Lenovo ThinkStation PGX GB10 1T LINUX Workstation (30KL0002US)

The Lenovo ThinkStation PGX is Lenovo’s first workstation built around the NVIDIA GB10 Grace Blackwell Superchip. It gives ML engineers, researchers and applied-AI teams data-center-class model development power on their own desk, backed by Lenovo’s workstation-grade build and enterprise support. If you rent cloud GPU instances just to fine-tune or evaluate open-weight models, it moves that work into a 1.13-litre box beside your keyboard.

Who it's for

• ML engineers running LoRA/QLoRA fine-tunes and evaluation harness runs on hardware they own
• Applied-AI teams prototyping RAG pipelines, agentic workflows, and computer-vision or multimodal models
• Organizations handling data that cannot leave the premises

Key features

• NVIDIA GB10 Grace Blackwell Superchip: a 20-core Arm CPU (10x Cortex-X925 performance cores + 10x Cortex-A725 efficiency cores) coupled to a Blackwell-architecture GPU with 5th Gen Tensor Cores and 4th Gen RT Cores over NVLink-C2C, rated for 1000 TOPS / 1 petaFLOP of FP4 AI performance with sparsity.
• 128GB unified LPDDR5x-8533 memory at 273GB/s: CPU and GPU share one coherent pool with no PCIe copy penalty, holding models up to roughly 200 billion parameters on a single unit without offloading to disk or splitting across GPUs.
• Dual 200Gb/s QSFP ports (NVIDIA ConnectX-7): link two PGX units to scale past a single unit’s memory ceiling.
• Self-encrypting NVMe storage: 1TB, 2TB or 4TB PCIe Gen4 TLC Opal SSD options, so you can size local storage to your dataset and checkpoint volume.
• NVIDIA DGX OS with the full AI stack pre-installed: CUDA 13, the complete NVIDIA AI software stack, the GB10 Dashboard and NVIDIA AI Workbench, using the same container stack as DGX systems, so a workload moves to a cluster or DGX Cloud unchanged.
• Enterprise-grade connectivity: 10 Gigabit Ethernet, Wi-Fi 7, Bluetooth 5.3, three USB4 ports carrying DisplayPort 2.1, and HDMI 2.1a.
• 1.13-litre chassis, from 1.2kg: 150 x 150 x 50.5mm, on a single 240W USB-C adapter.
• Workstation-grade security: AES SED NVMe, TPM 2.0, UEFI Secure Boot, an NVLink-C2C hardware enclave and NVIDIA Firmware Recovery.

Specifications

Good to know

• The 1 petaFLOP figure is a peak FP4-with-sparsity number. Real dense-precision throughput on a given model will be lower.
• This is a fine-tuning, inference and prototyping machine, not a foundation-model training rig. Pre-training multi-billion-parameter models from random initialization belongs on a cluster.
• A 70B-parameter model in FP4/INT4 quantization, or a 30B model at higher precision with generous context length, loads and runs without memory juggling.
• Pick 1TB, 2TB or 4TB by how many checkpoints, quantized variants and datasets you keep local.

Choose the ThinkStation PGX if you want Lenovo’s 1-year limited onsite service and a choice of 1TB, 2TB or 4TB storage.