Issues

• concat: ROW_MAJOR height-sharded width concat of two tensors hangs when a shard row exceeds NOC_MAX_BURST_SIZE

  bugMCWop_cat: TM
  Bug

  tenstorrent/tt-metal#60294 · Filter by author VirdhatchaniKN opened on Oct 12, 2026
  1

  

• [ttnn/experimental/quasar] transpose HC sharded factory derives cores from shard-grid bounding box, aborts on grids not anchored at (0,0) (#58407 item 9)

  bugMCWop_cat: TM
  Bug

  tenstorrent/tt-metal#60286 · Filter by author abdulla-TT opened on Oct 12, 2026
  

• [ttnn/experimental/quasar] tilize_with_val_padding single-core reader leaves row-tail bytes unwritten for widths not divisible by 4 (#58407 item 7)

  bugMCWop_cat: TM
  Bug

  tenstorrent/tt-metal#60285 · Filter by author abdulla-TT opened on Oct 12, 2026
  

• MoE Optimization - Only routing tokens to top2 experts

  tenstorrent/tt-metal#60284 · Filter by author avelickovTT opened on Oct 12, 2026

• [aw] /test — CI pipeline selector for tt-metal failed

  agentic-workflows
  tenstorrent/tt-metal#60281 · Filter by author github-actions opened on Oct 12, 2026

• [ttnn/experimental/quasar] untilize_with_unpadding ignores output_tensor_end for rank > 4 inputs

  bugMCWop_cat: TM
  Bug

  tenstorrent/tt-metal#60277 · Filter by author abdulla-TT opened on Oct 12, 2026

• Wormhole LLK perf: the isolate stand-in threads' RISC-V loops set about one in five isolate values

  tenstorrent/tt-metal#60276 · Filter by author mvlahovicTT opened on Oct 12, 2026

• Wormhole LLK perf: math INIT depends on which run type's kernel hosts the INIT launch

  tenstorrent/tt-metal#60275 · Filter by author mvlahovicTT opened on Oct 12, 2026

• Wormhole LLK perf: the other threads' start and end code runs inside the measured thread's TILE_LOOP

  tenstorrent/tt-metal#60274 · Filter by author mvlahovicTT opened on Oct 12, 2026

• Wormhole LLK perf: PACK_ISOLATE never releases dest, so the packer speed picked at the start holds for the whole loop

  tenstorrent/tt-metal#60273 · Filter by author mvlahovicTT opened on Oct 12, 2026

• ttnn.softmax on Blackhole after #53643: correctly rounded bfloat16 outputs, long-row sums fixed, at least 10% less time

  communityop_cat: reduces
  tenstorrent/tt-metal#60272 · Filter by author bvdyvncqr5-commits opened on Oct 12, 2026

• [ops] ttnn.sort: Single Row Single Core sort returns padding indices when the input contains the padding value

  bugcommunityop_cat: misc
  Bug

  tenstorrent/tt-metal#60262 · Filter by author ign-sudha opened on Oct 12, 2026

• Reduce validation skips output shard-grid checks for legacy sharded memory configs

  MCW
  Bug

  tenstorrent/tt-metal#60259 · Filter by author mcw-anasuya opened on Oct 12, 2026
  

• [squad-plan] Migrate round_up_to_mul32 callers to tt::round_up — ttnn/operations + tests/ttnn (part 4/5 of #59394)

  automationcopilot-ready
  tenstorrent/tt-metal#60256 · Filter by author github-actions opened on Oct 12, 2026

• [squad-plan] Deprecate tilize_utils rounding helpers in favor of tt::round_up — tt-metalium/tilize_utils.hpp (part 5/5 of #59394)

  automationcopilot-ready
  tenstorrent/tt-metal#60255 · Filter by author github-actions opened on Oct 12, 2026

• [squad-plan] Add integral-type constraints to div_up/round_up/round_down — tt-metalium/math.hpp (part 2/5 of #59394)

  automationcopilot-ready
  tenstorrent/tt-metal#60254 · Filter by author github-actions opened on Oct 12, 2026

• [squad-plan] Document and assert tt::align power-of-two precondition, add tests, use round_up in sockets — tt_align.hpp + tt_metal/distribute

  automationcopilot-ready
  tenstorrent/tt-metal#60253 · Filter by author github-actions opened on Oct 12, 2026

• [squad-plan] Fix round_up/round_down semantics, zero-divisor and noexcept consistency — tt-metalium/math.hpp (part 1/5 of #59394)

  automationcopilot-ready
  tenstorrent/tt-metal#60252 · Filter by author github-actions opened on Oct 12, 2026
  1

  

• matmul: 1D mcast sharded in0 validated against device grid, not factory core set

  tenstorrent/tt-metal#60249 · Filter by author mouliraj-mcw opened on Oct 12, 2026

• matmul in1 readers: NCRISC_RD_CMD_BUF NOC_CTRL VC leak across kernel launches

  tenstorrent/tt-metal#60248 · Filter by author mouliraj-mcw opened on Oct 12, 2026

• matmul: validate_matmul_bias_shape never checks logical bias h, so h in (1, M) silently reads padding

  tenstorrent/tt-metal#60247 · Filter by author mouliraj-mcw opened on Oct 12, 2026

• equal_within_n_sig_figs reports false matches (0.5 vs 5.5) and is asymmetric

  bugcommunity
  Bug

  tenstorrent/tt-metal#60246 · Filter by author CloneOfAlex opened on Oct 12, 2026

• convert_u32_to_bfp with is_exp_a=true encodes values below 2^-15 as ±2^-15 instead of 0

  bugcommunity
  Bug

  tenstorrent/tt-metal#60245 · Filter by author CloneOfAlex opened on Oct 12, 2026

• Tilize/untilize heap overflow and corrupted data for transposed tiles with non-square faces (e.g. Tile({16,32},{8,16},true))

  bugcommunity
  Bug

  tenstorrent/tt-metal#60244 · Filter by author CloneOfAlex opened on Oct 12, 2026

• FreeListOpt::allocate leaks the block when the address-limit OOM is thrown; clear() leaves a stale lowest_occupied_address

  bugcommunity
  Bug

  tenstorrent/tt-metal#60243 · Filter by author CloneOfAlex opened on Oct 12, 2026