[ad_1]
go through Hong Kong Economic Times August 29, 2024
originalPublished in Hong Kong Economic Times Financial NewsEJ Tech Innovation Lab“
On Tuesday (27th), the US company Cerebras Systems launched what it claims to be the world’s fastest AI solution.Brain ReasoningIn terms of model output speed, Llama 3.1-8B is 1800 tokens per second, while Llama 3.1-70B is 450 tokens per second.AllegedlyIt is 20 times faster than Nvidia’s GPU-based hyperscale cloud computing and is challenging the latter’s market leadership with “high-speed inference”.
Cerebras costs less and consumes less power than H100
Cerebras Inference is supported by four data centers in the United States.Third GenerationWafer-level engineWSE-3The cost and power consumption are only 20% of Nvidia H100. The new solution is open to any logged-in user for free through the application programming interface (API); the version for developers, taking the Llama 3.1-8B model as an example, charges 100 million tokens per month.10 US cents (approximately HK$0.78); The larger Llama-3.1 70B model charges 60 US cents (approximately HK$4.68) per million tokens.
The InformationNewsthe US company OpenAI is preparing to launch a new AI product with advanced features. The model is called “Strawberry” or Q* (pronounced Q-Star) internally to cope with problems and tasks that current AI models cannot solve. OpenAI CEO Sam AltmanPosted on social media X earlier this monthA photo of a strawberry potted plant has sparked speculation among netizens about the secret behind it.
Support EJ Tech
If you want to submit articles, report information, issue press releases or interview notices,Click here to contact us.
//
[ad_2]
Source link
