ASUS Ascent GX10 (NVIDIA GB10)
No editorial image yet — generic vendor mark shown. Credentials in spec table below.
The popular OEM twin of the NVIDIA DGX Spark, ~$1,000 cheaper (~$2,999). Same GB10 Grace Blackwell superchip, 128GB LPDDR5X unified, ~1 PFLOP FP4, ConnectX-7 for 2-node 256GB clustering. A direct, cheaper CUDA alternative buyers cross-shop against the Spark.
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.
Sub-scores sum to 310 / 1000. Headline = 310 × 0.70 (Estimated-confidence discount) = 217. 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 273 GB/s bandwidth — 32.8 tok/s estimated. No measured benchmarks yet.
Plain-English: Doesn't fit modern chat models usefully — vision models won't fit.
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.
What it does well
The ASUS Ascent GX10 is the value way onto NVIDIA's GB10 Grace Blackwell platform — the same superchip as the DGX Spark, 128GB unified LPDDR5X, ~1 PFLOP FP4, for about $1,000 less than Spark's price. Crucially it's CUDA, unlike the Strix Halo boxes, so the entire NVIDIA software stack (vLLM, TensorRT-LLM, CUDA llama.cpp, every research repo) runs natively — a big advantage for anyone who hit Strix Halo's ROCm friction. ConnectX-7 lets you cluster two units into a 256GB pool for very large models.
Where it struggles
Unified memory bandwidth (~273 GB/s) is the same Grace-Blackwell ceiling as the Spark — far below a discrete GPU — so this is a 'fit big models with CUDA' box, not a high-throughput one; token speed on 70B trails a multi-GPU rig. ASUS sells the GX10 'as-is' with no warranty, which is the real catch versus the Dell GB10 (full ProSupport) — factor that into the price gap. It's also a niche, supply-constrained product.
Bottom line
The best-value entry to CUDA-native 128GB unified local AI, and the obvious cross-shop against the pricier DGX Spark. Choose it over Strix Halo when you need CUDA; choose the Dell GB10 instead if you want a warranty.
Overview
The popular OEM twin of the NVIDIA DGX Spark, ~$1,000 cheaper (~$2,999). Same GB10 Grace Blackwell superchip, 128GB LPDDR5X unified, ~1 PFLOP FP4, ConnectX-7 for 2-node 256GB clustering. A direct, cheaper CUDA alternative buyers cross-shop against the Spark.
Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.
Specs
| System RAM (typical) | 128 GB |
| Power draw (peak) | 170 W |
| Released | 2025 |
| MSRP | $2999 |
| Backends | CUDA Vulkan |
Models that fit
Open-weight models small enough to run on ASUS Ascent GX10 (NVIDIA GB10) with usable context.
Frequently asked
Does ASUS Ascent GX10 (NVIDIA GB10) support CUDA?
Where next?
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify hardware specifications.