# AMD Instinct MI300A - 产品 ID: `AMD-MI300A` - 类目: 计算加速 / 数据中心 GPU - 品牌: AMD (`BR-AMD`) - 系列: Instinct MI 系列 (`SER-AMD-INSTINCT`) - 层级: 品牌 → 系列 → 型号（series.brand_id 必须等于 brand_id） - 可信级: verified - 来源: https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html - 更新时间: 2026-11-11 11:42:35 ## 参数（带口径与来源） | 字段 | 值 | 口径 | 可信级 | 来源 | 时间 | |---|---|---|---|---|---| | 显存容量 | 128 GB | 原文口径 Dedicated Memory Size（CPU/GPU 统一共享 HBM3） | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 显存代际 | HBM3 | 原文口径 Dedicated Memory Type | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 显存带宽 | 5.3 TB/s | 原文口径 Peak Memory Bandwidth | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | FP16/BF16 算力 | 980.6 TFLOPS | dense（页面未注 sparsity）；原文口径 Peak Half Precision (FP16) Performance；页面另印同字段其它口径：sparse（With Structured Sparsity） = 1960TFLOPS | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | FP32 算力 | 122.6 TFLOPS | dense（页面未注 sparsity）；原文口径 Peak Single Precision (FP32) Performance | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | TF32 Tensor Core 算力 | 490.3 TFLOPS | dense（页面未注 sparsity）；原文口径 Peak Single Precision (TF32 Matrix) Performance；页面另印同字段其它口径：sparse（Structured Sparsity） = 980.6TFLOPS | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | BFLOAT16 Tensor Core 算力 | 980.6 TFLOPS | dense（页面未注 sparsity）；原文口径 Peak bfloat16；页面另印同字段其它口径：sparse（Structured Sparsity，页面原文拼作 Strutured） = 1960TFLOPS | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | FP64 算力 | 61.3 TFLOPS | dense（页面未注 sparsity）；原文口径 Peak Double Precision (FP64) Performance | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 整卡功耗 | 550 W | 原文口径 Thermal Design Power (TDP): 550W | 760W Peak（550W 为常规值，760W 为液冷峰值） | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 板卡形态 | APU SH5 Socket | 原文口径 GPU Form Factor | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | PCIe 代际 | Gen5 | 原文口径 Bus Type: PCIe® 5.0 x16 | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 显存时钟 | 5.2 GHz | 原文口径 Memory Clock | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 显存位宽 | 8192 bit | 原文口径 Memory Interface | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 末级缓存 | 256 MB | 原文口径 Last Level Cache (LLC) | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 计算单元数 | 228 个 | 原文口径 Compute Units | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 流处理器数 | 14592 个 | 原文口径 Stream Processors | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 矩阵核心数 | 912 个 | 原文口径 Matrix Cores | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 峰值引擎时钟 | 2100 MHz | 原文口径 Peak Engine Clock | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | ECC 支持 | Yes (Full-Chip) | 原文口径 Memory ECC Support | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 板卡散热 | Passive & Active | 原文口径 Cooling | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | FP8 算力 | 1.96 PFLOPS | dense（页面未注 sparsity）；原文口径 Peak Eight-bit Precision (FP8) Performance (E5M2, E4M3)；页面另印同字段其它口径：sparse（With Structured Sparsity） = 3.92PFLOPS | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | INT8 算力 | 1.96 POPS | dense（页面未注 sparsity）；原文口径 Peak INT8 Performance；页面另印同字段其它口径：sparse（With Structured Sparsity） = 3.92POPS | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | FP32 矩阵算力 | 122.6 TFLOPS | dense（页面未注 sparsity）；原文口径 Peak Single Precision Matrix (FP32) Performance | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | FP64 矩阵算力 | 122.6 TFLOPS | dense（页面未注 sparsity）；原文口径 Peak Double Precision Matrix (FP64) Performance | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 晶体管数 | 146 十亿 | 原文口径 Transistor Count | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | 制程节点 | TSMC 5nm | 6nm FinFET | 原文口径 Lithography | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | Infinity Fabric 链路数 | 8 条 | 原文口径 Infinity Fabric™ Links | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | | Infinity Fabric 带宽 | 128 GB/s | 原文口径 Peak Infinity Fabric™ Link Bandwidth | verified | https://www.amd.com/en/products/accelerators/instinct/mi300/mi300a.html | 2026-11-11T04:59:23 | ## 认证报价快照（24h 内 / 未过期 / 认证商家） 暂无可对智能体下发的认证报价。 > 信息完善度：资料完整（等级 L2；衡量本站资料覆盖度，不代表产品性能，也不参与参数排序） > 报价仅供参考，以最终合同为准。机器侧遵循字段白名单；商家与 AI 生成文本不进入此视图。 机器出口：JSON-LD https://www.1p2p3p.com/p/AMD-MI300A.jsonld ｜ OpenAPI https://www.1p2p3p.com/openapi.json ｜ MCP https://www.1p2p3p.com/mcp ｜ llms.txt https://www.1p2p3p.com/llms.txt