I'm back to basics after an incredible #devopsdays, Singapore edition. Loved the city, the conference, the people.
I had a talk on #k0s and it was an Instant hit.
Meta's LLaMA-3 paper https://t.co/3MorAhOsRD: 16,384-GPU cluster, one job interruption every 3 hours over 54 days. 58.7% GPU, 17.2% HBM. Hardware reliability is now quantified, means everything above the hardware in AI cluster ops is the unbounded problem.
#NVIDIA#LLM#GPUs
NVIDIA's AI Cluster Runtime conformance suite is a catalog of Kubernetes Jobs that assert behavior on a live cluster. If youcan write a job that asserts behavior, you can be your own conformance authority. That model unbundles certification fromthe vendor.
#nvidia#ai#kubernetes
@khushbooverma I don't think it will come to survival. There's no going back to the era of learning languages and syntax. There is active work done both on the hardware and software to optimize token spending. I'm working with AI infra & I know first hand that optimizations are on top priority
Please don't make itineraries to the dot when travelling. It's fine to miss a few spots.
The word 'travel' itself comes from Old English 'travailen', which means to toil or to labor, & it was quite true in ancient times, when there was 50-50 chances that you'd return home.
Software development has never been about writing code.
This was obvious to many before, but now AI has made it crystal clear to everyone.
Building software is 99% thinking, 1% writing code, and 1% fighting off-by-1 errors.
OpenSource gives all the ingredients to build your custom business solution when you've the right building blocks. OSS is great for everyone & I've one such example to showcase the same.
Building FinOps With k0rdent Open Source https://t.co/nst7KBEKaB via @containerblog