AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S) vs Minisforum MS-R1 (as tested by Jeff Geerling)

AMDAI mini-PCNo licensed photo yet

(independent test)independent test systemNo licensed photo yet

AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S) is faster in all 3 like-for-like comparisons (largest gap: LLM generation, 69.35×).

	AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S)	Minisforum MS-R1 (as tested by Jeff Geerling)
Best LLM generation	249 tok/s C 2026-03-03 (published)	71.36 tok/s C 2025-11-10 (page last edited)
Power	120 W Chip TDP, Accelerator only, host not included, max of range 45-120 W	n/a
Perf / W	2.08 tok/s/W Chip TDP, Accelerator only, host not included, max of range 45-120 W	n/a
Price	n/a	n/a
Workloads with data	2 of 7	1 of 7

Best published LLM generation result per platform (may be from different like-for-like groups — see data). n/a = not published; never estimated.

Show data — all workloads, sources and like-for-like groups

object detection

Only AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S) has published object detection results.

Platform	Result	Model · precision · stack	Grade	Power	Source
AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S)	476 FPS inference_time · b=1	YOLO26n FP32 · Ultralytics 8.4.165 ONNX Runtime MIGraphX EP · nano (YOLO-n)	B	120 W Chip TDP, Accelerator only, host not included, max of range 45-120 W reported in result source — Chip TDP, Accelerator only, host not included, max of range 45-120 W	Ultralytics documentation: AMD ROCm/MIGraphX integration benchmark 2026-10-04 (page last edited)

Minisforum MS-R1 (as tested by Jeff Geerling):

◌

No published object detection results for Minisforum MS-R1 (as tested by Jeff Geerling) or AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S) yet.
We only show cited public figures — nothing is estimated to fill gaps. Requests tell us what to source or test next.

instance segmentation

◌

No published instance segmentation results for AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S) or Minisforum MS-R1 (as tested by Jeff Geerling) yet.
We only show cited public figures — nothing is estimated to fill gaps. Requests tell us what to source or test next.

pose estimation

◌

No published pose estimation results for AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S) or Minisforum MS-R1 (as tested by Jeff Geerling) yet.
We only show cited public figures — nothing is estimated to fill gaps. Requests tell us what to source or test next.

image classification

◌

No published image classification results for AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S) or Minisforum MS-R1 (as tested by Jeff Geerling) yet.
We only show cited public figures — nothing is estimated to fill gaps. Requests tell us what to source or test next.

semantic segmentation

◌

No published semantic segmentation results for AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S) or Minisforum MS-R1 (as tested by Jeff Geerling) yet.
We only show cited public figures — nothing is estimated to fill gaps. Requests tell us what to source or test next.

speech-to-text

◌

No published speech-to-text results for AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S) or Minisforum MS-R1 (as tested by Jeff Geerling) yet.
We only show cited public figures — nothing is estimated to fill gaps. Requests tell us what to source or test next.

LLM generation

✓ Like-for-like: model-size class LLM 5–14B, batch 1. Model, precision and stack may still differ.

Platform	Result	Model · precision · stack	Grade	Power	Source
AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S)	47.30 tok/s generation_throughput · b=1	Llama 2 7B Q4_K_M · llama.cpp (llama-bench; best backend per README, mostly Vulkan RADV / ROCm 7 nightlies) · LLM 5–14B	C	120 W Chip TDP, Accelerator only, host not included, max of range 45-120 W reported in result source — Chip TDP, Accelerator only, host not included, max of range 45-120 W	visorcraft/strix-halo-llm-perf README (single-host table) 2026-03-03 (published)
Minisforum MS-R1 (as tested by Jeff Geerling)	34.52 tok/s generation_throughput · b=1	llama3.1:8b Ollama default tag (4-bit unless tag says otherwise) · Ollama (geerlingguy/ollama-benchmark) · LLM 5–14B	C	unknown power not published not published	Jeff Geerling ai-benchmarks issue results (comments) 2025-11-10 (page last edited)

✓ Like-for-like: model-size class LLM ≤4B, batch 1. Model, precision and stack may still differ.

Platform	Result	Model · precision · stack	Grade	Power	Source
AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S)	249 tok/s generation_throughput · b=1	TinyLlama 1.1B Q4_K_M · llama.cpp (llama-bench; best backend per README, mostly Vulkan RADV / ROCm 7 nightlies) · LLM ≤4B	C	120 W Chip TDP, Accelerator only, host not included, max of range 45-120 W reported in result source — Chip TDP, Accelerator only, host not included, max of range 45-120 W	visorcraft/strix-halo-llm-perf README (single-host table) 2026-03-03 (published)
Minisforum MS-R1 (as tested by Jeff Geerling)	71.36 tok/s generation_throughput · b=1	llama3.2:3b Ollama default tag (4-bit unless tag says otherwise) · Ollama (geerlingguy/ollama-benchmark) · LLM ≤4B	C	unknown power not published not published	Jeff Geerling ai-benchmarks issue results (comments) 2025-11-10 (page last edited)

✓ Like-for-like: model-size class LLM >40B, batch 1. Model, precision and stack may still differ.

Platform	Result	Model · precision · stack	Grade	Power	Source
AMD Ryzen AI Max+ 395 / PRO 395 (Strix Halo, Radeon 8060S)	53.40 tok/s generation_throughput · b=1	GPT-OSS 120B Q4_K_M · llama.cpp (llama-server; best backend per README, mostly Vulkan RADV / ROCm 7 nightlies) · LLM >40B	C	120 W Chip TDP, Accelerator only, host not included, max of range 45-120 W reported in result source — Chip TDP, Accelerator only, host not included, max of range 45-120 W	visorcraft/strix-halo-llm-perf README (single-host table) 2026-03-03 (published)
Minisforum MS-R1 (as tested by Jeff Geerling) CPU-only	0.770 tok/s generation_throughput · b=1	Llama 3.1:70b as in linked issue (Ollama default tags are 4-bit) · Ollama / llama.cpp (see linked issue) · LLM >40B	C	38.20 W ● Measured, Whole system reported in result source — Measured, Whole system	Jeff Geerling ai-benchmarks README (Findings tables) 2026-09-22 (page last edited)

Grades: A audited MLPerf · B third-party · C community · D vendor. How we rank · JSON