Nvidia rivals building a different kind of AI chip
Amazon doubles down on AI startup Anthropic with $4bill
Nvidia rivals building a different kind of chip for AI
World's largest AI chip beats Amazon's fastest in head-to-head comparison
Amazon doubles down on AI startup Anthropic with $4bill
FreshBooks all-in-one accounting
FreshBooks is an easy-to-use billing, payments, and accounting platform that streamlines your bookkeeping and keeps you tax-time ready all year round. The ideal solution for small businesses, FreshBooks has everything you need to grow.
Save 80% on FreshBooks Black Friday Offer
***
Nvidia rivals building a different kind of chip for AI
Nvidia's specialized computer chips have been instrumental in building the latest generation of AI chatbots, establishing the company as a leader in the AI boom. However, the same features that make these graphics processor units (GPUs) ideal for developing powerful AI systems also render them less efficient for deploying AI products.
This inefficiency has created an opening for competitors in the AI chip industry. These rivals aim to challenge Nvidia by offering AI inference chips, which are better suited for the everyday operation of AI tools and help reduce the substantial computing costs associated with generative AI. "These companies see an opportunity for specialized hardware," said Jacob Feldgoise, an analyst at Georgetown University's Center for Security and Emerging Technology. "As the adoption of these models grows, so will the need for compute power for inference, driving demand for inference chips."
Creating an AI chatbot requires significant computing power, starting with a process known as training or pretraining, where AI systems learn from vast amounts of data. GPUs excel at this task due to their ability to perform numerous calculations simultaneously across a network of interconnected devices.
However, once trained, generative AI tools still need chips for tasks like composing documents or generating images. This is where inferencing comes into play. A trained AI model must process new information and make inferences based on its existing knowledge to generate responses. While Nvidia's GPUs can handle this, it is often overkill for the task.
"Training involves much heavier, more intensive work, whereas inferencing is lighter," explained Forrester analyst Alvin Nguyen. This has led startups like Cerebras, Groq, and d-Matrix, along with Nvidia's traditional rivals such as AMD and Intel, to develop chips optimized for inference. Meanwhile, Nvidia continues to focus on meeting the high demand from major tech companies for its advanced hardware....
***
World's largest AI chip beats Amazon's fastest in head-to-head comparison
In a remarkable feat, Cerebras has demonstrated that Meta’s Llama 3.1 405B large language model can run 75 times faster on its cloud AI service than on Amazon Web Services' (AWS) fastest AI service using GPUs. A video from Cerebras showcases AI writing code at a speed 75 times greater than the world's fastest AI GPU cloud.
This achievement was made possible through Cerebras’s cloud AI service, Cerebras Inference, which leverages the company's third-generation Wafer Scale Engines instead of GPUs from Nvidia or AMD. Cerebras has consistently touted its Inference service as the fastest for AI inference tasks. Upon its launch in August, Cerebras Inference was claimed to be approximately 20 times faster than Nvidia GPUs operating through cloud providers like AWS for models such as Llama 3.1 8B and Llama 3.1 70B.
When compared to Nvidia GPUs rented from AWS, Cerebras Inference was reportedly 75 times faster. The Wafer Scale Engine chips outperformed even the fastest Nvidia GPU implementations from Together AI by a factor of 12. Additionally, Cerebras Inference surpassed its closest competitor, AI processor designer SambaNova, by six times.
To illustrate this speed, Cerebras conducted a test where both Fireworks (the fastest AI cloud service equipped with GPUs) and Inference were tasked with creating a chess program in Python. Cerebras Inference completed the task in about three seconds, while Fireworks took 20 seconds....
***
Amazon doubles down on AI startup Anthropic with $4bill
In a strategic move to bolster its position in the generative artificial intelligence (AI) sector, Amazon has injected an additional $4 billion into Anthropic, a competitor to OpenAI. This latest investment doubles Amazon's financial commitment to the startup, renowned for its GenAI chatbot Claude, though Amazon remains a minority investor, as confirmed by Anthropic on Friday.
Amazon has steadily become Anthropic's primary cloud partner, vying with tech giants Microsoft and Alphabet's Google to provide AI-powered tools to its cloud clientele. Amazon Web Services (AWS) is a significant revenue source for Anthropic, distributing its latest AI models. "The investment in Anthropic is essential for Amazon to stay in a leadership position in AI," noted D.A. Davidson analyst Gil Luria.
Anthropic intends to train and deploy its foundational models using Amazon's Trainium and Inferentia chips. The demanding process of training AI models necessitates powerful processors, making the acquisition of high-cost AI chips a critical priority for startups. "This partnership also allows Amazon to promote its AI services, such as leveraging its AI chips for training and inferencing, which Anthropic is using," Luria added. Nvidia currently leads the AI processor market, with Amazon among its hyperscaler customers.
Amazon is advancing its chip development through its Annapurna Labs division, with which Anthropic is "working closely" to develop new processors. Additionally, Amazon is working on its own AI model, code-named "Olympus," which has yet to be released....
Enjoy! SBalley Team