WHAT IS GUSTAV ENTERPRISE LOW ENERGY CLOUD?
The Gustav enterprise low energy cloud is based on a power-saving architecture, focused on machine learning. Under the word machine learning, you often read the term AI, artificially intelligent. We don’t really like this term AI alone at REBOTNIX, because it doesn’t say what we understand by artificial intelligence, nor what products or services the customer can use to solve their problems or projects. Our solution consists of our own hardware, which is based on the host side on ARM CPU and powerful accelerators of an integrated graphics card, so-called GPU. Now we bring over 10 years of experience to our first cloud. This cloud and the provided software modules focus on the execution of machine learning models with the lowest power consumption.
During the development phase, we concentrated on accommodating as many instances in the smallest possible space. We can fit two modules in a half-inch in the vertical direction. In-depth, we can extend the modules up to 1 meter. So we find much more space in one device than standard servers or supercomputers ever will.
The result is a more efficient use of the rack within the data center. On the low power consumption, we generate less waste heat and therefore less CO2 emissions.
Per CPU and GPU modules we are between 3 – 15-watt power consumption. In the large modules with ARM CPU and GPU, which have up to 32 GB RAM, we consume from 10 – 50 watts. The power can be adjusted in real-time per module by the customer without rebooting. With the focus on artificial intelligence, it is even possible to increase or reduce the neural networks during runtime.
We are based on ARM, GPU and our products has low energy power consumption. We use Ubuntu 18.04 ARM 64 Bit for operating system.
THE RACE "NEURON PER WATT"
The race in artificial intelligence is which company can train and recognize which data at the lowest price. This price is in the CLOUD when it comes to AI, always around the GPU, storage and CPU instances. The truth about training neuronal networks is that many nets like ResNets50 or Resnet 101 and even smaller like mobilenet´s are not able to scale in parallel on all GPU cores due to their sequential architecture of their internal. The side effect is the very high power consumption of several hundred watts per second. Even if the GPU neither trains nor performs a detection, the standard operation with CPU and GPU is over 100 watts. Our GUSTAV cloud is capable of running over 800 ARM cores at 100 watts in standard operation and, depending on the instance, between 30,000 and 51,000 CUDA cores. The prerequisite is that you have booked enough GUSTAV instances of course.
NEW SECURITY LAYERS
Cloud instances, no matter how they are protected by a firewall, are publicly accessible via a domain. The DNS server resolves the IP address to the server. Our cloud instances can access internal public addresses of the booked cloud instance via their own DNS server. This means that we can access a private VPN line through a public DNS entry. Even if this entry is known to the domain, only the client that is in the VPN group can access it. Each client is activated by the administrator via a control interface. This activation can be temporary or permanent over a longer period of time.
The client software is completely open-source and available for Windows, Mac. Certified clients for Android and iOS are also available.
A direct instance is a hybrid model where our Gustav Enterprise instances are operated directly from the secure network of our customers. Secured by our own internal interline, a VPN path, we offer additional security via 4G or 5G modems, which immediately maintain the connection in case of failure. The instance is booked and used directly in a point-to-point connection. Access from our side is not possible. Employees can access the instances in groups or individually and work together.
FIELD OF APPLICATIONS
The areas of application lie in the rapid provision of embedded low energy modules. Immediate testing of the own application or prototypes.
The areas are
– Drive autonomously simulations
– Industrial applications
– Robot simulation
– Smart cities
– Predictive maintenance
– Model testing
GUSTAV ENTERPRISE Module A
ARM A57 Complex
500 GB or 1 TB SSD Storage
32 GB eMMC 5.1
8 GB MEMORY (SHARED)
NVIDIA PASCAL TX2 – 256 CUDA CORES
GUSTAV ENTERPRISE Module B
8 x ARM-v8.2-64-Bit-CPU-Kerne, 8 MB L2 + 4 MB L3
500 GB or 1 TB SSD Storage
32 GB eMMC 5.1
32 GB MEMORY (SHARED)
Volta graphics processor with 512 computing units and Tensor computing units.
GUSTAV ENTERPRISE Module C
THIS MODULE IS BASED ON NVIDIA JETSON NANO XAVIER
FOR WHO IS THIS PRODUCT SUITABLE?
The offer is currently aimed exclusively at industrial customers. For private customers, there will be a service offered at the end of the year.
WHERE ARE THE SERVER LOCATIONS?
We have currently opened two locations. One in Germany with on a Tier4 and the second in Ireland, both Europe. The locations can be tested for industrial customers now.