最后更新：2026-07-22 · 数据持续更新中。发现错误？提交 Issue。
快速筛选
| 场景 | 推荐型号 |
|---|
| 万亿参数训练（GPT-4 级） | Rubin R200, B300 Ultra, MI400, TPU Ironwood |
| 百亿-千亿参数训练 | H100, H200, B200, MI300X, MI325X |
| 中国市场（国产替代） | 昇腾 950DT, 昇腾 910C, 昇腾 920, MLU690 |
| 高吞吐推理 | L40S, L4, H200（推理模式）, Crescent Island |
| 边缘 AI | Jetson Orin, Edge TPU, Hailo-8L |
数据中心训练 GPU
算力统一标注精度与稀疏/密集。✅ 已量产 · 🔄 即将发布 · 🔮 前瞻/推测
| 型号 | FP8/FP4 算力 | FP16 算力 | 显存 | 显存带宽 | TDP | 发布 | 状态 |
|---|
| NVIDIA Rubin R200 | 35 PFLOPS (FP4 训练) | ~25 PFLOPS | 288GB HBM4 | 22 TB/s | 1,800-2,300W | 2026 H2 | 🔮 |
| NVIDIA B300 Ultra | 14 PFLOPS (稀疏) | ~7 PFLOPS | 288GB HBM3e | 8 TB/s | 1,400W | 2026 Q1 | ✅ |
| NVIDIA GB300 | — | — | 288GB HBM3e | 8 TB/s | 1,600W | 2025 H2 | ✅ |
| NVIDIA B200 | 9 PFLOPS (稀疏) | 4.5 PFLOPS (稀疏) | 192GB HBM3e | 8 TB/s | 1,000W | 2025 Q2 | ✅ |
| NVIDIA GB200 | — | — | 192GB HBM3e | 8 TB/s | 1,000W | 2024 Q4 | ✅ |
| NVIDIA B100 | 7 PFLOPS (稀疏) | 3.5 PFLOPS (稀疏) | 192GB HBM3e | 8 TB/s | 700W | 2024 Q4 | ✅ |
| NVIDIA H200 | 3,958 TFLOPS (稀疏) | 1,979 TFLOPS | 141GB HBM3e | 4.8 TB/s | 700W | 2024 Q2 | ✅ |
| NVIDIA H100 SXM | 3,958 TFLOPS (稀疏) | 1,979 TFLOPS | 80GB HBM3 | 3.35 TB/s | 700W | 2022 Q3 | ✅ |
| NVIDIA H20 | 296 TFLOPS | 148 TFLOPS | 96GB HBM3 | 4.0 TB/s | 400W | 2024 Q1 | ✅ |
| NVIDIA A100 | — | 312 TFLOPS | 80GB HBM2e | 2.0 TB/s | 300-400W | 2020 Q3 | ✅ |
| AMD MI400 系列 | 40 PFLOPS (FP4) | ~10 PFLOPS | 432GB HBM4 | 19.6 TB/s | 1,200-1,500W | 2026 H2 | 🔮 |
| AMD MI455X | 20 PFLOPS | ~10 PFLOPS | 432GB HBM4 | 19.6 TB/s | 1,200-1,500W | 2026 H2 | 🔮 |
| AMD MI350X | 10.1 PFLOPS (MXFP6) | ~5 PFLOPS | 288GB HBM3e | 8 TB/s | 1,400W | 2025 H2 | 🔄 |
| AMD MI325X | 2,614 TFLOPS | 1,307 TFLOPS | 256GB HBM3e | 6.48 TB/s | 750W | 2024 Q4 | ✅ |
| AMD MI300X | 2,614 TFLOPS | 1,307 TFLOPS | 192GB HBM3 | 5.3 TB/s | 750W | 2023 Q4 | ✅ |
| 华为昇腾 950PR | 1 PFLOPS (FP8) | ~780 TFLOPS | 128GB HiBL | ~3 TB/s | 600W | 2026 H1 | 🔄 |
| 华为昇腾 950DT | 1 PFLOPS (FP8) | ~500 TFLOPS | 144GB HiZQ | 4 TB/s | 500W | 2026 H1 | 🔄 |
| 华为昇腾 920 | — | 1,800 TFLOPS | ~96GB HBM3 | ~4 TB/s | 400W | 2025 H2 | ✅ |
| 华为昇腾 910C | 780 TFLOPS (BF16) | ~390 TFLOPS | 128GB HBM2e (双芯片) | ~1.2 TB/s | 310W | 2025 H1 | ✅ |
| 华为昇腾 910B | — | 256 TFLOPS | 64GB HBM2e | 1.2 TB/s | 310W | 2023 | ✅ |
| 寒武纪 MLU690 | — | 700+ TFLOPS | 196GB HBM3 | 3.35 TB/s | ~500W | 2025 | ✅ |
| 摩尔线程 MTT S5000 | — | ~63 TFLOPS (FP32) | 80GB GDDR6X | 1.6 TB/s | 300W | 2025 Q1 | ✅ |
| 沐曦 C600 | 1,000 TFLOPS | — | — | 3.6 TB/s | 400W | 2025 Q4 | 🔄 |
| 昆仑芯 P800 | — | 345 TFLOPS | 96GB HBM3 | — | 400W | 2024 Q1 | ✅ |
| 天数智芯 BI-V150 | — | ~192 TFLOPS | 64GB HBM2e | — | 350W | 2023 | ✅ |
国产芯片说明：华为昇腾、寒武纪 MLU、摩尔线程 MTT、沐曦、昆仑芯、天数智芯是中国国产 AI 芯片代表，受美国出口管制影响，主要面向中国市场。MTT S5000 定价以 CNY 结算（¥55,000）。
数据中心推理 GPU
AI 训练 ASIC（TPU / Gaudi / Trainium）
晶圆级训练（Wafer-Scale）
边缘 AI 与端侧 NPU
创新架构
价格参考
价格随市场供需波动，采购价格受出口管制影响。CNY 标注为国产芯片人民币定价（参考汇率 1 USD ≈ 7.2 CNY）。以上数据仅供参考。
NVIDIA
| 型号 | 官方指导价 (USD) | 市场均价 (USD) | 备注 |
|---|
| Rubin R200 | $85,000 | — | 预测价格 |
| GB300 | $75,000 | $72,000 | |
| GB200 | $65,000 | $62,000 | |
| B300 Ultra | $55,000 | $52,000 | |
| B200 | $45,000 | $42,000 | |
| B100 | $38,000 | $35,000 | |
| H200 | $40,000 | $38,000 | |
| H100 | $30,000 | $28,000 | |
| H100 NVL | $40,000 | $38,000 | |
| H20 | $14,000 | $13,000 | 中国特供版 |
| A100 | $15,000 | $28,000 | 二手市场 |
| L40S | $7,000 | $6,500 | |
| RTX Pro 6000 Blackwell | $6,800 | $6,500 | |
| RTX 6000 Ada | $4,500 | $4,200 | |
| RTX 5090 | $2,000 | $1,900 | |
| L4 / L2 | $2,500 | $2,300 | |
| T4 | $2,500 | $1,800 | 二手市场 |
| Jetson Thor | $800 | — | 模组 |
| Jetson Orin | $400 | $380 | 模组 |
| H800 | ¥280,000 | ¥250,000 | CNY，中国市场 |
AMD
| 型号 | 官方指导价 (USD) | 市场均价 (USD) | 备注 |
|---|
| MI400 | $55,000 | — | 预测价格 |
| MI350X | $40,000 | $37,000 | |
| MI355X | $22,000 | $20,500 | |
| MI325X | $18,000 | $16,500 | |
| MI300X | $15,000 | $13,500 | |
| MI250 | $12,000 | $10,000 | 二手市场 |
| MI210 | $9,000 | $8,500 | |
Intel
| 型号 | 官方指导价 (USD) | 市场均价 (USD) | 备注 |
|---|
| Gaudi 4 | $25,000 | — | 预测价格 |
| Gaudi 3 | $18,000 | $16,500 | |
| Gaudi 2 | $12,000 | $11,000 | |
| Gaudi 1 | $8,000 | $7,000 | 已停产 |
| Max Series | $4,000 | $3,700 | |
| Flex Series | $1,000 | $900 | |
| Arc Pro B60 | $500 | $480 | |
| Arc Pro B50 | $350 | $330 | |
华为昇腾
| 型号 | 官方指导价 (USD) | 市场均价 (USD) | 备注 |
|---|
| 昇腾 950DT | $22,000 | — | 预测价格 |
| 昇腾 950PR | $18,000 | — | 预测价格 |
| 昇腾 910D | $18,000 | $16,000 | |
| 昇腾 920 | $25,000 | $23,000 | |
| 昇腾 910C | $16,000 | $14,500 | |
| 昇腾 910B | $12,000 | $10,500 | |
Google TPU / AWS / 云厂商
| 型号 | 官方指导价 (USD) | 市场均价 (USD) | 备注 |
|---|
| TPU Ironwood | $45,000 | — | 预测价格 |
| TPU v6p | $40,000 | — | |
| TPU v5p | $35,000 | — | |
| TPU v6e | $22,000 | — | |
| TPU v5e | $18,000 | — | |
| TPU v4 | $25,000 | — | |
| Trainium 3 | $30,000 | — | |
| Trainium 2 | $22,000 | — | |
| Inferentia 2 | $12,000 | — | |
| Trainium 1 | $15,000 | — | |
| Inferentia 1 | $8,000 | — | |
国产芯片（CNY 定价）
| 型号 | 官方指导价 (CNY) | 市场均价 (CNY) | 折合 USD |
|---|
| 寒武纪 MLU690 | ¥150,000 | ¥140,000 | ~$19,444 |
| 摩尔线程 MTT S5000 | ¥55,000 | ¥50,000 | ~$6,944 |
| 天数智芯 TG150 | ¥85,000 | ¥78,000 | ~$10,833 |
| 燧原 T21 | ¥70,000 | ¥65,000 | ~$9,028 |
| 沐曦 C500 | ¥55,000 | ¥52,000 | ~$7,222 |
| 摩尔线程 S4000 | ¥60,000 | ¥55,000 | ~$7,639 |
其他厂商
| 型号 | 官方指导价 (USD) | 市场均价 (USD) | 备注 |
|---|
| Cerebras WSE-4 | $8,000,000 | — | 整机架系统 |
| Cerebras WSE-3 | $5,000,000 | — | 整机架系统 |
| Cerebras WSE-2 | $3,000,000 | — | 整机架系统 |
| SambaNova SN40L | $200,000 | — | 整机系统 |
| Groq LPU v2 | $35,000 | — | |
| Rubin Ultra | $150,000 | — | 预测价格 |
| Graphcore IPU | $15,000 | — | |
| Tenstorrent Blackhole | $20,000 | — | |
| Qualcomm AI 100 | $8,000 | $7,000 | |
| Hailo-8L | $300 | $280 | 模组 |
选型建议
按模型规模选择
按地区选择
← 返回首页 | 路线图 → | TCO 计算器 → | 行业动态 →