- Groq standalone chip processor provides flexible integration into compute-intensive applications.
- It is suitable for high-performance computing and features up to 80 TBs of on-die memory bandwidth.
Groq Chip processor accelerates AI, ML, and HPC workloads and removes extraneous circuitry from the chip. The processor reduces data movement for predictable low-latency performance. This standalone chip provides flexible integration into compute-intensive applications. The on-chip features up to 80 TBs of on-die memory bandwidth that facilitates massive concurrency and data parallelism needed for bandwidth-sensitive applications.
The chip has integrated software control units at strategic points to optimize data movement and processing. It offers up to 750 TOPs, 188 TFLOPs (INT8, FP16 @900 MHz). It also offers end-to-end on-chip protection that improves uptime and reliability with error correction code (ECC) protection throughout the entire GroqChip data path.
The Groq chip has a simplified architecture which leads to an easy-to-use software suite, accelerating developer velocity. The chip comes with 230 MB of on-die memory enabling high-bandwidth, low-latency access to model parameters without the need for external memory. It also improves the data movement which enables lower latency and provides flexible integration into compute-intensive applications. The chip has a maximum power consumption of up to 300 watts with an average power consumption of 185 watts. It also features end-to-end on-chip protection that improves uptime and reliability with error correction code (ECC) protection throughout the entire GroqChip data path.
The Groq chip offers 16 integrated RealScale chip-to-chip interconnects which allows for better connectivity. For I/O the chip offers an integrated PCIe Gen4 x16 controller. The GroqChip is suitable for use in applications in enterprise comm, oil and gas industries, high-level research and scientific computation, etc.
