AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。
Artificial intelligence startup Groq Inc. today announced that it has raised $350 million in funding. The Series A round was led by returning backer Disruptive. Grok stated that Nvidia Corp. plans to join the round later down the line, but didn’t specify how much the chip giant will invest. The cash infusion comes less than three months after the company closed a $650 million round. Groq launched in 2016 with a focus on developing AI accelerators. Last December, Nvidia inked a $20 billion deal to license the company’s chip technology and hire several members of its leadership team. Groq subsequently pivoted by launching a public cloud platform optimized for AI workloads. The platform is powered by the chip technology that the company developed prior to the Nvidia deal. Nvidia, for its part, used Groq’s technology to develop a chip called the Groq 3 LPU that debuted in March. It’s optimized for inference, the task of running AI models in production after they’ve been trained. A large language model comprises two main components: an attention mechanism and a feed-forward network, or FFN. The former module identifies the most important parts of the user’s prompt. The FFN, in turn, stores much of the knowledge that the LLM picks up during training. It also performs a sizable portion of calculations involved in generating a prompt response. The Groq 3 LPU is designed to be used alongside Nvidia Corp.’s Rubin graphics processing unit. Customers can run their models’ FFN modules on the former chip while sending attention-related calculations to the GPUs. According to Nvidia, that disaggregated processing approach is more efficient than running everything on graphics cards. There are also other situations where the Groq 3 LPU provides a performance boost. It can speed up LLMs that feature a mixture of expert architecture or use speculative decoding. The latter technology enables an LLM to speed up some calculations by offloading them to a second, more hardware-efficient model. Nvidia ships the Groq 3 LPU as part of racks that each contain 256 accelerators. Groq is using the racks to power its AI-optimized public cloud, which is called GroqCloud. It provides bare-metal environments that enable customers to customize the underlying hardware. For less tech-savvy users, the company offers a toolkit called GroqStack that automates infrastructure management tasks. Notably, Groq stated today that its platform lends itself to not only inference but also AI training. The Groq 3 LPU doesn’t support the letter use case. The company will presumably run training workloads on the Rubin graphics cards with which its Groq 3 LPU racks are deployed. Nvidia provides a software engine called Dynamo that automatically determines which chip should run what part of an inference workload. Groq will use the proceeds from its funding round to grow its public cloud. Currently, the platform runs on hardware hosted in 13 data centers worldwide. Groq plans to grow its cloud capacity from 57 megawatts to over 200 megawatts next year. Image: Groq A message from John Furrier, co-founder of SiliconANGLE: Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities. 15M+ viewers of theCUBE videos, powering conversations across AI, cloud, cybersecurity and more 11.4k+ theCUBE alumni — Connect with more than 11,400 tech and business leaders shaping the future through a unique trusted-based network Are you an AWS customer? Support SiliconANGLE financially by buying your AWS services from our Marketplace portal page and links: https://siliconangle.com/aws-marketplace/ About SiliconANGLE Media