Thanks again to everyone who investigated this in the past and made all these lovely tools and posts, I would never have figured this out on my own.
05.02.2026 23:08 โ ๐ 0 ๐ 0 ๐ฌ 0 ๐ 0@zaoo.bsky.social
Game programmer in New Zealand, previously HPC sysadmin in Sweden. My views are my own. he/him
Thanks again to everyone who investigated this in the past and made all these lovely tools and posts, I would never have figured this out on my own.
05.02.2026 23:08 โ ๐ 0 ๐ 0 ๐ฌ 0 ๐ 0I had 365k leaked processes for 11 GiB of page table (32 KiB each) and probably 11 GiB more assuming your 64 KiB figure holds.
Apart from just compiling software, this was primarily due to that if you idle a VS Code window with Docker installed, it spawns three processes every other second.
Memory seemed a bit cramped but nothing in user-land stood out.
Profiling with Superluminal showed D3D12's CreateCommittedResource1 stalling for several seconds rather than the usual sub-40ms.
It generally spent more time in creating the resource/heap than making it resident but both suffered.
The original symptoms that led me to dig into this over the last few months were that after a few days of system uptime, running a particular game had longer loads and the system outside become strangely slow to react, where UI kind of worked but actions could stall for minutes, even Ctrl-Alt-Del.
05.02.2026 23:08 โ ๐ 0 ๐ 0 ๐ฌ 1 ๐ 0AMD driver downloads has "-a" for RNDA2 and down, "-b" for RDNA3 and up, and "-c" as a combined package for "systems with RDNA series graphics products". The combined one is only obscurely linked from driver release notes.
05.02.2026 23:08 โ ๐ 2 ๐ 0 ๐ฌ 1 ๐ 0For extra fun, the Adrenalin drivers are bifurcated between RDNA2 (7800X3D) and RDNA3+ (9070XT), so despite having proper "Adrenalin drivers" from the AMD website installed for the dGPU, those didn't apply to the iGPU.
05.02.2026 23:08 โ ๐ 0 ๐ 0 ๐ฌ 1 ๐ 0Adding to the confusion when testing with a mate is that motherboard vendor matters, the same CPU model has different SUBSYS in the hardware ID.
My ASUS (PCI\VEN_1002&DEV_164E&SUBSYS_88771043&REV_CB) gets 31.0.24002.92 while an ASRock (PCI\VEN_1002&DEV_164E&SUBSYS_364E1849&REV_CB) gets 32.0.11024.2.
Thanks Bruce for figuring this out.
I've been chasing some system gremlins and I regret to inform you that it's still a problem to this day.
The latest WU driver for the iGPU on my 7800X3D (ASUS mobo) dates back to Jan 2024 and it leaks.
Gonna note down some key info for other unfortunate souls.