Predictable and repeatable performance with no run-to-run variation
Massive concurrency and data
parallelism for bandwidth sensitive applications
Near-linear multi-server and multi-rack scalability without the need for external switches
Improves uptime and reliability with error-correction code (ECC) protection throughout the entire GroqChip™ data path
Up to 31.5GB/s of bi-directional bandwidth in an industry standard interface for fast device and network connections
We offer the GroqCard for approved servers. Get in touch to learn more.
High-density servers with pre-integrated cards and a comprehensive warranty.
BittWare has 30+ years providing customized solutions and form factors.
For machine learning inference, GPUs suffer from inefficiencies leading to latency, low silicon resource usage, and unpredictable performance. Groq has designed its AI deep learning chip specifically to provide predictable, efficient, low-latency inference that’s easy to bring into your current workflow.
The GroqCard is a double-with PCIe form factor ML accelerator that’s hassle-free to integrate. The GroqWare suite follows a software-defined hardware approach, giving easy deployment paths for your PyTorch, TensorFlow, and ONNX-trained deep learning models.
Scalability is a core feature of the GroqCard, with 9 RealScale chip-to-chip connections that ensure deployment of multiple cards is as efficient as one. An internal software defined network provides predictable, repeatable performance with no run-to-run variations.
Simplify programming with
GroqWare Suite is a comprehensive and versatile software stack designed to accelerate a variety of HPC and ML workloads. Composed of Groq™ Compiler, Groq API, and Utilities, the suite eases deployment implementations with an open source driver/runtime and support for industry standard AI/ML frameworks.
GroqFlow™ Tool Chain (included in the GroqWare Suite) enables a single line of Pytorch or TensorFlow code to import and transform existing models through a fully automated tool chain to run on Groq hardware.
See your program as it travels across GroqChip
The revolutionary, fully deterministic GroqChip processor is the core of scalable performance. Built from the ground up to accelerate AI, ML, and HPC workloads, GroqChip reduces data movement for predictable low-latency performance, bottleneck-free. This standalone chip provides flexible integration into compute intensive applications.
The architecture is much simpler than a GPU and is designed with a software-first focus, making it easier to program and providing predictable performance with lower latency.
Form Factor
Dual width, full height, ¾ length PCI Express Gen4 x16 adapter
Performance
Up to 750 TOPs, 188 TFLOPs (INT8, FP16 @900 MHz)
Memory
230 MB SRAM per chip
Up to 80 TB/s on-die memory bandwidth
Chip Scaling
Up to 9 RealScale™ chip-to-chip connectors
Numerics
INT8, INT16, INT32 & TruePoint™ technology MXM: FP32
VXM: FP16, FP32
Power
Max: 375W; TDP: 275 ; Typical: 240W
BittWare can design and manufacture the right form factor for your deployment.
Fill out the form to get in touch for details on Groq Real-time AI Acceleration.
PCIe Gen4 data mover IP from Atomic Rules. Achieve up to 220 Gb/s using BittWare’s PCIe Gen4 cards, saving your development team when you need more performance than standard DMA. Features: DPDK and AXI standards, work with packets or any other data format, operate at any line rate up to 400 GbE.
BittWare Partner IP Query Processing Unit (QPU) Build FPGA-powered accelerators to query, analyze or reformat stored or streaming data at PCIe Gen4 speeds! Eideticom’s Query
REFERENCE DESIGN MAU Accelerator for AI Financial Trading Models Ultra-low Latency, High Throughput Machine Learning Inference Well suited to a range of applications in financial
White Paper FPGA Acceleration of Convolutional Neural Networks Overview Convolutional Neural Networks (CNNs) have been shown to be extremely effective at complex image recognition problems.
