Thu Sep 19 08:25:13 UTC 2024: ## SiFive Unveils Customizable AI Accelerator Design, Taking on Nvidia

**San Mateo, CA** – SiFive, the leading RISC-V processor company, has announced the launch of its Intelligence XM series, a customizable AI accelerator design for companies looking to develop their own AI chips. This marks a significant expansion for SiFive, which has previously supplied RISC-V cores for AI chips like Google’s Tensor Processing Units and Tenstorrent’s Blackhole accelerator.

The XM series offers a scalable building block for AI chip development, allowing companies to license the design and integrate it into their own processors and system-on-chips (SoCs). This can be used for a wide range of applications, from edge and IoT devices to data center servers.

The base XM cluster comprises four of SiFive’s Intelligence X RISC-V CPU cores connected to a specialized matrix math engine for neural network calculations. Each cluster supports up to 1TB/sec of memory bandwidth and delivers up to 16 TOPS (tera-operations per second) of INT8 or 8 teraFLOPS of BF16 performance per gigahertz.

While the performance of a single cluster might seem modest, SiFive expects most chips to utilize between four and eight clusters, resulting in peak memory bandwidth of 4-8TB/sec and BF16 performance of 32-64 teraFLOPS.

Furthermore, the design can be scaled up to 512 clusters, potentially achieving a staggering 4 petaFLOPS of AI performance. This places SiFive’s XM clusters in direct competition with Nvidia’s upcoming Blackwell accelerators, which boast around 2.5 petaFLOPS of BF16 performance.

SiFive’s offering is aimed at organizations seeking an off-the-shelf AI accelerator design that they can customize and send to fabrication. This provides an alternative to developing their own hardware from scratch, which can be costly and time-consuming.

The company is also releasing an open source reference implementation of its SiFive Kernel Library, further lowering barriers to adoption for RISC-V-based AI architectures.

SiFive’s move into the AI accelerator market is a significant development in the industry. It provides a compelling alternative for companies looking to develop their own AI chips, potentially challenging the dominance of Nvidia and other established players. This could foster greater competition and innovation in the AI hardware space.

Read More