At the 2023 US Consumer Electronics Show (CES), AMD brought a new product “gift package”, ranging from CPU to GPU, from mobile version to desktop version, including Ryzen 7000 series mobile processors, Ryzen 7000 X3D series Desktop CPU, mobile version RX7000 independent display and AMD’s most complex chip so far – Instinct MI300, etc.
PS: The 2023 Las Vegas Consumer Electronics Show (CES) is hosted by the Consumer Technology Association (CEA). It was founded in 1967 and has a history of 50 years. Held in Las Vegas, it is the world’s largest and most influential consumer electronics technology annual exhibition, and it is also the world’s largest consumer technology industry event.
As the world’s leading provider of computer and communication solutions, AMD will continue to launch a series of innovative products to help customers meet the challenges faced by the artificial intelligence era. These new technologies on display are all from AMD. AMD said it is developing the next generation of cores. This also means that more self-developed chips will be used in the future.
AMD’s most complex chip ever
Instinct MI300 is AMD’s first data center/HPC-class APU. CEO Lisa Su called it “AMD’s most complex chip to date”, with a total of 146 billion transistors. Compared with InstinctMI250X, InstinctMI300 can improve AI training by 8 times Computing power and 5 times AI energy efficiency.
AMD said that this series of processors has the advantages of low power consumption, high flexibility and low latency, and can meet customers’ growing demand for artificial intelligence applications. The processor is called “the core of artificial intelligence on a high-performance computer platform.”
What is the concept of 146 billion transistors? Intel’s server GPU Ponte Vecchio integrates 100 billion transistors, while Nvidia’s new nuclear bomb H100 has 80 billion transistors. It is worth noting that the Instinct MI300 adopts the currently hot advanced packaging technology – Chiplet, which uses 3D packaging technology to integrate the CPU and accelerated computing unit.
Disruptive Innovation – Chiplet
Chiplet is Chiplet in English, which refers to a chip (Die) that is prefabricated, has specific functions, and can be combined and integrated. Chiplet is also translated as “small chip”. Core grains”. Small chip technology is actually packaging chips with different functions into modules, and each small chip is responsible for the corresponding function, thus forming a complete SOC.
Compared with “big chips”, the yield rate of small chips is higher, and different chips can choose different processes, and the cost is lower. Small chips have broad application prospects. Small chips can be widely used in consumer electronics products, communication products, automotive electronic systems, etc. In general, chiplet technology is a “modular” design method for chips and a packaging technology for heterogeneous integration.
Instinct MI300 king fried debut
Instinct MI300 stacks nine 5nm computing chips on top of four 6nm chips, surrounded by HBM3 memory. All of these make Instinct’s computing power reach the world’s top level. Inst Inct MI300 will be equipped with Nvidia’s new GPU platform. The new processors are based on Intel’s Xeon Sandy architecture. It can support up to 10 times the processing power of the CPU, and can achieve faster computing speed.
Intel said that in future planning, it is expected to become one of the next-generation mainstream computer platforms. This is the first time Intel has released a new processor. This also means that while the Instinct M300 has the highest speed (600,000 Gb/s) and the fastest capacity (each processor can hold 16GB of data storage), the cost is only about half of that of a processor of the same size. On this basis, its power consumption is only 1/10 of the traditional Intel processor. This is the current mainstream trend of Intel. In terms of performance, it has improved by leaps and bounds.
The Instinct MI300 is expected to be delivered in the second half of 2023. It will first be deployed on the new generation of supercomputer El Capitan in the United States. In addition, according to Tom’s Hardwre, AMD also revealed that Instinct MI300 can shorten the training time of large models such as ChatGPT and DALL•E from months to weeks.
Intel has accumulated a certain amount in the GPU field, but further efforts are needed for the CPU. Hope to get a breakthrough soon. Intel announced as early as February 2021 that it plans to invest US$3 billion in high-performance computing business. Intel CEO Krzanich said: “We are working hard to develop a powerful and stable core.”
In addition to Instinct MI300, there is Genoa-X CPU
AMD CTO Papermaster also revealed that AMD is expected to launch the EPYC Bergamo processor, which competes with Arm servers such as the AWS Graviton series, in the first half of 2023, using the Zen 4C architecture. AMD EPYC Bergamo processors will feature up to 128 cores and will be optimized for HBM-powered Xeon chips, as well as server products from Apple, Amazon and Google, using the SP5 socket, optimized for higher-throughput workloads.
Additionally, the AMD Genoa-X CPU, expected to be in production late Q3/early Q1 2023 and launched around mid-2023, will feature a similar design method. The Genoa-X CPU will be equipped with more than 1GB of L3 cache and up to 96 cores based on the Zen 4 architecture.
AMD’s most complex chip ever
Instinct MI300 is AMD’s first data center/HPC-class APU. CEO Lisa Su called it “AMD’s most complex chip to date”, with a total of 146 billion transistors. Compared with InstinctMI250X, InstinctMI300 can improve AI training by 8 times Computing power and 5 times AI energy efficiency.
AMD said that this series of processors has the advantages of low power consumption, high flexibility and low latency, and can meet customers’ growing demand for artificial intelligence applications. The processor is called “the core of artificial intelligence on a high-performance computer platform.”
What is the concept of 146 billion transistors? Intel’s server GPU Ponte Vecchio integrates 100 billion transistors, while Nvidia’s new nuclear bomb H100 has 80 billion transistors. It is worth noting that the Instinct MI300 adopts the currently hot advanced packaging technology – Chiplet, which uses 3D packaging technology to integrate the CPU and accelerated computing unit.
Disruptive Innovation – Chiplet
Chiplet is Chiplet in English, which refers to a chip (Die) that is prefabricated, has specific functions, and can be combined and integrated. Chiplet is also translated as “small chip”. Core grains”. Small chip technology is actually packaging chips with different functions into modules, and each small chip is responsible for the corresponding function, thus forming a complete SOC.
Compared with “big chips”, the yield rate of small chips is higher, and different chips can choose different processes, and the cost is lower. Small chips have broad application prospects. Small chips can be widely used in consumer electronics products, communication products, automotive electronic systems, etc. In general, chiplet technology is a “modular” design method for chips and a packaging technology for heterogeneous integration.
Instinct MI300 king fried debut
Instinct MI300 stacks nine 5nm computing chips on top of four 6nm chips, surrounded by HBM3 memory. All of these make Instinct’s computing power reach the world’s top level. Inst Inct MI300 will be equipped with Nvidia’s new GPU platform. The new processors are based on Intel’s Xeon Sandy architecture. It can support up to 10 times the processing power of the CPU, and can achieve faster computing speed.
Intel said that in future planning, it is expected to become one of the next-generation mainstream computer platforms. This is the first time Intel has released a new processor. This also means that while the Instinct M300 has the highest speed (600,000 Gb/s) and the fastest capacity (each processor can hold 16GB of data storage), the cost is only about half of that of a processor of the same size. On this basis, its power consumption is only 1/10 of the traditional Intel processor. This is the current mainstream trend of Intel. In terms of performance, it has improved by leaps and bounds.
The Instinct MI300 is expected to be delivered in the second half of 2023. It will first be deployed on the new generation of supercomputer El Capitan in the United States. In addition, according to Tom’s Hardwre, AMD also revealed that Instinct MI300 can shorten the training time of large models such as ChatGPT and DALL•E from months to weeks.
Intel has accumulated a certain amount in the GPU field, but further efforts are needed for the CPU. Hope to get a breakthrough soon. Intel announced as early as February 2021 that it plans to invest US$3 billion in high-performance computing business. Intel CEO Krzanich said: “We are working hard to develop a powerful and stable core.”
In addition to Instinct MI300, there is Genoa-X CPU
AMD CTO Papermaster also revealed that AMD is expected to launch the EPYC Bergamo processor, which competes with Arm servers such as the AWS Graviton series, in the first half of 2023, using the Zen 4C architecture. AMD EPYC Bergamo processors will feature up to 128 cores and will be optimized for HBM-powered Xeon chips, as well as server products from Apple, Amazon and Google, using the SP5 socket, optimized for higher-throughput workloads.
Additionally, the AMD Genoa-X CPU, expected to be in production late Q3/early Q1 2023 and launched around mid-2023, will feature a similar design method. The Genoa-X CPU will be equipped with more than 1GB of L3 cache and up to 96 cores based on the Zen 4 architecture.