ITglobal.com: Physical Server Rental with GPU

The main articles are:

• Servers (Global Market)
• Servers (Russian market)

2025

Power increase by 3 times

ITGLOBAL.COM actively develops the GPU infrastructure DPCs Russia in and, Kazakhstan placing on these facilities projects related to: artificial intelligence the deployment of large language models (), LLM infantry and other AI services. This was announced by ITglobal.com on December 22, 2025.

To implement such projects, ITGLOBAL.COM provides GPU power in two formats: in the cloud and on dedicated servers. More here.

Renting Nvidia RTX PRO 6000 Blackwell Server Edition Graphics Accelerators

ITGLOBAL.COM (ITG International Cloud) has expanded its line of GPU servers with NVIDIA RTX PRO 6000 Server Edition graphics accelerators based on the Blackwell architecture. GPUs are available for lease both as part of a virtualized infrastructure and on dedicated physical servers.

The RTX PRO 6000, released in 2025, uses the Blackwell architecture (2024) and is equipped with 96GB of GDDR7 video memory with error correction support. The graphics accelerator contains 24,064 CUDA cores, 752 tensor cores and 188 RT cores, which provides performance up to 117 TFLOPS in single-precision calculations. The memory bandwidth is 1.6 TB/s on the PCIe Gen 5 interface.

A key feature of the RTX PRO 6000 is support for Multi-Instance GPU (MIG) technology with the ability to divide into up to 4 virtual instances. Compared to previously used solutions, this accelerator surpasses the A100 in terms of video memory (96 GB versus 80 GB), FP32 performance (117 TFLOPS versus 19.5 TFLOPS) and clock speed (2617 MHz versus 1410 MHz). Compared to L40S, the RTX PRO 6000 has twice as much video memory (96 GB versus 48 GB), almost twice the memory bandwidth (1.6 TB/s versus 864 GB/s) and additionally supports MIG.

The implementation of the RTX PRO 6000 Blackwell Server Edition highlights our commitment to giving customers access to NVIDIA solutions. The increased amount of video memory allows you to download larger AI models without allocating to multiple GPUs, and MIG support provides flexible resource utilization in the cloud. The graphics accelerator is a versatile solution that can handle a wide range of tasks and loads. In particular, it is suitable for use in VDI environments, where it was previously required to acquire specialized GPUs, - said Evgeny Svidersky, director of the ITGLOBAL.COM cloud business, ITG Corporation.

The RTX PRO 6000 complements the existing line of graphics accelerators without replacing current solutions. Customers can choose the optimal configuration depending on the requirements of the project.

Providing power based on Nvidia H200

ITGLOBAL.COM began providing computing power based on Nvidia H200 graphics accelerators. Access to AI technology will allow developers, researchers and companies to build competitive AI solutions without a multi-million dollar investment in their own infrastructure. The company announced this on July 21, 2025.

NVIDIA H200 - a graphics accelerator that overcame the bar of 100 GB of memory on a single chip - systems are equipped with 141 GB of HBM3E memory with a bandwidth of 4.8 TB/s. This is almost one and a half times higher than the previous generation of H100 and provides acceleration of infection and training of AI models up to two times.

Such characteristics open up opportunities for the development of more complex language models, computer vision systems and scientific simulations that can find application in medicine, education, industry and other industries. H200 is suitable for the tasks of generative AI, training large language models and high-performance computing in scientific and engineering projects, where the amount of memory has previously become a limiting factor in the development of technologies.

The systems support the SuperPOD architecture, allowing multiple HGX H200 nodes to be combined using high-speed InfiniBand networking solutions. Each computing node can have up to eight NVIDIA H200 accelerators, the total amount of HBM3e memory reaches 1128 GB (141 GB per accelerator). This allows you to scale resources to work with large AI models and large datasets, providing flexibility for large enterprises and research centers.

{{quote 'The amount of memory in 141 GB on one chip radically changes the approaches to AI development. Now models that used to require distribution between multiple GPUs can be completely housed in the memory of a single accelerator. This opens up opportunities to create ambitious solutions in medicine, industry, science and other fields, allowing the use of the same tools and capacities as the world's leading laboratories. At the same time, the purchase of equipment of this level remains a serious barrier for startups and research teams due to the high cost.

Service providers solve this problem by offering access to advanced computing resources according to the rental model. This approach allows organizations to work with technologies without capital investments and focus on their own developments without being distracted by infrastructure tasks, "comments Evgeny Svidersky, ITGLOBAL.COM Cloud Business Director, ITG Corporation.}}

Leased dedicated servers with GPU

ITGLOBAL.COM has launched a physical server rental service with graphics accelerators. The service allows you to use GPU servers both separately and in combination with the company's cloud infrastructure. This makes it possible to build hybrid IT solutions. The company announced this on May 26, 2025.

Previously, ITGLOBAL.COM provided the lease of graphics accelerators exclusively as part of the cloud infrastructure. However, the market has established a trend towards the use of allocated resources with GPUs due to the desire to optimize costs by eliminating virtualization and reducing administrative costs, as well as due to the specific requirements of high-load projects, where exclusive use of hardware resources and maximum performance are critical.

In practice, many companies combine different consumption patterns depending on the features of the project. Cloud GPUs are especially in demand for temporary projects, pilot launches and active development stages, where the load is difficult to predict, and flexible scaling and payment for use allows you to optimize costs. Dedicated GPU servers, in turn, are used for stable, long-term tasks with predictable load, where performance, full control over the infrastructure and compliance with security requirements are important. Leased dedicated servers with GPUs from ITGLOBAL.COM can be integrated into the client's cloud infrastructure, providing seamless operation of different environments within a single computing loop.

The launch of the physical server rental service with GPU is a continuation of our strategy to develop a complete set of infrastructure solutions based on graphics accelerators. In addition, this is a timely response to the growing trend towards hybrid IT infrastructures, where cloud and allocated resources are combined. Thanks to a wide selection of configurations and various models for providing resources, we can accurately adapt the infrastructure to specific client tasks, - commented Evgeny Svidersky, Director of the ITGLOBAL.COM Cloud Business, ITG Corporation.

For smaller companies and startups, ITGLOBAL.COM offers compact configurations with two graphics cards. For tasks that require high performance, servers are available with the ability to install up to eight GPUs combined using NVLink technology. All equipment is located in Tier III data centers located in nine countries. Graphics accelerators are presented in several main series from Nvidia, which allows you to choose a configuration for specific requirements.