News
The fight against Nvidia still depends on AMD's big move next week: the monster level graphics card MI300 is coming
Release time:
2023-07-19 17:28
June 6 news, now in the graphics card market, whether it is a game card or an acceleration card, NVIDIA is an extremely strong existence, AMD's share is constantly compressed, but it is only they who can fight against N cards, fortunately, at next week's data center product conference, AMD will come up with a graphics card that really makes NVIDIA feel pressure.
AMD will hold a special conference on June 6 (the early morning of the 13th in China), the main products are for data centers, and it is expected that there will be new EPYC products, including the 14-core Bergamo, cache-enhanced Genoa-X processor.
The graphics card will clarify the launch date of the Instinct MI300 released last year, which is expected to be in the second half of the year, the earliest autumn launch, and this product can compete with NVIDIA's high-performance computing card H100.
At present, NVIDIA has no rivals in the field of high-performance computing and AI, A100, H100 and other popular AI cards are in short supply, the delivery period is as long as half a year, the price has even increased by 40%, the original price of H100 is 25,<> yuan, and now it has risen to I don't know what the price, anyway, some companies have to grab how much money they want, the AI track is too hot.
MI300 graphics card is AMD's killer against NVIDIA, it adopts multi-chip, multi-IP integrated packaging design, 5nm advanced manufacturing process, the number of transistors is as high as 1460 billion!
It also integrates CDNA3 GPU units (the exact number of cores is not disclosed), 4 CPU cores of Zen24 architecture, large-capacity Infinity Cache unlimited cache, and HBM8192 high-bandwidth memory with 128-bit bit width and 3GB capacity.
In terms of technology, it supports the fourth-generation Infinity Fabric bus, CXL 3.0 bus, unified memory architecture, and new mathematical computing formats, claiming that AI performance is up to 8 times higher than the previous generation, which can meet the needs of tens of billions of calculations.
Recommended News

