FuriosaAI Launches RNGD Chip for Data Centers

<p><strong><span class&equals;"legendSpanClass"><span class&equals;"xn-location">SANTA CLARA<&sol;span><&sol;span><&sol;strong> &&num;8212&semi; <a href&equals;"https&colon;&sol;&sol;c212&period;net&sol;c&sol;link&sol;&quest;t&equals;0&amp&semi;l&equals;en&amp&semi;o&equals;4239861-1&amp&semi;h&equals;1999214758&amp&semi;u&equals;https&percnt;3A&percnt;2F&percnt;2Ffuriosa&period;ai&percnt;2F&amp&semi;a&equals;FuriosaAI" target&equals;"&lowbar;blank" rel&equals;"nofollow noopener">FuriosaAI<&sol;a>&comma; an emerging leader in the AI semiconductor space&comma; has unveiled RNGD &lpar;pronounced &&num;8220&semi;Renegade&&num;8221&semi;&rpar;&comma; a leading AI accelerator&comma; at Hot Chips 2024&period; RNGD is positioned to be the most efficient data center accelerator for high-performance large language model &lpar;LLM&rpar; and multimodal model inference&comma; disrupting an AI hardware landscape long defined by legacy chipmakers and high-profile startups&period; Founded in 2017 by three engineers with backgrounds at AMD&comma; Qualcomm&comma; and Samsung&comma; the company has pursued a strategy focused on rapid innovation and product delivery which has resulted in the unveiling and fast development of RNGD&period;<&sol;p>&NewLine;<div class&equals;"pull-right inline-gallery-container col-md-8 col-sm-7 col-xs-12">&NewLine;<div class&equals;"gallery inline-gallery">&NewLine;<div class&equals;"row">&NewLine;<div class&equals;"col-sm-12"><&sol;div>&NewLine;<div class&equals;"col-sm-12 clearfix">&NewLine;<figure>&NewLine;<div class&equals;"image lightbox-item " data-src&equals;"https&colon;&sol;&sol;mma&period;prnewswire&period;com&sol;media&sol;2489185&sol;FuriosaAI&lowbar;june&lowbar;ceo&period;jpg&quest;p&equals;publish" data-asset-type&equals;"photo" data-asset-id&equals;"Life&lowbar;After&lowbar;Debt&lowbar;Aug&lowbar;17&lowbar;Event&period;jpg" data-asset-label&equals;"General" data-sub-html&equals;"June Paik&comma; Co-Founder and CEO of FuriosaAI&period;" data-tweet-text&equals;"June Paik&comma; Co-Founder and CEO of FuriosaAI&period;" data-facebook-share-text&equals;"June Paik&comma; Co-Founder and CEO of FuriosaAI&period;" data-linkedin-text&equals;"June Paik&comma; Co-Founder and CEO of FuriosaAI&period;" data-download-url&equals;"https&colon;&sol;&sol;mma&period;prnewswire&period;com&sol;media&sol;2489185&sol;FuriosaAI&lowbar;june&lowbar;ceo&period;jpg&quest;p&equals;publish" data-pinterest-text&equals;"June Paik&comma; Co-Founder and CEO of FuriosaAI&period;" data-twitter-share-url&equals;"https&colon;&sol;&sol;mma&period;prnewswire&period;com&sol;media&sol;2489185&sol;FuriosaAI&lowbar;june&lowbar;ceo&period;jpg&quest;p&equals;twitter" data-linkedin-share-url&equals;"https&colon;&sol;&sol;mma&period;prnewswire&period;com&sol;media&sol;2489185&sol;FuriosaAI&lowbar;june&lowbar;ceo&period;jpg&quest;p&equals;linkedin" data-facebook-share-url&equals;"https&colon;&sol;&sol;mma&period;prnewswire&period;com&sol;media&sol;2489185&sol;FuriosaAI&lowbar;june&lowbar;ceo&period;jpg&quest;p&equals;facebook" data-pinterest-share-url&equals;"https&colon;&sol;&sol;mma&period;prnewswire&period;com&sol;media&sol;2489185&sol;FuriosaAI&lowbar;june&lowbar;ceo&period;jpg&quest;p&equals;facebook"><a class&equals;"tabfocus" role&equals;"button"><img id&equals;"imageid&lowbar;2" class&equals;"gallery-thumb img-responsive" title&equals;"June Paik&comma; Co-Founder and CEO of FuriosaAI&period;" src&equals;"https&colon;&sol;&sol;mma&period;prnewswire&period;com&sol;media&sol;2489185&sol;FuriosaAI&lowbar;june&lowbar;ceo&period;jpg&quest;w&equals;400" data-getimg&equals;"https&colon;&sol;&sol;mma&period;prnewswire&period;com&sol;media&sol;2489185&sol;FuriosaAI&lowbar;june&lowbar;ceo&period;jpg&quest;w&equals;400" &sol;><&sol;a><&sol;div><figcaption>June Paik&comma; Co-Founder and CEO of FuriosaAI&period;<&sol;figcaption><&sol;figure>&NewLine;<&sol;div>&NewLine;<&sol;div>&NewLine;<&sol;div>&NewLine;<&sol;div>&NewLine;<p>Furiosa successfully completed the full bring-up of RNGD after receiving the first silicon samples from their partner&comma; TSMC&period; This achievement reinforces the company&&num;8217&semi;s track record of fast and seamless technology development&period; With their first-generation chip&comma; introduced in 2021&comma; Furiosa submitted their first MLPerf benchmark results within 3 weeks of receiving silicon and achieved a 113&percnt; performance increase in the next submission through compiler enhancements&period;<&sol;p>&NewLine;<p>Early testing of RNGD has revealed promising results with large language models such as GPT-J and Llama 3&period;1&period; A single RNGD PCIe card delivers 2&comma;000 to 3&comma;000 tokens per second throughput performance &lpar;depending on context length&rpar; for models with around 10 billion parameters&period;<&sol;p>&NewLine;<p>&&num;8220&semi;The launch of RNGD is the result of years of innovation&comma; leading to a one-shot silicon success and exceptionally rapid bring-up process&period; RNGD is a sustainable and accessible AI computing solution that meets the industry&&num;8217&semi;s real-world needs for inference&comma;&&num;8221&semi; said <span class&equals;"xn-person">June Paik<&sol;span>&comma; Co-Founder and CEO of FuriosaAI&period; &&num;8220&semi;With our hardware now starting to run LLMs at high performance&comma; we&&num;8217&semi;re entering an exciting phase of continuous advancement&period; I am incredibly proud and grateful to the team for their hard work and continuous dedication&period;&&num;8221&semi;<&sol;p>&NewLine;<p><b>RNGD&&num;8217&semi;s key innovations include&colon;<&sol;b><&sol;p>&NewLine;<ul >&NewLine;<li>A non-matmul&comma; Tensor Contraction Processor &lpar;TCP&rpar; based architecture that enables a perfect balance of efficiency&comma; programmability and performance&period;<&sol;li>&NewLine;<li>Programmability through a robust compiler co-designed to be optimized for TCP that treats entire models as single-fused operations&period;<&sol;li>&NewLine;<li>Efficiency&comma; with a TDP of 150W compared to 1000W&plus; for leading GPUs<&sol;li>&NewLine;<li>High-performance&comma; with 48GB of HBM3 memory delivering the ability to run models like Llama 3&period;1 <span class&equals;"xn-money">8B<&sol;span> efficiently on a single card&period;<&sol;li>&NewLine;<&sol;ul>&NewLine;<p>The chip is currently sampling to early access customers&comma; with broader availability expected in early 2025&period;<&sol;p>&NewLine;

Editor

Wispr Scores $25 Million Series A Extension

SAN FRANCISCO -- Wispr, the voice-to-text AI that turns speech into clear, polished writing in every…

1 day

Numeric Dials Up $51 Million Series B

SAN FRANCISCO -- Numeric, an AI accounting automation platform, has raised a $51 million Series…

1 day

Apple Names 45 Finalists for App Store of the Year Awards

Apple has announced 45 finalists for this year’s App Store Awards, recognizing the best apps…

2 days

UC Reaches Agreement With Nurses, Strike Canceled

The University of California (UC) and the California Nurses Association (CNA) have reached a tentative…

4 days

HouseRX Rakes In $55 Million Series B

SAN FRANCISCO -- House Rx, a health tech company focused on making specialty medications more accessible and…

4 days

King Charles Honors NVIDIA’s Jensen Huang

Britain's King has given an award to the King of NVIDIA! NVIDIA founder and CEO…

4 days