Groq Selects Samsung Foundry to Bring Next-gen LPU™ to the AI Acceleration Market
MOUNTAIN VIEW, Calif. -- Aug. 24, 2023 -- Groq, an artificial intelligence (AI) inference systems innovator, today announced it has contracted with Samsung's growing Foundry business to be its next-gen silicon partner, solidifying Groq's product roadmap with a US-based foundry services provider.
Groq, an AI solutions innovator, announced it has contracted Samsung as its next-gen silicon partner, solidifying the startup's product roadmap with a US-based foundry services provider.
Jonathan Ross, CEO and founder of Groq, stated, "The first-gen Groq LPU™ was built for AI from the ground up, enabling Groq to consistently outperform graphics processors on AI and ML tasks. Our partnership with Samsung will allow us to continue that leap forward, by tapping into the most advanced semiconductor manufacturing technologies available."
Marco Chisari, EVP of Samsung Electronics, Head of Samsung's US Foundry business and Head of Samsung Semiconductor Innovation Center shared, "Samsung Foundry is committed to advancing semiconductor technology and bringing groundbreaking AI, HPC, and data center solutions to market. This relationship with Groq is another proof point of how we're using our advanced silicon manufacturing nodes to bring new AI innovation to market."
The current generation of Groq hardware solutions continue to achieve orders of magnitude performance improvements over existing solutions at scale with a 2-4X power advantage. The next-gen silicon, to be designed in partnership with Samsung's Foundry Design Service team and manufactured by Samsung on its SF4X process (4nm), will utilize the revolutionary first-gen Tensor Streaming architecture. It is set to improve throughput, latency, power consumption, hardware footprint, and memory capacity. Additional technology and architectural improvements will increase the Language Processor Unit™ system's interconnect bandwidth, enabling Groq to build systems ranging from 85,000 to more than 600,000 chips, without any external switches and with significantly more cross-sectional bandwidth than today's exascale supercomputers.
Samsung Electronics, a world leader in advanced semiconductor technology headquartered in South Korea, has a new fab under construction in Taylor, Texas, neighboring its first US semiconductor manufacturing facility in Austin, which broke ground in 1996. Working with Samsung further strengthens Groq's already completely North American-based operations for engineering and manufacturing, and will help maintain its high product availability and short lead times.
About Groq
Groq is an AI solutions company delivering ultra-low latency inference with the first ever LPU™. For more information, visit www.groq.com.
|
Related News
- Why UCIe is Key to Connectivity for Next-Gen AI Chiplets
- ADTechnology Partners with Arm, Samsung Foundry, and Rebellions on AI CPU Chiplet Platform
- DeepX Hints At Next-Gen AI Chips
- Cadence and Samsung Foundry Accelerate Chip Innovation for Advanced AI and 3D-IC Applications
- OPENEDGES' Memory Subsystem IPs Selected by ASICLAND for Next-gen AI Applications
Breaking News
- Breker RISC-V SystemVIP Deployed across 15 Commercial RISC-V Projects for Advanced Core and SoC Verification
- Veriest Solutions Strengthens North American Presence at DVCon US 2025
- Intel in advanced talks to sell Altera to Silverlake
- Logic Fruit Technologies to Showcase Innovations at Embedded World Europe 2025
- S2C Teams Up with Arm, Xylon, and ZC Technology to Drive Software-Defined Vehicle Evolution
Most Popular
- Intel in advanced talks to sell Altera to Silverlake
- Arteris Revolutionizes Semiconductor Design with FlexGen - Smart Network-on-Chip IP Delivering Unprecedented Productivity Improvements and Quality of Results
- RaiderChip NPU for LLM at the Edge supports DeepSeek-R1 reasoning models
- YorChip announces Low latency 100G ULTRA Ethernet ready MAC/PCS IP for Edge AI
- AccelerComm® announces 5G NR NTN Physical Layer Solution that delivers over 6Gbps, 128 beams and 4,096 user connections per chipset
![]() |
E-mail This Article | ![]() |
![]() |
Printer-Friendly Page |