Matrix

Gemma12-5060 GrokBuild matrix

2026-09-30 · speed stamp 20260930-150835-1 · quality stamp 20260930-121551 · RTX 5060 Ti · GrokBuild

  • 2026-09-30
  • 5060 Ti
  • GrokBuild
  • gemma4:12b Q4_K_M (speed)

Five matrix cells from Neo run-cards only. Speed refreshed to stamp 20260930-150835-1 (eng tok/s leads; wall tok/s secondary when present). Quality cells stay stamp 20260930-121551 with path noted (LMS/QAT path). Unknowns stay unknown/n/a. Keepers stay side/background chrome — not inside cells.

Stamp SoT (split): speed 20260930-150835-1 · weights gemma4:12b (Q4_K_M) · quality still 20260930-121551 · weights gemma4-12b-qat:latest (Q4_0) · cell prefix 5060/Gemma12-5060/*. JSON under .grok/notes/run-cards (no local machine paths on this page).

Lane × GPU matrix

Lane by GPU outcome matrix
Lane 5060 Ti CUDA
text mixed Ollama
describe not run no card this stamp
image-gen not run no card this stamp
video-gen not run no card this stamp

Text lane is mixed because bugfix + tooljson pass, scriptreview is partial 3/6, refactor is ok (ungraded). Logo-card keepers are side/background chrome only.

Ollama (CUDA)

Model × task table for this stamp. Hardware is a column. Tasks are prompts (hover headers). Cell chips link full run cards below.

Model Hardware bugfix tooljson refactor scriptreview
gemma4-12b-qat:latest 5060 Ti pass 1/1 pass 4/4 ok partial 3/6

Serving: Ollama · Engine: Ollama CUDA servergen1 · Harness: GrokBuild · host win

Ollama (CUDA) — speed cell

Speed suite from stamp 20260930-150835-1 only. Lead with engine tok/s + outcome; wall tok/s secondary when present (Neo card). Quality suites below remain stamp 20260930-121551.

Gemma12-5060 speed eng and wall tok/s
ModelQuantHardware Engine tok/s Wall tok/s ttft_msOutcomeCard
gemma4:12b Q4_K_M RTX 5060 Ti 16GB 88.18 39.59 24000 ok 20260930-150835-1

tok/s_gen by suite (split stamps)

Speed row = 150835-1. Quality rows = still 121551 with path noted — not claimed as 150835 measurements.

SuiteStampStatus / accuracyWall tok/sEngine tok/swall_secsttft_msrun_id
speed150835ok39.5988.18n/a2400020260930-150835-1
bugfix121551pass 1/11.7718.8885.27600020260930-121551-2
refactor121551ok1.9917.570.26100020260930-121551-3
tooljson121551pass 4/43.7917.5592.77100020260930-121551-4
scriptreview121551partial 3/618.4420.68321.93400020260930-121551-5

Full run cards (all columns)

Mandatory Experience→Silicon fields per llm-run-card-schema-2026-09-20. Values copied from stamp JSON / Neo handoff only. Speed row is 150835-1; quality rows remain 121551-2..5. LAN endpoints omitted on public HTML — open the linked JSON for stack detail. Perf set: lead with engine tok/s_gen; wall tok/s is secondary when both are present. Unknowns stay n/a/null; columns are not dropped.

Scroll horizontally for Silicon columns (UI → Hardware → Ctx → tok/s → notes). Heading says all columns because the markup keeps them — use the scrollbar under the table.

5060 Ti cards (5) — speed 150835 + quality 121551

Gemma12-5060 run cards
run_iddatehostcell UIHarnessTools/MCPGateway ServingEngineWeightsHardware OS_pathCtxMTP/spec load_sttft_ms tok/s_gen walltok/s_gen enginetok/s_prefill Tokens (p/c/t)concurrency custom_suiteJSON
20260930-150835-12026-09-30I9RUMBLE5060/Gemma12-5060/speed noneGrokBuildnonenone OllamaOllama CUDA servergen1gemma4:12b (Q4_K_M)NVIDIA RTX 5060 Ti 16GB Win nativeallocated=unknown; filled=unknownoff null24000 39.5988.18null 25829 / 1940 / 277691 speed; status=ok; accuracy=null; speed stamp 150835 (quality cells still 121551) JSON
20260930-121551-22026-09-30I9RUMBLE5060/Gemma12-5060/bugfix noneGrokBuildnonenone OllamaOllama CUDA servergen1gemma4-12b-qat:latest (Q4_0)NVIDIA RTX 5060 Ti 16GB Win nativeallocated=unknown; filled=unknownoff null76000 1.7718.88null 26022 / 151 / 261731 bugfix; status=pass; accuracy=pass 1/1 JSON
20260930-121551-32026-09-30I9RUMBLE5060/Gemma12-5060/refactor noneGrokBuildnonenone OllamaOllama CUDA servergen1gemma4-12b-qat:latest (Q4_0)NVIDIA RTX 5060 Ti 16GB Win nativeallocated=unknown; filled=unknownoff null61000 1.9917.5null 23698 / 140 / 238381 refactor; status=ok; accuracy=null JSON
20260930-121551-42026-09-30I9RUMBLE5060/Gemma12-5060/tooljson noneGrokBuildnonenone OllamaOllama CUDA servergen1gemma4-12b-qat:latest (Q4_0)NVIDIA RTX 5060 Ti 16GB Win nativeallocated=unknown; filled=unknownoff null71000 3.7917.55null 25995 / 351 / 263461 tooljson; status=pass; accuracy=pass 4/4 JSON
20260930-121551-52026-09-30I9RUMBLE5060/Gemma12-5060/scriptreview noneGrokBuildnonenone OllamaOllama CUDA servergen1gemma4-12b-qat:latest (Q4_0)NVIDIA RTX 5060 Ti 16GB Win nativeallocated=unknown; filled=unknownoff null34000 18.4420.68null 16289 / 5936 / 222251 scriptreview; status=partial; accuracy=partial 3/6 JSON

All Builds Tok Lanes numbers B70 vs 5080

Preview branch fill only until Matthew merges. No apex deploy from this change.