Possibly, but probably not either. To restate the OP a bit more specifically (and hopefully correctly),
- 1DPC 1R 6000 40-40-40 1.35V (Kingston 2x16) is stable
- 1DPC 2R 4800 30-38-38 ? V (G.Skill 2x32) is at least conditionally stable
- 1DPC 2R 6000 30-38-38 1.35V is unstable
(1DPC = 1 DIMM per channel, 1R = single rank, 2R = double rank)
AMD and Intel are both (very likely intentionally) vague about support but both Zen 4 and 5 are more capable of overclocks than Intel LGA1700 parts. Arrow’s headline spec is higher than Zen 5 but, in general, details are unclear.
Most reasonable interpretation I’m aware of is Raphael supports DDR5-5200 JEDEC A (38-38-38) for 1DPC 2R on quad socket motherboards and Granite Ridge 5600 A (40-40-40) for the same. Intel 1DPC quad socket support’s 4400 (LGA1700 parts). Haven’t been able to find anything as to whether Arrow’s 6400 applies beyond 1DPC 1R two socket but would guess probably not.
A default approach I’d suggest is a parts acceptance test pass to verify supported speeds, incrementally tightening and clocking up from there until short tests begin to show instability, back down and do another test pass to get another known good combination, and then start probing to locate the edge stability more specifically.
Potentially. 40-40-40 is looser than 30-38-38 but you need to consider also the rest of the primaries, the secondaries, and the tertiaries. ZenTimings is convenient but sometimes there are reporting bugs, so check what it says against the BIOS.
For any overclock, including XMP and EXPO, pretty much the first thing to do if it boots is stability validation. Everybody rolls their own protocol but the de facto standard’s short testing on incremental changes and periodic full test passes to establish known (pretty) good points. MemTest86 and MemTest86+ are ok, but probably not the most efficient at finding errors, and last I checked MemTest86 would utilize a max of eight cores so had limited or no dual CCD coverage.
Different folks like different tools. Personally I use 10 minutes of OCCT CPU+RAM for short testing. My current test pass is
- four hours OCCT + Furmark2 hot testing: different OCCT settings every hour, DDR fans that offset dGPU passthrough spun down
- 8-12 hours y-cruncher FFT, FFTv4, N63, and VT3
- four hours Prime95 variable with n threads = n cores - 2
with HWiNFO64 running the whole time to log hardware errors. I don’t generally then call a build stable at the support limit or on an overclock until it runs fine in real world use for a couple weeks.
I think my go to question here’s whether 2x32 5200 38-38-38 1.2V is stable with JEDEC tREFI and whatnot. If it is anything else is bonus and there’s no strong reason to try returns.
Higher clocks take a higher DDR voltage. The curve depends on the DRAMs. Kingston indirectly sort of specifies this (see the datasheet in the OP), with other manufacturers mostly you have to table up test speeds, voltages, and die types to make educated guesses.
IO die voltages are also influential but are more fiddly, even less documented, and mostly come up at higher overclocks than being attempted here. So I’d suggest not getting into those for now.
I’m leaning towards the former. Gigabyte claims 8000 MT/s for the X670 Aorus Elite AX, so hopefully it’s not too much of a limitation here. But @Goblin, which AGESA are you on? If the Elite AX is on a pre-1.2 BIOS I would update as more mature AGESAs seem to overclock a bit better.
Even with a 1R to 1R comparison, rather than 1R to 2R, I’m not sure this conclusion would hold.
QVL just means a combination worked with the parts the manufacturer had for testing, not that it’s so far inside the tolerance envelope all combinations of the same parts will work.
