LLMs on the Radxa ROCK 5B+ (RK3588)
4 sourced results. 2 configurations reached 5+ tokens/s generation.
Good for
- Small 1–3B (chat, simple agents): fast, feels like chat, median 17.0 t/s (≈13 words/s), best 17.0 t/s, 1 test
- Big 14B+: usable, median 6.8 t/s (≈5 words/s), best 9.6 t/s, 2 tests
- Not tested yet: Medium 7–8B. Help fill the gap
No Indian price found yet. Labels: ≥15 t/s fast · 5–15 usable · 2–5 slow · <2 too slow to chat. 1 token ≈ 0.75 words.
Real-world uses
Our data for this board: small (1–3B) models: median 17.0 t/s across 1 tests (fast, feels like chat); big (14B+) models: median 6.8 t/s across 2 tests (usable). Use cases that need chat-speed replies want ≥5 t/s; background jobs can live with 2 t/s.
- Home server + local chatbot: Low-power always-on box running a small model for family chat or Home Assistant.
- NPU-accelerated small models: RK3588's NPU (via rknn-llm or NPU forks) can speed up small models; see our NPU vs CPU rows.
- Private document Q&A: Small RAG setups on 8–32GB RAM variants.
- Media + AI kiosk: Multiple display outputs plus a small model for signage or info desks.
Your board here
Reach makers and engineers choosing edge-AI hardware. Sponsor this slot →
India price comparison
Prices as seen on retailer pages (via search snapshots) on the date shown. Not live; confirm on the retailer page. Buy links are plain links for now: affiliate tags will be added later where marked.
| Board ↕ | Model ↕ | Runtime & setup ↕ | Gen t/s ↕ | Price ₹ (India) ↕ | Source ↕ |
|---|---|---|---|---|---|
| Radxa ROCK 5B+ (RK3588)16 GB RAM | Gemma 3 1B IT1B params · Q8_0 | rk-llama.cpp fork (NPU) | 17.0prompt 201.6 t/s | No India listing yet | Community Mojo24x7/rk-llama.cpp2026 |
| Radxa ROCK 5B+ (RK3588)16 GB RAM | Qwen3 30B-A3B30B params · Q4_0 | rk-llama.cpp fork (NPU) | 9.6prompt 21.1 t/s | No India listing yet | Community Mojo24x7/rk-llama.cpp2026 |
| Radxa ROCK 5B+ (RK3588)16 GB RAM | Qwen3.6 35B-A3B35B params · Q4_0 | rk-llama.cpp fork (NPU) | 4.1prompt 18.5 t/s | No India listing yet | Community Mojo24x7/rk-llama.cpp2026 |
| Radxa ROCK 5B+ (RK3588)16 GB RAM | Gemma 4 E4B ITQ8_0 | rk-llama.cpp fork (NPU) | 4.0prompt 38.2 t/s | No India listing yet | Community Mojo24x7/rk-llama.cpp2026 |
Each row shows only what its source reported: RAM, quant, prompt speed and setup appear under the main value when published. Rows with full details are listed first, fastest first. Source badges: Community, Vendor, Reproduced; ⚠ marks a speed far from similar results (how we check).