Rockchip RK3576 / Qwen3.5 4B
Can a Rockchip RK3576 run Qwen3.5 4B?
Yes, usable but slower than reading speed. Measured generation speed: 3.03–5.02 tok/s across 3 public runs.
Rule of thumb: 10+ tok/s feels like a chat; 4–10 is usable for assistants and automations; under 4 is batch-job territory. Speeds depend on quantization, runtime, context length and cooling. Check each source.
| Board | Quant | Runtime | Gen tok/s | Prompt tok/s | TTFT ms | Source |
|---|---|---|---|---|---|---|
| Rockchip RK3576 Rockchip RK3576 (vendor reference board) | w4a16 | RKLLM v1.3.1 | 5.02 | – | 3976.41 | source ↗ commit 2026-09-29 |
| Rockchip RK3576 Rockchip RK3576 (vendor reference board) | w4a16_g128 | RKLLM v1.3.1 | 4.6 | – | 4476.37 | source ↗ commit 2026-09-29 |
| Rockchip RK3576 Rockchip RK3576 (vendor reference board) | w8a8 | RKLLM v1.3.1 | 3.03 | – | 3906.19 | source ↗ commit 2026-09-29 |
Qwen3.5 4B on other boards
- Jetson AGX Thor 68 tok/s best
- Jetson Orin Nano 18 tok/s best
- Rockchip RK3588 6.22 tok/s best
Buy a Rockchip RK3576
- Radxa ROCK 4D 8GB: $106 listed price, source ↗ (2026-10-17) Check price (affiliate slot)
Ran Qwen3.5 4B on a Rockchip RK3576? Submit your numbers →