TL;DR:
- SiFive introduces Performance P870 and Intelligence X390 RISC-V cores for AI and ML applications.
- P870 offers 50% higher single-thread performance, suitable for consumer and data center use.
- X390 boasts a 4x improvement in vector computation, doubling vector length and dual vector ALUs.
- VCIX enables customization with its own vector instructions and acceleration hardware.
Main AI News:
In the realm of AI and machine learning, SiFive has unveiled a formidable duo of high-end RISC-V cores poised to revolutionize the consumer, automotive, and infrastructure sectors. Dubbed the “Performance P870” and the “Intelligence X390,” these cutting-edge cores represent a significant leap in low-power compute density and vector compute capabilities. When harnessed in unison, they usher in a new era of performance for data-intensive computing applications.
Performance P870 stands as a pinnacle of computational prowess, catering to both consumer applications and data centers when coupled with a vector processor. This architectural marvel boasts an impressive 50% increase in peak single-thread performance (specINT2k6) compared to its predecessors. Anchored by a six-wide out-of-order core, it adheres rigorously to RVA 23 standards and flaunts a shared cluster cache accommodating up to 32 cores.
SiFive elaborates, “High execution throughput comes with more instruction sets per cycle, more ALU, and more branch units,” underlining the core’s formidable capabilities. Notably, the P870 core seamlessly aligns with the requirements of Google Android-on-RISC-V, and it incorporates essential features like x128b VLEN RVV, vector crypto, hypervisor extensions, IOMMU, AIA, non-inclusive L3 cache, and WorldGuard security. P870-A, a variant tailored for automotive applications, introduces additional features to bolster its automotive utility.
On the other hand, the Intelligence X390 heralds a remarkable evolution from its predecessor, the X280. In a single-core configuration, the X390 boasts a staggering 4x enhancement in vector computation, a doubling of vector length, and the inclusion of dual vector ALUs. This translates to a quadrupling of sustained data bandwidth, a feat that holds immense promise. SiFive also introduces VCIX (vector coprocessor interface extension), offering companies the freedom to incorporate their own vector instructions and acceleration hardware seamlessly.
VCIX, with its 2,048-bit outbound and 1,024-bit inbound architecture, is the cornerstone of this innovation. Additionally, the X390 core packs a punch with features such as 1,024-bit VLEN, 512-bit DLEN, and single-dual vector ALU, cementing its position as a trailblazing solution for AI and machine learning applications across various domains. SiFive’s Performance P870 and Intelligence X390 cores promise to be pivotal players in driving the future of high-performance computing, setting new standards in power-efficient, data-intensive processing.
Conclusion:
SiFive’s latest RISC-V cores, the P870 and X390, mark a significant leap in AI and machine learning capabilities. With enhanced performance and customization options, they are poised to drive innovation across consumer, automotive, and infrastructure markets, enabling more efficient and powerful data-intensive computing. These cores represent a promising advancement in the competitive landscape of high-performance computing solutions.
