Nobody can tell you what is wrong with your rig from a distance. What they can give you is an order to work in. Almost every fault on a multi-card rig falls into one of four groups, and each one has a different sensible first move.
No post. The rig powers on and never reaches the BIOS.
No boot. It posts but never reaches the operating system.
Missing GPU. It boots and one card does not show up.
Reboots. Everything runs, then it restarts hours later.
These need different steps. Naming yours first saves you from swapping parts at random.
Pull every GPU out. All of them.
Try to boot the bare board.
That single step separates a board fault from a card fault, and it takes two minutes.
If it still will not post, reseat the CPU.
Reseat the memory.
Clear the CMOS. Pop the coin battery out with the metal tab, leave it out for 20 seconds, put it back.
That returns the BIOS to defaults.
Then test the power supply.
Most of the time it is the CPU, the memory or one of the cards.
Start the same way. Unplug every riser.
Boot the bare board.
If it reaches the OS, add one GPU back.
Boot it.
Shut down, add the next one, boot again.
Keep going until it fails. The card you just added is the one hanging the boot.
It is slow and it always works. A rig that will not boot with eight cards tells you nothing. A rig that boots with three and fails on the fourth tells you everything.
You have found a slot that fails. You do not yet know whether the card or the riser is at fault.
Put a known good GPU on that same riser.
If it works, the card is bad.
If it still fails, the riser setup is bad. That means the riser board, the USB cable or the connector at either end.
It can also be power, because a weak rail feeding that riser looks identical from the outside.
Then test the suspect card on its own, in a full x16 slot on the board, with no riser involved.
That removes every variable except the card.
A card that fails there has usually had a BIOS flashed to it that did not take.
If none of this gets you booting, rebuild the operating system. Rewrite the USB with Etcher or Rufus and try again.
You reach the OS and one card is missing. Windows marks it in Device Manager. A mining OS shows it as a missing GPU.
Check drivers first, because this is usually drivers.
On Windows, run DDU to strip the display driver out of the registry completely, then install it fresh.
On a mining OS, run its driver update tool. A card newer than the image you installed will not be supported by the drivers that shipped with it.
A whole card generation showing as missing is a driver support gap, not eight dead cards. When the mining OS has not caught up with a generation yet, run those cards on Windows and keep them separate from the rest of the farm until it does.
If you flashed that card's BIOS, put the stock BIOS back and test again.
This is the part that wastes an hour if you do it by counting.
Slot numbering in software does not reliably match the order the cards sit in the frame. Boards label and enumerate them differently.
Use the fans instead.
Set every fan to 100 percent.
Set the fan on the card you are hunting to 10 percent.
Walk over and find the quiet one.
It works in HiveOS, and it works in Afterburner or whatever else you tune with on Windows.
Then swap it against a known good card as above.
Everything posts, everything is detected, everything mines. Then twenty minutes or three days later it restarts.
You need the log from before the reboot, so set that up first.
On Windows, add a log file switch to the miner's batch file so it writes output to disk. Most miners take one, such as a logfile argument followed by a filename.
In HiveOS, the red reboot warning is clickable and shows you the tail of the miner output.
Either way you are reading for the same thing.
Find the card that dropped to zero, or fell off a cliff in hashrate, just before the restart.
That is your offender.
Turn its overclock off.
Run it long enough to pass the point it normally fails at.
Stable without the overclock means the overclock was wrong for that particular card. Cards from the same box do not clock the same.
Still failing means it is hardware, so swap it against a known good card to rule out the riser.
If the card and the riser both check out, suspect the power supply.
Remove variables until the rig works, then add them back one at a time.
Swap against a known good part rather than guessing which one is bad.
Change one thing between tests.
None of this tells you the answer. It tells you where the answer is.