The V100 Volta Architecture remains particularly effective for training large language models and processing complex datasets that demand reliable, high-throughput parallel performance. These servers are available at considerably lower costs than at the big cloud providers for equivalent performance.
The NVIDIA Volta V100 is uniquely equipped with NVSwitch technology, linking all 16 GPUs in a non-blocking, all-to-all fabric. This design eliminates communication bottlenecks that can limit newer systems, enabling unified GPU memory access and ultra-fast data exchange. It’s ideal for training large-scale models or executing massive parallel simulations that demand seamless interconnect efficiency.
The V100 represents the most refined generation of NVIDIA’s CUDA, cuDNN, and NCCL stack—optimized and battle-tested over years of deployment. Many scientific, engineering, and enterprise workloads (TensorFlow 1.x, legacy PyTorch, Fortran/CUDA kernels, etc.) are tuned specifically for the Volta architecture, often delivering greater stability and reproducibility without the need for code modifications.
| g2.large | g2.medium | g2.xlarge | |
|---|---|---|---|
| Processor: | 2 x Intel Xeon 8168 | 2 x Intel Xeon E5-2680v4 | 2 x Intel Xeon 8174 |
| vCores: | 96 | 56 | 96 |
| Processor Speed: | 2.7GHz | 3.3GHz | 3.1GHz |
| Hourly1: | $11.1990 | $6.6290 | $26.9520 |
| Monthly2: | $4,905.16 | $2,903.50 | $11,804.98 |
| SLA: | 99.5% | 99.5% | 99.5% |
| Memory: | 512GB | 512GB | 1536GB |
| Network Speed: | 100Gbps | 25Gbps | 100Gbps |
| Data Transfer3: | $0.0025/GiB | $0.0025/GiB | $0.0025/GiB |
| Data Transfer Included: | 1TiB/month | 1TiB/month | 1TiB/month |
| Local Storage: | 6x 2TB SATA SSD | 1x 1TB SATA SSD | 2x 1TB NVMe, 2x 4TB NVMe |
| Attached Storage: | 2x 2TB NVMe | ||
| GPU: | 8x Nvidia V100 (32GB) | 8x Nvidia V100 (16GB) | 16x Nvidia V100 (32GB) |
| GPU Memory: | 256GB | 128GB | 512GB |
| Cuda Cores: | 40,960 | 40,960 | 81,920 |
| Tensor Cores: | 5,120 | 5,120 | 10,240 |
Using the Cato Metal Console you can start, stop, and configure things like server ports and networks, as well as accessing the server’s console.
Subject to availability, additional servers can be deployed within 4 hours during business hours, with emergency deployment available 24/7 for enterprise customers.
Our operating system offering varies, and is shown when you select a server for deployment. Currently it is limited to Linux family systems.
Our V100-based servers excel at deep learning, machine learning, high-performance computing, and advanced data analytics workloads. Learn more in The V100 Advantage.
