AMD made 146 billion transistors: 13 in one, performance soared 8 times
Views : 130
Update time : 2023-01-12 17:31:29
AMD unveiled Instinct MI300, its next-generation APU product for data centers at CES 2023, featuring a chiplet design with 13 small chips and up to 146 billion transistors. Specifically, the Instinct MI300 is built from 13 small chips, many of which are based on 3D stacks, have 24 Zen4 CPU cores, and incorporate the CDNA 3 graphics engine, as well as a shared unified memory pool, including Infinity Cache cache and 8 HBM shared memory designs.
Overall, the chip has 146 billion transistors, surpassing Intel's 100 billion Ponte Vecchio transistor, making it the largest AMD chip ever put into production. There are eight 128GB HBM3 chips on either side of the MI300, and multiple silicon wafers of small structures are placed between these HBM3 chips to ensure the stability of the cooling solution when tightened to the top of the package. The MI300's computing part consists of nine small chips based on TSMC's 5nm process, including CPU and GPU cores, but AMD did not provide details on each small chip. Since Zen 4 cores are typically deployed with eight cores, a 24-core CPU means that 3 small chips are CPU chips and the other 6 are GPU chips. GPU chips use AMD's CDNA 3 architecture, the third version of AMD's datacenter-specific graphics architecture. AMD has not yet specified the number of CUs, but official data shows that CDNA 3's AI performance per watt has reached five times that of the previous generation CDNA 2. The nine small chips are stacked on four small 6nm chips in a 3D package, and these chips are not just passive intermediaries – these chips are active and can handle I/O and various other functions. AMD claims that the MI300 offers 8x, 5x performance per watt of Instinct MI250's AI performance (using the sparsity FP8 benchmark). AMD says it can save millions of dollars in electricity by reducing the training time of very large AI models like ChatGPT and DALL-E from months to weeks. AMD says these MI300 chips for supercomputers will be expensive and relatively rare — these are not high-volume products, so they won't be as widely deployed as EPYC Genoa CPUs. However, technology will filter into multiple variants in different form factors. The chip will also compete with NVIDIA's Grace Hopper Superchip, which integrates Hopper GPUs and Grace CPUs on the same substrate. The chips are expected to be available this year. The Neoverse-based Grace CPU is based on Arm v9 Instruction Set Architecture and comes with two chips that blend with Nvidia's new NVLink-C2C interconnect technology. AMD's approach is designed to provide superior throughput and energy efficiency, as combining these devices into a single package often allows higher throughput between cells than when connecting two separate devices. The MI300 will also compete with Intel's Falcon Shores, which will feature varying numbers of computing modules, including x86 cores, GPU cores, and memory, with a dizzying array of possible configurations, but these won't arrive until 2024. AMD expects to deliver the Instinct MI300 in the second half of 2023, when the El Capitan supercomputer will debut the MI300, which is expected to be the fastest supercomputer in the world. It is worth mentioning that Intel and Argonne National Laboratory are also deploying the Aurora supercomputer, which is based on Intel's Ponte Vecchio data center graphics card with more than 100 billion transistors.
The LDK220 series of low-dropout linear regulators (LDOs) is a high-performance device designed specifically for low-power consumption and high-precision voltage regulation, widely used in scenarios such as consumer electronics, industrial control, and po
Xilinx Spartan®-7 FPGA Family stands as a defining solution in the mid-range FPGA landscape, blending high performance, energy efficiency, and cost-effectiveness to redefine versatility for industrial, IoT, and consumer electronics applications. Built on
The Altera FLEX series was more than a lineup of FPGAs—it was a blueprint for how programmable logic devices could evolve to meet diverse market needs. The FLEX 8000 laid the architectural groundwork, the FLEX 10K redefined functionality with embedded mem
Among TI standout offerings, the LM4765 and LM4766 are dual-channel amplifiers designed to cater to diverse audio needs—from compact setups to high-fidelity systems. While sharing the same product lineage, these chips differ significantly in power output,
