CPU-Z validation:
https://valid.x86.fr/xf3yer
(CPU-Z validation reports incorrect memory tRC, CPU-Z app, Ryzen Master, and BIOS all show 30-40-40-96-136 (correct, tRP + tRAS = tRC),
but validation link shows 30-40-40-96-127 (incorrect, tRP + tRAS != tRC). Not sure why.)
CPU: AMD Ryzen 7 7700X PBO curve -8 across cores (currently on 1.30, lost the more aggressive settings I had before)
Motherboard: MSI MEG X670E ACE
RAM: G.SKILL Trident Z5 NEO 64GB DDR5 6000 F5-6000J3040G32GX2-TZ5N
(Technically, this kit is not on the compatible memory list, but according to G.SKILL datasheets, it should be identical to the
F5-6000J3040G32GX2-TZ5NR version minus the RGB lights. Same timings, voltages, SK Hynix M-Die as TZ5NR. In any case, it
worked well on 7D69v13.)
CPU Cooler: Optimus foundation waterblock for AM5, Watercool MO-RA3 360 PRO, big reservoir.
(Overkill for just the processor, but GPU will eventually run through it as well.)
GPU: Nvidia 4090 FE
PSU: CORSAIR AX1600i
Monitor: Samsung C27HG7x
=====================
Tried the upgrade to 7D69v14, released December 21, 2022, last night.
Set BIOS up with:
- HD Audio disabled (using AE-9 instead)
- Integrated GFX disabled
- ASMedia SATA controller off
- Fans set to correct PWM/DC settings where applicable
- Mfg Supplied EXPO profile at 30-40-40-96
- UCLK/FCLK 1:1
- Memory Power Down DISABLED
- Memory Context Restore DISABLED
(Below, 7D69v13 will be referred to as v1.30, 7D69v14 will be referred to as v1.40)
=====================
Memory timings on v1.40 were EXPO profile defaults, 30-40-40-96-136, FCLK=UCLK, power down disabled, memory context restore disabled. System was not remotely stable.
v1.40 BIOS gave frequent BSODs, assuming it managed to get all the way into Windows. BSOD types were mixed, mostly MEMORY_MANAGEMENT, or IRQL_NOT_LESS_OR_EQUAL. There was at least one other type, but I don't have it written down.
(Also got the A6 error code that other users have mentioned, especially in the BIOS GUI or trying to save BIOS changes and reboot)
Once memory timings were set to EXPO, the BIOS interface itself also stopped working correctly; unable to consistently make changes without the BIOS GUI locking up (no reaction to mouse or keyboard). Even when the BIOS GUI did not lock up while using it, it would often freeze/hang trying to save settings and reboot; needed a manual reset to reboot, and settings were not saved.
I also noticed when looking at subtimings, that a couple of them were lower on v1.40 by default, though not by much:
- tWRWRSD: 9 on v1.40, 10 on v1.30
- tWRWRDD: 9 on v1.40, 10 on v1.30
-tRDWR: 20 on v1.40, 21 on v1.30
Unsure how much difference that makes, but seemed strange. Setting these manually to the v1.30 values didn't make any difference in stability.
Default PBO settings also showed a lower voltage, but don't have the exact number recorded, something like 1.262 on v1.40, according to BIOS GUI, instead of 1.340 on v1.30. System not stable enough to mess with PBO on v1.40.
=====================
Rolled back to v1.30, which initially exhibited the same behavior until I cleared CMOS and set everything back up again.
Original memory timings on v1.30, which I've now lost, were tighter than the default EXPO profile and stable with no issues, FCLK=UCLK, power down disabled, memory context restore disabled, something like 30-38-38-96-134 instead of default 30-40-40-96-136.
Had the TRFC1, TRFC2, and TRFCSB lower than default as well, but forgot to record the exact values, iirc, something like 525-385-313 instead of default 884-480-390.
PBO per-core curve settings were around -12 for some cores, with one or two at -8 and the rest at -10, stable for the last several weeks.
=====================
Not sure what the issue with v1.40 is, but something (memory related, I assume) is definitely very broken.
On v1.30, unstable timings or memory settings generally don't boot at all; I hadn't seen a BSOD on this machine prior to trying v1.40.
Uh... I think that's all of the info I've got, let me know if you need anything else, thanks!
