@AlizTheHax0r Ahh yes, I'm quite lucky, we have a cluster of beefy GPUs in house we can use when needed, in fact the power used by the GPUs is used to heat the campus!! Our hot water supply comes from heat exchanging our datacentre!
We got favourable results on Infiniband 40, so worth a trial
@AlizTheHax0r Another question - what size models are you loading? If you're looking for higher throughput you might find the spark bottlenecks as it uses shared ram - so you can load a massive model, but actually pairing it with an RTX 5090 or similar may be faster depending on size
@AlizTheHax0r Infiniband is really good for distributed models,it depends on your use case, for some things I find Infiniband is enough, for others NVLink is the only way to avoid the bottleneck.
@AlizTheHax0r Learning Styles is also good if you're teaching/mentoring as different people learn in different ways.
https://t.co/gqMNgpF6MY
- so yeah, both of those are my main "managers learning" things the rest - yeah, experience.
Also worth looking into generational differences too!
@AlizTheHax0r How someone may want information delivered to them, or how to help them/influence them.
Really good for that "why can't I connect with this person" 2/n
@AlizTheHax0r You need two numbers - how many cubic feet per minute (cfm) and at what pressure ? Most likely forp low flow rate an airbrush compressor with small cylinder would be fine. The noise isn't much more than your laptop fan at full speed and would only run when required