AMD has launched its MI325x accelerators designed for demanding AI tasks, significantly enhancing performance and efficiency compared to competitors in the AI training space.
During the Advancing AI conference, AMD emphasized that the MI325X accelerators feature 256GB HBM3E memory, boasting an impressive 6TB/s bandwidth, and provide industry-leading capabilities.
The MI325X accelerators aim for high performance in LLM inference, with AMD claiming up to 1.3x better performance on the Mistral 7B model compared to rival products.
AMD’s timeline for MI325X production shipments is set for Q4 2024, with plans for widespread availability from major platform providers beginning in Q1 2025.
[
Collection
]
[
|
...
]