problem here. Another is a smooth voltage output and less ripple. This problem cannot be solved with twin, triple or doubled phases but with a higher number of real phases or with a higher switching PWM controller. As higher switching decreases the efficiency of the VRM the solution to add more phases is better here. So ideally your VRM has for example an 8-phase capable PWM controller and you have 8 real phases. If one of those phases are discharging another of the 8 can set in with an offset and voltage drop gets smoothed. With 4 fake phases or even less and twice components there are only 4 or 3 in the game of charging/discharging and with 4 doubled phases, the PWM switching freq gets halved which downgrades voltage output also and makes transient response worse.
But if your VRM is too oversized for the daily load, the efficiency will also be a little worse. So it has to match the used cpu and if you are planning to oc, but we are talking here of an Ryzen 3900X, so not the smallest AMD cpu.
2, Regarding the used MOSFETs
Every VRM consists of High-Side and Low-Side MOSFETs and they both have a specific max current capability. Some boards have separate chips other boards have one chip as package of High/Low-Side MOSFET included, others include also the Gate Driver. Each MOSFET High/Low-Side has to handle a portion of the current load. How much determines the duty cycle of the PWM signal. The smaller the duty cycle the more current have to handle the Low-Side MOSFETs. As the duty cycle calculates to Vout/Vin you see that the less the cpu output voltage for a given input voltage is the smaller is the duty cycle and the more current have to handle the Low-Side MOSFETs. So they must be the better chips in the VRM or those chips that have the better current capability compared to the High-Side MOSFETs. So you can approximately calculate the VRMs current capability in counting the number of Low-Side FETs and multiply it with their current capability.
Additionally the power losses in the MOSFETs result in increasing temps. Power Losses are mainly conduction losses and switching losses. First is dependent of the MOSFETs Drain to Source resistance, can be found in their data sheets, second depends of the switching freq. Both result in increasing temps and the amount of heat, the MOSFET rises per W power loss, depends of the MOSFETs thermal resistance, can alo be found in their data sheets. The hotter the MOSFET the less its current handling capability.
So the less Drain to Source resistance, the less thermal resistance and the less switching loss, the less heat is produced and the better the efficiency and current capability.
3, The Boards
If you look for example at these posted boards:
- Asrock Fatality B450 Gaming ITX
- Asrock Fatality X470 Gaming ITX
- Gigabyte B450I Aorus
- Asus ROG STRIX B450-I Gaming
- Asus ROG STRIX X470-I Gaming
- MSI B450-I Gaming Plus
The Asrock boards:
Both have 6 fake phases, that is only 3 real phases and every of these 3 phases has its components twice. According to a list from Hardwareluxx, the B450 Gaming uses for High/Low-Side MOSFETs either the Onsemi FDPC5030 or the Sinopower SM7341EH . I think this is dependent of the board's revision. Assuming they use both the Onsemi chip, this chip consists of one package with High-Side and Low Side MOSFET included. Here is the data sheet:
The Ryzen 3900X draws 100A to 140A (Spikes) with stock setting, no oc here. At 100A every stage (3x2) has to handle 100/6=16,7A. With spikes up to 140/6=23.3A. With the data from the data sheet you can calculate the power losses in the MOSFETs (mainly conduction losses, switching losses and charging losses) to approximately (assuming 12V input/1.42V output, 300kHz switching freq):
Power Loss@16,7A=1.14W
Power Loss@23,3A=2.1W
According the data sheet, the chip has a thermal resistance junction to ambient of 55 degrees C/W, that is 1W power loss in the MOSFET results in 55 degrees celsius heat against ambient. So if ambient is about 25 degrees the junction temp will result in 25+55=80 degrees celsius.
1.14W power loss corresponds to 1.14x55=62.7C
2.1W power loss results in 2.1x55=115.5C
At an ambient of 25 degrees it will result in the first case to 62.7+25=87.7C and in the second case to 115.5+25=140.5C junction temp. The max junction temp of this chip is 150 degrees, so a heatsink is definitelyx needed to cool the MOSFETs down.
The Gigabyte B450I Aorus:
Gigabyte uses 4 real phases with IR3556 stages. The IR is a package with Low and High Side MOSFET and gate driver included. It's a 50A stage and as they don't use doubled or twin phases each phase has to handle 100/4=25A up to 140/4=35A.
Fortunately they have a power loss graph of the IR3551 as there is no data sheet of the 3556, but both are IR 50A stages.
The thermal resistance junction to ambient is here 21.1 C/W, that's way better than the Asrock stages thermal resistance. The power loss at 25A is at approximately 2.5W, so at 25 degrees ambient it would result in 2.5x21.1+25=77.75 degrees junction temp.
The power loss at 35A is at 4.5W, so 4.5x21.1+25=119.95C degrees junction temp. So although it only uses 4 power stages here it stays cooler than the two Asrock boards because of the thermal resistance of the used IR stages. Additional those IR stages have the Gate Driver included, which dissipates additional power loss, so they are very efficient.
The Asus B450I Gaming board:
6 real phases of IR3556 stages, the same Gigabyte uses but now 2 more. Each phase has to handle 16.7A up to 23.3A, the same as Asrock here.
Power loss at 16.7A at about 1.6W result in 1.6x21.1+25=58.76 degreees junction. Power loss at 23.3A approx 2.3W result in 2.3*21.1+25=73.53 degrees junction temp, so the best far now.
The Asus X470 board:
The same VRM as the B450I board.
The MSI B450I Gaming:
6 real phases of IR3555 power stages. Those are 60A power stages.
As there is no data sheet of the IR3555 here is one of the IR3550, also a 60A IR stage:
Thermal resistance junction to ambient is here 20.2 C/W. The power loss at 16.7A is at approx 1.3W results in 1.3x20.2+25=51.26 degrees junction temp. Power Loss at 23.3A is at 2W, results in 2x20.2+25=65.4 degrees junction.
All those temps are without heatsink, the chips mounted on a 1 in2 pad of 2 oz copper, as stated in their data sheets to get it comaprable. With heatsinks all of these boards are capable to handle a stock Ryzen 3900X, but nevertheless you can make a rating which VRM will be more or less capable or efficient, and this is my personal rating from top to bottom:
1, MSI B450I Gaming
2, The two Asus boards
3, The Gigabyte board
4, The two Asrock boards
And I would stay away from the Asrock boards if I had to choose a new B450/X470 board and if I have a Ryzen 3900X/3950X to put in. If I would own such a board already it will do fine also, no question, but with less headroom than the others.