Gavin Baker+ Your Authors @GavinSBaker Husband, Becky Painter. CIO, Atreides Management. Former PM, Fidelity OTC fund. No investment advice, views all my own. gavin-s-baker.com/ May. 14, 2020 1 min read + Your Authors

1) Is Ampere being 20x better than Volta good or bad for all of the AI silicon startups that are “10x” better than Volta on a single algorithm on one framework? 😀

Especially given that *none* of them except Habana have posted the 10x results on MLPerf.

 https://venturebeat.com/2020/05/14/nvidia-unveils-monstrous-a100-ai-chip-with-54-billion-transistors-and-5-petaflops-of-performance/ 

2) Now the startups have to hope that their next chip at 5 nm can beat Ampere’s 5 nm successor.

Very few of them have any revenue.

Designing and building the next chip will likely cost $100m plus which means they will need to raise another round if they can.

3) Jensen has likely already spent over $1 billion designing the Ampere successor. And he will have much lower costs per wafer than any startup.

4) Jensen can cut price and lower his 70% gross margins on Ampere whenever he wants which would really change the performance per watt per $.

Obviously that would be negative for Nvidia, but it would be a disaster for competitors.

5) Thoughts on why algorithmic diversity is NVDA’s friend while making life much harder for silicon startups.

And why investors should be suspicious of cherry picked “10” results on one algorithm on one framework.

 https://medium.com/datadriveninvestor/longer-term-thoughts-on-nvidia-the-importance-of-algorithmic-diversity-in-ai-and-dlss-in-gaming-ce6b29527f68 

6) The modern semiconductor industry is a brutal, expensive treadmill. Semiconductor VC is *very* different from internet or software VC and much more akin to biotech VC in the required fund raising and attendant dilution.

 https://medium.com/@gavin_baker/thoughts-on-nvidia-ray-tracing-and-modern-semiconductor-realities-436d161b8817 

7) Ampere’s *real* competitors will be Habana, TPU4, the internal chips yet to come from AWS and Cerebras if they can convince a hyperscaler to rewrite their software.

And all of those companies are spending hundreds of million per year designing their next chip.

8) Really curious to see how the TPU4 and the 7 nm Gaudi and Goya’s from Habana fare vs. Ampere in the MLPerf benchmarks.

Those chips are the primary near term risk to Nvidia - both revenue and margins - not the 15-20 startups ex Cerebras.

9) Important to note that Intel almost certainly looked at *every* startup before buying Habana.


You can follow @GavinSBaker.



Bookmark

____
Tip: mention @threader_app on a Twitter thread with the keyword “compile” to get a link to it.

Enjoy Threader? Sign up.

Since you’re here...

... we’re asking visitors like you to make a contribution to support this independent project. In these uncertain times, access to information is vital. Threader gets 1,000,000+ visits a month and our iOS Twitter client was featured as an App of the Day by Apple. Your financial support will help two developers to keep working on this app. Everyone’s contribution, big or small, is so valuable. Support Threader by becoming premium or by donating on PayPal. Thank you.


Follow Threader