I was working on my VMware Explore 2024 session about networking and needed a demo. You should come and see me at explore at my session or the Hands On Labs. I went through the “normal” things and got one demo done. My session is about networking so a lot of the demos I can do involve tcpdump or show commands on routers. Now don’t get me wrong that kind of thing makes me excited but for some reason I think perhaps “normal” people don’t really get going in the morning reading pcaps. Maybe something is wrong with me? No must be the rest of the world.
Anyways. AI is the shiny new thing. So of course if I can do something with AI or Machine Learning people will like it right? Well wouldn’t you know it I don’t have any servers in AVS with GPUs. In fact I don’t know of any VMware Cloud (Should that be Broadcom Cloud now??) that has GPUs. But I think all of the cloud providers have GPU IaaS instances. The great part about having GPU instances close to your VMware instances is that you get low latency and high throughput to them.
So what is the point of this post? First remember this is my blog and this is for personal non commercial use. I’m not endorsing any particular cloud or anything. “Kenyon why did you say that??” Well legal said I couldn’t include the demo because it constituted commercial use. So instead you should all go check out Juice Labs. https://www.juicelabs.co. The community version here: https://github.com/Juice-Labs/Juice-Labs. I tried it out by deploying an Azure VM with an NVIDIA GPU in it and followed the instructions to host the Juice Server in it. Then ran the client in an AVS VM. Worked pretty well. With this I was able to run workloads that needed GPU acceleration in AVS. You can even share one GPU across multiple instances. I’m pretty sure you could setup a scale set with a load balancer infront of it and use the metrics from nvidia-smi to scale your GPU compute automatically. Seems like a cool way to keep costs under control while still providing all the GPU compute you need. Another neat trick is that Juice not only supports compute workloads but also graphical ones. It really reminds me of Bitfusion which VMware bought and promptly EOLd. And yes it runs Doom.
vSkeeBall 