F894 — GPU Working Cortex Cache
F894 covers “GPU Working Cortex Cache”. The canonical record classifies it as validated, based on 2 source artifacts and 1 original source figure. The highlighted measurements are Pass = Yes, Cpu Passed = 6, and Cpu Pass Rate = 1.
Purpose and research question
F894 evaluated a GPU working-cortex cache candidate under cold and warm-cache conditions, asking whether repeated recall could offset GPU setup cost without changing the language path or promoting a new runtime default.
Method
F894 simulated 20,000 working-cortex rows and ran six probes in each CPU, cold-GPU, and warm-cache condition. It counted cold GPU executions and hot cache hits, measured average latency for all three paths, and added a two-call reason-memory smoke at a 32-row threshold; the default 128-row threshold and runtime promotion state remained unchanged.
Documentation basis
2 source artifacts and 1 phase-specific figure are recorded for this phase.
Evidence highlights
- Rows simulated: 20,000
- CPU average: 26.597 ms
- GPU cold average: 151.391 ms
- GPU hot average: 6.09 ms
- GPU hot cache hits: 6/6
- Promoted: false
Recorded evidence
| Pass | Yes |
|---|---|
| Cpu Passed | 6 |
| Cpu Pass Rate | 1 |
| Cpu Avg Latency Ms | 26.597 |
| Gpu Cold Passed | 6 |
| Gpu Cold Pass Rate | 1 |
| Gpu Cold Avg Latency Ms | 151.391 |
| Gpu Hot Passed | 6 |
Public evidence and source figures
Original source figure · 1190 × 630 px
These are original source figures from the existing public evidence set, not editorial HTML visualizations.
Result and interpretation
CPU average latency was 26.597 ms, GPU cold latency 151.391 ms, and GPU hot latency 6.09 ms. All 6 cold probes used the GPU and all 6 hot probes hit the cache. The candidate passed as a warm-cache optimization, but the report explicitly recorded no promotion.
Key results
- Pass
- Yes
- Cpu Passed
- 6
- Cpu Pass Rate
- 1
- Cpu Avg Latency Ms
- 26.597
- Gpu Cold Passed
- 6
- Gpu Cold Pass Rate
- 1
- Gpu Cold Avg Latency Ms
- 151.391
- Gpu Hot Passed
- 6
Evidence sources
Values on this page are reproduced from the archived phase record and the source artifacts below. No new benchmark was run to create this page.
- GPU WORKING CORTEX CACHE 894 (JSON)
- GPU WORKING CORTEX CACHE 894 (MD)
Research topics
Word Memory · Language · Memory · Reasoning