Active/passive HA is a common approach to mitigate server downtime. However, when constructing back-up nodes, you’ll have to count the GPUs in as well. The problem is that these expensive GPUs in the back-up machines will be idled most of the time.
Given disaggregated GPU chassis, you can assign GPUs to your active nodes, and simply reassign them to back-up nodes when the original machine is down. This way, the number of GPUs needed in HA systems is halved without sacrificing reliability of your systems.
Not only that your GPU idle rate will be almost zero percent, it is also more convenient in terms of replacement. It is quicker to replace one GPU than switch to another server. With a disaggregated chassis, you can prepare all your redundant GPU resource in one place. It only takes seconds to assign redundant GPUs to your hosts, mitigating any performance drop and potential server downtime due to GPU failure.
If you want to apply for any product display, please write a form and we will contact you after receiving the message.