Problemsby GlyphTrouble with NVIDIA slower
3 posts from 3 people3d activeSteady, +1 this week
Posts per day
The posts behind it
CUDA training loop runs ~50x slower while a DataLoader iterates in a background thread, since 2.10.0
### 🐛 Describe the bug PyTorch appears to be slower in newer versions. From my testing, slowdown starts at 2.10.0. It slows down when dataloader is involved. I have attached repro steps. #### AI-generated report > The section below was written by Claude Code (Anthropic) from...
[CUDA][cuDNN] conv1d float32 about 27x slower on nightly (cuDNN 9.26) than on 2.9.1; the native kernel is 100x faster
### 🐛 Describe the bug `F.conv1d` on float32 CUDA inputs is about 27x slower on a recent nightly than on 2.9.1 on an RTX 3050 Laptop GPU (sm_86). With cuDNN disabled, the same call takes 8 ms on both builds, so the time goes into the cuDNN path that the nightly selects....
layer_norm / rms_norm forward launches idle warps for narrow rows (up to ~9x slower, e.g. per-head q/k norm)
### 🐛 Describe the bug `F.rms_norm` and `F.layer_norm` forward on GPU are much slower for narrow rows than for wide rows with the same total size. Narrow rows are the per-head q/k norm in many current models (head dim 64/128: Qwen3, Flux, Wan, MiniMax-H3, ...). On a Radeon AI...
Companies and products named
About this problem
Evidence
3 posts from 3 people in 1 place, about 2 a week over 13 days. Mostly on pytorch/pytorch. Tools named alongside: PyTorch, MiniMax, Tesla, Claude Code.
Frustration Frustration 0 of 3· Seen on GitHub issues
How it was grouped
Posts about NVIDIA with no known issue, sharing the word "slower". First post Oct 12, 2026, latest Oct 24, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method
Other NVIDIA problems
NVIDIA users keep filing bugs
Top post: "`clef-flash` decision model always fails on `/v1/systemone`, Clef: non-finite logit (CUDA) / Clef: cannot", one of 39 posts from 36 people on GitHub issues, Reddit, Community forums.
NVIDIA upgrades break working setups
11 posts from 11 people on GitHub issues.
Trouble with NVIDIA account
Top post: "NVIDIA Careers Workday Account Inactive: Unable to Log In or Reset Password", one of 7 posts from 6 people on Community forums.
NVIDIA lock-in makes switching costly
5 posts from 5 people on GitHub issues, Hacker News, Bluesky.
NVIDIA vulnerabilities are hard to triage and patch
77 posts from 77 people on Glyph Ibis.
Running NVIDIA locally hits memory and speed limits
10 posts from 10 people on GitHub issues.
NVIDIA GPU access and drivers are painful
9 posts from 9 people on GitHub issues, Mastodon, Reddit.
NVIDIA runs out of memory
5 posts from 5 people on GitHub issues.
Trouble with NVIDIA tensor
Top post: "DFlash2 is not working with `--split-mode tensor`", one of 3 posts from 3 people on GitHub issues.
NVIDIA error messages are cryptic
3 posts from 3 people on GitHub issues, Reddit.
Trouble with NVIDIA phone
Top post: "Unable to receive OTP for Vietnam (+84) phone verification on build.nvidia.com", one of 3 posts from 3 people on Community forums.
Trouble with NVIDIA inductor
Top post: "Inductor template lookup keys are tied to CUDA architecture names", one of 3 posts from 3 people on GitHub issues.
History