Issues

• [BUG][Metal] Grouped conv with C_per_group = 8 is ~10x slower than C_per_group = 4 / 16

  ml-explore/mlx#4616 · Filter by author zsun6 opened on Oct 3, 2026

• mx.fast.gated_delta_update VJP: wrong gradient w.r.t. g (and ~1e-3 error on all inputs) on the NAX path for Hk=16

  bug
  ml-explore/mlx#4614 · Filter by author kevincaldwellgordon-bot opened on Oct 2, 2026
  

• Split-K quantized_matmul rounds partial sums to the input dtype

  ml-explore/mlx#4613 · Filter by author mattt opened on Oct 2, 2026

• [BUG] [Metal] Wired memory is released shortly after GPU idle (residency never renewed)

  ml-explore/mlx#4609 · Filter by author namolnad opened on Oct 2, 2026

• Custom kernel (mx.fast.metal_kernel, MathModeSafe) returns wrong output rows on M3 Ultra / macOS 27.0 — exact under Relaxed and on M2; MLX-free reproducer

  ml-explore/mlx#4603 · Filter by author Pietroski opened on Oct 1, 2026

• [BUG] CPU Float16 NEON comparisons broadcast the first lane’s result

  ml-explore/mlx#4601 · Filter by author crabel99 opened on Sep 30, 2026

• Option to opt out of TF32 for fp32 conv: small-kD conv3d is ~5000× less precise on M5 since 0.32

  ml-explore/mlx#4595 · Filter by author xocialize opened on Sep 29, 2026

• [BUG] custom_function.jvp and .vmap are ignored when the body contains a Metal kernel

  ml-explore/mlx#4580 · Filter by author alford-auld opened on Sep 28, 2026
  2

• fp32 matmul with small N is ~5.5x slower at M=2..8 than M=1 (GEMM path vs gemv): (M,16384)@(16384,24) 120us vs 22us on M5 Ultra

  ml-explore/mlx#4571 · Filter by author sethforprivacy opened on Sep 27, 2026

• Feature: Adding minmax, similar to how numpy and jax recently added.

  ml-explore/mlx#4569 · Filter by author aaishwarymishra opened on Sep 26, 2026

• First run after a local mlx build: prefill up to 2.3x slower on small models (Metal shader cache cold), negligible in absolute time

  ml-explore/mlx#4566 · Filter by author jrp2014 opened on Sep 25, 2026

• Expose precompiled_metal_kernel to load offline-compiled .metallib (-Ofast) for custom kernels

  ml-explore/mlx#4541 · Filter by author dahai80 opened on Sep 21, 2026

• [BUG] mlx.sort return value incorrect for shape (65537, 65536) on GPU

  buglow priority
  ml-explore/mlx#4538 · Filter by author kvcache670 opened on Sep 20, 2026

• Metal command-buffer limits (50 ops / 50 MB) cost 5–8% decode throughput on launch-heavy models (M3 Ultra measurements)

  performance
  ml-explore/mlx#4521 · Filter by author freddyhaddad opened on Sep 16, 2026
  1

• [Performance] gather_qmm at M=1 streams 4-bit expert weights at ~half the bandwidth of M=4 (MoE decode)

  ml-explore/mlx#4509 · Filter by author freddyhaddad opened on Sep 14, 2026
  1

• [BUG] threads and streams issues

  bug
  ml-explore/mlx#4506 · Filter by author davidkoski opened on Sep 14, 2026
  2

• quantized_matmul is slower than dense bf16 when the output dimension is not a multiple of 8, and ~7x slower below 8

  performance
  ml-explore/mlx#4498 · Filter by author ursk opened on Sep 13, 2026

• mx.distributed ring backend: gradient-shaped CPU-stream all_sum deadlocks probabilistically (~3-4%/call) across a training loop; GPU-stream variant crashes deterministically at ~5s

  bugdistributed
  ml-explore/mlx#4475 · Filter by author ictashik opened on Sep 8, 2026
  1

• Metal: distributed collectives run on the CPU stream, and with the default fence each crossing costs ~180 µs (MLX_METAL_FAST_SYNCH: ~4 µs)

  distributedperformance
  ml-explore/mlx#4438 · Filter by author avlp12 opened on Aug 31, 2026
  2

• nn.Upsample(mode="linear") is 2-4x slower than a matmul-based bilinear resize when upsampling

  performance
  ml-explore/mlx#4435 · Filter by author bahaehmimdi opened on Aug 30, 2026
  5

• [BUG] NAX MoE simdgroup skipping in 0.32.2 triggers lost JACCL completion under heterogeneous TP

  bugdistributed
  ml-explore/mlx#4403 · Filter by author jonathan308 opened on Aug 25, 2026
  1

• [BUG] mx.var / mx.std wrong for offset data

  buglow priority
  ml-explore/mlx#4379 · Filter by author JasonHonKL opened on Aug 23, 2026
  3

• Support fusing element-wise operations into reductions

  enhancement
  ml-explore/mlx#4355 · Filter by author sakupan102 opened on Aug 19, 2026
  4

• [CUDA] gather_qmm ~24x slower than equal-FLOPs quantized_matmul on GB10 (sm_121) — makes quantized MoE prefill unusable

  cudaperformance
  ml-explore/mlx#4339 · Filter by author NORDIT opened on Aug 18, 2026
  2

• [BUG] JACCL SIGSEGV in tbt_post_recv during unique all_sum (no unplug, no model, ports still up)

  bugdistributed
  ml-explore/mlx#4319 · Filter by author guruswami-ai opened on Aug 17, 2026
  1