本帖最後由 s84292 於 2024-5-6 08:58 編輯
回覆 s84292
Lunar lake 總和 100TOPS (NPU 40T)
唔通都是靠 INT4 跑出來
usei 發表於 2024-5-6 08:44
Lunar lake 講40tops個發佈會,
intel 冇寫拎咩格式,冇回應媒體使用int8 int4 或者混合精度跑出黎的問題
基本上都幾肯定係int4, 畢竟佢都叫npu 4.0
依家intel core npu同內顯只支持int8,但arc a個xmx 已經支持int2同4
下一代npu 加返int4同內顯升級做arc a又可以話自己幾倍性能
Intel did not specify the data type it used to calculate Lunar Lake's TOPS count.
Most orgs in this space default to INT8, but we've asked Intel to clarify its numbers. INT4 trades more TOPS for less accuracy, but using INT4 to measure the NPU's performance wouldn't necessarily be a bad thing – it would lower memory requirements to run local AI models. But it would mean the performance comparison of Meteor Lake and Lunar Lake is not really apples to apples. |