NetwinuX & CyberPragmatic local AI inference fleet

--:--

Chatbot Arena

STRIX HALO OFFLINE — the machine is not answering. Values below are the last known readings.

Hardware status ~2200€ CAPEX & ~35€/month OPEX — electricity

Power Settings

▲ PPT

L M H

▼ CPU

Instant APU Power

—

GPU load

—

GPU clock

—

GPU temperature

—

CPU load

—

CPU clock

—

CPU temperature

—

RAM Unified LPDDR5X — — / 128 GiB total (—%)

LLM runtimes (distrobox CT image)

#	CT	ROCm Runtimes	LLM	Slots	Ctx/slot	RAM + GTT	Throughput Live (tok/s)

Nightly Benchmarks (prefill 0; cold prompt 2048 tokens; output 128 tokens)

#	LLM	Architecture	Weight (GiB)	Max context	Vision	PP (tok/s)	TG (tok/s)

RTX-4090 OFFLINE — the machine is not answering. Values below are the last known readings.

Hardware status ~5500€ CAPEX & ~80€/month OPEX — electricity

Power Profile

▲ PROFILE

S B P

▼ CPU

▼ EPP

Instant GPU Power

—

GPU load

—

GPU clock

—

GPU temperature

—

CPU load

—

CPU clock

—

CPU temperature

—

RAM system DDR5 — — / 128 GiB total (—%)

VRAM dedicated GDDR6X — — / 24 GiB

LLM runtimes (podman CT image)

#	CT	CUDA Runtimes	LLM	Slots	Ctx/slot	VRAM	Throughput Live (tok/s)

Benchmarks (prefill 0; cold prompt 2048 tokens; output 128 tokens)

#	LLM	Architecture	Weight (GiB)	Max context	Vision	PP (tok/s)	TG (tok/s)

DGX OFFLINE — the machine is not answering. Values below are the last known readings.

Hardware status ~4800€ CAPEX & ~25€/month OPEX — electricity

Power Profile disabled

S B P

NVMe Storage

—

Instant GPU Power

—

GPU load

—

GPU clock

—

GPU temperature

—

CPU load

—

CPU clock

—

CPU temperature

—

RAM Unified LPDDR5X — — / 128 GiB total (—%)

LLM runtimes (Docker CUDA/ARM64 CT image)

#	CT	CUDA Runtimes	LLM	Slots	Ctx/slot	RAM	Throughput Live (tok/s)

Benchmarks (prefill 0; cold prompt 2048 tokens; output 128 tokens)

#	LLM	Architecture	Weight (GiB)	Max context	Vision	PP (tok/s)	TG (tok/s)