Issues

• Hardware/agent report: Maya-M on RTX 3090 + RTX 3060 + dual V100, 256K context (v1.1.11)

  mw00/project-maya#220 · Filter by author techrian opened on Oct 16, 2026
  1

• Speed report: RTX PRO 5000 72GB + RTX PRO 5000 48GB, Epyc 7532, 512gb DDR4 2400 - Unsloth UD-Q4_K_XL, UD-Q6_K_XL, Q8_0 - 1M Context

  mw00/project-maya#218 · Filter by author Xar1c opened on Oct 16, 2026

• Validation on Radeon AI PRO R9700 (gfx1201, ROCm 7.0): v1.1.12 +15% decode, 384K context tested, thinking control confirmed

  mw00/project-maya#213 · Filter by author potatopote-debug opened on Oct 16, 2026
  2

• Speed report: 1 / 2 / 3x Tesla P40 (Pascal, sm_61) - Maya-S24 28 tok/s and Maya-L 17 tok/s on two P40s, once global loads are cached in L1 (-Xptxas -dlcm=ca, 1.7-3x decode)

  mw00/project-maya#212 · Filter by author Jw88888 opened on Oct 16, 2026
  1

• 3x Tesla V100-PCIE-16GB (48 GB VRAM, 62 GB RAM): Maya-S at 26.6 tok/s decode, 497 tok/s prefill, 128K context

  mw00/project-maya#211 · Filter by author saatvik-viniak opened on Oct 15, 2026
  1

• Speed report: Maya-M and Maya-L on 2× V100 32 GB, Ryzen 9 7945HX3D, 96 GB RAM (v1.1.11)

  mw00/project-maya#208 · Filter by author nkhaooat opened on Oct 15, 2026
  1

• project Maya - Windows - RAM usage 39GB out of 64GB for Experts 🙈

  mw00/project-maya#207 · Filter by author technerd-official opened on Oct 15, 2026
  1

• Strix Halo 96 GB on Windows (8060S, ROCm 10.0): v1.1.11 Prefill speed slows down once the context length exceeds a certain threshold

  mw00/project-maya#206 · Filter by author YoutechA320U opened on Oct 15, 2026
  2

• Please add support for RDNA-2 cards - gfx1030 (v620 32gb)

  mw00/project-maya#205 · Filter by author slavap opened on Oct 15, 2026
  2

• Strix Halo 128 GB on Linux (8060S, ROCm 10.2): v1.1.11 report - MTP decode +20-27%, prompts need a gfx1151 hipBLASLt 100500 table

  mw00/project-maya#202 · Filter by author boxwrench opened on Oct 15, 2026
  2

• Maya-S24 on a 32 GB 4080 SUPER + 3970X: 262k context and tuning results

  mw00/project-maya#200 · Filter by author Micro-ATP opened on Oct 15, 2026
  3

• Speed report: 4x Instinct MI50 32 GB (gfx906 wave64) — Maya-L, 26.9 tok/s decode / 193.2 tok/s prefill

  mw00/project-maya#193 · Filter by author nicholas631 opened on Oct 14, 2026
  1

• Speed report: 3x Tesla P40 24 GB (2x Xeon E5-2658 v2, 110 GB RAM) - Maya-S24, 9.2 tok/s decode / 204 tok/s prefill (8K), Linux + Pascal build questions

  mw00/project-maya#189 · Filter by author hengwen1 opened on Oct 14, 2026
  1

• Speed report: EXL3-3.05bpw on 2x RTX 3090 (Threadripper PRO 3955WX, 247 GB 8-ch DDR4-2400): 29.2 tok/s decode, first 2-GPU EXL3 datapoint + CPU lane / DFlash2 findings

  mw00/project-maya#187 · Filter by author catonooka opened on Oct 14, 2026
  4

• Windows: RAM tier stops at ~45 GB on a 64 GB PC, ~7 GB stays free (pinning cap) - can it be raised?

  mw00/project-maya#185 · Filter by author Hosein20000 opened on Oct 14, 2026
  4

• Checkpoints (#146): never kept when the disk slots are on, and a new chat with the same system prompt is read from token 0 (1.1.7, Windows, 7900 XTX)

  mw00/project-maya#179 · Filter by author jerem91150 opened on Oct 14, 2026
  1

• Speed report: RTX 3090 Ti + EPYC 7302P (KVM guest, 16 vCPUs), Maya-L: decode +23% with STRATA_GLM_CPU_LANE=14 + STRATA_GLM_VRAM_EVICT=lru

  mw00/project-maya#177 · Filter by author Jw88888 opened on Oct 14, 2026

• MI100 CDNA 1 GFX908 TESTING

  mw00/project-maya#176 · Filter by author Elefant-Freeciv opened on Oct 14, 2026

• Speed report: RTX 3090 24 GB (Ryzen 7 5800X, 62 GB RAM) - Maya-S24, 9.0 tok/s decode / 507 tok/s prefill (8K), Linux + disk-bound decode question

  mw00/project-maya#175 · Filter by author itsmetamike opened on Oct 14, 2026
  2

• RTX PRO 4000 Blackwell 24 GB + i9-13900K + 128 GB (3600 MT/s): Maya-S24 at 24.7 tok/s decode, 1146 tok/s prefill @8K

  mw00/project-maya#174 · Filter by author morrisfamily opened on Oct 14, 2026

• DFlash2 bf16 drafter: deterministic 'illegal memory access' at fast_boundary table update on first decode request (Q4_K_M drafter works)

  mw00/project-maya#173 · Filter by author waxinz opened on Oct 13, 2026
  1

• Speed report: RTX PRO 6000 Blackwell 96 GB (Ryzen 7 9800X3D, 61 GB RAM) - Maya-M, 81.0 tok/s decode / 3102 tok/s prefill (8K), Linux

  mw00/project-maya#172 · Filter by author yundddd opened on Oct 13, 2026
  1

• Speed report: Tesla V100-PCIE-32GB in an HP Z8 G4 (2x Xeon Gold 6138, 183 GB RAM) - Maya-S24, 26.8 tok/s decode / 541 tok/s prefill (8K), Linux + three items to review

  mw00/project-maya#171 · Filter by author mw00 opened on Oct 13, 2026

• Speed report: Maya-M 44 / Maya-L 30 tok/s on 2x RTX 3090 (EPYC 7452, 256 GB) + note on v1.1.5 long-prompt prefill with checkpoints

  mw00/project-maya#165 · Filter by author orochifj opened on Oct 13, 2026
  2

• RX 9070 XT (gfx1201, 16 GB) on Windows 11, HIP: speed report (Maya-S / Maya-S24) + two small issues

  mw00/project-maya#156 · Filter by author pros02 opened on Oct 13, 2026
  2