Builds / AMD Radeon 8060S (integrated, Strix Halo)
HP ZBook Ultra G1a (Ryzen AI Max+ 395, 64 GB): gpt-oss 20B at 67.2 tok/s, Qwen3.6 35B-A3B at 52.3 (llama.cpp Vulkan)
Last updated:
- GPU(s)
- AMD Radeon 8060S (integrated, Strix Halo)
Amazon · eBay (used) · Newegg - VRAM / unified memory (GB)
- 64
- CPU
- AMD Ryzen AI Max+ 395 (16 cores / 32 threads)
Amazon · eBay (used) · Newegg - Motherboard
- HP ZBook Ultra G1a 14-inch laptop
Amazon · eBay (used) · Newegg - RAM
- 64 GB LPDDR5X-8000 soldered (16 GB set aside for graphics; Windows sees 47.8 GB)
Amazon · eBay (used) · Newegg - PSU
- not stated
- Other
- Windows 11 Pro, timed on mains power; llama.cpp b11669 Vulkan build, whole model on the Radeon 8060S. ROCm 7.2.1 in WSL 2 was slower on decode for every model (noted per row)
- Reported total cost
- not stated
- Power
- not stated
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| gpt-oss 20B | MXFP4 (12.1 GB GGUF), reasoning low | llama.cpp b11669 (Vulkan) | 67.2 | 1,220 @ 15,000-token prompt est. TTFT @8k ≈ 6.7 s (calculated from measured prefill) | context: code prompt, 32K window📏 measuredlocally test: one user, thinking off, same code prompt on every machine; middle of three runs (idle machine, 8 quiet seconds before each run); llama.cpp b11669, 32,768-token window; Windows power mode Balanced; ROCm in WSL 2: 51.6 |
| Qwen3.6 35B-A3B | UD-Q4_K_M (22.1 GB) | llama.cpp b11669 (Vulkan) | 52.3 | 1,006 @ 19,000-token prompt est. TTFT @8k ≈ 8.1 s (calculated from measured prefill) | context: code prompt, 32K window📏 measuredlocally test: one user, thinking off, same code prompt on every machine; middle of three runs (idle machine, 8 quiet seconds before each run); llama.cpp b11669, 32,768-token window; Windows power mode Best performance; ROCm in WSL 2: 39.2; 16 users at once: 75 tok/s in total |
| Gemma 4 26B-A4B | UD-IQ4_XS (13.6 GB) | llama.cpp b11669 (Vulkan) | 43.3 | 271 @ 19,000-token prompt est. TTFT @8k ≈ 30.2 s (calculated from measured prefill) | context: code prompt, 32K window📏 measuredlocally test: one user, thinking off, same code prompt on every machine; middle of three runs (idle machine, 8 quiet seconds before each run); llama.cpp b11669, 32,768-token window; Windows power mode Balanced; ROCm in WSL 2: 34.6 (prefill 596) |
| Qwen3.5 9B | Q4_K_M (5.7 GB) | llama.cpp b11669 (Vulkan) | 38.1 | 775 @ 19,000-token prompt est. TTFT @8k ≈ 10.6 s (calculated from measured prefill) | context: code prompt, 32K window📏 measuredlocally test: one user, thinking off, same code prompt on every machine; middle of three runs (idle machine, 8 quiet seconds before each run); llama.cpp b11669, 32,768-token window; Windows power mode Best performance; ROCm in WSL 2: 32.9 (prefill 935) |
| Gemma 4 12B | QAT Q4_0 (7.0 GB, Google's file) | llama.cpp b11669 (Vulkan) | 25.9 | 182 @ 19,000-token prompt est. TTFT @8k ≈ 45.0 s (calculated from measured prefill) | context: code prompt, 32K window📏 measuredlocally test: one user, thinking off, same code prompt on every machine; middle of three runs (idle machine, 8 quiet seconds before each run); llama.cpp b11669, 32,768-token window; Windows power mode Balanced; ROCm in WSL 2: 22.6 (prefill 380) |
| Qwen3.8 27B | Q4_K_M (17.1 GB, Unsloth) | llama.cpp b11669 (Vulkan) | 12.1 | 205 @ 19,000-token prompt est. TTFT @8k ≈ 40.0 s (calculated from measured prefill) | context: code prompt, 32K window📏 measuredlocally test: one user, thinking off, same code prompt on every machine; middle of three runs (idle machine, 8 quiet seconds before each run); llama.cpp b11669, 32,768-token window; Windows power mode Balanced; ROCm in WSL 2: 10.2 (prefill 259) |
| Qwen3.8 27B | Q4_K_M (17.1 GB, Unsloth) + MTP (2 draft tokens) | llama.cpp b11669 (Vulkan) | 25.0 | 215 @ 19,000-token prompt est. TTFT @8k ≈ 38.1 s (calculated from measured prefill) | ⚡ spec-decode (MTP)context: code prompt, 32K window📏 measuredlocally test: one user, thinking off, same code prompt on every machine; middle of three runs (idle machine, 8 quiet seconds before each run); llama.cpp b11669, 32,768-token window; Windows power mode Best performance; 18.5 on the prose prompt; ROCm in WSL 2: 20.2 |
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.
measuredlocally.com (Alex, UK), tested 22-27 Oct 2026 on his own laptop. Vulkan decoded faster than ROCm on every model, but ROCm read long prompts faster on the dense models. Some Vulkan runs were in Windows' Balanced power mode (noted per row). Artificial Analysis's AgentPerf-Local replay with Qwen3.8 27B + MTP took 43.2 min (Vulkan). A ~2-bit Nemotron 3.5 Bolt run (28.6 tok/s) is below Q4 and not listed.
https://measuredlocally.com/hardware-leaderboard.html
Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.
Copy into a README or blog. Links back to this build. See badge docs (CC BY 4.0 attribution). The SVG does not invent speeds — numbers stay on the build page.
Markdown
[](https://getrigbase.com/builds/hp-zbook-ultra-g1a-strix-halo-64gb-measuredlocally/)
HTML
<a href="https://getrigbase.com/builds/hp-zbook-ultra-g1a-strix-halo-64gb-measuredlocally/" rel="noopener"><img src="https://getrigbase.com/badge/hp-zbook-ultra-g1a-strix-halo-64gb-measuredlocally.svg" alt="Measured on RigBase" height="20"/></a>
Preview: 
· How to use
Quick answers: Can Strix Halo (Radeon 8060S) run Gemma 4 12B? · Can Strix Halo (Radeon 8060S) run Gemma 4 26B-A4B? · Can Strix Halo (Radeon 8060S) run Qwen3.5 9B? · Can Strix Halo (Radeon 8060S) run Qwen3.6 35B-A3B? · Can Strix Halo (Radeon 8060S) run Qwen3.8 27B? · Can Strix Halo (Radeon 8060S) run gpt-oss 20B?