Dgx specs

1TB NVMe SSD, 128GB LPDDR5X, GB10 Grace Blackwell

Dgx story

[{"type":"hero","eyebrow":"ASUS Ascent GX10 · NVIDIA GB10","heading":"A petaFLOP of AI.\nOn your desk.","body":"The GB10 Grace Blackwell Superchip and 128 GB of unified memory, in a chassis the size of a Mac mini — the DGX platform, at the smartest way in.","image":"//platinummicro.com/cdn/shop/files/512mn_LKArL._AC_UF894_1000_QL80_AIweblab1381794_T1_-removebg-preview.png","image_alt":"ASUS Ascent GX10 desktop AI supercomputer, front three-quarter view","pills":["GB10 Grace Blackwell","128 GB unified memory","1 PetaFLOP FP4","DGX OS"],"callouts":[{"value":"1","unit":"PF","label":"FP4 AI compute","x":55,"y":25,"side":"right"},{"value":"128","unit":"GB","label":"Unified memory","x":39,"y":45,"side":"left"},{"value":"200","unit":"Gb/s","label":"ConnectX-7","x":61,"y":63,"side":"right"},{"value":"20","unit":"core","label":"Arm CPU","x":42,"y":74,"side":"left"}],"cta_label":"View buying options"},{"type":"stat","icon":"code","value":"70","unit":"B","label":"Fine-tune range (30–70B)"},{"type":"stat","icon":"bolt","value":"200","unit":"B","label":"Inference reach, on-prem"},{"type":"stat","icon":"network","value":"2","unit":"node","label":"Validated cluster"},{"type":"stat","icon":"cloud","value":"$0","unit":"","label":"Recurring cloud cost"},{"type":"feature","icon":"chip","eyebrow":"The silicon","ground":"navy","heading":"GB10 Grace Blackwell Superchip","body":"A 20-core Arm CPU and a Blackwell GPU share one memory space over NVLink-C2C — no PCIe bottleneck between them. It is the architecture NVIDIA ships in DGX systems, on your desk.","pills":["20-core Arm CPU","Blackwell GPU","NVLink-C2C"]},{"type":"why","eyebrow":"Capacity","heading":"Room to run — and to scale","items":[{"icon":"memory","heading":"128 GB unified memory","body":"Fine-tune models in the 30–70B range and run inference up to ~200B parameters — one pool, addressed by CPU and GPU alike, nothing to copy or partition."},{"icon":"network","heading":"Cluster when you outgrow one","body":"The built-in NVIDIA ConnectX-7 NIC direct-connects two units — and scales to a four-unit cluster (per ASUS) — when a single box isn’t enough."}]},{"type":"who","eyebrow":"Who it's for","heading":"Built for teams doing real AI work","body":"Owned, local compute for the people who were paying for it by the hour.","items":[{"icon":"code","heading":"Developers & ML researchers","body":"Prototype and fine-tune in the 30–70B range; trade per-run cloud cost for a fixed, owned asset."},{"icon":"shield","heading":"Regulated industries","body":"Healthcare, finance and government teams that need on-prem, air-gapped-capable processing."},{"icon":"school","heading":"Research & academia","body":"Capex-budget compute for imaging, genomics, astrophysics and coursework — no recurring cloud bill."},{"icon":"robot","heading":"Robotics & edge vision","body":"A local target for NVIDIA Isaac, Metropolis and Holoscan, from sensor processing to assistive tech."}]},{"type":"feature","icon":"stack","eyebrow":"Software","ground":"navy","heading":"The NVIDIA AI stack, preloaded","body":"Runs NVIDIA DGX OS on Ubuntu, with CUDA, PyTorch, TensorFlow, TensorRT, Jupyter and NVIDIA NIM microservices already set up. Power on, open a notebook, start building.","pills":["CUDA","PyTorch","TensorFlow","TensorRT","NVIDIA NIMs","Jupyter"]},{"type":"ports","eyebrow":"Connectivity","heading":"Everything plugs in around the back","body":"Four USB-C ports (all 20 Gbps, all DisplayPort 2.1), HDMI 2.1b, 10 GbE wired networking — and the headline act, a dual-port NVIDIA ConnectX-7 for stitching units together. It even takes its power over USB-C.","image":"https://dlcdnwebimgs.asus.com/files/media/202506/5c0fb57c-4e48-4e96-8c97-04bf8df2677c/v1/img/line_pd.png","image_alt":"ASUS Ascent GX10 rear I/O — Kensington lock, four USB-C 20 Gbps ports with DisplayPort 2.1, HDMI 2.1b, 10 GbE LAN, and an NVIDIA ConnectX-7 NIC","caption":"Rear I/O — ASUS Ascent GX10","items":[["4 × USB-C · 20 Gbps","USB 3.2 Gen 2x2 — every port does DisplayPort 2.1"],["Powered over USB-C","One port takes 180 W in (EPR PD 3.1)"],["HDMI 2.1b","Direct display output"],["10 GbE LAN","Fast wired networking, standard"],["NVIDIA ConnectX-7","Dual-port cluster NIC — link units for bigger models"],["Kensington lock","Physical anti-theft anchor"]]},{"type":"spec-table","eyebrow":"Specifications","heading":"Technical specifications","rows":[["Superchip","NVIDIA GB10 Grace Blackwell"],["CPU","20-core Arm"],["GPU","NVIDIA Blackwell — 5th-gen Tensor Cores, FP4"],["AI performance","Up to 1 petaFLOP (FP4)"],["Unified memory","128 GB LPDDR5X, coherent — NVLink-C2C (5× PCIe 5.0 bandwidth)"],["Storage","NVMe SSD — capacity by configuration (choose above)"],["Networking","NVIDIA ConnectX-7 NIC + 10 GbE LAN"],["Rear I/O","4× USB-C 20 Gbps (DisplayPort 2.1) · HDMI 2.1b · Kensington lock"],["Power","180 W in over USB-C (EPR PD 3.1)"],["Operating system","NVIDIA DGX OS (Ubuntu Linux)"],["Dimensions","150 × 150 × 51 mm · 1.6 L toolless chassis"]]},{"type":"why","eyebrow":"Why Platinum Micro","variant":"band","heading":"More than a box on a shelf","items":[{"icon":"bolt","heading":"Real allocation, now","body":"One of a handful of reliable sources while GB10 memory supply is tight — not a backorder."},{"icon":"wrench","heading":"Setup & deployment","body":"Optional imaging, model-deployment consulting and support tiers so it’s productive on day one."},{"icon":"doc","heading":"PO & financing","body":"Purchase orders, EDU pricing and monthly financing for teams weighing capex against cloud."},{"icon":"refresh","heading":"Trade-in ready","body":"Upgrade credit as new GB10-generation systems land — a relationship, not a one-off."}]},{"type":"faq","eyebrow":"Questions","heading":"Frequently asked questions","items":[{"q":"Who is the ASUS Ascent GX10 for?","a":"AI developers, ML researchers, regulated-industry data scientists, research labs and robotics teams who want persistent, local AI compute — for fine-tuning in the 30–70B range and inference up to ~200B parameters."},{"q":"Can it train large models from scratch?","a":"It’s a fine-tuning and inference machine, not a from-scratch training cluster. For pretraining above ~70B parameters, look at NVIDIA DGX Station or a cloud cluster — we can advise on the right step up."},{"q":"How does it compare to cloud GPUs on cost?","a":"Cloud GPU sessions run roughly $2–$15 each and never stop. The GX10 turns that recurring opex into a fixed, owned asset — and keeps sensitive data on-prem."},{"q":"What comes in the box?","a":"The GX10 itself — GB10 Superchip, 128 GB unified memory, the built-in ConnectX-7 NIC, and NVIDIA DGX OS (Ubuntu) with the AI stack preloaded. Ask us to confirm exact in-box contents for your configuration."}]},{"type":"cta","heading":"Configure your GX10","body":"In stock and allocation-backed. Choose a configuration above, or request a quote for volume pricing and PO terms.","label":"View buying options"}]

Detailed specifications for this part are not yet published. Reach out to your account manager — dimensions, voltage, capacity, and supported configurations will be confirmed within one business day.