Close

Lenovo thinkstation pgx 30kl0002gf ai desktop nvidia gb10 grace blackwell 128GB unified 1TB SSD dgx os black

Product image: Lenovo thinkstation pgx 30kl0002gf ai desktop nvidia gb10 grace blackwell 128GB unified 1TB SSD dgx os black-null
Product image: Lenovo thinkstation pgx 30kl0002gf ai desktop nvidia gb10 grace blackwell 128GB unified 1TB SSD dgx os black-null
LENOVO
🔥Selling fast!

Lenovo thinkstation pgx 30kl0002gf ai desktop nvidia gb10 grace blackwell 128GB unified 1TB SSD dgx os black

€8,054.97
or
Pay in 3, 6 or 9 interest-free monthly installments and 0 service fees with Fleksy image
Select Color: Black
In stock
–+
–+
13054
Points earned
You can earn Antamivi points every time you make a payment using your Bank of Cyprus cards.
The number of Antamivi points and the corresponding amount of money displayed are indicative and may vary from time to time. Want to know more? Check out our FAQs here.
If you are an eligible member of the Bank of Cyprus loyalty scheme Pronomia, you will receive additional Antamivi points based on your assigned tier.
From: Singular Computers Systems LTD
Buy more from this shop
Delivery options
Pick-up points / Lockers
Est. delivery: 23 Jun - 24 Jun
Free
Delivery to your door
FREE shipping on orders over €85.00 from Singular Computers Systems LTD
Est. delivery: 23 Jun - 25 Jun
Free

Description

Item No. e639f51a-dcbb-46eb-91ec-25bec0513684
Lenovo ThinkStation PGX — GB10 Grace Blackwell AI Desktop Up to 1 petaFLOP FP4 AI128GB Unified Memory1TB NVMe SSDNVIDIA DGX OS The Lenovo ThinkStation PGX, part number 30KL0002GF, is a compact NVIDIA GB10 Grace Blackwell AI desktop supercomputer with 128GB of unified memory, built for AI developers, researchers and data scientists who want to prototype, fine-tune and run large language models locally on their desk. Powered by the NVIDIA GB10 Grace Blackwell superchip — a 20-core Arm CPU paired

Attributes

Pc Characteristics

  • Primary Disk Capacity: 1TB

Details

Lenovo ThinkStation PGX — GB10 Grace Blackwell AI Desktop

Up to 1 petaFLOP FP4 AI128GB Unified Memory1TB NVMe SSDNVIDIA DGX OS

The Lenovo ThinkStation PGX, part number 30KL0002GF, is a compact NVIDIA GB10 Grace Blackwell AI desktop supercomputer with 128GB of unified memory, built for AI developers, researchers and data scientists who want to prototype, fine-tune and run large language models locally on their desk.

Powered by the NVIDIA GB10 Grace Blackwell superchip — a 20-core Arm CPU paired with a Blackwell GPU and 5th-generation Tensor Cores — the PGX delivers up to 1 petaFLOP of FP4 AI performance from a chassis that measures just 150 × 150 × 50.5 mm. Its 128GB of coherent LPDDR5X memory is shared between CPU and GPU, so you can hold models up to roughly 200 billion parameters in memory on a single unit, or link two units over NVIDIA ConnectX networking to handle models up to around 405 billion parameters.

This is Lenovo’s equivalent of the NVIDIA DGX Spark — a purpose-built AI development appliance rather than a General-purpose Windows PC. Typically in stock with fast delivery across Cyprus.

Before you buy — software & OS compatibility
The ThinkStation PGX runs NVIDIA DGX OS (an Ubuntu-based Linux distribution) on an Arm-based GB10 superchip, not Windows on x86. It is designed for the NVIDIA AI software stack (CUDA, NeMo, NIM microservices and containers). Confirm that your frameworks, drivers and any external peripherals support Arm64 Linux / DGX OS before purchase.

Best For

✓Local prototyping and development of large language models
✓Fine-tuning and LoRA training of models up to ~200B parameters
✓On-device inference without sending data to the cloud
✓AI researchers and data-science teams needing desk-side compute
✓Edge and lab deployments where space and power are limited
✓Scaling to ~405B-parameter models by linking two units via ConnectX

Key Features

 
NVIDIA GB10 Grace Blackwell superchip
A 20-core Arm CPU (10× Cortex-X925 + 10× Cortex-A725) fused with a Blackwell GPU and 5th-gen Tensor Cores, delivering up to 1 petaFLOP of FP4 AI compute in a desktop form factor.
 
128GB coherent unified memory
LPDDR5X memory shared between CPU and GPU eliminates copy overhead and lets you keep very large models resident — up to roughly 200 billion parameters on a single PGX.
 
Scales with NVIDIA ConnectX
Built-in ConnectX networking plus 10GBE lets you link two PGX units to run models up to around 405 billion parameters, with Wi-Fi 6/7 and Bluetooth 5.3 for connectivity.
 
Turnkey NVIDIA DGX OS
Ships with NVIDIA DGX OS and the full NVIDIA AI software stack, so you can move from unboxing to model development without assembling a Linux/CUDA environment from scratch.

Technical Specifications

Identification
Part Number 30KL0002GF
EAN / GTIN 0199274974237
MPN 30KL0002GF
Brand Lenovo
Series ThinkStation PGX
Processor & AI Performance
Superchip NVIDIA GB10 Grace Blackwell
CPU 20-core Arm (10× Cortex-X925 + 10× Cortex-A725)
GPU NVIDIA Blackwell architecture with 5th-gen Tensor Cores
AI Performance Up to 1 petaFLOP (FP4)
Memory & Storage
Memory 128GB LPDDR5X coherent unified memory (CPU + GPU shared)
Storage 1TB PCIe 4.0 NVMe SSD (Opal 2.0 self-encrypting)
Connectivity
Networking NVIDIA ConnectX + 10 Gigabit Ethernet
Wireless Wi-Fi 6/7 + Bluetooth 5.3
Scaling Link two units via ConnectX for models up to ~405B parameters
Software & General
Operating System NVIDIA DGX OS (Ubuntu-based Linux)
Model capacity (single unit) Up to ~200 billion parameters
Dimensions 150 × 150 × 50.5 mm
Colour Black

Frequently Asked Questions

Does the ThinkStation PGX run Windows or my existing x86 software?
No. The PGX runs NVIDIA DGX OS, an Ubuntu-based Linux distribution, on the Arm-based GB10 superchip. It is built for the NVIDIA AI stack (CUDA, NeMo, NIM, containers). Confirm your tools support Arm64 Linux before buying; it is not a drop-in replacement for a Windows workstation.
How large a model can the PGX run locally?
A single ThinkStation PGX with its 128GB of unified memory can hold and run models up to roughly 200 billion parameters. By linking two PGX units over NVIDIA ConnectX you can scale to models of around 405 billion parameters.
What is the difference between the ThinkStation PGX and the NVIDIA DGX Spark?
The ThinkStation PGX is Lenovo’s implementation of the same NVIDIA GB10 Grace Blackwell platform that powers the DGX Spark class of desktop AI supercomputers. You get the same GB10 superchip, 128GB unified memory and DGX OS, backed by Lenovo ThinkStation engineering and support.
What do I need to know before buying the ThinkStation PGX?
It is a purpose-built AI development appliance, not a General PC: Arm + DGX OS Linux, 128GB shared CPU/GPU memory, 1TB NVMe storage and a tiny 150×150mm footprint. Make sure your AI frameworks, drivers and peripherals are Arm64/DGX OS compatible, and plan for ConnectX networking if you intend to scale to two units.
Singular Expert Insight
If your work involves developing or fine-tuning large models and you want to keep data on-premise, the ThinkStation PGX gives you cloud-class AI compute on your desk without ongoing GPU-rental costs. For teams already standardised on Windows/x86 tooling or running only smaller models, a conventional RTX workstation may be a better fit — the PGX’s value is its 128GB unified memory and DGX OS stack for Genuinely large models. Choose it when you need to iterate locally on 70B–200B+ parameter models, and consider a second unit when you outgrow a single one.

Buyer’s Guide

Buy it if
You fine-tune or run 70B–200B+ parameter models locally, want 128GB unified memory and DGX OS out of the box, and need a silent 150mm desk-side AI box instead of cloud GPUs.
Skip it if
You rely on Windows or x86-only software, only run smaller models, or need a General-purpose workstation — the Arm + DGX OS Linux platform would limit you.
Also consider
A second ThinkStation PGX linked via ConnectX for models up to ~405B parameters, or a Lenovo ThinkStation with RTX GPUs if you need Windows and CUDA on x86.