Builds / RX 9070 16GB
RX 9070 16 GB on Windows (HIP): Qwen3-Coder 30B-A3B Q3_K_L at 83.3 tok/s with an 8K prompt, 50.1 at 32K
Last updated:
- GPU(s)
- RX 9070 16GB
Amazon · eBay (used) · Newegg - VRAM / unified memory (GB)
- 16
- CPU
- AMD Ryzen 5 7600
Amazon · eBay (used) · Newegg - Motherboard
- not stated
- RAM
- about 32 GB
Amazon · eBay (used) · Newegg - PSU
- not stated
- Other
- Windows, llama.cpp b11146 HIP backend (ROCm 10.0.0), one slot, flash attention, q4_0 K/V cache
- Reported total cost
- not stated
- Power
- not stated
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| Qwen3-Coder 30B-A3B legacy | Q3_K_L, q4_0 KV | llama.cpp b11146 (HIP, ROCm 10.0.0) | 83.3 | 1,802 @ 8K-token prompt est. TTFT @8k ≈ 4.5 s (calculated from measured prefill) | context: 8,192 ctx, prompt about ctx - 256, 128 generated📏 Owner-run (minjunpark.blog, 4 Nov 2026), one run per size; TTFT 4.4 s |
| Qwen3-Coder 30B-A3B legacy | Q3_K_L, q4_0 KV | llama.cpp b11146 (HIP, ROCm 10.0.0) | 50.1 | 1,790 @ 32K-token prompt est. TTFT @8k ≈ 4.6 s (calculated from measured prefill) | context: 32,768 ctx, prompt about 32.5K, 128 generated📏 Same sweep: 16K 70.7 tok/s, 24K 59.4, 32K 50.1; TTFT 18.2 s at 32K |
Prompt processing = prefill speed (how fast the prompt is read) as reported by the source, with the prompt length it was measured at; blank = the source does not report it. est. TTFT @8k is calculated from the measured prefill speed (8,192 ÷ prefill tok/s), shown only when prefill was measured on a prompt of 512+ tokens. It is not a measurement and ignores load time and caching. measured TTFT is the source's own time-to-first-token. Why prefill matters for coding assistants.
Owner-run on a Windows desktop (published 4 Nov 2026). Q3_K_L is below Q4. The same model passed 17 of 19 tasks on the author's own local coding suite at 4K context, peak server VRAM about 14.3 GB. The page also has a small Qwen3.8 27B Q3_K_M vs UD-IQ4_XS quality pilot without speed numbers.
Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.