The Open Telekom Cloud is being extended with the addition of FPGAs (Field Programmable Gate Arrays). This new type of server will make the public cloud fit for new application scenarios that require particularly high performance. The new flavors will be available in November. Developers can apply for a closed beta phase that will start in July.

With the individual cloud turbo via FPGA, applications such as machine learning for image recognition can be accelerated more than tenfold. Data analysis speeds are 50 times faster than regular CPUs. When it comes to genome sequencing, the cloud FPGA can even increase the speed by a factor of 100. That means that a genome analysis that would otherwise take 33 hours could be carried out in less than 20 minutes.

The Open Telekom Cloud is offering a solution from the public cloud to address a fundamental IT problem: processor architectures will soon reach their physical limits. Moore’s Law, which stipulates that the number of transistors within an integrated circuit doubles every 18 to 24 months, was valid for decades. Now the physically feasible seems to be exhausted and the end of this development is approaching because it won’t be possible to make transistors any smaller in the foreseeable future.

Up to now, there have been two options to achieve a further increase: "Either you use several CPUs in parallel, or you use graphics processors," says Max Guhl, Product Manager Open Telekom Cloud. "However, both methods have the disadvantage that they are not equally suitable for every application."

The reason: normal vCPUs of the x86 generation can do everything – but not everything equally well or equally fast. They are all-rounders and therefore highly flexible. But: "They are similar to all-season tires that never work as well on snow as a real winter tire," says Guhl. A specialized processor can perform certain tasks much better, such as a graphics processing unit (GPU). That's why, for example, GPUs are preferred for applications in the area of Artificial Intelligence (AI).

However, even GPUs have their limits, because they are primarily suitable for the parallel calculation of the same processes. Hardware acceleration solves this problem: The architecture of FPGAs is not defined by their hardware, but by software. This makes them incredibly flexible. Furthermore, in contrast to a GPU, the packing density of the processor cores on a volume unit is significantly higher (3,000 for GPU versus 2 million for FPGA).

"With FPGA boards, I can stipulate how they work, in which order they should execute processes and thus define how fast they work. This makes their application scenarios limitless. Whether as a switch, router, AI server or IoT backend," says Max Guhl. "In this way, FPGA boards also relieve the CPU, which can devote itself to other tasks, while FPGAs take over complex, highly specialized processes. Even the bit rates are individually selectable. This means that developers are no longer limited to 8,16 or 24-bit scenarios. For example, 9 bits are no problem with FPGAs."

Acquiring individual FPGAs is costly. In future, they will be available via the Open Telekom Cloud in the established "Pay as you use" subscription model. This means that companies that cannot or do not want to purchase their own FPGAs will soon have access to the innovative technology. They only pay if they actually use the technology.

Interested developers can apply now for the closed beta phase of the FPGA flavors in the Open Telekom Cloud, which ends in February 2019. During this closed beta phase, the FPGA boards are free of charge. The only costs are the servers that the developers use, which they book in the Open Telekom Cloud. Xilinx VU9P cards are available at the start of the offer. The flavor includes eight vCPUs, 88 GB RAM, a Xilinx VU9P and CentOS 7.5 64 bit as the operating system.

Secure your beta access now: opentelekomcloud(at)telekom.de