2025-11-12
I often wonder about the gpu depreciation and what happens to those A100s and V100s that are still just fine, but nobody wants them. You're not going to hear about the gpu depreciation math from a big CSP, but thankfully Jon @HotAisle has shared the math and the reality check.
Bloomberg
Michael Burry's warning that hyperscalers are underestimating AI chip depreciation shows that tech investors may have forgotten how much depreciation can hurt
Costing tens of thousands of dollars each, Nvidia Corp.'s pioneering AI chips make up a hefty chunk of the $400 billion …
2025-10-11
If you're trying to choose which gpus to use for your inference workloads @SemiAnalysis_ started to publish frequently updated throughput and latency performance benchmarks for high end NVIDIA and AMD gpus https://inferencemax.semianalysis.com/ - open source code repo https://github.com/... - [image]
SemiAnalysis
SemiAnalysis launches InferenceMAX, an open-source benchmark that automatically tracks LLM inference performance across AI models and frameworks every night
vendor-neutral suite runs nightly and tracks performance changes over time Tae Kim / Barron's Online : Nvidia Touts Software Advantage in Beating Rivals Like AMD Dion Harris / NVID...
2024-10-06
A must read for anybody who uses AI clouds and wants to understand how the different components work together, the costs, and a lot more: https://www.semianalysis.com/ ... It'd be extremely practical if you need to shop for a cloud solution to understand the many underlying costs, which
SemiAnalysis
A deep dive into “AI Neoclouds”, a new breed of cloud compute providers built to offer GPU rentals, like Crusoe, Lambda Labs, and CoreWeave, and their economics
We are thrilled to be recognized as the “gold standard” in the latest SemiAnalysis report on the AI cloud landscape. …