← Return to Fleet

Neon Future

Status: benchmarking
A prototype build simulating the next generation of consumer AI hardware. Focused on high-bandwidth memory and neural-specific instruction sets. **Update (02/14/26):** Upgraded from RTX 5060 (8GB) to RTX 5070 (12GB) to test higher VRAM capacity models.
GPUNVIDIA RTX 5070 (12GB)
VRAM12GB GDDR7
CPUIntel Core Ultra 5
RAM96GB DDR5
Storage2TB NVMe Gen5
OSUbuntu
Shop This Rig ↗

Performance Log

DateModelHardwareParamsContextPrompt EvalToken Gen
2026-02-15Gemma-3-4.3BRTX 5070 (12GB)Q4_K_M4096 tk8944.5 t/s141.8 t/s
2026-02-15Gemma-3-4.3B (32k Context)RTX 5070 (12GB)Q4_K_M32768 tk7953.6 t/s122.7 t/s
2026-02-15Gemma-3-4.3B (64k Context)RTX 5070 (12GB)Q4_K_M65536 tk6968.6 t/s109.5 t/s
2026-01-29Gemma-3-4.3BRTX 5060 (8GB)Q4_K_M4096 tk5545.6 t/s115.4 t/s
2026-01-29Gemma-3-4.3B (32k Context)RTX 5060 (8GB)Q4_K_M32768 tk4809.1 t/s95.6 t/s
SPONSORED TESTS// AD_SLOT: 1234567890 // FORMAT: AUTO

Performance Analysis

Trend Visualization

Historical Archive

Filter:
Showing 5 of 5 runs
DateModelQuantContextVRAMPrompt (t/s)Gen (t/s)
2026-02-15Gemma-3-4.3BQ4_K_M4,096-8944.5141.8
2026-02-15Gemma-3-4.3B (32k Context)Q4_K_M32,768-7953.6122.7
2026-02-15Gemma-3-4.3B (64k Context)Q4_K_M65,536-6968.6109.5
2026-01-29Gemma-3-4.3BQ4_K_M4,096-5545.6115.4
2026-01-29Gemma-3-4.3B (32k Context)Q4_K_M32,768-4809.195.6