 |
Photo courtesy of Yonhap News |
[Alpha Biz= Paul Lee] FuriosaAI has secured Samsung SDS as a new customer, with plans to deploy its second-generation neural processing unit (NPU), Renegade (RNGD), in Samsung SDS servers starting in July.
The announcement was made on April 2 at the “Renegade Commit” event held at SJ Kunsthalle. CEO Baek Jun-ho stated, “We received the first batch of 4,000 Renegade units in January and plan to mass-produce up to 20,000 units by the end of this year.”
Renegade is designed with a strong emphasis on power efficiency. The chip operates with a thermal design power (TDP) of under 180W, while delivering up to 512 teraflops (TFLOPS) in floating-point performance and 1 petaflop (PFLOPS) in INT4 precision. It also incorporates fourth-generation HBM3 memory from SK Hynix, helping to alleviate memory bottlenecks.
According to FuriosaAI’s internal benchmarking, Renegade can process up to 7.4 times more data than NVIDIA’s RTX PRO 6000 at equivalent power levels. The company also claims that total cost of ownership (TCO) for AI infrastructure can be reduced by approximately 40% compared to conventional GPU-based systems.
FuriosaAI confirmed that Samsung SDS has been secured as a new customer for Renegade. The IT services provider plans to begin deploying the chips in its servers starting in July. The two companies have been collaborating since September last year.
Samsung SDS is also preparing a subscription-based cloud server sharing service and is considering integrating Renegade chips into its cloud platform.
The deal follows FuriosaAI’s earlier partnership with LG, to which it currently supplies NPU chips for its EXAONE AI model.
AlphaBIZ Paul Lee(hoondork1977@alphabiz.co.kr)
