More
Best rig by budgetRent before you buySubmit a buildNewsletterFree datasetAdvertise
Submit build

Builds / RTX 5060 Ti 16GB

RTX 5060 Ti 16 GB as a USB4 eGPU on an M1 Max MacBook Pro (Omarchy Linux): Qwen3.8 27B Q4_0 at 24 tok/s, pp512 849

Last updated:

Source: joshuawarren.com ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
GPU(s)
RTX 5060 Ti 16GB
Amazon · eBay (used) · Newegg
VRAM / unified memory (GB)
16
CPU
Apple M1 Max (16-inch MacBook Pro running Omarchy Linux)
Motherboard
not stated
RAM
not stated
PSU
not stated
Other
External GPU over USB4 on a MacBook Pro running Omarchy (Arch-based) Linux, kernel 7.1.12-2-20261029.2-lab-ARCH; llama.cpp 65840ed CUDA; an Arc Pro B60 24 GB (Vulkan, ANV 26.2.5) is attached as a second eGPU and was measured separately
Reported total cost
not stated
Power
not stated

Measured speed (generation tokens/sec)

ModelQuantEnginetok/sPrompt processing (tok/s)Details
Qwen3.8 27BQ4_0llama.cpp 65840ed (CUDA)24849 @ 512-token prompt
est. TTFT @8k ≈ 9.6 s (calculated from measured prefill)		context: llama-bench defaults📏 llama-bench tg128, 3 repetitions (± 0); measured Nov 2026 (post dated 2 Nov 2026)

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 post by Joshua Warren (joshuawarren.com, 2 Nov 2026), all rows separate llama.cpp runs on the same host. A 27B Q4_0 file is larger than 16 GB, so part of the model likely sits outside VRAM (the post does not say how it was split). Same post, same host: Arc Pro B60 24 GB over USB4 on Vulkan read Qwen3.8 27B Q4_0 at 203 tok/s (pp512) but decoded only about 3 tok/s; the M1 Max's own GPU on Vulkan (Honeykrisp) 63 / 9 tok/s. Llama-2-7B Q4_0: 5060 Ti 2,466 / 95, B60 1,117 / 20, M1 Max 240 / 34 tok/s.

Current GPU street price (estimate, not the builder's cost)Estimated GPU cost for this build today: ≈€739
New retail prices in Germany incl. VAT; used cards are often much cheaper. Prices change daily.
Source: https://joshuawarren.com/blog/two-external-gpus-and-the-m1-max-gpu-llama-cpp-benchmarks/

Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.

Embed this result

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

[![Measured on RigBase](https://getrigbase.com/badge/rtx-5060-ti-usb4-egpu-m1-max-macbook-omarchy-qwen38-27b-joshuawarren.svg)](https://getrigbase.com/builds/rtx-5060-ti-usb4-egpu-m1-max-macbook-omarchy-qwen38-27b-joshuawarren/)

HTML

<a href="https://getrigbase.com/builds/rtx-5060-ti-usb4-egpu-m1-max-macbook-omarchy-qwen38-27b-joshuawarren/" rel="noopener"><img src="https://getrigbase.com/badge/rtx-5060-ti-usb4-egpu-m1-max-macbook-omarchy-qwen38-27b-joshuawarren.svg" alt="Measured on RigBase" height="20"/></a>

Preview: Measured on RigBase · How to use

Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)

Quick answers: Can RTX 5060 Ti 16GB run Qwen3.8 27B?