AWS Set to Unveil Enhanced Graviton4 CPU and New Trainium AI Chip

AWS to introduce updated server CPU, new AI training chip - SiliconANGLE

Amazon Web Services Inc. is gearing up to refresh two of its custom processor families, CNBC reported late Tuesday. The cloud giant plans to release an updated version of its AWS Graviton4 central processing unit. It will also debut a new addition to the AWS Trainium chip series, which is optimized for training artificial intelligence models. Both chip families are developed by AWS' Annapurna Labs unit, which became part of the cloud provider through a 2015 acquisition. When Graviton4 debuted last June, AWS detailed that it has deployed more than 50 million Graviton-series processors to power 150-plus instance types. The Trainium series, in turn, powers an AI training cluster with 400,000 chips that AWS started building for Anthropic PBC last year. Graviton4 is the fourth iteration of the Graviton CPU series. It features 96 cores based on Arm Holdings Inc.'s server-optimized Neoverse N2 core design. The onboard L2 cache can hold 192 megabytes of data, up from the 8 megabytes offered by the first-generation Graviton chip. AWS says that the Graviton4 contains a total of 73 billion transistors across its components. The processor also includes cybersecurity optimizations. An Arm technology called BIT, or Branch Target Identification, blocks cyberattacks that attempt to use Graviton4's branch prediction mechanism. This is a CPU component that completes some calculations before the results are needed to save time. Additionally, the hardware interfaces through which Graviton4 sends data traffic to other components of AWS' infrastructure are encrypted. The updated version of Graviton4 that the cloud giant plans to announce will reportedly offer 600 gigabits per second, or Gbps, of network bandwidth. Current Graviton4-powered instances provide up to 50 Gbps. Each chip also offers up to 536.7 Gbps of memory bandwidth, which determines the rate at which data moves to and from RAM. That metric can heavily influences application performance. According to CNBC, AWS will announce the release schedule of Graviton4 later this month. Trainium2, the Amazon.com Inc.'s other new chip, is expected to launch by year's end. AWS' current-generation AI training chip, Trainium2, provides 1.29 petaflops of performance when processing FP8 floating point numbers. Those are data units that neural networks commonly use to perform calculations. One petaflop corresponds to one quadrillion, or a thousand trillion, computations per second. Trainium2's processing capacity is provided by eight cores based on a design called NeuronCore-V3. They're supported by 96 gibibytes of onboard memory. Developers may optically combine the compute and memory resources of multiple cores into a single virtual core, an arrangement that streamlines some large-scale AI training projects. AWS expects its next-generation Trainium3 chip to provide double the performance of Trainium2. Additionally, it's expected to be 50% more energy-efficient. AWS also uses other custom chips besides Graviton and Trainium processors to power its cloud platform. The Amazon unit's Nitro Security Chip, for example, protects its servers from tempering attempts. AWS data centers also contain Nitro Cards, devices that use custom processors to manage the flow of network traffic.

Amazon's Powerful AWS Custom Chip To Get An Upgrade - Why Is It Important? - Amazon.com (NASDAQ:AMZN)

Amazon.com, Inc AMZN Amazon Web Services could soon announce an update to its Graviton4 chip, sparking a rivalry with Intel Corp INTC and Advanced Micro Devices AMD. The chip could include 600 gigabits per second of network bandwidth, CNBC reported on Wednesday. Benzinga has reached out to Amazon for its comment. Hyperscalars like Amazon, Microsoft Corp MSFT, Alphabet Inc GOOGL, and Meta Platforms META migrated towards custom chips to curtail costs, tackle compatibility issues, and address supply crises. Also Read: Amazon Expands Global Data Centers, Boosts Access To Nvidia AI Chips For Cloud Customers What is the Graviton4 Chip? According to Amazon, its latest cloud computing chip, Graviton4, offers significant advancements in performance and energy efficiency. It's the fourth generation of AWS's custom-engineered data center chips, first introduced in 2018. Graviton4, based on Arm-architecture, boasts four times the performance of Graviton1 and is even more energy-efficient than its predecessor, Graviton3. With 73 billion transistors, it's being adopted by major customers like SAP and Epic Games for high-performance and reliable cloud experiences. This innovation, alongside other custom AWS chips like Trainium and Inferentia for AI, underscores Amazon's leadership in cloud computing technology. Amazon manufactures Graviton4 central processing unit at its Annapurna Labs in Austin, Texas. Why Does Having A Custom Chip Matter? Rami Sinno, director of engineering at Amazon Web Services (AWS) Annapurna Labs, cited strong demand for chips that outpaced the supply. AWS Senior Director Gadi Hutt told CNBC about Amazon's initiatives to provide an alternative to Nvidia Corp's NVDA costly graphics processing units. Hutt told CNBC. that Anthropic's Claude Opus 4 AI model launched on Trainium2 GPUs and Project Rainier. Trainium3 is coming up in 2025, he said. AWS has committed $8 billion to Anthropic. Also, Amazon chief Andy Jassy recently highlighted the company's extensive leveraging of Generative AI to drive value. Jassy highlighted Generative AI's crucial role within AWS for developers and optimizing internal functions such as fulfillment and customer service. He also pointed to developing AI agents to automate tasks and accelerate innovation. As Amazon commits huge spending and Capex to AI, it is imperative that it design its own chips based on custom needs rather than relying on NVIDIA. Jassy stated that this AI-driven transformation might reduce the corporate workforce in the coming years. Price Action: AMZN stock is up 0.92% at $216.80 at the last check on Wednesday. Read Next: Marvell Launches New Power Regulator And 2nm Memory Chips To Boost AI, Cloud Efficiency Photo by Deep Pixel via Shutterstock AMZNAmazon.com Inc$216.900.97%Stock Score Locked: Want to See it? Benzinga Rankings give you vital metrics on any stock - anytime. Reveal Full ScoreEdge RankingsMomentum60.73Growth97.14Quality73.37Value49.29Price TrendShortMediumLongOverviewAMDAdvanced Micro Devices Inc$126.89-0.17%ARMARM Holdings PLC$146.461.20%GOOGLAlphabet Inc$175.31-0.36%INTCIntel Corp$21.272.24%METAMeta Platforms Inc$701.280.58%MSFTMicrosoft Corp$478.990.20%NVDANVIDIA Corp$144.880.53%Market News and Data brought to you by Benzinga APIs