GPU SuperComputer

Technical characteristics

NVIDIA hardware and software systems

NVIDIA BigData DGX Supercomputers with NVIDIA Tesla V100 32G Graphics Cards
Use of NVIDIA KVM virtualization to create cloud servers with required number of video cards.
Use of NVIDIA's Ubuntu-based DGX OS operating system on hosts and virtual machines
NVIDIA Docker containerization system is pre-installed on each virtual machine
Full support of optimized GPU–accelerated NVIDIA GPU Cloud application container catalog

Amazing performance

Productivity up to 1000 TERAFLOPS (for deep learning tasks)
40.960 CUDA cores to speed up video card performance
5120 Tensor Cores to accelerate matrix operations that lie at core of Artificial Intelligence
Up to 10 times faster than connecting video cards through PCIe interface because of service bus NVidia NVLink with capacity of 300 Gbit/s
Up to 4 times faster training of deep learning algorithms compared to other GPU-based systems

Creating a platform for deep learning goes far beyond than just a choice of a provider and graphics processors. The desire to introduce artificial intelligence in work of enterprise implies careful selection and integration of a complex set of software and hardware hardware. GPU SuperComputer accelerates this process, providing a ready-made solution that allows to process data and get results within the shortest possible time.

Technical characteristics

NVIDIA hardware and software systems

NVIDIA BigData DGX Supercomputers with NVIDIA Tesla V100 32G Graphics Cards
Use of NVIDIA KVM virtualization to create cloud servers with required number of video cards.
Use of NVIDIA's Ubuntu-based DGX OS operating system on hosts and virtual machines
NVIDIA Docker containerization system is pre-installed on each virtual machine
Full support of optimized GPU–accelerated NVIDIA GPU Cloud application container catalog

Amazing performance

Productivity up to 1000 TERAFLOPS (for deep learning tasks)
40.960 CUDA cores to speed up video card performance
5120 Tensor Cores to accelerate matrix operations that lie at core of Artificial Intelligence
Up to 10 times faster than connecting video cards through PCIe interface because of service bus NVidia NVLink with capacity of 300 Gbit/s
Up to 4 times faster training of deep learning algorithms compared to other GPU-based systems

Configurations and pricing

1 x Tesla V100 32GB

10 cores
61GB RAM
250GB SSD

1 999 rubles / day

49 000 rubles / month

2 x Tesla V100 32GB

20 cores
122GB RAM
500GB SSD

3 999 rubles / day

99 000 rubles / month

4 x Tesla V100 32GB

40 cores
215GB RAM
1000GB SSD

5 999 rubles / day

199 000 rubles / month

8 x Tesla V100 32GB

80 cores
490GB RAM
2000GB SSD

9 999 rubles / day

259 000 rubles / month

There are many GPU-accelerated solutions on the market today. But only NVIDIA BigData DGX-based SuperComputer GPU unlocks the full potential of the most advanced NVIDIA Tesla V100 accelerators and uses the new NVIDIA NVLink technology and the Tensor core architecture. The service provides 4-fold acceleration of deep learning algorithms training compared to other GPU-based systems. This is possible through the use of NVIDIA GPU Cloud catalog that provides access to optimized versions of currently most popular frameworks.

Main supported technologies

High performance computing Deep learning and machine learning Visualization Inference Smart cities Infrastructure Medicine

HOOMD-blue

RELION

PIConGPU

QMCPACK

Parabricks

NAMD

MILC

Microvolution

LAMMPS

GROMACS

GAMESS

CHROMA

BigDFT

PGI Compilers

Lattice Microbes

CANDLE

MATLAB

VMD

RAPIDS

TensorRT Inference Server

PyTorch

TensorRT

TensorFlow

DIGITS

MXNet

NVCaffe

Kaldi

Deep Cognition Studio

Dotscience Runner

MapR PACC

Cuda

Cuda GL

MATLAB

Torch

Theano

Microsoft Cognitive Toolkit

Caffe2

Kinetica

H2O Driverless AI

Chainer

PaddlePaddle

OmniSci [MapD]

VMD

ParaView IndeX

Clara Render Server

ParaView

ParaView OptiX

ParaView Holodeck

IndeX

Smart Parking Detection

DeepStream

CT Organ Segmentation AI

Cuda

Cuda GL

TensorRT

TensorRT Inference Server

Kaldi

Transfer Learning Toolkit

Smart Parking Detection

DeepStream

Dotscience Runner

MapR PACC

Cuda

Cuda GL

DCGM Exporter

Device Plugin

CT Organ Segmentation AI

Clara Render Server

HOOMD-blue

RELION

PIConGPU

QMCPACK

Parabricks

NAMD

MILC

Microvolution

LAMMPS

GROMACS

GAMESS

CHROMA

BigDFT

PGI Compilers

Lattice Microbes

CANDLE

MATLAB

VMD

RAPIDS

TensorRT Inference Server

PyTorch

TensorRT

TensorFlow

DIGITS

MXNet

NVCaffe

Kaldi

Deep Cognition Studio

Dotscience Runner

MapR PACC

Cuda

Cuda GL

MATLAB

Torch

Theano

Microsoft Cognitive Toolkit

Caffe2

Kinetica

H2O Driverless AI

Chainer

PaddlePaddle

OmniSci [MapD]

VMD

ParaView IndeX

Clara Render Server

ParaView

ParaView OptiX

ParaView Holodeck

IndeX

Smart Parking Detection

DeepStream

CT Organ Segmentation AI

Cuda

Cuda GL

TensorRT

TensorRT Inference Server

Kaldi

Transfer Learning Toolkit

Smart Parking Detection

DeepStream

Dotscience Runner

MapR PACC

Cuda

Cuda GL

DCGM Exporter

Device Plugin

CT Organ Segmentation AI

Clara Render Server

Use NVIDIA and MTS experience in the field of deep learning for your project and there will be no need to spend extra time and money to get the needed results. Spend less time on setup and optimization and do more research.

Make innovations happen faster

High-performance training speeds up your productivity meaning that you can spend less time on launch of your solution in production use.

NVIDIA DGX-1 Delivers 140X Faster Deep Learning Training

Technical characteristics

NVIDIA hardware and software systems

NVIDIA BigData DGX Supercomputers with NVIDIA Tesla V100 32G Graphics Cards
Use of NVIDIA KVM virtualization to create cloud servers with required number of video cards.
Use of NVIDIA's Ubuntu-based DGX OS operating system on hosts and virtual machines
NVIDIA Docker containerization system is pre-installed on each virtual machine
Full support of optimized GPU–accelerated NVIDIA GPU Cloud application container catalog

Amazing performance

Productivity up to 1000 TERAFLOPS (for deep learning tasks)
40.960 CUDA cores to speed up video card performance
5120 Tensor Cores to accelerate matrix operations that lie at core of Artificial Intelligence
Up to 10 times faster than connecting video cards through PCIe interface because of service bus NVidia NVLink with capacity of 300 Gbit/s
Up to 4 times faster training of deep learning algorithms compared to other GPU-based systems

Customer feedback

Nowadays artificial intelligence technologies open up space for a variety of areas, ranging from finance, telecom, manufacturing, and ending with healthcare and education. Thanks to #СloudMTS, Russian companies and researchers will be able to use the NVIDIA DGX platform which is a powerful tool for developing neural networks.

Sergey Shatov

So far we are enjoying our experience. Instant access to latest version of Docker. We already played around with machine learning, pattern recognition and other typical tasks. The hardware works, there were no problems with software, the network did not fail. The pricing is reasonable. So far so good.

Ivan Muratov