Price of the H100s is gonna start dropping
this post was submitted on 15 Nov 2023
1 points (100.0% liked)
Hardware
47 readers
1 users here now
A place for quality hardware news, reviews, and intelligent discussion.
founded 1 year ago
MODERATORS
Amen. Sooner or later. What's worse, people will think "Nvidia will just sell to others" but if Microsoft has an advantage, buying more H100s to compete will just dig you into a capex hole.
Do you really think NVIDIA just found out that MS was doing their own chips?
Can you tell us a bit more about the process? How much of it is under NDA?
Considering what was announced, all the other things related to the process will be under nda
That's very likely, but it doesn't hurt to ask
Trust. We’re very proud of what we made! So I hope the future excites you just as much
I can't believe it took this long. Google is releasing their fifth generation of their TPUs.
They started on it in 2014 or 9 years ago.
Because they have been using FPGAs until now.
What are the practical uses of AI for ordinary customers for us to be excited about?
Showing you more advertisements
What you already see: conversation bots, art generation and manipulation...
But I think most of that power will go towards language models, and in a few years it will be standard to talk to computers using natural language. However, the other functionality will be included in this, like the computer being able to illustrate what you tell it, teach you things, create works of art (songs, pictures, videos) for you, ... But for a start it will be mainly talk.
At work GitHub co-pilot usually makes decent suggestions for boilerplate code like if I need a new endpoint for our API, or retrieve an object from the repository, or just to fill out the constructor on a new class or object instantiation of an existing class. It’s not always perfect so you have to have the experience to spot the issues but it gets you close real fast.
The most impressive honestly was I had made retrieval logic in one format but it turns out I needed a different format so I comment that whole code block and got ready to redo it and co-pilot suggested the same thing in the new format so I just had to hit tab and delete the commented code after.
Then there’s chatGPT for bouncing ideas off of like if I should use a factory, also our front end framework is both unpopular and poorly documented so if I have a question there wasn’t really a place to turn somehow chatGPT does a decent job of clearing things up, I simply wouldn’t find that many people talking about it on stack overflow.
At home I’ve been having fun prompting images on Dalle-3. Considering making the plunge to comfyUI and a local stable diffusion.
This is the greatest gift by TSMC. They have made it their life mission that designing chips should not be some black magic and any company should be able to do it. All the work they have put in over the last 2 decades, working with EDA vendors, universities, companies. Making pdks realistic, getting their silicon to hit the promised targets, enabling IP vendors to quickly iterate and get to GDS as quickly as possible.
This was enabled by entire mobile ecosystem ARM. Open source arch + TSMC’s design ecosystem made designing chips to be a relative cake walk.
Look at Tesla, they don’t design most of their IP. They focus on compute IP, system. All the others, they buy it off the shelf from IP vendors. And they don’t have to worry because IP vendors knows their IP will work on the process already.
This era enabled companies to focus on what’s important to them, like compute and not on IP building.
Intel robbed the industry of progress by pretending chip design needs a 20K+ people, IP teams, multiple BUs, architects.
40 years of incredible computing progress leads to unprecedented technology usage
Intel robbed the industry of progress by pretending chip design needs a 20K+ people, IP teams, multiple BUs, architects.
wat?
Yeah such a great comment overall but with that stinker at the end out of nowhere lol
The comment was silly all around. It's just emotional nonsense by people, who are adding weird drama to a field they don't understand.
LOL. You really think TSMC invented the for hire fab model?
This has been a thing since the 80s mate ;-)
Something else Microsoft can trail behind in.
bragging on reddit?? how sad...🤣🤣
Awesome work! Custom AI chips are definitely the future. Keep pushing the envelope.