Xeon 696X Dual RTX PRO 6000 AI Workstation

Xeon 696X Dual RTX PRO 6000 AI Workstation – Built in the UK

The Xeon 696X Dual RTX PRO 6000 AI Workstation is a high-end dual-GPU workstation for local large language model work and multi-GPU AI development, pairing a 64-core Intel Xeon 696X with two NVIDIA RTX PRO 6000 96GB GPUs for a combined 192GB of GPU memory in a single desk-side system.

At the heart of this dual RTX PRO 6000 AI workstation is the Intel Xeon 696X, a 64-core, 128-thread processor running at 2.4GHz base and up to 4.8GHz boost with 336MB of L3 cache. It exists for one reason: to run genuinely large models and multi-GPU AI workloads locally, without renting cloud GPU time or splitting a job across separate machines.

The Xeon 696X Dual RTX PRO 6000 AI Workstation is built on an Intel Xeon 696X with 64 cores and 128 threads, two NVIDIA RTX PRO 6000 Blackwell Workstation Edition 96GB GDDR7 graphics cards and 384GB of DDR5 ECC RDIMM memory at 5600MHz. Storage is two 512GB Innodisk Enterprise M.2 NVMe SSDs and two 7.68TB Micron 7500 PRO U.3 NVMe SSDs, and it comes with Ubuntu Linux 26.04 LTS pre-installed. It is housed in a Phanteks Enthoo Pro II Server Edition case with a 2500W 80 Plus Platinum power supply and comes with a 3-year parts and labour warranty.

Storage

A 512GB Innodisk Enterprise M.2 NVMe SSD (2600MB/s read, 1800MB/s write) holds the operating system and applications, with a second identical 512GB drive alongside it. Working storage comes from two 7.68TB Micron 7500 PRO enterprise U.3 PCIe 4.0 NVMe SSDs (7000MB/s read, 5900MB/s write), giving the system 15.36TB of enterprise-grade scratch space for datasets, model checkpoints and active projects.

Graphics

Two NVIDIA RTX PRO 6000 Blackwell Workstation Edition 96GB GPUs provide a combined 192GB of GDDR7 memory – enough to load very large language models locally, split a large model across both GPUs, or run two substantial models side by side for parallel experimentation. Each card has 24,064 cores and draws up to 600W, and the motherboard’s PCIe 5.0 x16 slots give each card its own full-length slot. Display outputs are 4 x DisplayPort 2.1 on the graphics card, supporting up to four monitors. The Xeon 696X has no integrated graphics, so every monitor connects to a graphics card.

Case, cooling and power

The Phanteks Enthoo Pro II Server Edition is a full tower with a tempered glass panel, measuring 560mm deep, 240mm wide and 580mm high, and it is designed around multi-GPU, multi-drive builds like this one. The Arctic Freezer WS360-4710 is a 360mm AIO liquid cooler rated to dissipate up to 800W, giving substantial headroom for the 64-core Xeon 696X. Four Be Quiet! Silent Wings 4 PWM 140mm fans are added for chassis airflow, and an FSP Cannon Pro 2500W 80 Plus Platinum ATX 3.1 fully modular PSU delivers the headroom that two 600W GPUs and a 64-core Xeon need, with capacity to spare.

Connectivity

The ASUS Pro WS W890E-SAGE SE motherboard supplies dual 10GbE networking (Intel E610-XAT2) plus a dedicated 1GbE BMC/IPMI port for remote management. That is useful for both fast dataset transfers and headless server-style administration. Networking is wired only, with no built-in Wi-Fi. The rear panel has 2 x USB4 Type-C and 6 x Type-A 10Gbps ports, and the front panel has 1 x USB-C Gen2, 4 x USB 3.0 and headphone and microphone jacks. Audio is a built-in 7.1 HD Audio codec (Realtek ALC1220P). With 384GB of ECC DDR5 RDIMM populating all eight memory slots, the system has the memory capacity to keep a 64-core CPU and two data-hungry GPUs fed at the same time.

Ideal for

• Local fine-tuning and inference of large language models across dual GPUs
• Running multiple substantial models in parallel for research and experimentation
• AI infrastructure teams standardising on-premises compute before scaling to cluster deployment
• Data-heavy scientific and engineering simulation requiring ECC memory
• Organisations that need model work to stay on hardware they control

The Xeon 696X Dual RTX PRO 6000 AI Workstation is built for teams and researchers whose model work has outgrown a single-GPU system. It puts 192GB of GPU memory, 384GB of ECC system memory and enterprise NVMe storage into one desk-side tower, without the complexity of moving to a rack-mounted server. For a dual RTX PRO 6000 AI workstation, the balance of CPU, memory, storage and power is matched to what two 600W GPUs and a 64-core Xeon demand.

Built to order and tested in the UK by Punch Technology, every system comes with a 3-year parts and labour warranty, backed by over 30 years of PC building experience. The estimated lead time for your configuration is shown when you order.

What is the Xeon 696X Dual RTX PRO 6000 AI Workstation best for?

The Xeon 696X Dual RTX PRO 6000 AI Workstation is primarily a multi-GPU local AI development platform, built for teams running large language model fine-tuning, multi-GPU inference serving, or parallel model experimentation that a single-GPU workstation can’t handle. 192GB of combined GPU memory is the headline capability. It is enough to load very large models, split a large model across both cards, or run two mid-sized models simultaneously for comparison work.

It also suits data science and scientific computing teams running GPU-accelerated simulation and analytics at scale, and organisations that need dual-GPU compute with enterprise-grade reliability. That means ECC system memory, enterprise NVMe storage, and dual 10GbE networking for fast dataset transfers from shared storage.

Where does the Xeon 696X Dual RTX PRO 6000 AI Workstation sit in the range?

Below it in the range, the Dual AMD EPYC CPU Pedestal Server is a pedestal server built around dual AMD EPYC 9684X processors and 768GB of DDR5 ECC memory, aimed at CPU-heavy workloads. The 2U Multi-GPU Server with 2 NVidia RTX PRO 6000 96GB GPUs offers the same number of GPUs in a rack-mounted 2U chassis, with an AMD EPYC 9355 32-core CPU and 192GB of memory.

Above it, the High End Quad RTX PRO 6000 Blackwell Rendering / AI Workstation steps up to four RTX PRO 6000 cards on an AMD Ryzen Threadripper Pro platform. The 2U Multi-GPU Server with 4 NVidia RTX PRO 6000 96GB GPUs does the same in a rack with an AMD EPYC 9655P 96-core CPU. This workstation is the choice for teams who want two 96GB GPUs on a 64-core Intel Xeon 696X and 384GB of DDR5 ECC memory in a desk-side tower.

Frequently Asked Questions

What software does the Xeon 696X Dual RTX PRO 6000 AI Workstation support?

The system ships with Ubuntu Linux 26.04 LTS and supports CUDA, cuDNN and the NVIDIA Container Toolkit, along with multi-GPU frameworks such as PyTorch (FSDP/DeepSpeed), TensorFlow, vLLM and TensorRT-LLM. The 64-core Xeon 696X and dual RTX PRO 6000 GPUs are equally capable running professional simulation, rendering and data science software that supports Linux.

Why two GPUs instead of one larger one?

Each RTX PRO 6000 Blackwell carries 96GB of memory, so two cards give 192GB combined in a single desk-side system. Two GPUs also enable multi-GPU distributed training and parallel inference serving that a single card, however powerful, can’t do on its own.

How does this compare to the 2U Multi-GPU Server with 2 RTX PRO 6000 GPUs?

Both carry two RTX PRO 6000 Blackwell cards, but this one is a desk-side tower on an Intel Xeon 696X with 64 cores and 384GB of DDR5 ECC memory. The 2U Multi-GPU Server is a rack-mounted system with an AMD EPYC 9355 32-core CPU and 192GB of memory. Choose the tower if the machine sits at a desk or in an office, and the 2U server if it is going into a rack.

Is there room to expand this system later?

Yes. Three of the five PCIe 5.0 x16 slots are free after the two graphics cards, and the four SATA 6Gb/s ports are unused, with case mounts for 10 x 3.5″ and 11 x 2.5″ drives. All four M.2 slots are used in the default build, and all eight memory slots are filled, so a memory upgrade means replacing modules. The 2500W power supply has been specified with that headroom in mind.

Who is this PC best suited for?

The Xeon 696X Dual RTX PRO 6000 AI Workstation suits AI engineering teams, research groups and organisations running multi-GPU model work who need that capacity on their own hardware rather than in the cloud. It is a step up from a single-GPU system, not a starting point.

Can I customise this system?

Yes. Use our configurator to adjust storage capacity, RAM configuration or operating system as needed. The processor, GPUs, cooling and power supply have been matched carefully to this platform’s power and thermal requirements, so we’d recommend speaking to our team on 0151 317 9860 before changing those specifically.