A compact Dell desktop AI system built on the NVIDIA GB10 Grace Blackwell Superchip, with 128GB of unified memory, a 2TB SSD and NVIDIA DGX OS 7. Designed for local AI development, prototyping, fine-tuning and inference with the NVIDIA AI software stack.
24-month Percepta return-to-base hardware warranty from delivery for items sold as New. Subject to the Hardware Warranty Policy. Your statutory rights are unaffected.
128GB of unified memory for AI development at your desk.
The Dell Pro Max with GB10 (FCM1253) is a complete desktop computer, 150mm square, built around the NVIDIA GB10 Grace Blackwell Superchip. A 20-core Arm CPU and a Blackwell GPU share one 128GB pool of LPDDR5X memory. The system comes with NVIDIA DGX OS 7 and NVIDIA's AI software stack.
It is designed for developers, researchers and data scientists who want to prototype, fine-tune and run AI models on local hardware. Projects can later move to larger NVIDIA infrastructure. This configuration is Dell part number T9WMV, with a 2TB SSD, Wi-Fi 7 and a 280W USB-C power adapter.
128GBLPDDR5X unified memory shared by CPU and GPU
273GB/sUnified memory bandwidth
1 PFLOPPeak FP4 AI compute, with sparsity
2x 200GbpsConnectX-7 QSFP network ports
Grace Blackwell in a compact desktop
One memory pool for large models
The GB10 links its CPU and GPU with NVLink-C2C, so both work from the same 128GB of memory rather than a separate, smaller pool of graphics memory. Dell quotes models of up to 200 billion parameters for this system; NVIDIA gives the same figure for FP4-precision inference on its GB10-based DGX Spark, and up to 70 billion parameters for fine-tuning. What fits in practice depends on precision, context length, the framework and the memory used by the operating system and other applications.
NVIDIA DGX OS and AI software
DGX OS 7 is NVIDIA's Ubuntu-based operating system, set up with the GPU drivers and NVIDIA AI software stack. NVIDIA names familiar tools such as PyTorch, Jupyter and Ollama for prototyping, fine-tuning and local inference on GB10 systems, and Dell positions this model for data science and agentic AI development. NVIDIA AI Enterprise, where required, is licensed separately.
Link two systems over ConnectX-7
An NVIDIA ConnectX-7 SmartNIC provides two 200Gbps QSFP ports; the system also has 10GbE and Wi-Fi 7 for everyday networking. Dell specifies one QSFP port for connecting a second Dell Pro Max with GB10 and quotes models of up to 400 billion parameters across the pair. Dell lists RDMA and GPUDirect support for clustered workloads. The software must support distributed inference or training to use both.
Technical specification
Dell part number T9WMV, with Dell and NVIDIA manufacturer specifications
01 / Supplied configuration
System
Dell Pro Max with GB10, model FCM1253, micro desktop
20-core Arm: 10x Cortex-X925 and 10x Cortex-A725; 16MB cache
GPU
Integrated NVIDIA Blackwell GPU with fifth-generation Tensor Cores and fourth-generation RT cores
Memory
128GB LPDDR5X unified system memory, 8533MT/s, up to 273GB/s; onboard, no memory slots
Storage
2TB M.2 2230 PCIe Gen4 NVMe SSD
Operating system
NVIDIA DGX OS 7 (Ubuntu-based Linux)
Wireless
Wi-Fi 7 (802.11be) and Bluetooth 5.4
Power
280W USB-C AC adapter
02 / GB10 compute, memory and video
AI performance
Up to 1 PFLOP FP4 Tensor performance, theoretical peak with sparsity
CUDA cores
6,144, NVIDIA Blackwell generation
Memory interface
256-bit LPDDR5X; coherent CPU-GPU memory over NVLink-C2C
Video engines
1x NVENC; 1x NVDEC
Display output
HDMI 2.1 up to 7680 x 4320 at 30Hz; DisplayPort 1.4a alt mode on the 20Gbps USB-C ports, up to 7680 x 4320 at 60Hz for a single display. DisplayPort MST is not supported.
FP4 performance is a manufacturer peak using sparsity, not an application benchmark. Model capacity and speed depend on precision, framework, context length and workload.
03 / Connections and physical format
High-speed networking
NVIDIA ConnectX-7 SmartNIC with 2x QSFP ports, 200Gbps each
Ethernet
1x RJ-45 10GbE
USB
3x USB-C 20Gbps (USB 3.2 Gen 2x2) with DisplayPort 1.4a alt mode and power out; 1x USB-C power input
Video port
1x HDMI 2.1
Expansion
No expansion slots; one M.2 2230/2242 SSD position, fitted with the 2TB drive
Dimensions
150mm wide x 150mm deep; 45.5mm high, 51mm at the highest point
Weight
1.22kg to 1.34kg, depending on configuration
Power adapter
280W USB-C AC adapter, connected to the dedicated USB-C power input
Operating temperature
0°C to 35°C
All ports are on the rear panel. Linking two systems requires a compatible QSFP cable and a second Dell Pro Max with GB10.
Supplied with this listing
What's included
Dell Pro Max with GB10 FCM1253 system, Dell part number T9WMV
128GB LPDDR5X unified memory and 2TB M.2 NVMe SSD, installed
NVIDIA DGX OS 7
Wi-Fi 7 and Bluetooth 5.4 wireless module
280W USB-C power adapter
Plan your setup
What you'll need
An internet connection during first setup; Dell's setup steps use the 10GbE RJ-45 port
A display, keyboard and mouse, or a Wi-Fi device to complete Dell's wireless setup
HDMI or USB-C DisplayPort cabling suited to your display
A compatible QSFP cable and a second Dell Pro Max with GB10 to link two systems
Software and containers built for Linux on 64-bit Arm (aarch64), plus the model weights and any licences your workload requires
What kind of AI work is it suited to?
Dell and NVIDIA position GB10 systems for local AI development: prototyping applications and agents, fine-tuning, inference and data science, as well as developing edge and robotics applications. The shared 128GB memory can hold larger models than a graphics card with less dedicated memory, and the NVIDIA software stack helps projects move to larger NVIDIA systems later.
It is a development and inference system rather than a replacement for multi-GPU servers used for large-scale training.
How large a model can I run in 128GB?
Dell quotes models of up to 200 billion parameters for this system. NVIDIA gives the same figure for FP4-precision inference on GB10, and up to 70 billion parameters for fine-tuning. These are upper limits, not guarantees. Memory is shared with the operating system and your applications, and requirements grow with higher precision, longer context and larger batches. Check your model's memory needs in the framework you plan to use.
Can I run Windows or x86 software on it?
It comes with NVIDIA DGX OS 7, the operating system Dell lists for this model. DGX OS is Ubuntu-based Linux running on Arm, so applications and containers need 64-bit Arm (aarch64) builds. It is not a Windows PC; check your toolchain before ordering.
Can I connect two systems together?
Yes. Dell specifies one of the two 200Gbps QSFP ports for connecting a second Dell Pro Max with GB10, and quotes models of up to 400 billion parameters across the pair. You will need a compatible QSFP cable and software that supports distributed inference or training.
Can I upgrade the memory or storage?
The 128GB of memory is onboard and cannot be expanded. The system has one M.2 2230/2242 SSD position, already fitted with the 2TB drive, and no expansion slots. Plan additional storage for models and datasets through the USB-C ports or the network.
How does the Dell manufacturer warranty relate to Percepta's warranty?
Percepta's warranty for this listing is shown in the buying information. Dell's manufacturer warranty is separate and is provided by Dell under Dell's own terms. Your statutory rights are unaffected.
