GPU or a Graphics Processing Unit is a specific type of processors able to work on large data volumes. They are mostly used for the AI processes, gaming and graphics rendering, while most users know them as video cards.
Contrary to CPU, a central processor used for basic operations and performing computing operations one-by-one in a row, GPU is used whenever there is a need for simultaneous computing performed all at once. Thanks to that, processes requiring a high-load can be faster.
To accelerate them even more, one can order resources in the GPU Cloud, hosted on servers with several powerful GPUs. GigaCloud uses NVIDIA T4 and NVIDIA A40, created specifically for virtualization. One such T4 has 16 GB GDDR6 memory, while A40 has 48 GB. In the cloud, the processor data is virtualized, and its capacities are divided between users, with every one getting processing power in a virtualized mode, vGPU.
Cloud video cards advantages when compared to physical ones:
- Ability to quickly scale up, allowing more resources in an instant;
- Saving costs on maintenance, as there is no need to buy and install elements on your own;
- Availability of resources from any device.
Below, we describe three scenarios of what tasks a GPU-enhanced cloud can solve. Unfortunately, due to NDAs with other clients, we can’t disclose some company names, so we will simply make a summary of the basis sense of their cooperation with our cloud.
Artificial intelligence projects
Earlier, GPUs were mostly used for graphics, but with the pace of artificial intelligence, their use has broadened to general-purpose computing on graphics processing units (GPGPU), including AI training. Since 2007, a software has been made, which specifically catered for the combination of the two. And the cloud technologies development made GPU for neural networks even more accessible.
For example, UniTalk uses public cloud with GPU by GigaCloud for their projects with artificial intelligence. Namely, when creating their voice robot to automate clients communication. In just a few minutes, the system can process up to several thousands of calls, having a wholesome conversation with the customers instead of managers.
The other company, which is a GigaCloud client, uses cloud capacity to adjust its generative Stable Diffusion AI model to its tasks. Having migrated everything it needs for learning into the cloud, including a vast data set, they have automated the image creation for their corporate style characters. This solution has allowed for time-saving on drawing each image for the website in detail, product packaging and advertising, as well as for decreasing the project budget due to having no need to purchase and maintain the expensive equipment.
Video and graphics processing
The most popular operations with GPU — editing visuals, from 3D images to SFX. Namely, all GigaCloud About YouTube channel videos are edited in Adobe Premiere Pro in the public cloud.
One of our clients, a video production studio, edit videos in DaVinci Resolve and Adobe Premiere Pro. The PCs usually ran this software very slowly and always lagged due to the system overload. However, after moving them to GigaCloud, the company got a chance to render video without failure and also work on the same project simultaneously, connecting to one virtual remote desktop. It accelerated the editing of one 30-minute video, making it 2 weeks instead of a month.
Data analytics
Big Data operations, creation of models, calculations, and simulations is impossible without video cards, and it is relevant for any field. For example:
- Retail
One of the retailers we cooperate with as a cloud provider is using Apache Spark for researching customer moods, detecting sales performance, providing personal offers for loyal buyers etc. GPU, cloud and AI together allowed for a 5 times faster processing of these purchases.
- Finances
Thanks to a powerful processor, it’s possible to more efficiently monitor risks for system protection or prevent errors. This is particularly important in the financial sector, where suspicious transactions need to be quickly identified among large data sets. By processing vast amounts of customer data, financial institutions can train neural networks to detect transaction fraud, and some banks hosted in our infrastructure are already utilizing this.
- Scientific research
Among GigaCloud’s clients, there are educational institutions that host scientific programs and models in our GPU cloud. One of them uses the ArcGIS Pro geographic information system for geodesic research — with the help of a processor, it’s possible to create detailed map images and make them interactive. Another client uses the International Futures tool to calculate scenarios in international relations. In the cloud, it calculates economic, sociopolitical, and geographical indicators faster and provides forecasts for the future.
- Medicine
Processing patient data, which requires deep analysis and quick screening, also demands significant resources. One of our clients, a medical center, hosts FotoFinder software in GigaCloud, which creates mole maps. Some of the program’s functions are performed by artificial intelligence, which gradually learns to detect abnormal changes in skin pigmentation faster and more accurately. The software requires numerous widescreen images, so a graphics card is especially useful here. Another clinic uses a Graphics Processing Unit to examine dental data, create 3D models of the oral cavity, and simulate treatment scenarios.
