Breaking
December 4, 2024

OpenAI spent $80M to $100M training GPT-4; Chinese firm claims it trained its rival AI model for $3 million using just 2,000 GPUs waynewilliams@onmail.com (Wayne Williams) | usagoldmines.com


  • 01.ai trained an AI model for $3 million using 2000 unnamed GPUS
  • “Efficient engineering” allows 01.ai to compete globally, company claims
  • 01.ai reduced inference costs to 10 cents per million tokens

Tech companies in China face a number of challenges due to the American export ban, which restricts access to advanced hardware from US manufacturers.

This includes cutting-edge GPUs from Nvidia, critical for training large-scale AI models, forcing Chinese firms to rely on older or less efficient alternatives, making it difficult to compete globally in the rapidly evolving AI industry.

However, as we’ve seen time and again, these seemingly insurmountable challenges are increasingly being overcome through innovative solutions and Chinese ingenuity. Kai-Fu Lee, founder and CEO of 01.ai, recently revealed that his team successfully trained its high-performing model, Yi-Lightning, with a budget of just $3 million and 2,000 GPUs. In comparison, OpenAI reportedly spent $80-$100 million to train GPT-4 and is rumored to have allocated up to $1 billion for GPT-5.

Making inference fast too

“The thing that shocks my friends in the Silicon Valley is not just our performance, but that we trained the model with only $3 million,” Lee said (via @tsarnick).

“We believe in scaling law, but when you do excellent detailed engineering, it is not the case you have to spend a billion dollars to train a great model. As a company in China, first, we have limited access to GPUs due to the US regulations, and secondly, Chinese companies are not valued what the American companies are. So when we have less money and difficulty to get GPUs, I truly believe that necessity is the mother of invention.”

Lee explained the company’s innovations include reducing computational bottlenecks, developing multi-layer caching, and designing a specialized inference engine. These advancements, he claims, result in more efficient memory usage and optimized training processes.

“When we only have 2,000 GPUs, the team has to figure out how to use it,” Kai-Fu Lee said, without disclosing the type of GPUs used. “I, as the CEO, have to figure out how to prioritize it, and then not only do we have to make training fast, we have to make inference fast… The bottom line is our inference cost is 10 cents per million tokens.”

For context, that’s about 1/30th of the typical rate charged by comparable models, highlighting the efficiency of 01.ai’s approach.

Some people may be skeptical about the claims that you can train an AI model with limited resources and “excellent engineering”, but according to UC Berkeley’s LMSIS, Yi-Lightning is ranked sixth globally in performance, suggesting that however it has done it, 01.ai has indeed found a way to be competitive with a minuscule budget and limited GPU access.

You might also like

​ 

This articles is written by : Nermeen Nabil Khear Abdelmalak

All rights reserved to : USAGOLDMIES . www.usagoldmines.com

You can Enjoy surfing our website categories and read more content in many fields you may like .

Why USAGoldMines ?

USAGoldMines is a comprehensive website offering the latest in financial, crypto, and technical news. With specialized sections for each category, it provides readers with up-to-date market insights, investment trends, and technological advancements, making it a valuable resource for investors and enthusiasts in the fast-paced financial world.

Recent:

The EU proposal to scan all your WhatsApp chats is back on the agenda chiara.castro@futurenet.com (C...
Best gaming headsets 2024: Reviews and buying advice | usagoldmines.com
Apple Watch dominates Strava running devices for 2024, even beating out Garmin stephen.warwick@futur...
Everything announced at AWS re:Invent 2024 you might have missed | usagoldmines.com
Miss the MacBook Pro's Touch Bar? Check Out the Flexbar Juli Clover | usagoldmines.com
Gemini AI might be ready for you to roll the prompt dice erichs211@gmail.com (Eric Hal Schwartz) | u...
NYT Strands today — hints, answers and spangram for Wednesday, December 4 (game #276) | usagoldmine...
Quordle today – hints and answers for Wednesday, December 4 (game #1045) | usagoldmines.com
NYT Connections today — hints and answers for Wednesday, December 4 (game #542) | usagoldmines.com
Marvel SNAP’s Latest Season ‘Rivals’ Now Live, Intros Daughter of Galactus Tim | usagoldmines.com
You Can Now Get the Classic PS1 Boot Screen on Your PS5 Michelle Ehrhardt | usagoldmines.com
Amazon Music Finally Has Its Own Version of Spotify Wrapped Jake Peterson | usagoldmines.com
Fastest VPN 2024: We identify the speediest performers | usagoldmines.com
Intel: ‘We have invested gi-namic resources into addressing’ Arc GPU driver issues | usagoldmines.c...
Eve finally fulfills a promise to Android smart home users | usagoldmines.com
This Tool Lets You Build Your Own Version of Google Justin Pot | usagoldmines.com
Prince of Persia: The Lost Crown Launches on Mac App Store Juli Clover | usagoldmines.com
"It's actually quite difficult to build a really good generative AI application" - Amazon CEO outlin...
“Nightmare” Zipcar outage is a warning against complete app dependency Scharon Harding | usagoldmine...
Splash pads really are fountains of fecal material; CDC reports 10K illnesses Beth Mole | usagoldmin...
Intel’s $249 Arc B580 is the GPU we’ve begged for since the pandemic | usagoldmines.com
This Sonos One Deal Is a Great Way to Upgrade to a Surround-Sound Setup Mark Knapp | usagoldmines.co...
The First Two ‘Warcraft’ Games Are Being Delisted Soon Michelle Ehrhardt | usagoldmines.com
Apple Releases First Firmware Update for Beats Solo Buds Juli Clover | usagoldmines.com
Would you watch a foreign film dubbed with AI to sound like the original actors? erichs211@gmail.com...
In-memory processing using Python promises faster and more efficient computing by skipping the CPU w...
Intel’s second-generation Arc B580 GPU beats Nvidia’s RTX 4060 for $249 Andrew Cunningham | usagoldm...
Wear OS to Get Hotel Key, Campus ID, and Corporate Badge Support Kellen | usagoldmines.com
Apple Uses Amazon's Custom AI Chips for Search Services Juli Clover | usagoldmines.com
Four desk-organizing gifts you don’t technically need but might very much want Kevin Purdy | usagold...
Yale Code keypad lock review: Svelte & affordable, just not smart | usagoldmines.com
Beats Cyber Week Sale Includes Big Savings on Earbuds, Headphones, and Speakers Mitchel Broussard | ...
Apple employee sues company for allegedly spying on personal worker devices benedict.collins@futuren...
"AI requires cold data to be warmer" — energy-efficient Tape-as-a-Service (TaaS) combines the benefi...
The Touch Bar is back, sort of...and it looks terrible lance.ulanoff@futurenet.com (Lance Ulanoff) |...
The best Christmas sales 2024: deals from Amazon, Walmart, Target and more | usagoldmines.com
Amazon unveils surprise new video and image AI models to compete with the best on the market | usag...
T-Mobile has a Free Shake Shack Burger for You Kellen | usagoldmines.com
Pixel Owners Can Now Check Their Device Temperature Tim | usagoldmines.com
Apple Seeds Fourth Beta of visionOS 2.2 to Developers With Ultrawide Mac Virtual Display Juli Clover...
Join us tomorrow for Ars Live: How Asahi Linux ports open software to Apple’s hardware Andrew Cunnin...
A peek inside the restoration of the iconic Notre Dame cathedral Jennifer Ouellette | usagoldmines.c...
Missed Cyber Monday? Get 45% off this hot OLED monitor anyway | usagoldmines.com
Apple just pushed a key smart home feature into 2025 | usagoldmines.com
Your Apple Music Replay 2024 Is Live Jake Peterson | usagoldmines.com
The rumored foldable iPhone could save a stagnant folding phone market, according to a new report ja...
Russian censorship is getting tougher – and Tor needs your help chiara.castro@futurenet.com (Chiara ...
3 new Paramount Plus movies with over 94% on Rotten Tomatoes that you won't want to miss in December...
Corrupted Microsoft Word files used to launch phishing attacks | usagoldmines.com
Nintendo Switch Online will get the NES version of Tetris next week | usagoldmines.com
US plan to protect consumers from data brokers faces dim future under Trump Jon Brodkin | usagoldmin...
The Raspberry Pi 5 now works as a smaller, faster kind of Steam Link Kevin Purdy | usagoldmines.com
A new ‘File Search’ feature is coming to the Windows 11 taskbar | usagoldmines.com
Google smart speakers are starting to sound like Gemini | usagoldmines.com
NZXT accused of ‘predatory scam’ gaming PC rental program | usagoldmines.com
How to solve RAM problems with Windows memory diagnostics | usagoldmines.com
8BitDo’s new extra-green Xbox keyboard gives me 2001 vibes | usagoldmines.com
iPhone SE Now Over 1,000 Days Old as New Model Edges Closer Hartley Charlton | usagoldmines.com
Samsung Cyber Week Sale Has Year's Best Prices on Monitors, TVs, Fridges, and More Mitchel Broussard...
MOVEit breach chaos continues, data on hundreds of thousands leaked from Nokia, Morgan Stanley | us...
Google’s AI podcast creator NotebookLM could be coming to the Gemini app on your phone | usagoldmin...
Cheerios effect inspires novel robot design Jennifer Ouellette | usagoldmines.com
China hits US with ban on critical minerals used in tech manufacturing Ashley Belanger | usagoldmine...
The makers of Arc show off new AI-driven ‘smart browser’ called Dia | usagoldmines.com
Watch Intel talk about Arc Battlemage GPUs on The Full Nerd today! | usagoldmines.com
This Smartwatch and Fitness Tracker for Kids Is 22% Off Right Now Pradershika Sharma | usagoldmines....
Creature Commandos is full of social outcasts and grieving misfits, but the voice actor for Rick Fla...
Code written by OpenAI and praised by GitHub may not be as good as Github says | usagoldmines.com
How businesses can break barriers to entry in integrating AI into operations | usagoldmines.com
Lessons in cybersecurity from the Internet Archive Breaches | usagoldmines.com
Javascript files loaded with RATs hits thousands of victims | usagoldmines.com
New website shows you how much Google AI can learn from your photos Paresh Dave, wired.com | usagold...
Fix your spotty home Wi-Fi signal with this simple $27 gadget | usagoldmines.com
Today’s best laptop deals: Save big on work, school, home use, and gaming | usagoldmines.com
The 4TB Samsung 990 Pro SSD with heatsink just dropped to 40% off | usagoldmines.com
Samsung’s 49-inch 240Hz ultrawide monitor is cheaper now than it was on Black Friday | usagoldmines...
How to Connect Windows or macOS to Your Roku David Nield | usagoldmines.com
Apple Podcasts Reveals 2024 Show of the Year Joe Rossignol | usagoldmines.com
Microsoft plans to make searching in Windows 11 better - I just hope it doesn't screw it up | usago...
Indiana Jones and the Great Circle's official launch trailer showcases new gameplay ahead of release...
Intel announces its new Battlemage graphics cards, and they might just be the 1440p budget champions...
Microsoft’s claim that Arm-based Copilot+ PCs are “fastest, most intelligent Windows PCs” is debunke...
Everything new on Paramount Plus in December 2024 | usagoldmines.com
Linux devices are being hit by LogoFAIL vulnerability, Bootkitty installed | usagoldmines.com
Stop Live Activities Taking Over Your Apple Watch Face Tim Hardwick | usagoldmines.com
Apple Fails to Block $995M UK App Store Commission Lawsuit Tim Hardwick | usagoldmines.com
Apple Raises Indonesia Investment Offer to $1B Amid iPhone Ban Tim Hardwick | usagoldmines.com
3 new movies on Max with over 90% on Rotten Tomatoes | usagoldmines.com
Insta360 Flow 2 Pro spotted on sale, even though the iPhone gimbal hasn’t launched yet | usagoldmin...
AI reckons it can do all jobs, even those thought previously 'safe' | usagoldmines.com
Two decades after Enron’s bankruptcy, the company is back as a crypto firm? Eric Berger | usagoldmin...
Dell G15 review: A ‘retro’ laptop that’s all about performance | usagoldmines.com
Windows Copilot+ PCs aren’t there yet: 8 must-change upgrades for 2025 | usagoldmines.com
Jaguar's striking Type 00 concept is a bold statement of intent, but it needs more to restore its pa...
The iPhone 17 Pro and Pro Max could get a display upgrade and avoid a frame downgrade | usagoldmine...
AI impact is only minor in many workplaces, employees believe | usagoldmines.com
Apple Music Replay beats Spotify Wrapped to the recap punch – here's how to get it | usagoldmines.c...
Got an older iPhone? WhatsApp won’t work on it for much longer alexblake.techradar@gmail.com (Alex B...
AMD RX 8800 XT could match RTX 4080’s performance – and easily outgun Nvidia’s GPU for ray tracing ...
PC Gaming Show: Most Wanted 2024 airs this week, here's how to watch it | usagoldmines.com

Leave a Reply