SiFive’s XM Series Gen 2 offers an extremely scalable and efficient AI compute engine to meet the needs of a wide range of end market applications. By integrating scalar, vector, and matrix engines, XM Series customers can take advantage of very efficient memory bandwidth. The XM Series also continues SiFive’s legacy of offering extremely high performance per watt for compute-intensive applications across a wide range of workloads and is heavily tuned for LLMs. To speed up development time, SiFive is also open sourcing the SiFive Kernel Library.
SiFive Intelligence
XM Series
SiFive Intelligence
XM Series Gen 2
SiFive Intelligence
XM Series Gen 2 Key Features
New in Second Generation
- Highly scalable to hit a wide range of end market applications
- 4 Integrated 2nd Gen X300 cores with 1-4 acting as Accelerator Control Units to the XM matrix Engine
- New XM Gen 2 matrix engine upgrades
- Improved Performance across a range of workloads
- New Datatype support
- Now Heavily tuned for LLMs
SiFive Matrix Engine
- Fat Outer Product design
- Tightly integrated with 4 X-Cores
- Deep fusion with vector units
4 X-Cores per cluster
- Each with dual vector units
- Executes all other layers e.g. activation functions
- New exponential acceleration instructions
New matrix instructions
- Fetched by scalar unit
- Source data comes from vector registers
- Destination to each matrix accumulator
1 Cluster = 16 TOPS (INT8), 8 TFLOPS (BF16) per GHz
1TB/s sustained bandwidth per XM Series cluster
XM clusters connect to memory in 2 ways:
- Shared Memory Port(s) for cached accesses which maintain coherency between the 4 internal X-cores
- Dedicated High Bandwidth Port per X-core for uncached accesses with guaranteed full data bandwidth availability
Host CPU can be RISC-V, x86 or Arm (or not present)
SiFive Intelligence Family
X100 Series
Intelligence
- 32-bit or 64-bit CPU
- 128-bit Vector Length
- SSCI and VCIX
X200 Series
Intelligence
- 512-bit Vector length
- Single Vector ALU
- SSCI and VCIX (1024-bit)
X300 Series
Intelligence
- 1024-bit Vector Length
- Single/Dual Vector ALU
- SSCI and VCIX (2048-bit)
Contact Sales
Core IP Families
Automotive
SiFive Automotive™ solutions provide a range of high-end application and real-time processors, delivering industry-leading performance, minimal area, and low power consumption. ISO 21434, ISO26262 ASIL B, ASIL D and split-lock certifications.
Intelligence
Intelligence embraces a software-first approach to processor design, addressing the future needs of deploying ML and AI at the edge. It provides high-performance, scalable vector and matrix compute capabilities, adaptable to evolving workloads.
Performance
The SiFive Performance™ family of RISC-V processors is designed for maximum throughput and performance efficiency for a range of application from datacenter workloads, ranging from Web servers and multimedia processing, Networking/Storage to consumer applications like TV, wearables and other smart devices requiring AI functionality.