NVIDIA authorization for DGX-1
Despite the very hectic end of the year, we were able to meet all the authorization conditions of NVIDIA for the area of machine learning and those for its flagship DGX-1, which boasts the designation of “the fastest way to machine learning”. We are thus building on our partnership with NVIDIA in the area Accelerated computing (GPGPU accelerators for the area of HPC in particular).
Machine learning is in times of rapid development, and NVIDIA is one of the world’s leaders in the area of hardware, and now, also software for accelerated machine learning. The dedicated NVIDIA DGX-1 server and the powerful workstation NVIDIA DGX Station are the result of its work. Both machines are based on NVIDIA’s most powerful iron and a fine-tuned software environment that includes a pre-installed and fine-tuned machine learning environment. And it is these two technological jewels that can be purchased in the Czech and Slovak Republics thanks to our company.
Acceleration of the learning phase against the traditional CPU computing environment and GPU environment vs. NVIDIA DGX-1.
Hardware DGX-1
Let’s take a more detailed look at the NVIDA machines, first in terms of hardware.
Parametr | DGX-1 | DGX Station |
---|---|---|
GPUs | 8× Tesla V100 | 4× Tesla V100 |
Výkon (GPU FP16) | 1 PetaFLOPS | 0,5 PetaFLOPS |
GPU paměť | 128 GB celkem | 64 GB celkem |
CPU | 2× E5-2698 v4 2.2GHz (20 jader) | E5-2698 v4 2.2GHz (20 jader) |
NVIDIA CUDA cores | 40 960 | 20 480 |
NVIDIA Tensor cores | 5 120 | 2 560 |
RAM | 512 GB | 256 GB |
HDD | 4× 1,92TB SSD | 4× 1,92TB SSD |
Network | Dual 10GbE, 4× 100Gb EDR Infiniband | Dual 10GbE |
Maximální příkon | 3 200 W | 1 500 W |
Both NVIDIA DGX-1 and NVIDIA DGX Station are equipped with the latest and fastest accelerators in existence – NVIDA Tesla V100 (NVLink version) – DGX Station with four cards, DGX-1 with as much as eight cards!
Software package DGX-1
What is a lot more interesting though is the already-mentioned software package of the offered NVIDIA machines. Both offer the same machine learning environments (e.g. Caffe, Cafe 2, Theano, TensorFlow, Torch, or MXNet) that are pre-installed and, more importantly, performance-tuned and an intuitive environment for data analysts (NVIDIA Digits). All this is elegantly packaged in Docker containers. According to NVIDIA, an environment that is tuned thus has a 30 % higher performance for applications in the area of machine learning as compared to applications deployment only on NVIDIA hardware as such. The main advantage of the pre-installed environment is its deployment speed, which is a matter of hours.
If you would like more information or are interested in trying and testing the latest NVIDIA technology, please do not hesitate to contact us