What am I talking about when I say “juiced” GPUs? Great question!
How about the ability to grab multiple GPUs, combine their vRAM and present them to a single endpoint?
What if the NVidia H100 and A100 video cards remain sold out for months, but your business still needs the same (or better) performance that those cards can provide ASAP? What if better performance costs you less in the long run, and offers the flexibility to be used on demand by multiple sources?
Luckily, all those options are fully available to you! Juice Labs just completed their latest feature release which does exactly that. Combine your GPU resources and offer them to a single device (or multiple, the choice is yours).
Quick example of instant ROI:
– Purchase 5x Nvidia A6000 video cards (receive in a week, no lead time delays)
– Purchase Juice Labs licensing
– Install software on servers and endpoints
– Provide 240GB of combined Video Memory
– Spend the same amount a single H100 would have cost you, get an extra 35K CUDA cores and 160GB vMem right away
– Have the flexibility to move and/or share these resources around as other projects may need them
If you have never heard of Juice Labs, check out my previous blog here
Or their official site here
- Quick recap:
- Use any GPU currently in stock (lead times are not usually an issue for lower tier cards)
- Combine resources from multiple GPUs and provide them to a single endpoint
- Reduce cost by leveraging cheaper GPU models, while providing increased (combined) performance
- Reduce complexity by accessing these GPUs over the network (less hardware/cost)
- Leverage GPUs from anywhere (Datacenter, cloud, Edge, home, etc)
- Save cost, offer better performance and more flexibility!
