DRAM issues on 9900K and Asus Z390-E

[EclipseVV]

Hi just RMA'd my 9900K that worked perfectly for 6 months on a Z390-E Gaming due to several WHEA warnings appearing recently in the Event Log; Intel troubleshooted this as a defective CPU. The CPU was working fine otherwise, and my DRAM was working fine with VCCIO/VCCSA set at 1.15 with 3333MHz sticks.

However, as soon as I tried another 9900K, something worse happened: the new CPU seems to work only with really high VCCIO/VCCSA voltages, at least 1.35V/1.40V (that's what the board automatically set up after a few unsuccessful reboots). As soon as I try something lower (even at stock 2666MHz), I get hundreds of errors in all DRAM test programs.

I also tried with other DRAM sticks and the situation is unchanged.

What would you recommend? This sound again like a defective CPU / IMC, but it still seems quite strange... could it be a motherboard issue?

The CPU itself with small FFT tests is 100% stable. The issue is with DRAM access only.

 

[EclipseVV]

Yes, absolutely. I also tried the previous one, no difference.

It also seems quite picky with DRAM speed/timings: VCCIO 1.325 VCCSA 1.375 is stable at 3500MHz, but not at 2666, 3200 or 3600 (with DDR-3600 sticks and identical timings). Tested with Karhu and 2000% coverage.

VCCIO 1.30 VCCSA 1.35 is not stable at every speed I tested (with identical timings).

It took several hours to find one setup that actually worked: all XMP/JEDEC profiles do not work!

Basically, I passed my whole sunday testing the DDR-3600 XMP timings (17-19-19-39), and lowering frequency:
- 3600MHz: not stable at 1.35/1.40 or lower
- 3500MHz: stable at 1.325/1.375
- 3200Mhz: not stable at 1.35/1.40 or lower
- 2666MHz: not stable at 1.35/1.40 and auto voltages

I decided not to try raising VCCSA above 1.40 for fear or destroying my CPU.

Not stable = error almost immediately with Karhu (below 100% coverage).

Can you recommend something else, or do I need to RMA again? And in that case, CPU or MB?

 

[EclipseVV]

Pins seemed to be fine when I mounted the second CPU, but I cannot exclude it 100%.

The CPU itself is on pretty normal vcore values, and on CPU-only tests (Prime95, Linpack, OCCT small, etc) there are zero errors.

What puzzles me is that raising VCCIO/VCCSA everything works for certains speeds (I started another Karhu Ram Test a few hours ago and I'm at 3000% coverage with no errors). That would seem to rule out motherboard... or not?

 

[EclipseVV]

A quick update: on monday I decided to RMA the second 9900K and mounted the new one a few hours ago: the new CPU seems to be stable with DRAM at 3600MHz (XMP profile without any adjustment) and DRAM = 1.35V, VCCIO = 1.225V, VCCSA = 1.200V.

So, indeed, it seems I had two defective CPUs in a row. The VCCIO/VCCSA seems a bit on the high side for 3600 (or not?), but I'd say I can use these voltages and enjoy my PC