AMD Ryzen AI Halo vs NVIDIA DGX Spark 2026: Which 128GB AI Dev Kit Actually Pays Off
TL;DR: Both boxes carry 128GB of unified memory and both are bandwidth-bound, so token generation is nearly tied — ~34 tok/s (Halo) vs ~39 tok/s (Spark) on gpt-oss 120B. DGX Spark wins prompt processing by roughly 5× and brings CUDA; AMD’s Halo gives 2TB of storage and Linux-first ROCm for $3,999 against a DGX Spark that now lists at $6,950. For pure inference, any Strix Halo box wins — and as of October 2026 the Halo dev kit is itself one of the cheaper ones.
| AMD Ryzen AI Halo | NVIDIA DGX Spark (Founders) | ASUS Ascent GX10 | |
|---|---|---|---|
| Price (Oct 2026) | $3,999 (2TB) | $6,950 (4TB) | $5,999 (1TB)** |
| Chip | Ryzen AI Max+ 395 | GB10 Grace Blackwell | GB10 Grace Blackwell |
| Memory / bandwidth | 128GB LPDDR5X / 256 GB/s | 128GB LPDDR5x / 273 GB/s | 128GB LPDDR5x / 273 GB/s |
| Token gen (gpt-oss 120B) | ~34 tok/s | ~39 tok/s | ~39 tok/s |
| Prompt processing | ~340 tok/s | ~1,720 tok/s | ~1,720 tok/s |
| Software stack | ROCm / Vulkan / Ollama | CUDA (full ecosystem) | CUDA (full ecosystem) |
| The catch | Weaker prefill, ROCm maturity | $6,950 — up 74% since launch | Only 1TB SSD, now $5,999 |
*NVIDIA raised the Founders Edition from $3,999 to $4,699 on February 23, 2026, and its own Marketplace listing read $6,950 on October 19, 2026. **ASUS raised the GX10 from $2,999 to $3,999 on August 18, 2026 and to $5,999 by late October 2026.
Prices re-verified October 27, 2026. The whole comparison moved this fall. The DGX Spark Founders Edition is $6,950 on NVIDIA’s Marketplace, the ASUS Ascent GX10 is $5,999, and NVIDIA’s partners opened a cheaper 64GB GB10 tier at $4,999 on October 23. On the AMD side the Ryzen AI Halo dev kit is still $3,999 at Micro Center (a 192GB Ryzen AI Max+ PRO 495 version arrived in October at $7,299), while the GMKtec EVO-X2 128GB has settled at $3,649 after the DRAM repricing. The “same chip for half the money” argument below is gone: the EVO-X2 is now only ~$350 under the Halo. Benchmarks are unchanged.
Honest take: If you only run inference on MoE models, both are bandwidth-bound and roughly tied on the tokens you actually feel — so the cheaper Strix Halo box, the GMKtec EVO-X2 at ~$3,649 (our full review), or the $3,999 Halo dev kit itself, is the smarter buy. Pay the DGX Spark premium only if you fine-tune, write CUDA, or run prefill-heavy agentic workloads. Jump to what to actually buy if you already know which camp you’re in.
The matchup isn’t $3,999 vs $2,999 (and that gap has since closed)
The headline that’s been circulating — AMD’s $3,999 dev kit “tackling” a $2,999 DGX Spark — quietly compares two different things. NVIDIA’s own DGX Spark Founders Edition launched at $3,999 with a 4TB SSD, and NVIDIA raised it to $4,699 in February 2026 and to $6,950 by October 2026 because of the same memory-price spike that’s hitting DDR5 and SSD buyers. The $2,999 figure belongs to the ASUS Ascent GX10, an OEM variant of the DGX Spark with the same GB10 Grace Blackwell Superchip and 128GB of memory — but only 1TB of storage instead of 4TB.
So the real picture, dollar for dollar, is:
- AMD Ryzen AI Halo — $3,999, Ryzen AI Max+ 395, 128GB, 2TB SSD
- DGX Spark (ASUS GX10) — launched $2,999, $3,999 in August, $5,999 as of October 2026, GB10, 128GB, 1TB SSD
- DGX Spark (Founders) — launched $3,999, $4,699 in February, $6,950 as of October 2026, GB10, 128GB, 4TB SSD
- DGX Spark 64GB (OEM tier) — $4,999, on sale from October 23, 2026 through Acer, ASUS, Dell, HP, Gigabyte and MSI; same GB10, half the memory
AMD slots its dev kit between the two NVIDIA configs on price, gives you double the storage of the cheap Spark, and matched the Founders Edition exactly at $3,999 before NVIDIA’s hikes. Today it undercuts the Founders Edition by about $2,950. That’s the framing AMD wants: less money, more SSD, and — they claim — leadership tokens-per-dollar.
What’s actually in each box
The AMD Ryzen AI Halo Developer Platform is built on the Ryzen AI Max+ 395 — the same “Strix Halo” APU in the GMKtec EVO-X2 and the chip we tore down in our Strix Halo deep dive. It pairs 16 Zen 5 cores (3.0GHz base, 5.1GHz boost) with a Radeon 8060S iGPU (40 RDNA 3.5 compute units) and an XDNA 2 NPU rated at 50 TOPS, for a platform total AMD quotes at 126 TOPS. Memory is 128GB of LPDDR5X-8000 on a 256-bit bus — 256 GB/s — and the kit ships with a 2TB PCIe 4 SSD, 10GbE LAN, Wi-Fi 7, four USB-C ports, and an aluminum chassis the size of a paperback (149 × 149 × 43mm). It sells through Micro Center (in-store pickup), shipping since July 2026, and you pick Linux or Windows at no price difference — a tell that AMD is aiming this squarely at developers.
The NVIDIA DGX Spark is a different animal under the hood. Its GB10 Grace Blackwell Superchip glues a 20-core Arm CPU (10 Cortex-X925 + 10 Cortex-A725) to a Blackwell GPU that NVIDIA rates at up to 1 petaFLOP of sparse FP4 tensor performance. Memory is also 128GB of unified LPDDR5x, but at a slightly higher 273 GB/s. It runs DGX OS (Ubuntu-based) and, crucially, the full CUDA stack.
The spec sheets converge on the thing that matters most for local LLMs: 128GB of unified memory on both. That’s enough to hold models no single 24GB consumer GPU can — gpt-oss 120B, Qwen3-235B at lower quant, dense 70B with room to spare. The question is how fast each one actually moves tokens through that memory.
Token generation: nearly a tie
Here’s the result that surprises people. On token generation — the decode phase, where the model streams one token at a time and you watch words appear — the two boxes are within a few tokens per second of each other.
On gpt-oss 120B (the MoE model both vendors lean on for demos), independent testing puts the Ryzen AI Max+ 395 at 34.13 tok/s against the DGX Spark’s 38.55 tok/s. That’s a 13% NVIDIA lead, not a generational gap. The reason is simple and it’s the same reason every box in this class behaves the way it does: decode is memory-bandwidth-bound, not compute-bound. The GB10’s 273 GB/s and Strix Halo’s 256 GB/s are within 7% of each other, so the tokens-per-second they can sustain on the same model are within 7% too. NVIDIA’s enormous FP4 compute advantage simply doesn’t get used during decode.
Per-model llama.cpp numbers fill in the rest of the picture on the DGX Spark side:
| Model (DGX Spark) | Prompt processing | Token generation |
|---|---|---|
| gpt-oss 20B (MXFP4) | ~2,000 tok/s | ~60 tok/s |
| gpt-oss 120B (MXFP4) | ~1,200 tok/s | ~35 tok/s |
| Qwen3-Coder 30B (Q8_0) | ~1,650 tok/s | ~44 tok/s |
| Llama 3.3 70B (Q8_0, dense) | low | ~2.6 tok/s |
On the AMD side, gpt-oss 20B on the Ryzen AI Max+ 395 lands around 30–33 tok/s generation with roughly 400 tok/s prompt processing. Notice the pattern across both platforms: MoE models (gpt-oss, Qwen3-30B-A3B) fly because only a few billion parameters activate per token, while dense 70B craters to ~2.6 tok/s on the Spark — the same single-digit decode we measured in our GMKtec EVO-X2 review. Neither of these machines is a good dense-70B box. Both are MoE boxes that happen to have enough memory for big models.
For reference, human reading speed is about 7–10 tok/s, so anything in the 30–60 range feels comfortably interactive and the dense-70B ~2.6 tok/s is a “start it and walk away” experience on either platform.
Prompt processing: where NVIDIA earns its badge
The gap that actually separates these two boxes isn’t decode — it’s prefill, the prompt-processing phase that runs before the first token appears. On gpt-oss 120B, the DGX Spark processes the prompt at roughly 1,723 tok/s versus Strix Halo’s 339.87 tok/s — about 5× faster. Prefill is compute-bound, and this is exactly where Blackwell’s tensor cores and the mature CUDA kernels do their work while Strix Halo’s RDNA 3.5 iGPU falls behind.
This matters more than the spec-sheet symmetry suggests, and it maps directly to your workload:
- Short prompts, chat, casual coding autocomplete → prefill is a rounding error. The two boxes feel identical.
- Long context: RAG over big documents, full-repo code analysis, 32K+ token agentic loops → prefill dominates time-to-first-token. A 5× prefill advantage is the difference between a 3-second wait and a 15-second one, every turn. Here the DGX Spark pulls clearly ahead.
- Fine-tuning / training → not a contest. CUDA, cuDNN, and the entire PyTorch training ecosystem run first-class on the Spark; a QLoRA pass on Llama 3.3 70B has been measured at over 5,000 tok/s throughput on the GB10. ROCm fine-tuning on Strix Halo works but is rougher and slower.
ROCm vs CUDA: the real tax
The benchmark you can’t put in a table is software friction. NVIDIA is selling a decade of CUDA momentum — PyTorch wheels that just work, every inference runtime tested on it first, Docker images, NIM microservices, forum answers for every error. If your workflow touches custom CUDA kernels, vLLM tensor-parallel, NeMo, or any “pip install and it runs” expectation, the Spark removes friction you’d otherwise spend evenings on.
AMD’s stack has genuinely improved. For pure inference, ROCm, Vulkan, llama.cpp, and Ollama all run gpt-oss and Qwen3 on Strix Halo today without drama — and AMD’s own Lemonade server is the one option that also puts the Halo’s otherwise-idle XDNA 2 NPU to work for prompt processing — and in some single-batch llama.cpp tests the Vulkan backend even edges ahead of the Spark on raw decode. But step off the inference path into training or exotic libraries and you’ll feel the maturity gap. The honest framing: for inference, ROCm is fine in 2026; for everything else, CUDA still wins. If you’re choosing a box to run coding agents, our coverage of local coding stacks on aicoderscope.com is worth a read before you commit either way.
The “pays for itself in 6 months” claim
AMD markets the Halo as recovering its cost in roughly six months of cloud API savings. That math only holds if you’re genuinely running a heavy, sustained workload. Renting a comparable GPU on RunPod runs about $1/hr for a 24GB-class card; at 4 hours a day that’s ~$120/month, so a $3,999 box “pays back” in about 33 months — not six. To hit six months you’d need to be running it ~8+ hours a day at higher cloud rates, which describes a small team or a power user, not a hobbyist. The break-even is real, but it’s a workload claim, not a universal one. We worked through the full rent-vs-buy spreadsheet in RunPod vs local GPU, and our local vs cloud cost calculator will run the same math against your own hours-per-day so you can see where a $3,999 box actually crosses over for you.
Who should buy which
The short version: most people reading this comparison should buy neither headline box. Both are bandwidth-bound and effectively tied on the speed you actually feel, so the $3,999–$6,950 price tags only earn themselves in two narrow cases — CUDA/fine-tuning work (Spark) or an official AMD dev platform with vendor support (Halo). If you’re a home-lab user running MoE inference, a $3,649 Strix Halo mini PC gives you the same chip and the same decode numbers — the Spark premium over that is now about $3,300.
Buy the DGX Spark (or ASUS GX10) if: you fine-tune models, write CUDA, run vLLM/NeMo, or your inference is long-context and prefill-heavy. The 5× prompt-processing lead and the CUDA ecosystem are worth real money to you. Take the $5,999 ASUS Ascent GX10 if 1TB is enough storage; take the DGX Spark Founders Edition only if you need the 4TB and can stomach the hiked price.
Buy the Ryzen AI Halo if: you want an official, Linux-first AMD developer platform with 2TB of storage, 10GbE, and ROCm support out of the box, and your work is inference plus light experimentation. The Halo configuration discussed here has 2TB of storage, while NVIDIA specifies 4TB for the 128GB DGX Spark Founders Edition; the dated prices above are not equal. It is in stock at Micro Center (in-store), and in October 2026 AMD added a 192GB Ryzen AI Max+ PRO 495 version at $7,299 for people whose models outgrow 128GB; if there is no Micro Center near you, Ryzen AI Max+ 395 mini PCs with the same silicon ship today.
Buy neither — get a Strix Halo mini PC instead — if: you’re a home-lab user who mostly runs MoE inference. The GMKtec EVO-X2 has the same Ryzen AI Max+ 395 chip and 128GB for ~$3,649 and posts the same decode numbers — we took it apart in detail here. As of October 2026 the Halo dev kit is only ~$350 more, and for that you get official-dev-kit status, 2TB, 10GbE, and AMD’s support commitment. The EVO-X2 still wins if you want 2.5GbE and Amazon delivery instead of a Micro Center trip; it no longer wins on price by much.
Buy neither of these, period — get a used GPU — if: every model you run fits in 24GB. A used RTX 3090 pushes 936 GB/s and ~95 tok/s on a 7B model — roughly 3× the decode speed of either 128GB box — for $1,456–$1,558. These mini-supercomputers only justify themselves once your models physically don’t fit on a real GPU. If you’re not sure which side of that line you’re on, our VRAM calculator will tell you whether your target models actually need more than 24GB before you spend $4,000 solving a problem you may not have.
What to actually buy
Prices below are dated October 27, 2026 and have not been re-verified today. Check current configuration, price, and availability before buying. Editorial correction November 7: the Founders Edition listing link has been restored to its own row; the 64GB OEM link remains separate.
| Your situation | The machine | Price | Where |
|---|---|---|---|
| CUDA, fine-tuning, prefill-heavy — 1TB is fine | ASUS Ascent GX10 (GB10, 128GB) | $5,999 | Check listings |
| Same, but you need 4TB of storage | DGX Spark Founders Edition | $6,950 | Check listings |
| CUDA, but 64GB covers your models | DGX Spark 64GB OEM tier (Acer/ASUS/Dell/HP/MSI, from Oct 23) | $4,999 | Check listings |
| You want AMD’s official dev platform + support | Ryzen AI Halo (128GB, 2TB) | $3,999 | Micro Center, in-store |
| MoE inference at home — the value pick | GMKtec EVO-X2 (same 395 chip, 128GB) | $3,649 (64GB: $2,199) | Check price |
| Everything you run fits in 24GB | Used RTX 3090 | $1,703–$1,823 | Check price |
| Undecided — want to test the workload first | Rented RTX 3090 or 4090 on Vast.ai | from $0.07-0.14/hr | Check availability |
If you’re on the fence between the last two rows, spending $20 on cloud hours before spending $4,000 on a box is the cheapest decision you can make — a few evenings on a rented card tells you exactly which model sizes you actually reach for, and that answer picks the machine for you.
FAQ
Is the DGX Spark really only $2,999? No — that number is from launch coverage and is gone. The $2,999 figure was the ASUS Ascent GX10, an OEM DGX Spark with 1TB of storage; its US list price went to $3,999 in August and $5,999 by late October 2026. NVIDIA’s own Founders Edition launched at $3,999, was raised to $4,699 in February 2026, and listed at $6,950 on NVIDIA’s Marketplace in October. The cheapest way into GB10 is now the 64GB OEM tier at $4,999. Budget $5,000–$7,000 depending on which box and which trim.
Which is faster for everyday local chat? Effectively a tie. On MoE models like gpt-oss 120B, token generation is ~34 tok/s (Halo) vs ~39 tok/s (Spark) — both bandwidth-bound and within ~13% of each other.
Can either run dense Llama 3.3 70B well? No. Dense 70B decodes at ~2.6 tok/s on the DGX Spark and similarly slowly on Strix Halo. Both shine on MoE models, not dense ones.
Does the Ryzen AI Halo run Linux? Yes — AMD ships it with Linux or Windows at no price difference, and it’s positioned as a Linux-first developer platform with ROCm support.
Should I just buy a GMKtec EVO-X2 instead? For pure inference it was the obvious answer at ~$1,999; at $3,649 (October 2026) it is only ~$350 under the Halo dev kit. Both run the same chip at the same speed, so the question is whether official AMD support, 2TB and 10GbE are worth $350 to you — for most people who can reach a Micro Center, yes.
Sources
- AMD Ryzen AI Halo Developer Platform — official product page
- AMD Opens Pre-Orders For The Linux-Friendly Ryzen AI Halo Developer Platform — Phoronix
- $4,000 AMD Ryzen AI Halo Developer Platform features 126 TOPS Ryzen AI Max+ 395 — CNX Software
- AMD’s Ryzen AI Halo AI mini PC launches with 128GB and a $3,999 price tag — TweakTown
- NVIDIA DGX Spark — Hardware Overview (User Guide)
- NVIDIA DGX Spark In-Depth Review — LMSYS Org
- Nvidia DGX Spark gets $700 price hike to $4,699 — Tom’s Hardware
- ASUS announces Ascent GX10 Mini AI Supercomputer, costs $2999 — VideoCardz
- Tested: AMD’s Strix Halo vs Nvidia’s DGX Spark — The Register
- Compact AI workstations in comparison: DGX Spark meets Ryzen AI Max+ 395 — Notebookcheck
- Performance of llama.cpp on NVIDIA DGX Spark — ggml-org/llama.cpp Discussion #16578
- DGX Spark 128GB / 4TB specifications — NVIDIA Marketplace, checked November 7, 2026
- NVIDIA DGX Spark price (Q4 2026): $6,950 on NVIDIA Marketplace, October 19, 2026 — pi3g
- ASUS Ascent GX10 vs DGX Spark: GB10 boxes compared, 64GB OEM tier at $4,999 from Oct 23, 2026 — Spheron
- AMD Ryzen AI Halo with Ryzen AI Max+ PRO 495 launches at $7,299 with 192GB — VideoCardz
- GMKtec EVO-X2 128GB/2TB, $3,649.99 — Micro Center
Last updated October 28, 2026 (prices re-verified October 27). Prices and specs change; verify current rates before purchasing.
Recommended Gear
- GMKtec EVO-X2 (Ryzen AI Max+ 395, 128GB) — same chip as the Halo dev kit, ~$3,649
- ASUS Ascent GX10 (GB10, 128GB, 1TB) — the 1TB DGX Spark variant, $5,999 as of October 2026
- NVIDIA DGX Spark — Founders Edition, 4TB
- Ryzen AI Max+ 395 mini PCs — Strix Halo boxes shipping now
- Used RTX 3090 24GB — faster than either box if your models fit in 24GB
Was this article helpful?
Thanks for the feedback — it helps improve future articles.
Need hands-on help?
I offer 1-on-1 technical consulting for local AI setup, GPU selection, and AI coding tool configuration — same topics covered on this site.
Book a session — $49 / hour →Get the numbers before you buy
New GPU and mini-PC benchmarks, VRAM thresholds, and price checks — sent when there's something worth acting on, not on a schedule. No spam, unsubscribe anytime.