Hyperstack

In an era where data-driven innovation is reshaping industries, the choice of computing infrastructure has never been more critical. As businesses and researchers push the boundaries of what's possible with artificial intelligence, rendering, and high-performance computing, access to powerful graphics processing units (GPUs) has become a competitive necessity. Hyperstack stands at the forefront of this transformation, offering a European-based GPU-as-a-Service platform that combines exceptional performance with sustainable energy practices. This article examines Hyperstack's comprehensive suite of services, from its green energy-powered data centers to its scalable GPU offerings, revealing how the company is redefining cloud computing for the modern era.

Overview of Hyperstack's Services and Infrastructure

Hyperstack operates on 100% renewable energy, with its data centers powered by hydroelectric sources (NexGen Cloud Launches, NVIDIA Inception program collaboration document). This commitment to sustainable energy aligns with the company's goal of addressing cloud carbon footprint and waste, which contributes to 3.7% of global carbon emissions (NVIDIA Inception program collaboration document).

The company's infrastructure demonstrates exceptional performance efficiency through its use of NVIDIA GPUs and optimized architecture. Each GPU is 26 times more energy-efficient than standard CPUs, and the company's server equipment operates at 20x greater efficiency than traditional computing approaches (NVIDIA Inception program collaboration document).

For virtual machine deployment, Hyperstack offers a range of options optimized for different computing needs. Entry-level VMs support basic GPU acceleration, while higher-tier options like the NVIDIA A100 and H100 series deliver leading performance for AI, rendering, and HPC workloads (Hyperstack documentation). Each GPU configuration includes optimized networking support, with speeds up to 10 Gbps and support for SR-IOV technology to reduce latency (Network Types Offered by Hyperstack document).

The platform supports multiple GPU models including A100, L40, and RTX A6000, with detailed specifications for each variant. These include configuration options for 48GB to 80GB VRAM, support for 28 to 31 processing cores per GPU, and memory configurations up to 58GB per GPU (GPU Models and Specifications document).

GPU offerings and specifications

The company provides several NVIDIA GPU models including A100, L40, and RTX A6000, offering configurations with 48GB to 80GB of VRAM and various processing capabilities. Each GPU configuration includes optimized networking support, with speeds up to 10 Gbps and support for SR-IOV technology to reduce latency.

Technical specifications vary between configurations:

• NVIDIA H100 models offer 80GB of VRAM across different forms factors, supporting 24 to 31 processing cores per GPU and up to 180 to 240GB of RAM per GPU, depending on the model variant. Pricing ranges from $1.33 to $2.40 per hour.

• The company's A100 family includes 80GB VRAM configurations supporting 28 to 31 processing cores, with 120 to 240GB of RAM per GPU. Pricing for these configurations ranges from $0.95 to $1.40 per hour.

• The L40 model provides 48GB of VRAM, supporting 28 processing cores and 58GB of RAM per GPU, available for $0.70 to $1.00 per hour.

• RTX A6000/A40 configurations offer 48GB of VRAM, supporting 28 processing cores and 58GB of RAM per GPU, priced at $0.35 to $0.50 per hour.

The company's GPU variants include various networking capabilities:

• NVIDIA H100 models support networking speeds up to 1.8 Tb/s per GPU, with options for direct NVLink connections and standard PCIe networking. These models are available in both 80GB and 40GB variants with different core counts and RAM configurations.

• A100 models support similar networking capabilities, offering 1.8 Tb/s per GPU with options for NVLink and PCIe networking, available in 80GB and 40GB variants.

• L40 models support 10 Gbps networking through standard PCIe connections, providing reliable and affordable connectivity for virtual machines.

Performance and Networking

Hyperstack's networking solutions support up to 10 Gbps transfer speeds with options for SR-IOV technology to further reduce latency. Their VM configurations enable up to 31 pCPUs per GPU and provide scalable memory configurations.

The company's virtual machines support multiple network types, including standard Ethernet for reliable and affordable connectivity, and high-performance Ethernet with SR-IOV technology for reduced latency and improved network performance. The networking solutions are designed to meet various workload demands, with the ability to support up to 1.8 Tb/s per GPU through optional NVLink connections.

Each GPU configuration supports fine-grained network control and direct hardware resource assignment to individual virtual machines, reducing CPU cycles required for processing network packets and improving overall system efficiency. The company's NVIDIA GPU variants include both standard PCIe networking and high-performance Ethernet options with SR-IOV technology, enabling users to select the optimal configuration for their specific workloads.

Hyperstack's network solutions are available in multiple regions, including the CANADA-1 region for L40, A100, and H100 PCIe GPU virtual machines. The company's infrastructure enables up to 10 Gbps transfer speeds across all supported VM configurations, with the option to scale memory and processing resources independently for optimal performance.

Pricing and Usage Model

Pricing for Hyperstack's GPU-accelerated virtual machines (VMs) begins at $0.35 per hour for the entry-level RTX A6000/A40 configuration, while high-end H100-based VMs cost up to $2.40 per hour. The company's billing system tracks usage on a minute-by-minute basis, allowing users to pay only for the time their VMs are active (NVIDIA H100, A100, L40 pricing document).

The cost structure varies significantly by GPU model, with the lowest-cost option being the RTX A6000/A40 configuration, priced at $0.35 per hour. Mid-range options include the NVIDIA A100 series, available at $0.95 to $1.40 per hour, while the high-end NVIDIA H100 series ranges from $1.33 to $2.40 per hour (NVIDIA H100, A100, L40 pricing document).

Hyperstack offers multiple configurations for each GPU model, allowing customers to select the optimal balance of performance and cost for their workloads. For example, the company provides both standard PCIe and NVLink configurations for the H100 series, with pricing adjustments of $0.55 to $0.95 per hour depending on the selected variant (NVIDIA H100, A100, L40 pricing document).

The company's pricing model also includes cost efficiencies through its architecture and infrastructure. By operating at 75% lower cost than traditional cloud providers and using data centers powered by 100% renewable energy, Hyperstack is able to offer competitive rates while maintaining sustainable operations (NVIDIA Inception program collaboration document).

Applications and Support

Hyperstack's platform supports a variety of applications, including AI, rendering, machine learning, and data analytics, through its comprehensive API and enterprise-grade features. The company's solution enables users to deploy any workload in the cloud on their latest infrastructure, with the option to pay only for the resources consumed (NVIDIA Inception program collaboration document).

The platform's technical capabilities include direct-to-compute, GPU-accelerated cloud access through managed Kubernetes, which automates software deployment, scaling, and management. Users can create GPU-accelerated cloud environments with pre-configured or customized virtual machines in minutes (NVIDIA Inception program collaboration document).

The company's ecosystem includes extensive support for AI development, with capabilities for both model training and inference. The platform enables users to train models on high-performance GPUs while focusing on their core research, with the infrastructure managed by Hyperstack (NVIDIA Inception program collaboration document).

Rendering workloads can also be efficiently handled through Hyperstack's platform, which provides purpose-built solutions for render-based workloads using NVIDIA's RTX family. The company's infrastructure enables real-time rendering capabilities for 3D visuals, game development, photorealism, and film production (NVIDIA Inception program collaboration document).

For machine learning applications, the platform offers scalable GPU resources that enable users to handle complex workloads without resource constraints. The company's solutions support both local and distributed training approaches, allowing users to scale their operations as needed (NVIDIA Inception program collaboration document).

The data analytics capabilities of Hyperstack's platform leverage its GPU-accelerated infrastructure to provide significant performance improvements over CPU-based alternatives. Users can take advantage of the platform's scalable memory and processing resources to handle large-scale data processing tasks efficiently (NVIDIA Inception program collaboration document).

Hyperstack's technical support team is available to assist users with deployment and operations, providing expertise in managing the company's GPU-accelerated virtual machines and related resources (NVIDIA Inception program collaboration document). The service operates across multiple regions, with specific GPU variants available in different data center locations to optimize performance for various workloads (NVIDIA Inception program collaboration document).