Deep Learning Appliance

DEEP Gadget provides the best deep learning platform powered by seven water-cooled GPUs, ready-to-use software stack, and huge virtual GPU memory for DNNs.


Train Faster, Spend Less, and Stay Focused.

Incredible Computing Power from 7 GPUs

7 GPUs in a single DEEP Gadget machine provides up to 70 TFLOPS single-precision peak performance. A closed-circuit water cooling system keeps high-density GPUs at a low temperature. DEEP Gadget works quietly even at full load and can even be placed in your office.

Cost-Effective Gaming GPUs with Enterprise-Level Reliability

Water cooling ensures the long lifetime and reliability of gaming GPUs as with expensive high-end GPUs. Users can choose the appropriate GPU option based on their demands, applications, and budgets.

Ready-To-Use Deep Learning Software Stack and Virtual GPU Memory for DNNs

DEEP Gadget is delivered with pre-installed software stacks including an operating system and deep learning frameworks, such as Caffe and TensorFlow. In addition, the VMDNN library develped by ManyCoreSoft is bundled with DEEP Gadget. Users do not need to worry about the small size of GPU memory because VMDNN virtually expands the memory space on demand during the training period of deep neural networks.

Water Cooling Solution

DEEP Gadget is created by ManyCoreSoft, who developed the first water-cooled GPU supercomputer in the world.

Supercomputer Chundoong Emerging Technologies in SC15
Supercomputer Chundoong of Seoul National University

ManyCoreSoft designed and built Chundoong, the first water-cooled GPU supercomputer in the TOP500 list, in October 2012.

Emerging Technology in SC15

The liquid cooling solution of ManyCoreSoft is accepted as one of 10 Emerging Technologies in SC15.

CPU 1x or 2x Intel Xeon E5-2600 v4 series CPUs

One of the following:

  • 7x NVIDIA Tesla P100 PCIe 16 GB
  • 7x NVIDIA Tesla P100 PCIe 12 GB
  • 7x NVIDIA Tesla P40
  • 7x NVIDIA GeForce GTX Titan Xp
  • 7x NVIDIA GeForce GTX 1080 Ti
  • 7x AMD Radeon RX Vega 64
  • 7x NVIDIA Tesla V100 for PCIe (coming soon)
  • 7x AMD Radeon Instinct MI25 (coming soon)
PCIe 2x PCIe 3.0 x16, 5x PCIe 3.0 x8
Main Memory 64 GB, 128 GB, or 256 GB DDR4 2400 MHz ECC Reg
Primary Sotrage 250 GB M.2 NVMe SSD or higher
Secondary Storage

One of the following:

  • 4 TB SATA3 HDD
  • 8 TB SATA3 HDD
  • 16 TB RAID 6 HDD storage (6x 4 TB SATA3 HDD, Software RAID 6 4+2)
  • 32 TB RAID 6 HDD storage (6x 8 TB SATA3 HDD, Software RAID 6 4+2)
Motherboard ASUS Z10PE-D8 WS or ASUS X99-E-10G WS
Power Supply 2x 1200 W or 2x 1300 W (80 PLUS Platinum)
Operating System Ubuntu 16.04 LTS
  • For NVIDIA GPUs: CUDA Toolkit, cuBLAS, cuDNN, NCCL, NCCL 2, VMDNN Library
  • For AMD GPUs: ROCm, MIOpen
Deep Learning Frameworks
  • For NVIDIA GPUs: Caffe, TensorFlow, Caffe2, Torch, PyTorch, Theano, MXNet, Cognitive Toolkit
  • For AMD GPUs: hipCaffe, hipTensorFlow

DEEP Gadget will be released in December 2017.