COMPUTER RECORD · SOURCED SPECIFICATIONS

Seeed reComputer J3010: specifications and local LLMs

4 GB unified memory and 51 GB/s rated memory bandwidth. Compact CUDA edge device with a 128GB NVMe SSD.

Catalog reviewed 2026-10-08. Listing evidence: 2026-09-25. Price and performance records retain their own dates.

As an Amazon Associate I earn from qualifying purchases.

CPU, GPU, memory and storage

CPU: Arm Cortex-A78AE · Orin Nano · 6 cores / 6 threads
GPU: Ampere · 512 CUDA cores
Installed memory: 4 GB · LPDDR5 · 64-bit · Owner-supplied Amazon.com screenshot
Unified memory, not separate VRAM: 4 GB shared with CPU and operating system
Memory bandwidth: 51 GB/s rated / theoretical · NVIDIA rates the original Orin Nano 4GB module at 51 GB/s; shared by CPU and GPU.
Storage: 128 GB · Owner-supplied Amazon.com screenshot
Operating system: JetPack 5.1.1 / Linux (screenshot)
Expansion: M.2 storage; fixed memory
Edge AI advertised peak: 20 TOPS · not LLM tokens/s

4GB limits model size and context. Confirm the power adapter before purchase.

Screenshot confirms 4GB / 128GB SSD and JetPack 5.1.1. Advertised AI performance is 20 TOPS, or 34 TOPS with an upgrade; the table retains the original 20 TOPS. Prior listing bullets conflict about the power adapter.

Owner-supplied Amazon.com screenshot reviewed 25 Sep 2026; capture time not supplied. J3010 / Orin Nano 4GB; 128GB SSD; JetPack 5.1.1; 20 TOPS, advertised upgrade to 34 TOPS; 130x120x58.5 mm; 4x USB 3.2, Wi-Fi/BT, M.2 Key E. ASIN not visible; matched to the stored record by product name and visible configuration. No live Amazon access. Current seller, stock and checkout price can change.

What local models could Seeed reComputer J3010 run?

The 4 GB unified-memory pool gives a 2 GB initial weight budget using our 30% (minimum 2 GB) reserve. The CPU, GPU and OS share this capacity; software allocation limits still apply.

28 catalog files pass this size check. The largest is TwIL LM3 at 1.915 GB (Q4_K_M). This is a catalog weight limit, not a measured maximum practical model size. Context length, runtime and architecture support can rule out a candidate.

Model snapshot: 14 October 2026. The examples below use MMLU-Pro; models without that score are not ranked in this table.

MMLU-Pro memory candidates · GB = 1 billion bytes. Published scores may use different settings or precision.
Model / quantization	Weights	Parameters	Score / 100	Score / GB	Budget left	Sources
LFM2.5 230MQ4_K_M · Pareto choice	0.153 GB	0.23B	20.25	132.002	1.847 GB	Get weights Score source
MiniCPM5 2BQ4_K_M · Pareto choice	1.561 GB	2.517B	70.8	45.346	0.439 GB	Get weights Score source
LFM2.5 2.6BQ4_K_M	1.674 GB	2.697B	65.2	38.938	0.326 GB	Get weights Score source

Choose this computer in the memory tool · Compare its memory tier

Published LLM performance

Local AI Stats has not tested this computer. No matching token-speed result is recorded for this configuration.

Advertised compute peaks and bandwidth are not tokens per second. Compare the same model, quantization, context and concurrency.

Price and value

ASIN: B0D9Y1SP5C · Amazon.com listing

$599 USD reference. Amazon.com base price visible in an owner-supplied screenshot reviewed 25 Sep 2026; capture time not supplied. Dated reference, not a live quote. Excludes card signup offers, coupons, tax and shipping.

Reference price per installed-memory GB: $149.75. This measures cost per GB, not speed or total running cost.

As an Amazon Associate I earn from qualifying purchases. Check the exact configuration, seller, marketplace and current price before buying.

GET NOW · Amazon

Related hardware

Seeed reComputer Industrial J4012 · 16 GB unified memory
Apple Mac mini M4 Pro (2024) · 24 GB unified memory
Apple Mac mini M6 · 24GB · 24 GB unified memory
Apple Mac mini M5 Pro · 24GB · 24 GB unified memory

Compare all hardware · Compare budgets