Hey Pau, when is it good to roll your own GPU vs just using existing? Maybe some specific case that absolutely will give one an edge? Just asking. Nothing more. I have attended your classes.
The driving factor behind this choice is "How much work do you plan to do with this GPU in the next year?"
For individuals like you and me I am almost sure the best option is to rent a GPU on demand. You pay-as-you-go and you can also optimize your costs by picking the GPU you want to spin. For example, I pick an A4000 to develop at low cost, and then to run at full speed I pick an H100.
For companies, the best option is to rent GPUs, add them to their Kubernetes platform, build their solutions, put them on production and monitor impact and cost. If the thing works, maybe it makes sense to reduce long-term costs by buying some of the GPUs.
Hey Pau, when is it good to roll your own GPU vs just using existing? Maybe some specific case that absolutely will give one an edge? Just asking. Nothing more. I have attended your classes.
The driving factor behind this choice is "How much work do you plan to do with this GPU in the next year?"
For individuals like you and me I am almost sure the best option is to rent a GPU on demand. You pay-as-you-go and you can also optimize your costs by picking the GPU you want to spin. For example, I pick an A4000 to develop at low cost, and then to run at full speed I pick an H100.
For companies, the best option is to rent GPUs, add them to their Kubernetes platform, build their solutions, put them on production and monitor impact and cost. If the thing works, maybe it makes sense to reduce long-term costs by buying some of the GPUs.