ASUS Ascent GX10 64 GB specifications
Specification | Detail |
|---|---|
Platform | NVIDIA GB10 Grace Blackwell |
CPU | 20 Arm cores: 10 Cortex-X925 + 10 Cortex-A725 |
GPU | Integrated NVIDIA Blackwell, 5th-generation Tensor cores |
Unified CPU–GPU memory | 64 GB LPDDR5X-8533 |
Stated memory bandwidth | 273 GB/s |
Theoretical AI compute | Up to 1 petaFLOP in FP4 with sparsity; not measured application throughput |
SSD slot | 1 × M.2; up to 4 TB |
Wired networking | 1 × 10 Gb/s Ethernet; NVIDIA ConnectX-7 SmartNIC for interconnection |
Wireless | Wi-Fi 7 2×2; Bluetooth 5.4 |
USB | 4 × USB-C 20 Gb/s: 3 with DisplayPort Alt Mode and 1 with PD 3.1 EPR power input |
Video | HDMI 2.1a; DisplayPort 2.1 through compatible USB-C ports |
System | NVIDIA DGX OS, Linux environment for Arm |
Power adapter | 240 W; rated power, not measured consumption |
Dimensions and weight | 150 × 150 × 51 mm; 1.48 kg |
Finish | Stellar Grey |
Capacity changes, the family stays the same
ASUS's early October 2026 announcement presents this configuration as a new entry point into the GX10 range. It retains an Arm CPU paired with a Blackwell GPU. In the catalogue consulted, GX10-GG0062BN combines 64 GB of memory with a 1 TB SSD.
Choosing this version does not mean all workloads will run at half the speed of the 128 GB model. Capacity primarily determines what can remain in memory. As long as a workload fits, speed also depends on the runtime, operations performed and data exchanges.
Start with the model, then allow headroom
For a defined project, measure model memory use at the target precision, then add context, concurrent-request and system requirements. The downloaded file size alone is not enough: it does not necessarily represent all memory used during inference.
A 64 GB configuration can therefore suit a well-defined application. Conversely, teams looking to compare many models or rapidly expand usage should assess the headroom of the 128 GB version. The cheapest machine only makes sense if it can deliver the planned project.
Assess the configuration as a whole
The 1 TB storage capacity must also be considered from the outset. Models, software environments, caches and datasets may share the same SSD. Storage organisation and backups remain necessary even when all computation takes place locally.
64 GB does not mean half a GPU
ASUS specifies 273 GB/s memory bandwidth and retains GB10. Two comparisons must therefore be kept separate: how much data the machine can host and how long it takes to process it. Reduced capacity does not mean every throughput figure is halved.
ASUS particularly targets models with 30–35 billion parameters. That guidance still depends on the model and precision: it does not guarantee that every model of that size fits with maximum context. ASUS also mentions linking two units through ConnectX-7. This requires suitable distributed execution; it does not let any program treat both machines' memory as one pool.
Key points
Criterion | GX10 64 GB | GX10 128 GB |
|---|---|---|
Platform | NVIDIA GB10 | NVIDIA GB10 |
Unified CPU–GPU memory | 64 GB | 128 GB |
Range operating system | NVIDIA DGX OS | NVIDIA DGX OS |
64 GB configuration consulted | GX10-GG0062BN, 1 TB SSD | See the 1, 2 and 4 TB variants |
Decision point | Known model and context | More memory headroom |
SKUs and manufacturer part numbers
Model | Manufacturer part number | SSD |
|---|---|---|
GX10-GG0062BN | 90MS0371-M001Y0 | 1 TB NVMe SSD, PCIe 4.0 |
Explore further
ASUS Ascent GX10 128 GB: a compact machine for local AI development
ASUS ExpertCenter Pro ET900N G3: a GB300 workstation for large-scale AI projects
GX10 or ET900N G3: how to choose your ASUS machine for local AI
Key takeaways
This variant therefore broadens the choice without automatically replacing the GX10 128 GB. It makes sizing more important: buying suitable capacity today while retaining realistic headroom for tomorrow. A retailer listing must also be distinguished from confirmed stock and delivery times.