Any idea why Intel revenue does get a boost by gaudi2? Performance is within H100 as per this article https://www.databricks.com/blog/llm-training-and-inference-intel-gaudi2-ai-accelerators Trying to understand what Nvidia is doing different? #hardware #ai #semiconductor
Marketing is very far from reality. FYI google has to add nvidia gpu in cloud when customer start churning. Nobody want TPU as it has bad performance.
Absolutely fake news. Google cloud sold out all their TPUs… source: I plan their capacity. Dang… these people just talk shit without knowing what they’re saying
Nvidia ceo is damn scared from google right now then. Sell nvidia fast
CUDA
Gaudi per this article is second best, and the first one H100 is better by >2x, and easily maintains performance at large scale of thousands of chips.
There's also a big diff between technical marketing engineers doing testing and user experience for everyone else.
Customers are used to nvida and locked in. to have them move away from nvidia performance is only the first step. Other things to consider are 1. Sustained performance on multiple workloads 2. Ease of use 3. Continuous support in the next few years 4. Clear roadmap I think 2 is the biggest challenge. If a company has to hire 100 extra engineers to move their workloads to gaudi, there is zero chance anyone will risk the move Finally if the first product which is good, and there is no history of goodness and no clear roadmap Why would anyone jump only to come back the next year or so. 2 is an engineers marketing role 4. Roadmap is a leadership problem
The dev ecosystem is a big challenge for anyone not named Nvidia
Nvidia is making buckets of money, where intel is not
I know. Trying to understand why Gaudi 2 isn’t a hot product where it has performed so well
Intel is trying to catch few drops from NVIDIA's bucket that overflow.