AMD Instinct MI300X
芯片
GPU
发货中
AMD's serious challenge to Hopper — more memory than H100, adopted by Microsoft and Meta.
您正在查看一条芯片记录：该处理器是什么、其关键规格、谁设计它、谁制造它、谁供应其零部件，以及它被用在哪里。
The MI300X stacked 8 CDNA3 compute chiplets on 4 base dies — 153 billion transistors, more than any GPU before it — with 192 GB of HBM3 when the H100 had 80. That memory advantage won real deployments: Azure ran GPT-4-class serving on it, Meta served Llama on it, and it proved the ROCm software stack at scale. The knock: software maturity, not silicon, remained AMD's gap.
规格与关键参数
每个数据均标注来源依据：厂商规格、厂商声明、公开报道、估算或官方数据。"未公开披露"即字面含义——我们从不捏造数值。
Overview
Designer
AMD
Architecture
CDNA 3, chiplet (8 XCD + 4 IOD)
Announced
Launched December 2023
Status
Shipping
Manufacturing
Foundry / process
TSMC N5 compute dies on N6 base diesVendor spec
Transistors
153 billionVendor spec
Packaging
CoWoS + SoIC 3D hybrid bondingVendor spec
内存
Memory
192 GB HBM3 (8 stacks)Vendor spec
Memory bandwidth
5.3 TB/sVendor spec
算力
BF16/FP16 (dense)
1,307 TFLOPSVendor spec
FP8 (dense)
2,615 TFLOPSVendor spec
FP64 matrix
163.4 TFLOPSVendor spec
电力
TDP
750 WVendor spec
Physical
Form factor
OAM; 8-GPU UBB platformVendor spec
Interconnect
Infinity Fabric — 896 GB/s aggregate per GPUVendor spec
价格
Price
Reported meaningfully below H100 street pricing; AMD does not publish list pricesReported
它如何与整个技术栈相连
本记录的关联链接一览——点击分组可直接跳转。 +6 更多分组见下方。
设计方
由……制造
内存技术
内存供应商
支持方
后继产品
前代产品
竞品
来源
记录审阅截至 2026-01-15 — 查看 方法论.