|
|
|
|
|
by dwrodri
1504 days ago
|
|
The idea of setting up my own cluster seems really cool, but it always seems that I'm paying more money for less compute vs. just buying a single more powerful computer. I suppose this is worth it for people who'd like to physically own a cluster where they can test their cluster code with a real network layer before deploying that code to a data center? The MSRP for Nvidia's Jetson Xavier NX 16GB module is USD500. So 4 Xaviers + board is USD2200 for a combined 64GB of LPDDR4 memory and 84TOPs of Compute. On the other hand, I can buy an Nvidia RTX3080 with 10GB of memory on the card and 238 INT8 TOPS for ~USD1000, 64GB of DDR4 for USD250 and the last USD1000 on whatever other parts I need to get something working. The only counterexample I can think of is that the cluster could potentially be easier to upgrade/purchase in pieces, which would be more ideal if you're broke and would like to get started on a budget. And of course, there are probably some use cases where the topology of the Jetson Cluster leverages its hardware more effectively than the single GPU. Can someone help this make sense for me? |
|
> On the other hand, I can buy an Nvidia RTX3080
Good luck with that in the UK. I know someone who ordered on launch date, still hasn't got one, and has been told there's no more coming from that brand...
[edit: to clarify that I have a TPi 1]