Fedora 43 / niri-wayland / Firefox 148.0.2 / NVIDIA 專有 580.126.18

https://lite.ip2location.com

By the way this is actually the major reason why I defect to AMD from Intel camp since the Zen 4 generation.

Intel used to offer affordable ECC support for their small workstation line of product, e.g. Xeon-E. Those are basically priced the same as the Core product line, sans the bundled heatsink, which you'd upgrade anyway. However Xeon-E discontinued around 2023 and merged into the Core line. Now the Core line CPUs support ECC, but are gated behind the overpriced Wxxx chipsets. This is purely market discrimination since all modern CPUs have builtin memory controllers and ECC should have nothing to do with the southbridge.

That said, I don't think having ECC DIMM is as big of a problem in DDR5 age though, as DDR5 already have on-die ECC. On-die ECC is not ideal because errors were silently recovered on the chip so you can't detect DIMMs that have weak cells and preemptively replace them, but it worked well enough for most users. However I think the prevalent problems nowadays are factory overclocked DIMMs marketed as premium products. You see DIMMs marketed as DDR5-8000 that are really built from JEDEC DDR5-5600 dies pumped up to 1.45V, and average consumers would happy enable XMP without realizing they just made a unstable computer.

I thought I have bad luck with DIMMs. I've encountered maybe 2~3 faulty DIMMs in my life throughout ~10 PCs I've built for myself. Nowadays I just buy Kingston ValueRAM at the highest JEDEC speed available and run stock settings and never encountered a bad DIMM since.

看來BUG不在顯示卡driver，而是 firefox 處理 「DrawTargetWebgl instances share one SharedContextWebgl」這一段。