AMD offers a major viable and more cost-effective solution to meet users' reasoning needs. The fourth generation of AMD EPYC 9004 series processors, codenamed Genoa, implements a core architecture upgrade that increases the total number of instructions per clock cycle by approximately 14 per cent. In addition, the EPYC 9004 series achieves up to 96 cores and 192 threads, all of which are full-sized cores, while offering different core counts of 84/64/48/32/24/16 to choose from. The large number of cores, coupled with synchronous multi-threading technology, can simultaneously handle data inference requests from thousands of sources, greatly improving efficiency.
The AI inference process is closely integrated with the enterprise business, which requires high-speed, frequent reading and processing of the corresponding data, which requires the chip to have a high-frequency computing power, with the help of PCIe Gen 5 I/O, the processor's I/O throughput has doubled, which enables faster data acquisition, accelerating the speed of AI reasoning. The AMD EPYC processor is equipped with DDR5 memory, with a total memory throughput of 2.25 times that of AMD's previous generation, which increases data transfer rates and reduces the latency of task response. EPYC processors also support AI-optimised instruction sets such as AVX-512 extensions, BF16 data types, etc., which are dedicated to AI computation, making AI reasoning even more efficient.
AMD EPYC processor has become one of the solutions to the current dilemma. In the face of the continuous escalation of AI arithmetic demand, AMD is not only accelerating the upgrading of its products to solve the arithmetic problems faced by AI reasoning, but also joining hands with domestic head server vendors such as Xinhuanet San, to further unleash the value of the AMD EPYC processor to the whole process of AI training + reasoning, so that more enterprises can meet the big models brought by infinite opportunities brought by big models.