Ampere Computing has provided an update that highlights the company’s newest achievements in the semiconductor business, with a particular emphasis on the AI and Cloud sectors. This update highlights a substantial move towards sustainable and efficient computing. A significant partnership with Qualcomm Technologies was announced as part of Ampere’s annual product and milestone update. The goal of the partnership is to develop AI inferencing solutions that are low power and high performing.
Under the direction of CEO Renee James, a seasoned veteran of the semiconductor business, Ampere Computing is becoming more and more concerned with the growing power requirements and environmental effects of developing AI technology. Renee James said, “The current path is unsustainable,” underscoring the need of tackling these issues immediately. In her vision, new data centers will be built with environmental sustainability and grid capacity in mind, and existing data centers would be renovated to allow enhanced computation capabilities in the future.
“Low power used to be synonymous with low performance. Ampere has proven that isn’t true,” said Renee James, CEO at Ampere Computing.The collaboration with Qualcomm involves the development of a joint solution that integrates Qualcomm’s Cloud AI 100 Ultra with Ampere’s CPUs to enhance AI inferencing on large generative AI models. This partnership is set to scale out solutions that handle extensive tasks such as LLM inferencing, combining Qualcomm’s innovative AI inference chips with Ampere’s powerful processors.
Highlighting the company’s forward-looking approach, Jeff Wittich, Chief Product Officer at Ampere, introduced the concept of ‘AI Compute.’ This initiative extends Ampere’s traditional cloud-native capabilities to encompass a broader range of AI applications, integrating AI with conventional cloud-native applications like data processing and media delivery. Mr. Wittich’s vision would represent a significant expansion in the versatility and application of Ampere’s CPU solutions.
Collaboration with Oracle Cloud, NETINT Technologies
Ampere Computing also announced exciting developments in its hardware lineup. The new 12-channel 256 core CPU, built on the N3 process node, is poised to deliver unprecedented performance levels. This upcoming CPU is expected to offer more than 40% greater performance than any existing market offering without the need for exotic platform designs. Furthermore, Ampere’s ongoing projects include the 192-core 12-channel memory platform, slated for release later this year.
The company also shared updates from its collaboration with Oracle Cloud, where Meta’s Llama 3 AI model is running on Ampere CPUs. The performance data revealed that Llama 3, operating on a 128-core Ampere Altra CPU without additional GPU support, matches the performance of systems utilizing Nvidia A10 GPUs paired with x86 CPUs, but with only a third of the power usage.
Ampere is also enhancing its technological ecosystem through initiatives like the formation of a UCIe working group within the AI Platform Alliance. This move is aimed at leveraging open interface technology to incorporate customer IP into future CPUs, thereby increasing the flexibility and application range of Ampere’s solutions.
In terms of product performance, the new AmpereOne platform has demonstrated significant advancements, according to Ampere. The company also noted that the new AmpereOne OEM and ODM platforms are expected to start shipping in the coming months.
Additionally, Ampere announced a collaborative venture with NETINT Technologies, integrating Ampere CPUs with NETINT’s Quadra T1U video processing chips. This solution is designed to handle demanding tasks such as transcoding 360 live channels and providing real-time subtitling for 40 streams across multiple languages, utilizing OpenAI’s Whisper model.
The update concluded with the introduction of a new FlexSKU feature, allowing data centers to adapt the same SKU to various scale-out and scale-up scenarios, demonstrating Ampere’s commitment to providing versatile and scalable solutions.
