RUNLOCALAIv38
->Will it run?Best GPUCompareTroubleshootStartLearnPulseModelsHardwareToolsBench
Run check
RUNLOCALAI

Independently operated catalog for local-AI hardware and software. Hand-written verdicts. Source-cited claims. Reproducible commands when we have them.

OP·Eruo Fredoline
DIR
  • Models
  • Hardware
  • Tools
  • Benchmarks
TOOLS
  • Will it run?
  • Compare hardware
  • Cost vs cloud
  • Choose my GPU
  • Prompting kits
  • Quick answers
REF
  • All buyer guides
  • Learn local AI
  • Methodology
  • Glossary
  • Errors KB
  • Trust
EDITOR
  • About
  • Author
  • How we make money
  • Editorial policy
  • Contact
LEGAL
  • Privacy
  • Terms
  • Sitemap
MAIL · MONTHLY DIGEST
Get monthly local AI changes
Monthly recap. No spam.
DISCLOSURE

Some links on this site are affiliate links (Amazon Associates and other first-class retailers). When you buy through them, we earn a small commission at no extra cost to you. Affiliate links do not influence our verdicts — there are cards we rate highly that we don't have affiliate relationships with, and cards that sell well that we refuse to recommend. Read more →

© 2026 runlocalai.coIndependently operated
RUNLOCALAI · v38
  1. >
  2. Home
  3. /Hardware
  4. /AMD Ryzen AI Halo
UNIT · AMD · DESKTOP
128 GB UNIFIEDworkstation·Reviewed July 2026

AMD Ryzen AI Halo

AMD · HARDWARE
AMD Ryzen AI Halo

No editorial image yet — generic vendor mark shown. Credentials in spec table below.

The Ryzen AI Halo is AMD's first-party answer to NVIDIA's DGX Spark: a 150 x 150 x 45.4 mm mini-workstation built on the Ryzen AI Max+ 395 (16 Zen 5 cores, Radeon 8060S with 40 RDNA 3.5 CUs, 50-TOPS XDNA 2 NPU) with 128 GB of LPDDR5X-8000 unified memory at 256 GB/s. AMD says that pool holds models up to roughly 200B parameters, and Variable Graphics Memory ships pre-set to maximum so large models load without tuning. It boots Windows 11 or AMD's Linux developer image — the only box in the Spark's class that does both — and pairs a standard M.2 2280 bay (2 TB included, 8 TB aftermarket) with 10GbE, Wi-Fi 7, three USB-C data ports, and HDMI 2.1b, all inside a 120 W USB-C-powered envelope. It reached retail July 10, 2026 at $3,999, exclusively through Micro Center.

Released 2026·256 GB/s memory bandwidth
▼ CHECK CURRENT PRICE· 1 retailer
AMD Ryzen AI Halo
Check on Amazon→

Affiliate disclosure: as an Amazon Associate and partner of other retailers, we earn from qualifying purchases. The verdict on this page is our editorial opinion; affiliate links never influence what we recommend.

RUNLOCALAI SCORE
See full leaderboard →
155/ 1000
DD-tier
Estimated
Throughput
74/ 500
VRAM-fit
0/ 200
Ecosystem
130/ 200
Efficiency
17/ 100

Sub-scores sum to 221 / 1000. Headline = 221 × 0.70 (Estimated-confidence discount) = 155. This is an algorithmic performance-tier score — distinct from, and often lower than, the editorial “Our verdict” below, which weighs value and real-world fit (especially for hardware we haven’t measured yet). How scoring works →

Extrapolated from 256 GB/s bandwidth — 25.6 tok/s estimated. No measured benchmarks yet.

WORKLOAD FIT
Try other hardware →

Plain-English: Doesn't fit modern chat models usefully — vision models won't fit.

7B chat✗
Doesn't fit
14B chat✗
Doesn't fit
32B chat✗
Doesn't fit
70B chat✗
Doesn't fit
Coding agent✗
Doesn't fit
Vision (≤8B VLM)✗
Doesn't fit
Long context (32K)✗
Doesn't fit
✓Comfortable — fits with headroom
~Tight — works, no slack
△Marginal — needs aggressive quant
✗Doesn't fit usefully

Verdicts extrapolated from catalog VRAM + bandwidth + ecosystem flags. Hover any chip for the rationale. Want measured numbers? Submit your own run with runlocalai-bench --submit.

BLK · VERDICT

Our verdict

OP · Eruo Fredoline|VERIFIED JUL 10, 2026
unrated

Positioning

This is the box to buy if you want one machine that is both a serious x86 workstation and a 128 GB local-AI node — and you should be clear-eyed about what it is not. In StorageReview's testing it was the strongest Ryzen AI Max+ 395 implementation yet (37,316 Cinebench R23 multi-core, top Procyon AI text and image scores), and it beat the DGX Spark outright in CPU work. But in vLLM serving the Spark led by 2-4x at higher concurrency, stretching to 8.8x in prefill-heavy GPT-OSS-120B runs; the Halo only got within ~10-12% in decode-heavy scenarios. If maximum tokens per second or multi-node clustering is the goal, the Spark's 200G fabric versus the Halo's lone 10GbE port settles it — buy the Spark. If raw decode bandwidth per dollar matters most, a used M2 Ultra Mac Studio (800 GB/s, up to 192 GB, MLX) is the stronger pure-inference play, minus x86 and ROCm. Against the GMKtec EVO-X3 — same silicon, $400 less — the Halo buys 10GbE, first-party dual-OS images, pre-tuned VGM, and Micro Center support, but gives up the EVO-X3's OCuLink eGPU escape hatch. For ROCm developers, Windows-in-the-loop teams, and single-box pragmatists, this is the most complete Strix Halo desktop shipped so far: $3,999 with 2 TB, $700 under the Spark Founders Edition's $4,699.

BLK · OVERVIEW

Overview

The Ryzen AI Halo is AMD's first-party answer to NVIDIA's DGX Spark: a 150 x 150 x 45.4 mm mini-workstation built on the Ryzen AI Max+ 395 (16 Zen 5 cores, Radeon 8060S with 40 RDNA 3.5 CUs, 50-TOPS XDNA 2 NPU) with 128 GB of LPDDR5X-8000 unified memory at 256 GB/s. AMD says that pool holds models up to roughly 200B parameters, and Variable Graphics Memory ships pre-set to maximum so large models load without tuning. It boots Windows 11 or AMD's Linux developer image — the only box in the Spark's class that does both — and pairs a standard M.2 2280 bay (2 TB included, 8 TB aftermarket) with 10GbE, Wi-Fi 7, three USB-C data ports, and HDMI 2.1b, all inside a 120 W USB-C-powered envelope. It reached retail July 10, 2026 at $3,999, exclusively through Micro Center.

Retailers we'd check:Amazon

Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.

BLK · SPECS

Specs

System RAM (typical)128 GB
Power draw (peak)120 W
Released2026
MSRP$3999
Backends
ROCm
Vulkan

Models that fit

Open-weight models small enough to run on AMD Ryzen AI Halo with usable context.

all-MiniLM-L6-v2
0.022B · other
FLUX.1 [dev]
12B · other
Qwen 3 0.6B
0.6B · qwen
BGE Large EN v1.5
0.335B · other
Nomic Embed Text v1.5
0.137B · other
Kokoro 82M
0.082B · other
Llama 3.1 8B Instruct
8B · llama
Qwen 3 30B-A3B
30B · qwen

Frequently asked

Does AMD Ryzen AI Halo support CUDA?

No — AMD Ryzen AI Halo is an AMD card. Use ROCm (Linux) or the Vulkan backend in llama.cpp instead. CUDA-only tools won't work.

Where next?

Buyer guides
  • Best GPU for local AI →
  • Best laptop for local AI →
  • Best Mac for local AI →
  • Best used GPU for local AI →
Troubleshooting
  • CUDA out of memory →
  • Ollama running slowly →
  • ROCm not detected →
  • Model keeps crashing →

Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify hardware specifications.

Compare alternatives

Hardware worth comparing

The closest alternatives by price, memory bandwidth, and form factor, plus a step up and down — so you can frame the buying decision against real options.

Closest matches
Similar price, bandwidth & form factor
  • GMKtec EVO-X3
    other · 256 GB/s
    unrated
  • NVIDIA DGX Spark (Project Digits)
    nvidia · 273 GB/s
    10.0/10
  • ASUS Ascent GX10 (NVIDIA GB10)
    nvidia · 273 GB/s
    8.1/10
  • NVIDIA L4
    nvidia · 24 GB VRAM
    9.0/10
  • Framework Desktop (Ryzen AI Max+ 395)
    amd · 256 GB/s
    8.2/10
  • Intel Arc Pro B60 24GB
    intel · 24 GB VRAM
    7.6/10
Step up
More capable — more memory or a higher tier
  • NVIDIA RTX 6000 Ada Generation
    nvidia · 48 GB VRAM
    10.0/10
  • NVIDIA L40
    nvidia · 48 GB VRAM
    10.0/10
  • NVIDIA L40S
    nvidia · 48 GB VRAM
    10.0/10
Step down
Lighter — cheaper or more constrained
  • Framework Desktop (Ryzen AI Max+ 395)
    amd · 256 GB/s
    8.2/10
  • NVIDIA RTX PRO 4000 Blackwell
    nvidia · 24 GB VRAM
    7.3/10
  • Intel Arc Pro B60 24GB
    intel · 24 GB VRAM
    7.6/10