403
Sorry!!
Error! We're sorry, but the page you were looking for doesn't exist.
OpenAI Tests Custom Chip Against Nvidia Systems
(MENAFN) OpenAI says its first internally developed artificial intelligence processor, Jalapeno, performs better than several Nvidia systems in company evaluations, with stronger energy efficiency and quicker response times across multiple large language models.
According to the reported results, Jalapeno generates between 1.5 and 1.9 times more AI workload per watt at peak throughput than the Nvidia systems used for comparison. The chip also cuts overall processing latency by between 1.7 and 3.6 times.
Performance is even higher for highly interactive applications, where Jalapeno records gains ranging from 2.1 to 4.1 times, based on the published testing results.
OpenAI conducts the evaluation through InferenceX, a publicly available benchmark created by semiconductor research firm SemiAnalysis. The processor is tested with GPT-OSS 120B, DeepSeek R1 and Kimi K2.5.
The assessment compares Jalapeno with Nvidia’s GB200 and GB300 platforms. However, the company does not include Nvidia’s newer Vera Rubin generation in the comparison.
While running GPT-OSS 120B, Jalapeno achieves roughly 1.9 times greater peak throughput per kilowatt than the GB200. Tests involving DeepSeek R1 show about 1.7 times better performance per watt than the GB300, while the chip records a 1.5-fold advantage when running Kimi K2.5, according to OpenAI.
According to the reported results, Jalapeno generates between 1.5 and 1.9 times more AI workload per watt at peak throughput than the Nvidia systems used for comparison. The chip also cuts overall processing latency by between 1.7 and 3.6 times.
Performance is even higher for highly interactive applications, where Jalapeno records gains ranging from 2.1 to 4.1 times, based on the published testing results.
OpenAI conducts the evaluation through InferenceX, a publicly available benchmark created by semiconductor research firm SemiAnalysis. The processor is tested with GPT-OSS 120B, DeepSeek R1 and Kimi K2.5.
The assessment compares Jalapeno with Nvidia’s GB200 and GB300 platforms. However, the company does not include Nvidia’s newer Vera Rubin generation in the comparison.
While running GPT-OSS 120B, Jalapeno achieves roughly 1.9 times greater peak throughput per kilowatt than the GB200. Tests involving DeepSeek R1 show about 1.7 times better performance per watt than the GB300, while the chip records a 1.5-fold advantage when running Kimi K2.5, according to OpenAI.
Legal Disclaimer:
MENAFN provides the
information “as is” without warranty of any kind. We do not accept any
responsibility or liability for the accuracy, content, images, videos,
licenses, completeness, legality, or reliability of the information
contained in this article. If you have any complaints or copyright issues
related to this article, kindly contact the provider above.
Comments
No comment