JARVIS@WORK
Sign inBook a discovery call→
JX—001 · Pre-loaded AI appliance

The brain
behind Jarvis.

AMD AI Max+ 395 · 96 GB unified memory · 50+ TOPS neural processing. Everything needed to run 70B-parameter models locally — in a 1-litre aluminium enclosure.

Jarvis JX—001 appliance — angled view
16 coresZen 5 CPU
96 GBUnified memory
50+ TOPSNeural processing
160 WTDP sustained
~1 LForm factor
10G LANWired networking
§ 01 The Chip

AMD AI Max+ 395 Strix Halo

CPU16-core Zen 5 · 3.0–5.1 GHz
GPU40 RDNA 4 CUs · 50+ TOPS
Memory96 GB unified LPDDR5X · 256 GB/s
NPUXDNA 2 · dedicated AI acceleration
AMD AI Max+ 395 Strix Halo
§ 02 Performance

Benchmark results

Cinebench R24 MT40,825 pts
Time Spy11,850 pts
LLM inference60–90 tok/s @ 70B Q4
Context128k tokens in unified memory
Benchmark results
§ 03 Connectivity

I/O & network

USB440 Gbps (×2) + Oculink 64 Gbps
LAN10G RJ45 (Intel i226-V)
Wi-Fi7 (802.11be) + Bluetooth 5.4
USB-A2× USB 3.2 Gen2 · 1× USB 2.0
I/O & network
§ 04 Form Factor

Physical specs

Dimensions220 × 133 × 35 mm
Volume~1 litre
Weight< 1 kg
MaterialCNC aluminium enclosure
Physical specs
§ 05 Thermal

Cooling system

Stage 1Vapour chamber direct-die contact
Stage 2Twin turbofan array
Stage 3Phase-change conduction pad
TDP160 W sustained · fanless burst capable
Cooling system
§ 06 · Model compatibility

What runs on the JX—001

ModelParamsQuantRAMSpeed
Qwen3-235B-A22B235BQ2_K80 GB18 tok/s
Llama4-Scout-17B17BQ8_020 GB110 tok/s
Mistral-Large-2123BQ4_K_M68 GB32 tok/s
Llama-3.3-70B70BQ6_K62 GB55 tok/s
Phi-4-Mini3.8BF168 GB200+ tok/s

Ready to see it
in your building?

Book a 30-minute discovery call. We'll scope a deployment plan tailored to your team and data sources.

Book a discovery call →← Back to Jarvis