CARINA Ryzen AI Max 1L Mini AI Workstation

Overview

A one-litre box that holds a model your laptop cannot. The AMD Ryzen AI Max+ 395 pairs a 16-core CPU, a 40-compute-unit Radeon 8060S and a 50 TOPS NPU against a single 128 GB unified memory pool — so the graphics engine draws on system memory instead of a fixed VRAM budget. At 160 × 160 × 47 mm it sits under a monitor, runs on a 230 W adapter, and keeps your data in the building.

Key highlights

• AMD Ryzen AI Max+ 395 — 16 cores / 32 threads, 5.1 GHz boost, 80 MB cache.
• Up to 128 GB LPDDR5X-8000 on a 256-bit bus (~256 GB/s) — up to 96 GB assignable to graphics.
• 50 TOPS NPU plus a 40-CU Radeon 8060S — a genuine Copilot+ class AI PC.
• x86 and Windows 11 — no Arm port, no new toolchain, drops into an existing estate.
• 2× USB4 40 Gbps, 2.5GbE, Wi-Fi 7, SD 4.0 — docking and capture without adapters.
• 1 L chassis, 160 × 160 × 47 mm, external 230 W adapter — desk-side, not server-room.
• Built and supported in India by RDP, with GST invoice and pan-India onsite service.

AI workload fit

• Local inference on quantised open models at desk scale.
• RAG over private documents that must not leave the building.
• Agentic AI and application development against a local model.
• Light fine-tuning and LoRA work on small to mid models.
• Generative AI and NLP & speech prototyping.
• A per-desk AI seat for an engineer, analyst or clinician.

AI workload positioning

This is an inference and development machine, and we would rather say so than let you discover it. Unified memory decides which model fits; it does not make this a training box. A 50 TOPS NPU with an integrated GPU is not in the class of a discrete accelerator for sustained training, and anyone telling you otherwise is selling. What it does exceptionally well is hold a large quantised model in memory, on your desk, with no per-token meter and no data leaving your premises.

Industry use cases

• Enterprise & GCCs: a per-desk AI seat for platform and applied-AI teams on a Windows estate.
• BFSI: local inference over confidential material with nothing leaving the network.
• Healthcare: clinical document and note workflows kept on-premises under DPDP.
• Public sector & sovereign: on-soil AI at department scale, no cloud dependency.
• Research & education: a per-researcher machine that runs real models without cluster time.

Performance & how to be sure

The number that matters on this class of machine is memory capacity and bandwidth — 128 GB at roughly 256 GB/s decides which model fits and how fast it decodes, far more than any headline TOPS figure. Rather than quote a tokens/sec figure that will not match your workload, RDP offers a “benchmark your model” session: bring the model, precision and context length you actually intend to run, and we will validate it on this exact configuration before you commit.

Series & upgrade path

CARINA entry tier. Beside: the 2L Compact AI PC for a slightly higher power envelope. Above: the QUASAR 2L Liquid-Cooled AI PC for sustained load, then RDP AI Workstations with discrete RTX PRO GPUs when you need real fine-tuning throughput, and GPU servers beyond that. Compare with NVIDIA DGX Spark if your team needs CUDA parity with the servers they deploy to — we sell both and will tell you which fits.

On-prem vs cloud (TCO)

For sustained daily AI work this configuration removes per-GPU-hour billing, queueing for scarce instances, and egress charges on your own data — and keeps everything on-premises for DPDP and data-residency obligations. Cloud still wins for burst capacity and one-off very large training runs; on-prem wins on sustained utilisation, control and predictable INR capital cost. We model the crossover with your actual usage rather than assert it.

Software & day-one readiness

Ships with Windows 11 or Ubuntu, and the AMD AI stack — ROCm, DirectML and ONNX Runtime — with local model runtimes (LM Studio, Ollama, llama.cpp) validated before dispatch. Because the GPU draws from the same unified pool as the CPU, you assign memory to the model rather than shopping for a card with enough VRAM. Migration from a CUDA workflow is a real exercise and RDP scopes it honestly rather than calling it a drop-in swap.

Power, thermal & acoustics

Single external 230 W adapter (20 V / 11.5 A) on an ordinary socket. Air-cooled and quiet enough for a shared desk; no dedicated circuit, no rack, no facility work. Exact wattage, BTU and dB(A) figures come from RDP bench measurement rather than estimates — ask for the site-readiness sheet with your quote.

Deployment, warranty & support

Built to order by RDP Technologies. Supplied with GST invoice, pan-India onsite support, and availability through GeM for public-sector procurement. Built to order — typically 4–6 weeks.

Why RDP

RDP Technologies is a Make-in-India OEM with 14+ years and 300,000+ units shipped, supplying AI infrastructure from desk-side systems to rack-scale AI factories — with predictable INR pricing, GST invoicing and pan-India onsite service.

Buy with confidence

Use Request a Quote to reach an RDP solution architect for sizing, a benchmark-your-model session, financing options and a delivery plan. No obligation.

Specifications