Intel announces first AI-optimized Stratix 10 NX FPGA: INT8 performance soars 15x

Intel has invested heavily in AI artificial intelligence over the years, especially in the data center market, with the most complete solutions in the industry, xeon CPUs, Xe GPU (under development), Agilex FPGA, Movidius, Habana, eASIC… With a variety of efforts to respond flexibly to a wide variety of workloads, optimized acceleration.

Today, the Stratix 10 FPGA family welcomes the newest member, the Stratix 10 NX, the first FPGA optimized for AI, with a high-performance AI integrated with custom hardware that delivers high bandwidth, low-latency AI acceleration.

Intel announces first AI-optimized Stratix 10 NX FPGA: INT8 performance soars 15x

Stratix 10 NX integrates core chips, HBM storage, I/O expansion sits in EMBI integrated package, with a 14nm process manufacturing and a high-performance “AI Tensor Block” (tens or block) supporting IN T4, INT8, FP16, FP32 and other formats, in which INT8 integer computing performance can reach up to 15 times the existing Stratix 10 MX scheme, and for different AI workloads, can also be hardware programming, flexible and efficient response.

It also embeds high-capacity HBM high-bandwidth memory, has an ultra-high bandwidth network, up to 5.8G PAM4 transceivers, and supports PCIe 3.0 x16, 10/25/100G Ethernet blocks, providing flexible, customizable interconnect channels across multiple nodes.

Extreme scalability is also a highlight, with a small chip (chiplet) design that makes it easy to access a wide range of custom chips, modules, and ASIC extensions, greatly expanding their processing power to accommodate a wide variety of workloads.

Intel announces first AI-optimized Stratix 10 NX FPGA: INT8 performance soars 15x

Stratix 10 NX is widely used in natural language processing such as speech recognition and speech synthesis, security such as deep package detection, congestion control recognition, anti-fraud, content recognition, video pre-processing, and post-processing.

Microsoft says that in designing real-time multi-node AI solutions, there is an urgent need to provide ASIC-level zhangon performance, high-capacity storage, ultra-high connectivity bandwidth, and very low latency elastic processing devices, stratix 10 NX FPGAs meet Microsoft’s very high needs, and is working closely with Intel to develop next-generation solutions to meet the needs of customers for large-scale AI applications.

The Stratix 10 NX will be available later this year.

Intel announces first AI-optimized Stratix 10 NX FPGA: INT8 performance soars 15x

Intel announces first AI-optimized Stratix 10 NX FPGA: INT8 performance soars 15x

Intel announces first AI-optimized Stratix 10 NX FPGA: INT8 performance soars 15x

Intel announces first AI-optimized Stratix 10 NX FPGA: INT8 performance soars 15x

Intel announces first AI-optimized Stratix 10 NX FPGA: INT8 performance soars 15x

Intel announces first AI-optimized Stratix 10 NX FPGA: INT8 performance soars 15x