A lot of industries rely heavily on machine learning nowadays. You can find AI driven application in fields such as weather forecast, industrial automation, financial analysis…etc. One key component that backs it up is the HPC system behind the scene, and the more powerful your HPC system is the easier it is to achieve your goals.
As time goes on, more and more data are being processed, any system will meet the point where you must scale it up or out, in order to handle the workloads that are getting heavier and heavier.
H3 Platform knows how difficult it is, so we are providing solutions that help you to easily scale up or scale out your systems when it comes.
The GPUs are extracted from the compute nodes and pooled with our GPU expansion chassis, then the chassis is connected to any CPU hosts. We made the GPU devices software-defined so that users can assign/reassign any number of GPUs to any connected host with lines of code rather than physically configure the machines.
The easiest way to scale up your system is to increase the number of GPUs. H3 Falcon GPU expansion chassis can carry up to 16 GPUs. Say good-bye to system migration and bare-metal server configurations. With H3 Platform solutions, simply connect the chassis to your compute nodes and you will be able to boost up the computing performance for your applications painlessly
It is also easier to upgrade your GPUs. As the GPU devices are pooled together, you can manage all device upgrades in one place rather than go to every single compute node.
What if you decide to scale out your system? H3 solutions still got you. The unique port bifurcation design allows multiple host connections to one single GPU expansion chassis. With the GPU dynamic provisioning function, you would not have to worry about how many GPUs you should have for your new compute nodes. Connect any CPU servers to Falcon GPU expansion chassis and assign GPUs to them according to performance requirements.
H3’s GPU solutions uses PCIe cables for interconnections so you wouldn’t have to modify much BIOS setting. Different from Ethernet and InfiniBand, the data would not go through another layer of encoding and decoding when transmitted to another node via PCIe interface, and this ensures higher bandwidth and lower latency at the same time.
If you want to apply for any product display, please write a form and we will contact you after receiving the message.