Ultra-small AI supercomputer
ASUS Ascent GX10 AI Supercomputer
ASUS Ascent GX10 is a 150 mm square desktop that puts a petaflop of AI compute and 128GB of unified memory on your desk. It uses the same NVIDIA GB10 Grace Blackwell Superchip as the NVIDIA DGX Spark, ships with NVIDIA DGX OS and the NVIDIA AI software stack, and runs models up to 200 billion parameters locally.
YourShoppy is an authorised ASUS dealer in Koramangala, Bangalore, and we list the GX10 in 1TB, 2TB and 4TB versions. WhatsApp 97392 21133 or 98869 19198 for today’s price.
Design
Petaflop-class AI that fits in one hand
Until recently, running a 70B or 200B model locally meant a rack server or a stack of graphics cards. The Ascent GX10 does it in a box about the size of a hardback book. It sits on a desk, runs off a single USB-C power supply and is quiet enough for an office.
That makes it a practical machine for developers, AI researchers, data scientists and college labs who want to work on large models without booking time on a shared cluster.
Performance
NVIDIA GB10 Grace Blackwell inside
The GB10 superchip combines a Blackwell GPU with 5th-generation Tensor Cores and FP4 support, and a 20-core Arm CPU for data preparation and orchestration. NVLink-C2C joins the two with a single shared memory pool, at roughly five times the bandwidth of PCIe 5.0.
The result: up to 1 petaFLOP (about 1,000 AI TOPS) of AI performance and 128GB of LPDDR5x memory that both the CPU and GPU can use. That shared memory is why the GX10 can load models a gaming graphics card never could.
Large models
Run 200-billion-parameter models at your desk
One GX10 lets you prototype, fine-tune and run inference on generative AI models of up to 200 billion parameters. ASUS rates DeepSeek R1 inference at up to 70B on a single unit.
Need more? Connect two GX10 units over the built-in NVIDIA ConnectX-7 link and you can work with models like Llama 3.1 405B.
AI networking
NVIDIA ConnectX-7 built in
The GX10 has NVIDIA ConnectX-7 networking on board, which is what makes two-unit clustering and fast data movement possible.
Fast data throughput
High-bandwidth links between nodes for distributed AI workloads.
Encrypted by hardware
TLS, IPsec and MACsec acceleration, so encryption doesn’t eat into CPU time.
Precise time sync
IEEE 1588v2 PTP support for time-sensitive edge AI and robotics work.
AI software
Ready to work out of the box
Preloaded AI platform
NVIDIA DGX OS (Ubuntu-based) and the NVIDIA AI software stack come installed, with frameworks, SDKs and tools ready to use.
The tools you already use
CUDA, PyTorch, TensorFlow, JAX, Jupyter and Ollama, plus NVIDIA TensorRT for inference and NVIDIA NIM microservices and Blueprints.
Open model support
Run Llama, DeepSeek, Qwen, NVIDIA Nemotron and other open models locally, from 7B up to 200B on one unit.
Agentic AI
Run AI agents locally, around the clock
With 128GB of memory and a petaflop of compute, the GX10 can keep agent frameworks such as OpenClaw or Hermes Agent running all day on open models. NVIDIA’s NemoClaw stack adds privacy and security guardrails on top.
Running models on your own hardware means more predictable costs than paying per API token, and your data stays in your office. Paid third-party models and services still charge separately.
Thermal
Cooling built for long AI runs
Fine-tuning and inference keep the chip busy for hours. ASUS uses a dual-fan design with 7-level control, and says it gives 1.6× more efficient thermal coverage than comparable compact systems, so performance holds steady under sustained load.
Scalable
Small footprint, room to grow
Start with one unit on a desk. When your models outgrow it, add a second GX10 and cluster them over ConnectX-7. Because it runs the NVIDIA AI software platform, work you build here moves to NVIDIA DGX Cloud or a data centre with minimal code changes.
Connectivity
I/O ports
- Kensington lock slot
- 1× USB 3.2 Gen 2×2 Type-C (20Gbps) with DisplayPort 2.1 and 180W power input
- 3× USB 3.2 Gen 2×2 Type-C (20Gbps) with DisplayPort 2.1
- HDMI 2.1b
- 10GbE LAN
- NVIDIA ConnectX-7 network port
Applications
Build locally, deploy anywhere
Move to the cloud when you’re ready. Models built on the GX10 run on the same NVIDIA software platform as DGX Cloud and data-centre GPUs.
Experiment cheaply. Do the trial-and-error on your desk and keep expensive cluster time for production training.
Prototyping
Try new models and agent ideas without waiting for cloud quota.
Fine-tuning and inference
Fine-tune on your own data and serve it to your team.
Data science
GPU-accelerated analytics, computer vision and forecasting.
Buy in Bangalore
ASUS Ascent GX10 price and storage options
All three versions have the same GB10 superchip and 128GB of memory. The only difference is SSD storage: 1TB, 2TB or 4TB. Pick 4TB if you’ll keep several large models and datasets on the machine.
- Same-day delivery in Bangalore if you order by 2 PM, or pick it up from our Koramangala store.
- ASUS extended warranty care packs available. Ask us for the options and prices.
ASUS Ascent GX10 AI Supercomputer, GB10 Grace Blackwell, 128GB RAM, 1TB SSD,
- NVIDIA® GB10 Grace Blackwell Superchip — 20-core Armv9.2 CPU + Blackwell GPU | up to 1000 AI TOPS (1 PFLOP FP4)
- 128 GB LPDDR5x Unified Memory (CPU + GPU share one pool — no bandwidth bottleneck)
- 1 TB M.2 NVMe PCIe 4.0 SSD | Ubuntu Linux | NVIDIA DGX Spark platform
- Wi-Fi 7 (802.11be) + Bluetooth 5.4 | 10G Ethernet + NVIDIA ConnectX-7 SmartNIC
- 4× USB-C 3.2 Gen 2×2 (20 Gbps) — 3× with DisplayPort 2.1 | 1× 180W PD-in
- HDMI 2.1 | up to 4 simultaneous displays | 150 × 150 × 51 mm | 1.48 kg
- Run 200B+ parameter AI models locally — scale to 256 GB via NVLink (2× GX10)
- 1-Year Warranty
Asus Ascent GX10 AI Supercomputer — GB10 Grace Blackwell, 128GB, 2TB
- NVIDIA® GB10 Grace Blackwell Superchip — 20-core Armv9.2 CPU + Blackwell GPU | up to 1000 AI TOPS (1 PFLOP FP4)
- 128 GB LPDDR5x Unified Memory (CPU + GPU share one pool — no bandwidth bottleneck)
- 2 TB M.2 NVMe PCIe 4.0 SSD | Ubuntu Linux | NVIDIA DGX Spark platform
- Wi-Fi 7 (802.11be) + Bluetooth 5.4 | 10G Ethernet + NVIDIA ConnectX-7 SmartNIC
- 4× USB-C 3.2 Gen 2×2 (20 Gbps) — 3× with DisplayPort 2.1 | 1× 180W PD-in
- HDMI 2.1 | up to 4 simultaneous displays | 150 × 150 × 51 mm | 1.48 kg
- Run 200B+ parameter AI models locally — scale to 256 GB via NVLink (2× GX10)
- 1-Year Warranty
ASUS Ascent GX10 AI Supercomputer- GB10 Grace Blackwell,128GB,4TB
- NVIDIA® GB10 Grace Blackwell Superchip — 20-core Armv9.2 CPU + Blackwell GPU | up to 1000 AI TOPS (1 PFLOP FP4)
- 128 GB LPDDR5x Unified Memory (CPU + GPU share one pool — no bandwidth bottleneck)
- 4 TB M.2 NVMe PCIe 5.0 SSD | Ubuntu Linux | NVIDIA DGX Spark platform
- Wi-Fi 7 (802.11be) + Bluetooth 5.4 | 10G Ethernet + NVIDIA ConnectX-7 SmartNIC
- 4× USB-C 3.2 Gen 2×2 (20 Gbps) — 3× with DisplayPort 2.1 | 1× 180W PD-in
- HDMI 2.1 | up to 4 simultaneous displays | 150 × 150 × 51 mm | 1.48 kg
- Run 200B+ parameter AI models locally — scale to 256 GB via NVLink (2× GX10)
- 1-Year Warranty
Full specifications
| Superchip | NVIDIA GB10 Grace Blackwell |
|---|---|
| CPU | 20-core Arm |
| GPU | NVIDIA Blackwell, 5th-gen Tensor Cores, FP4 |
| AI performance | Up to 1 petaFLOP (FP4), about 1,000 AI TOPS |
| Memory | 128GB LPDDR5x coherent unified memory |
| Storage | 1TB, 2TB or 4TB NVMe SSD |
| Networking | NVIDIA ConnectX-7, 10GbE LAN, Wi-Fi 7 |
| Ports | 4× USB 3.2 Gen 2×2 Type-C (DP 2.1, one with 180W PD in), HDMI 2.1b |
| Operating system | NVIDIA DGX OS (Ubuntu Linux) |
| Software | NVIDIA AI software stack, NIM, Blueprints, CUDA, PyTorch, Jupyter, Ollama |
| Clustering | Up to two units over ConnectX-7 |
| Size | 150 × 150 × 51 mm (about 1.6L) |
| Warranty | 1-year ASUS limited hardware warranty. Extended warranty care packs available, ask us. |
Specifications from ASUS. Product images and video courtesy ASUS. Specs may change.
Why buy the ASUS Ascent GX10 from YourShoppy
- Authorised ASUS dealer in 1st Block Koramangala since 2010, so you get genuine sealed stock with ASUS India warranty.
- GST invoice for your company. Share your GSTIN and we’ll bill it to the company. Check input tax credit with your CA.
- Same-day delivery in Bangalore when you order by 2 PM. Prefer to collect? Pick it up from our Koramangala store.
- ASUS extended warranty care packs on top of the standard 1-year warranty. WhatsApp us for the plans and prices.
- EMI options on select cards. See our EMI page.
- Compare in one place: see the NVIDIA DGX Spark and all GB10 AI supercomputers we stock.
Address: YourShoppy, 339, 1st Floor, 1st Main Cross, next to Essar Petrol Bunk, 1st Block Koramangala, Bengaluru 560034
Hours: Monday to Sunday, 10:30 AM to 9 PM
Phone:97392 21133 · 98869 19198 · 080 4141 0072
WhatsApp:97392 21133 or 98869 19198
Common questions
What is the ASUS Ascent GX10?
A compact AI supercomputer built on NVIDIA DGX Spark, with the NVIDIA GB10 Grace Blackwell Superchip, 128GB of unified memory and up to 1 petaFLOP of AI performance.
What is the difference between ASUS Ascent GX10 and NVIDIA DGX Spark?
Both use the same GB10 superchip and 128GB of memory, so AI performance is the same. The GX10 is ASUS’s own design with its own cooling, storage options of 1TB, 2TB or 4TB, and ASUS warranty and support.
How large an AI model can the ASUS Ascent GX10 run?
One unit runs models up to 200 billion parameters. Two units linked over ConnectX-7 can handle models such as Llama 3.1 405B.
What operating system does the ASUS Ascent GX10 use?
NVIDIA DGX OS, which is based on Ubuntu Linux, with the NVIDIA AI software stack preinstalled.
What warranty does the ASUS Ascent GX10 have?
A 1-year ASUS limited hardware warranty from the date of purchase. ASUS extended warranty care packs are also available. Ask us for the details.
Do you deliver the ASUS Ascent GX10 the same day in Bangalore?
Yes. Order by 2 PM for same-day delivery anywhere in Bangalore, or pick it up from our Koramangala store.
Where can I buy the ASUS Ascent GX10 in Bangalore?
At YourShoppy, an authorised ASUS dealer at 339, 1st Floor, 1st Main Cross, 1st Block Koramangala. WhatsApp 97392 21133 or 98869 19198 for today’s price.