ASUS Ascent GX10
The NVIDIA GB10 Grace Blackwell Superchip delivers petaflop-scale AI compute with 128GB of unified LPDDR5x memory, enabling large model inference in a compact, stackable metal chassis. Its 10GbE networking and Wi-Fi 7 connectivity, combined with a 2TB NVMe SSD, provide high-speed data throughput for secure, on-device agentic workflows. This workstation is best for AI developers running private, long-duration agentic tasks with frameworks like OpenClaw and NemoClaw.
요약
The 30-Second Version
The ASUS Ascent GX10 is a specialized AI beast with best-in-class CPU and memory for running massive models locally. It's dead silent, surprisingly compact, and absolutely useless for gaming. Worth it for AI devs who can find it near the $4,700 mark.
Pros & Cons
장점
- 128GB of unified memory lets you run enormous AI models locally. 100th
- The GB10 Superchip delivers best-in-class AI compute in a compact chassis. 99th
- Near-silent operation and low power draw for this level of performance. 92nd
- Stackable design and 10GbE networking make it easy to cluster. 75th
단점
- Gaming performance is essentially nonexistent, don't even try.
- Port selection is mediocre with only four USB-C and one HDMI 2.1.
- Reliability scores are below average based on early data.
- No pricing listed, but expect a hefty premium over a standard desktop.
사용자 의견
The Word on the Street
근거 자료
Performance
The CPU sits at the absolute top of our charts, and the 128GB of LPDDR5x RAM is best-in-class for this category. That unified memory is the real star here, letting you load massive 200B-parameter models that would choke most desktops. The 2TB NVMe SSD is fast and well above average. But the integrated GPU, while purpose-built for AI compute, lands in the bottom 10% for traditional graphics tasks. Don't expect to game on this thing at all, it scored a 13.4 out of 100 in our gaming tests. For AI inference and training, though, it's a petaflop-class monster in a shockingly small footprint.
Specifications
Full Specifications
Processor
| CPU | 20-Core: Up to 3.25 GHz Cortex A725 (10 Cores)Up to 3.8 GHz Cortex X925 (10 Cores) |
| Cores | 20 |
| Frequency | 3.3 GHz |
Graphics
| GPU | NVIDIA GB10 Grace Blackwell Superchip |
| Type | Integrated |
Memory & Storage
| RAM | 128 GB |
| RAM Generation | DDR5 |
| Storage | 2 TB |
| Storage Type | NVMe SSD |
Build
| Form Factor | workstation |
| Weight | 3.2 kg / 7.0 lbs |
Connectivity
| USB-C Ports | 4 |
| HDMI | HDMI 2.1 |
| Wi-Fi | Wi-Fi 7 |
| Bluetooth | Bluetooth 5.4 |
| Ethernet | 10GbE LAN |
vs Competition
Stacked against the Apple Mac Studio M4 Max, the GX10 trades blows depending on your workflow. The Mac has better GPU versatility, more ports, and a mature ecosystem, but it can't touch the GX10's 128GB of unified memory for massive model inference. The Lenovo Legion 34IAS10 and HP Omen 45L are gaming-first machines that completely outclass the GX10 in graphics but can't even load the models this thing handles natively. The MSI EdgeXpert and Dell Tower Plus are more traditional workstations with better reliability scores and port selection, but they lack the specialized AI silicon. Pick your poison: general purpose or AI-focused.
| Spec | ASUS Ascent GX10 | Lenovo Legion 34IAS10 | HP Omen 45L | Apple Mac Studio M4 Max | MSI EdgeXpert EdgeXpert-11SUS | CLX SET TGMSETRTU5204BM |
|---|---|---|---|---|---|---|
| CPU | 20-Core: Up to 3.25 GHz Cortex A725 (10 Cores)Up to 3.8 GHz Cortex X925 (10 Cores) | Intel Core Ultra 9 | Intel Core Ultra 9 285K | Apple M4 Max | NVIDIA GB | Intel Core i9 14900KF |
| RAM (GB) | 128 | 64 | 64 | 36 | 128 | 64 |
| Storage (GB) | 2048 | 3072 | 8096 | 512 | 4000 | 8000 |
| GPU | NVIDIA GB10 Grace Blackwell Superchip | NVIDIA GeForce RTX 5080 | NVIDIA GeForce RTX 5080 | Apple M4 Max 32-core | NVIDIA Blackwell Architecture | NVIDIA GeForce RTX 5070 |
| Form Factor | workstation | mid-tower | mid-tower | sff | mini | mid-tower |
| Psu W | - | 1200 | - | - | 240 | 850 |
| OS | - | Windows 11 Pro | Windows 11 Home | macOS | NVIDIA DGX OS | Windows 11 Home |
| Compare | Compare | Compare | Compare | Compare |
| 제품 | CPU | GPU | RAM | 포트 | 저장 공간 | 신뢰성 | 사용자 평판 |
|---|---|---|---|---|---|---|---|
| ASUS Ascent GX10 | 99.6 | 9.6 | 98.8 | 46.9 | 91.6 | 36.4 | 75.2 |
| Lenovo Legion 34IAS10 Compare | 97.6 | 88 | 96.7 | 91.7 | 96.5 | 69.7 | 84.8 |
| HP Omen 45L Compare | 97.6 | 88 | 95.5 | 98 | 99.4 | 69.7 | 87.2 |
| Apple Mac Studio M4 Max Compare | 85.5 | 65.5 | 69.6 | 94.5 | 30.2 | 99.4 | 99.9 |
| MSI EdgeXpert EdgeXpert-11SUS Compare | 99.7 | 95.2 | 98.8 | 87.1 | 97.9 | 36.4 | 84.5 |
| CLX SET TGMSETRTU5204BM Compare | 94.3 | 81.4 | 96.7 | 86.3 | 99.2 | 11.1 | 95.7 |
가격
Value & Pricing
ASUS hasn't published an official price yet, but third-party listings show a wild spread from $4,699 to $6,579. At the low end, it's competitive with a loaded Mac Studio M4 Max for AI workloads, though the Mac is far more versatile. At the high end, you're paying a serious early-adopter tax. If you can snag it closer to $4,700, the petaflop-level AI performance in a quiet, power-sipping box is genuinely impressive. Just know you're buying a specialized tool, not a do-everything computer.
B&H Photo 가격 1개 최저 US$4,699
Amazon 가격 1개 최저 US$4,699
Price History
더 보기
Overview
The ASUS Ascent GX10 is basically a baby NVIDIA DGX Spark in a tidy metal box. It packs the GB10 Grace Blackwell Superchip, which combines a 20-core ARM CPU and a Blackwell GPU on one package, with 128GB of unified memory. This isn't a gaming rig or a general purpose desktop. It's a focused AI workstation for developers who want to run big models locally without renting cloud GPUs.
Common Questions
Q: Can I use the ASUS Ascent GX10 for gaming or video editing?
No, the integrated GPU is optimized for AI compute and scores a 13.4 out of 100 in our gaming tests. It's not built for traditional graphics workloads.
Q: What size AI models can this run locally?
With 128GB of unified memory, it can handle models up to around 200 billion parameters, depending on quantization. That's far beyond what most consumer GPUs can manage.
Q: Does it support clustering multiple units together?
Yes, the stackable chassis and 10GbE networking via the ConnectX-7 SmartNIC are designed for linking multiple GX10s to scale up workloads.
Who Should Skip This
If you need a machine that can game, render 3D graphics, or run traditional creative apps, skip this entirely. The GPU is bottom-of-the-barrel for anything outside AI compute. You'd be better off with a Mac Studio or a high-end gaming desktop that can moonlight as an AI rig.
Verdict
This is for AI developers and researchers who need to run large language models, agentic workflows, or fine-tuning jobs locally and value a quiet, compact setup. If that's you, the GX10 is a standout. If you need a machine that can also game, edit video, or drive multiple high-res displays, look elsewhere. The single-minded focus is both its superpower and its limitation.