
A Speed That Would Make Even Hermione Jealous
Right gather 'round because I've got news that's almost as exciting as finding a stash of Chocolate Frogs! Amazon Web Services or AWS as they're calling it is upgrading its Graviton4 chip and it's supposed to have network bandwidth of 600 gigabits per second. Apparently some bloke named Ali Saidi at AWS reckons that's like a machine reading 100 music CDs every second. Crikey that's faster than Ron eating a treacle tart! I can only imagine what Hermione would do with that type of power!
Annapurna Labs: Not Your Average Potions Class
This Graviton4 chip which is a central processing unit (CPU) comes from Amazon's Annapurna Labs in Austin Texas. They're really coming into their own aren't they? It's all part of Amazon's plan to make their own custom stuff. It's putting them up against some of the big boys like Intel and AMD which is a bold move even for someone who's faced down Voldemort more than once.
Project Rainier: The AI Supercomputer That Could
But here's the real kicker: Amazon is getting into the artificial intelligence game and they're taking on Nvidia. AWS even announced Project Rainier at their re:Invent 2024 conference last December. It's an AI supercomputer built for a startup called Anthropic and Amazon's thrown a whopping $8 billion into backing them! It's a bit like Dumbledore backing me against all the odds eh?
Nvidia's GPUs Facing Some Serious Competition
Gadi Hutt someone high up at AWS said they're trying to cut down AI training costs and give people an alternative to Nvidia's expensive graphics processing units or GPUs. Apparently Anthropic's Claude Opus 4 AI model launched on Trainium2 GPUs and Project Rainier is powered by over half a million of these chips. Hutt even mentioned that while Nvidia's Blackwell chip is better than Trainium2 the AWS chip is cheaper to use. 'Cheaper' is always a good word especially if it means being able to buy more Weasley's Wizard Wheezes products.
Trainium3: Faster Than a Firebolt
And hold onto your broomsticks folks because Trainium3 is coming out this year and it's twice as good as Trainium2 and uses 50% less energy. That's a bit like improving a Firebolt broomstick to be even faster and more efficient – a dream come true! This chap Rami Sinno director of engineering at AWS' Annapurna Labs even said they can barely keep up with the demand for these things.
Amazon's Grand Plan: A New Era in AI
So with Graviton4 getting an upgrade and Project Rainier's Trainium chips Amazon is really trying to control the whole AI shebang from the networking to the training. And since big AI models like Claude 4 can train on non Nvidia stuff the big question is how much of the market Amazon can steal away. An AWS spokesperson said we'll find out the release schedule for Graviton4 by the end of June. I reckon things are about to get very interesting – almost as interesting as a game of Quidditch against Slytherin!
Comments
- No comments yet. Become a member to post your comments.