Jedox GPU Accelerator requires CUDA-capable NVIDIA Tesla™ GPUs with Compute Capability 3.0 or higher. If you are unsure about the compute capability of your NVIDIA device, please check the following website and/or contact your graphics card vendor:
The minimum recommended GPUs are devices of the NVIDIA Tesla™ K-series (code-named “Kepler”, NVIDIA Tesla™ K80). We recommend Tesla™ V100 (32GB) or P100 (16GB). Current mainboards offer up to 8 PCIe slots and can thus house up to 8 GPUs. If you use more than one GPU, all cards must be of the same type.
Non-Tesla™ GPUs are supported for testing purposes (viable support); thus, any CUDA-capable NVIDIA GPU with Compute Capability 3.0 or higher can be used. Note that non-Tesla™ GPUs have a driver-side CUDA kernel execution timeout (normally 5 seconds) that can be reached in extremely computation-intensive operations, such as aggregations or dimension filters. This timeout might be configurable, but Jedox does not provide any support for these types of configuration changes.
Purchasing GPU hardware
While it is possible to buy GPUs and other server components separately, Jedox strongly recommends that customers buy pre-configured GPU hardware from a specialized dealer to avoid incompatibilities or hardware instability. If you would like to start with one GPU but keep your upgrade options open, we recommend telling this to your hardware dealer, who can ensure you will have an appropriate mainboard (with multiple PCIe slots) and an adequate power supply. We recommend connecting the graphics card to undivided 16-bit PCI slots. We also recommend connecting all graphics cards on a single PCIe bus controller.