The 64 GB GX10 broadens access to the GB10 platform
Announced in early October 2026, the 64 GB ASUS Ascent GX10 variant complements the 128 GB versions. It retains NVIDIA GB10, which combines a 20-core Arm processor — 10 Cortex-X925 and 10 Cortex-A725 cores — with an integrated Blackwell GPU. LPDDR5X memory is unified across the CPU and GPU, with a stated bandwidth of 273 GB/s.
The main change therefore concerns the capacity available for models and their working data. It does not mean performance is automatically halved. ASUS targets models with around 30 to 35 billion parameters, with actual capabilities depending on quantisation, context and the runtime engine.
The GX10-GG0062BN model, identified by SKU 90MS0371-M001Y0, pairs 64 GB of memory with a 1 TB PCIe 4.0 NVMe SSD. It retains the 150 × 150 × 51 mm form factor, 10 GbE networking, Wi-Fi 7 and ConnectX-7. The stated operating system is NVIDIA DGX OS: this machine is designed for development and computing on Linux on Arm.
128 GB: more headroom in the same enclosure
The 128 GB version remains relevant when the model, context cache and supporting services take up more memory. The three listed models differ in their SSD: GX10-GG0003BN with 1 TB, GX10-GG0026BN with 2 TB and GX10-GG0027BN with 4 TB. The first two are specified as PCIe 4.0, and the last as PCIe 5.0.
The 64 or 128 GB is shared by the CPU and GPU: it does not come on top of a separate VRAM pool. The one-petaFLOP figure advertised for GB10 refers to FP4 with sparsity, not a universal performance level for all models. The choice therefore depends as much on precision and software as on memory capacity.
The ET900N G3 moves up to GB300 and HBM3e
At the other end of the range, the ExpertCenter Pro ET900N G3 combines an NVIDIA Grace processor with 72 Arm cores and a Blackwell Ultra GPU. Its 748 GB coherent memory pool comprises 496 GB of LPDDR5X on the CPU side and 252 GB of HBM3e on the GPU side. The stated bandwidths are 396 GB/s and 7.1 TB/s respectively: the total memory figure should not obscure this difference.
The workstation has four M.2 slots, two ConnectX-8 ports rated at up to 400 Gb/s each, 10 GbE and a BMC management controller. Its roughly 27 kg chassis and 1,600 W power supply put it in a different installation category from the GX10.
Ten SKUs were identified in the catalogue consulted, with 4, 6 or 8 TB of raw SSD capacity. Some add a 16 GB RTX PRO 2000 Blackwell, a 24 GB RTX PRO 4000 Blackwell SFF or a 96 GB RTX PRO 6000 Blackwell Max-Q. Their GDDR7 memory is separate from GB300's coherent memory pool. Wi-Fi and QSFP accessories also vary by model.
The specifications that distinguish the three tiers
Specification | GX10 64 GB | GX10 128 GB | ET900N G3 |
|---|---|---|---|
Superchip | NVIDIA GB10 | NVIDIA GB10 | NVIDIA GB300 |
CPU | 20 Arm cores | 20 Arm cores | 72 Grace Arm cores |
Memory | 64 GB unified | 128 GB unified | 496 GB LPDDR5X + 252 GB HBM3e |
Listed storage options | 1 TB | 1, 2 or 4 TB | 4, 6 or 8 TB raw |
Compute networking | ConnectX-7 | ConnectX-7 | ConnectX-8, up to 2 × 400 Gb/s |
Positioning | Local project sized to fit | Development with more headroom | Specialised workstation for more demanding projects |
Understanding the range before ordering
The new development is the arrival of the 64 GB version; the 128 GB GX10 and ET900N G3 provide context for the range, rather than representing three simultaneous launches. A commercial listing does not guarantee immediate stock. The ET900N offer consulted is sold on a project basis subject to NVIDIA's approval, with terms that must be confirmed with the supplier.
To go beyond the catalogue, our guide explains how to size the machine around model weights, context, concurrent workloads and software. The dedicated product profiles detail each machine's connectivity, storage and SKUs.
Explore further
GX10 or ET900N G3: how to choose your ASUS machine for local AI
ASUS Ascent GX10 128 GB: a compact machine for local AI development
ASUS Ascent GX10 64 GB: who is this new configuration for?
ASUS ExpertCenter Pro ET900N G3: a GB300 workstation for large-scale AI projects
Key takeaways
The 64 GB GX10 broadens the options for projects sized to fit its capabilities; the 128 GB model retains more headroom, while the GB300 addresses an entirely different scale of computing and installation. The detailed product profiles and guide help turn these specifications into a choice suited to the project.