01 OVERVIEW
CLUSTER STABLEPREFILL SPEED +151%
1,265.7tok/s
▲ UNLOCK前 503 tok/s (2.51倍)
TTFT LATENCY -58%
820ms
▼ 1,962ms → 1秒未満へ半減
DECODE THROUGHPUT MEM-BOUND
84.8tok/s
+23.1% (物理帯域504GB/s上限)
CLOUD API COST ECONOMICS
$0.00/DAY
Strataローカル完全オフロード
GPU CLUSTER HARDWARE & gpu-unlock STATE
NVIDIA 595.99.02 / cuBLAS 13
GPU 0: RTX 4070 Ti SUPER (16GB GDDR6X)
Gen4 x16
158.5 TFLOPS
TDP 285W
UNLOCKED (90.1%)
GPU 2: RTX 3060 12GB (12GB GDDR6)
Gen4 x4
41.4 TFLOPS
TDP 170W
UNLOCKED (81.2%)
GPU 1: RTX 4060 8GB (8GB GDDR6)
Gen1 x4 (DEGRADED)
-- TFLOPS
TDP [N/A]
ISOLATED (Xid 79 SAFE)
LIVE AUTONOMOUS EXECUTION CONSOLE
AUTO-STREAMING
[22:36:12]STRATAKernel initialized with dual GPU pipeline [2, 0]
[22:38:44]UNLOCKRate throttle bit 0x88 cleared on PCIe 0000:01:00.0 (RTX 4070 Ti S)
[22:39:01]GEMM2048^3 benchmark completed: 158.5 TFLOPS achieved (cuBLAS 13 native)
[22:41:20]PREFILLqwen3.8-flash-next-iq3_xxs prompt evaluated @ 1265.7 tok/s
[22:42:08]SAFEGUARDRTX 4060 link check: AER error caught, device quarantined from active lanes
[22:43:00]TAILSCALEDirect peer tunnel active: kazukimacbook-air <-> z790-al (4ms)
