Всем привет,

There is Sapphire Nitro+ rx580 for repair. I have already brought it back to life. There was a shorted (KZ) ceramic capacitor, nothing special, PEX power was missing. But this card reports 83% of memory controller utilization.
When memory clock is set to its default 2000MHz there is 83% but when I increase it even by 1MHz the memory controller utilization drops to 60-70%, when I decrease MHz there is no change to the certain point and it starts increasing to 91%.
This card works normally, Furmark passes, this card can mine but slowly up to 22.5Mh and the hash rate reported by pool is lower.
All VRMs work and have proper scope shapes, everything looks good but this card does nor want to eat power and des not want to work faster. GPU frequency and power limit work and only the memory clock does not react properly when adjusted in afterburner. I reflashed bios several times and this is not the bios issue.

PWM controller NCP81022

What could happen to this card?

Check GPU caps of 3-4-5-6 Chokes. I have cards that crashed after 1800mhz-agresive straps where applied.. and problem was in gpu.

This card does not crash. The only problem is such that it does not want more power and there is strange reaction on the memory clock adjustment. I can increase core voltage and the card immediately becomes hotter and draw more watts but it is not followed by memory controller utilization. It looks like memory controller makes gaps between access cycles and you see these "teeth" on the GPU-Z memory controller bar, I circled it by yellow frame.

It does not look like faulty NCP81022 because it controls only voltages, not the memory frequency. Memory frequency is set inside the GPU by its PLL clock generator.

This card is based on Hynix memories but I already saw such symptoms some time ago on another RX580 XFX with Samsung FB memories.