Cart
Molecule Air Configure
Renderboxes Molecule Air Workstation

When
nothing else
is enough.

Molecule Air, Uncompromising compute, designed to run 24/7.

Molecule Air

For relentless industrial muscle

Molecule Air handles the most demanding VFX, virtual production and AI work, where uptime and throughput matter most. Dual AMD EPYC CPUs and up to 3TB of ECC memory support massive parallel workloads, heavy simulations and large datasets. Its cooling sustains output under constant load. That keeps performance steady across long renders and training runs. Studios use it to power Omniverse, virtual production stages and 24/7 departments.

Configure Help Me Choose Explore RBOS
CPU icon
2
CPUs
RAM (Slots) icon
24
RAM (Slots)
PCIe slots (16x) icon
9
PCIe slots (16x)
NVMe M.2 Drives (Gen 4) icon
2
NVMe M.2 Drives (Gen 4)
U.2 Drives icon
4
U.2 Drives
2
Front drive bay
Interior view of Renderboxes Molecule Air Workstation highlighting GPU

8-GPU Gen5 Expansion

Up to eight dual-slot blower GPUs on our custom PCIe Gen5 MCIO daughterboard. Every GPU runs at full Gen5 x16.

Interior view of Renderboxes Molecule Air Workstation highlighting airflow

Engineered airflow

Front and top fan arrays direct airflow for efficient, low-noise cooling under sustained load.

Eight GPUs at full Gen5 bandwidth, beside your desk, without a server room.

Eight GPUs at full Gen5 bandwidth, beside your desk, without a server room.
8 GPUs on dedicated Gen5 x16 links
768 GB VRAM across eight GPUs1
56.4 GB/s GPU to GPU, each way
0 errors PCIe errors under sustained load
  1. 1With eight NVIDIA RTX PRO 6000 Blackwell Max-Q GPUs, 96 GB each. Total VRAM depends on the GPUs specified.

Enterprise-grade PCIe connectivity, across eight GPUs.

The Renderboxes Gen5 daughterboard gives all eight GPUs a full PCIe Gen5 x16 link, wired directly to the motherboard through server-standard interconnects.

Eight GPUs on the Renderboxes Gen5 PCIe daughterboard, each on its own x16 link
Enterprise-standard MCIO interconnect to all eight GPUs.
A dedicated PCIe Gen5 x16 link to every GPU, direct to the motherboard.
Decoupled GPU architecture, with a dedicated GPU cooling zone.
Chassis designed and manufactured in-house, from the GPU out.

Overview

Front view of the Molecule Air Workstation
View gallery

Overview

  • Dimensions (W × H × D): 71 × 50.6 × 50 cm
  • Weight: 50kg (80kg in shipping crate)

Technical

  • CPU icon
    Dual AMD EPYC 9005 Turin Series
  • GPU icon
    NVIDIA RTX PRO (Multi GPU Max 8x)
  • Storage icon
    NVMe Gen 4/5 M.2 / U.2 (Max 4x)
  • RAM icon
    DDR5 ECC (Max 3.0TB)
  • Cooling System icon
    Air/AIO
Configure your Molecule Air Help Me Choose

Primary use

The Molecule Air is designed to deliver unparalleled GPU compute power in a rugged and highly serviceable 8-way multi-GPU form factor. Initially designed to be the ultimate virtual production rig, the Air is incredibly versatile and built to handle non-stop render action. With advanced cooling technology, including up to 384 CPU Cores via dual AMD EPYC 9005 technology with dual closed loop liquid cooling, the Molecule Air keeps CPU temperatures under control, enabling the system to operate at maximum performance without any thermal throttling. Whether you are a 3D artist, game developer, or a team in need of extreme GPU power, the Molecule Air is the ultimate choice for your high-performance computing needs.
Molecule Air
Molecule Air
Front view of the Molecule Air Workstation
Front view of the Molecule Air Workstation
Rear view of the Molecule Air Workstation
Top down view of the Molecule Air Workstation with cover
Top down view of the Molecule Air Workstation with fans
Top down view of the Molecule Air Workstation
Covered side view of the Molecule Air Workstation
Open side view of the Molecule Air Workstation
Corner view of the Molecule Air Workstation

CPU

  • AMD EPYC 9375F 32 Core (2xCPU 3.8GHz, 4.8GHz Turbo, 256MB, 320W)
  • AMD EPYC 9575F 64 Core (2xCPU 3.3GHz, 5.0GHz Turbo, 256MB, 400W)
  • AMD EPYC 9755 128 Core (2xCPU 2.7GHz, 4.1GHz Turbo, 512MB, 500W)
  • AMD EPYC 9965 192 Core (2xCPU 2.25GHz, 3.7GHz Turbo, 384MB, 500W)

GPU

  • 8x Full Length PCIe GPU (maximum 2 slot wide, 400 W)

Memory

  • 24x DDR5 ECC RDIMM Slots for maximum of 3.0TB

Storage

  • 4x x4 U.2 Gen 4.0
  • 2x x4 M.2 Gen 3.0

System Cooling

  • 18x Noctua Industrial 120mm
  • 5x Noctua Cromax 80mm
  • 2x 360mm All in one liquid cooler CPU

Connectivity

  • 4x USB 3.2 Gen 2 Type-A ports
  • 2x USB 3.2 Gen 1 Type-A ports (front, 5Gbps)
  • 2x 1GbE LAN
  • 1x 1GbE IPMI Management LAN
  • VGA port

Dimensions + Weight

  • W × H × D: 71 × 50.6 × 50 cm
  • Weight: 50kg (80kg in shipping crate)

Power Supply

  • 3x 2000 watts Platinum ATX

Warranty

  • 3 years on all parts
  • 1 year return to base on all labour
  • 3 years technical support via remote/email/telephone
Front view of the Photon Workstation, with a mesh grill and handle on top.

Photon
  • AMD Ryzen 9000 Series
  • NVIDIA RTX PRO / AMD Radeon AI PRO (Max 1x)
  • NVMe Gen 5/4 M.2 (Max 2x)
  • DDR5 (Max 256GB)
  • Air/AIO
BUILD YOUR WORKSTATION
Renderboxes Atom Workstation

Atom
  • AMD Threadripper / Threadripper PRO 9000 Series
  • NVIDIA RTX (Multi GPU Max 4x)
  • NVMe Gen 5 M.2 (Max 3x) / U.2 (Max 2x)
  • DDR5 ECC (Max 2.0TB)
  • Air/AIO
BUILD YOUR WORKSTATION
Front view of the Molecule Air Workstation

Molecule Air
  • Dual AMD EPYC 9005 Turin Series
  • NVIDIA RTX PRO (Multi GPU Max 8x)
  • NVMe Gen 4/5 M.2 / U.2 (Max 4x)
  • DDR5 ECC (Max 3.0TB)
  • Air/AIO
BUILD YOUR WORKSTATION
Renderboxes Electron Workstation

Electron
  • AMD Ryzen 9000 Series
  • NVIDIA RTX (Max 1x)
  • NVMe Gen 5 M.2 (Max 2x) / U.2 (Max 1x)
  • DDR5 (Max 256GB)
  • Air/AIO
BUILD YOUR WORKSTATION
Front view of Renderboxes Nano Pro Workstation

Nano Pro
  • AMD Threadripper PRO 9000 Series
  • NVIDIA RTX (Multi GPU Max 4x)
  • NVMe Gen 4/5 M.2 / U.2 (Max 2x Hotswap)
  • DDR5 ECC (Max 2.0TB)
  • Air/AIO
BUILD YOUR WORKSTATION
Front view of Renderboxes Atom Rack Workstation

Atom Rack
  • AMD Threadripper / Threadripper PRO 9000 Series
  • NVIDIA RTX (Multi GPU Max 4x)
  • NVMe Gen 5 M.2 / U.2 (Max 2x)
  • DDR5 ECC (Max 2.0TB)
  • Air/AIO
BUILD YOUR WORKSTATION

Molecule Air is for briefs that once looked impossible. Think virtual production stages running in real time, and AI models trained on terabyte-scale datasets. Think VFX sequences that render overnight, not over weeks. It isn’t a scaled-up workstation. It’s industrial compute designed for creative environments.

Configure

Configure your Molecule Air, or talk to our team about building the right system for the way you work.

CONFIGURE BOOK A CALL

What system is right for you?

Renderboxes Workstations

Answer a few simple questions and we’ll recommend a system that is right for your creative needs

What system is right for you?

Render at the speed of imagination wherever imagination takes you.

Before you buy

As a rule of thumb, a model needs about 2 GB of GPU memory per billion parameters at 16-bit precision, and about half that at 8-bit. With 768 GB of GPU memory1, the Molecule Air holds models of several hundred billion parameters at 8-bit, or serves several 70-billion-parameter models side by side.

We have measured 128 simultaneous chat users on one Molecule Air, at 1,245 tokens/s in total. For a team of 10 to 30, every reply starts almost immediately.

Yes. It runs on three 2000 W power supplies, each on its own standard wall socket, and sits beside a desk at 71 × 50.6 × 50 cm.

Yes, with RBOS. The Molecule Air can run up to eight virtual workstations, each with a dedicated GPU, its own operating system, USB devices and network port. For eight workstations we recommend at least 512 GB of system memory, 64 GB per workstation.

Yes. On Linux, the GPUs exchange data peer to peer, measured at 56.4 GB/s between GPUs in each direction, which is why training throughput reaches 7.9× with eight GPUs.

For a team, usually. You pay for one chassis, CPU platform, memory and power system rather than one per user; GPUs are reallocated rather than left idle; and there is one system to secure, update and support.

It weighs 50 kg and ships in a shipping crate at 80 kg, ready to connect.

Up to 3 TB.

No. Models and data run on the machine, on your own network.

Three years of parts cover, one year of return-to-base labour and three years of technical support.