Skip to content

GitLab

"git@developer.sourcefind.cn:gaoqiong/flash-attention.git" did not exist on "5badfb78485adf1333f04c46510d99ac56a17622"
Commit 24b257f1 authored by sunzhq2's avatar sunzhq2
Browse files

init

parent 920b3c0f
Hide whitespace changes
Inline Side-by-side
Showing with 638 additions and 0 deletions
ByteMLPerf/vendor_zoo/Moffett/S30.json 0 → 100644
View file @ 24b257f1
{
"SKU参数": {
"厂商": "Moffett",
"型号": "S30",
"用途": "Inference",
"照片": "vendor_zoo/Moffett/image_9.png"
},
"硬件参数": {
"制程(NM)": 12,
"尺寸": "FHFL, Dual Slot Card",
"接口": "PCIe 4.0x16",
"功耗(W/TDP)": 250
},
"内存参数": {
"内存层次架构图": "vendor_zoo/Moffett/mem_9.png",
"内存": {
"内存类型": "LPDDR4x",
"内存容量(GB)": 60,
"内存带宽(GB/s)": 246
},
"一级缓存": {
"缓存类型": "Distributed Buffer(x12)",
"缓存容量(MB)": null,
"缓存带宽(TB/s)": null
},
"二级缓存": {
"缓存类型": null,
"缓存容量(MB)": null,
"缓存带宽(TB/s)": null
}
},
"算力参数": {
"PE层次架构图": "vendor_zoo/Moffett/pe_9.png",
"PE参数": {
"算力架构": "异构多核",
"并行方式": "SIMD/MIMT",
"通信带宽(GB/s)": null
},
"标量参数": {
"标量精度": null,
"INT8标量算力(TFLOPS)": null,
"FP16标量算力(TFLOPS)": null,
"FP32标量算力(TFLOPS)": null
},
"向量参数": {
"向量精度": null,
"INT8量算力(TFLOPS)": null,
"FP16向量算力(TFLOPS)": null,
"FP32向量算力(TFLOPS)": null
},
"张量参数": {
"张量精度": null,
"INT8张量算力(TFLOPS)": null,
"FP16张量算力(TFLOPS)": null,
"FP32张量算力(TFLOPS)": null
}
},
"卡间通信参数": {
"通信方式": null,
"端口数量": null,
"RDMA协议": null,
"下行带宽(GB/s)": null,
"上行带宽(GB/s)": null
}
}
\ No newline at end of file
ByteMLPerf/vendor_zoo/Moffett/S4.json 0 → 100644
View file @ 24b257f1
{
"SKU参数": {
"厂商": "Moffett",
"型号": "S4",
"用途": "Inference",
"照片": "vendor_zoo/Moffett/image_7.png"
},
"硬件参数": {
"制程(NM)": 12,
"尺寸": "FHFL, Single Slot Card",
"接口": "PCIe 3.0x16",
"功耗(W/TDP)": 70
},
"内存参数": {
"内存层次架构图": "vendor_zoo/Moffett/mem_7.png",
"内存": {
"内存类型": "LPDDR4x",
"内存容量(GB)": 20,
"内存带宽(GB/s)": 82
},
"一级缓存": {
"缓存类型": "Distributed Buffer(x4)",
"缓存容量(MB)": 1.8,
"缓存带宽(TB/s)": null
},
"二级缓存": {
"缓存类型": null,
"缓存容量(MB)": 82,
"缓存带宽(TB/s)": 11.7
}
},
"算力参数": {
"PE层次架构图": "vendor_zoo/Moffett/pe_7.png",
"PE参数": {
"算力架构": "异构多核",
"并行方式": "SIMD/MIMT",
"通信带宽(GB/s)": 204.8
},
"标量参数": {
"标量精度": null,
"INT8标量算力(TFLOPS)": null,
"FP16标量算力(TFLOPS)": null,
"FP32标量算力(TFLOPS)": null
},
"向量参数": {
"向量精度": "BF16, INT8",
"INT8量算力(TFLOPS)": 3.2,
"FP16向量算力(TFLOPS)": null,
"FP32向量算力(TFLOPS)": null
},
"张量参数": {
"张量精度": "BF16, INT8",
"INT8张量算力(TFLOPS)": 943.6,
"FP16张量算力(TFLOPS)": null,
"FP32张量算力(TFLOPS)": null
}
},
"卡间通信参数": {
"通信方式": null,
"端口数量": null,
"RDMA协议": null,
"下行带宽(GB/s)": null,
"上行带宽(GB/s)": null
}
}
\ No newline at end of file
ByteMLPerf/vendor_zoo/NVIDIA/A10-PCIe.json 0 → 100644
View file @ 24b257f1
{
"SKU参数": {
"厂商": "NVIDIA",
"型号": "A10 PCIe",
"用途": "Training/Inference",
"照片": "vendor_zoo/NVIDIA/image_1.png"
},
"硬件参数": {
"制程(NM)": 8,
"尺寸": "FHFL, Single Slot Card",
"接口": "PCIe 4.0x16",
"功耗(W/TDP)": 150
},
"内存参数": {
"内存层次架构图": null,
"内存": {
"内存类型": "GDDR6",
"内存容量(GB)": 24,
"内存带宽(GB/s)": 600.2
},
"一级缓存": {
"缓存类型": "Cache(x72)",
"缓存容量(MB)": 9.126,
"缓存带宽(TB/s)": null
},
"二级缓存": {
"缓存类型": null,
"缓存容量(MB)": 6,
"缓存带宽(TB/s)": null
}
},
"算力参数": {
"PE层次架构图": null,
"PE参数": {
"算力架构": "同构众核",
"并行方式": "SIMT",
"通信带宽(GB/s)": null
},
"标量参数": {
"标量精度": null,
"INT8标量算力(TOPS)": null,
"FP16标量算力(TFLOPS)": null,
"FP32标量算力(TFLOPS)": null
},
"向量参数": {
"向量精度": " FP32",
"INT8量算力(TOPS)": null,
"FP16向量算力(TFLOPS)": null,
"FP32向量算力(TFLOPS)": 31.2
},
"张量参数": {
"张量精度": "TF32, FP16, BF16, INT8, INT4",
"INT4张量算力(TOPS)": 500.0,
"INT8张量算力(TOPS)": 250.0,
"BF16张量算力(TFLOPS)": 125.0,
"FP16张量算力(TFLOPS)": 125.0,
"TF32张量算力(TFLOPS)": 62.5,
"FP32张量算力(TFLOPS)": null
}
},
"卡间通信参数": {
"通信方式": null,
"端口数量": null,
"RDMA协议": null,
"下行带宽(GB/s)": null,
"上行带宽(GB/s)": null
},
"数据源": [
"https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a10/pdf/datasheet-new/nvidia-a10-datasheet.pdf",
"https://www.techpowerup.com/gpu-specs/a10-pcie.c3793"
]
}
\ No newline at end of file
ByteMLPerf/vendor_zoo/NVIDIA/A100-PCIe.json 0 → 100644
View file @ 24b257f1
{
"SKU参数": {
"厂商": "NVIDIA",
"型号": "A100 PCIe",
"用途": "Training/Inference",
"照片": "vendor_zoo/NVIDIA/image_3.png"
},
"硬件参数": {
"制程(NM)": 7,
"尺寸": "FHFL, Dual Slot Card",
"接口": "PCIe 4.0x16",
"功耗(W/TDP)": 300
},
"内存参数": {
"内存层次架构图": null,
"内存": {
"内存类型": "HBM2e",
"内存容量(GB)": 80,
"内存带宽(GB/s)": 1935
},
"一级缓存": {
"缓存类型": "Cache(x108)",
"缓存容量(MB)": 20.736,
"缓存带宽(TB/s)": null
},
"二级缓存": {
"缓存类型": null,
"缓存容量(MB)": 40,
"缓存带宽(TB/s)": null
}
},
"算力参数": {
"PE层次架构图": null,
"PE参数": {
"算力架构": "同构众核",
"并行方式": "SIMT",
"通信带宽(GB/s)": null
},
"标量参数": {
"标量精度": null,
"INT8标量算力(TOPS)": null,
"FP16标量算力(TFLOPS)": null,
"FP32标量算力(TFLOPS)": null
},
"向量参数": {
"向量精度": "FP64, FP32, TF32",
"INT8量算力(TOPS)": null,
"FP16向量算力(TFLOPS)": null,
"FP32向量算力(TFLOPS)": 19.5,
"FP64向量算力(TFLOPS)": 9.7
},
"张量参数": {
"张量精度": "FP64, FP32, TF32, FP16, BF16, INT8",
"INT8张量算力(TOPS)": 624.0,
"BF16张量算力(TFLOPS)": 312.0,
"FP16张量算力(TFLOPS)": 312.0,
"TF32张量算力(TFLOPS)": 156.0,
"FP32张量算力(TFLOPS)": null,
"FP64张量算力(TFLOPS)": 19.5
}
},
"卡间通信参数": {
"通信方式": "NV-Link",
"端口数量": null,
"RDMA协议": null,
"下行带宽(GB/s)": null,
"上行带宽(GB/s)": null
},
"数据源": [
"https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/nvidia-a100-datasheet-nvidia-us-2188504-web.pdf",
"https://www.techpowerup.com/gpu-specs/a100-pcie-80-gb.c3821"
]
}
\ No newline at end of file
ByteMLPerf/vendor_zoo/NVIDIA/A100-SXM4.json 0 → 100644
View file @ 24b257f1
{
"SKU参数": {
"厂商": "NVIDIA",
"型号": "A100 SXM4",
"用途": "Training/Inference",
"照片": "vendor_zoo/NVIDIA/image_4.png"
},
"硬件参数": {
"制程(NM)": 7,
"尺寸": "N/A",
"接口": "SXM",
"功耗(W/TDP)": 400
},
"内存参数": {
"内存层次架构图": null,
"内存": {
"内存类型": "HBM2e",
"内存容量(GB)": 80,
"内存带宽(GB/s)": 2039
},
"一级缓存": {
"缓存类型": "Cache(x108)",
"缓存容量(MB)": 20.736,
"缓存带宽(TB/s)": null
},
"二级缓存": {
"缓存类型": null,
"缓存容量(MB)": 40,
"缓存带宽(TB/s)": null
}
},
"算力参数": {
"PE层次架构图": null,
"PE参数": {
"算力架构": "同构众核",
"并行方式": "SIMT",
"通信带宽(GB/s)": null
},
"标量参数": {
"标量精度": null,
"INT8标量算力(TOPS)": null,
"FP16标量算力(TFLOPS)": null,
"FP32标量算力(TFLOPS)": null
},
"向量参数": {
"向量精度": "FP64, FP32, TF32",
"INT8量算力(TOPS)": null,
"FP16向量算力(TFLOPS)": null,
"FP32向量算力(TFLOPS)": 19.5,
"FP64向量算力(TFLOPS)": 9.7
},
"张量参数": {
"张量精度": "FP64, FP32, TF32, FP16, BF16, INT8",
"INT8张量算力(TOPS)": 624.0,
"BF16张量算力(TFLOPS)": 312.0,
"FP16张量算力(TFLOPS)": 312.0,
"TF32张量算力(TFLOPS)": 156.0,
"FP32张量算力(TFLOPS)": null,
"FP64张量算力(TFLOPS)": 19.5
}
},
"卡间通信参数": {
"通信方式": "NV-Link",
"端口数量": null,
"RDMA协议": null,
"下行带宽(GB/s)": null,
"上行带宽(GB/s)": null
},
"数据源": [
"https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/nvidia-a100-datasheet-nvidia-us-2188504-web.pdf",
"https://www.techpowerup.com/gpu-specs/a100-sxm4-80-gb.c3746"
]
}
\ No newline at end of file
ByteMLPerf/vendor_zoo/NVIDIA/A30-PCIe.json 0 → 100644
View file @ 24b257f1
{
"SKU参数": {
"厂商": "NVIDIA",
"型号": "A30 PCIe",
"用途": "Training/Inference",
"照片": "vendor_zoo/NVIDIA/image_2.png"
},
"硬件参数": {
"制程(NM)": 7,
"尺寸": "FHFL, Dual Slot Card",
"接口": "PCIe 4.0x16",
"功耗(W/TDP)": 165
},
"内存参数": {
"内存层次架构图": null,
"内存": {
"内存类型": "HBM2e",
"内存容量(GB)": 24,
"内存带宽(GB/s)": 933.1
},
"一级缓存": {
"缓存类型": "Cache(x56)",
"缓存容量(MB)": 10.752,
"缓存带宽(TB/s)": null
},
"二级缓存": {
"缓存类型": null,
"缓存容量(MB)": 24,
"缓存带宽(TB/s)": null
}
},
"算力参数": {
"PE层次架构图": null,
"PE参数": {
"算力架构": "同构众核",
"并行方式": "SIMT",
"通信带宽(GB/s)": null
},
"标量参数": {
"标量精度": null,
"INT8标量算力(TOPS)": null,
"FP16标量算力(TFLOPS)": null,
"FP32标量算力(TFLOPS)": null
},
"向量参数": {
"向量精度": "FP64, FP32",
"INT8量算力(TOPS)": null,
"FP16向量算力(TFLOPS)": null,
"FP32向量算力(TFLOPS)": 10.3,
"FP64向量算力(TFLOPS)": 5.2
},
"张量参数": {
"张量精度": "TF32, FP16, BF16, INT8, INT4",
"INT4张量算力(TOPS)": 661.0,
"INT8张量算力(TOPS)": 330.0,
"BF16张量算力(TFLOPS)": 165.0,
"FP16张量算力(TFLOPS)": 165.0,
"TF32张量算力(TFLOPS)": 82.0,
"FP32张量算力(TFLOPS)": null,
"FP64张量算力(TFLOPS)": 10.3
}
},
"卡间通信参数": {
"通信方式": "NV-Link",
"端口数量": null,
"RDMA协议": null,
"下行带宽(GB/s)": null,
"上行带宽(GB/s)": null
},
"数据源": [
"https://www.nvidia.com/content/dam/en-zz/Solutions/data-center/products/a30-gpu/pdf/a30-datasheet.pdf",
"https://www.techpowerup.com/gpu-specs/a30-pcie.c3792"
]
}
\ No newline at end of file
ByteMLPerf/vendor_zoo/NVIDIA/H100-PCIe.json 0 → 100644
View file @ 24b257f1
{
"SKU参数": {
"厂商": "NVIDIA",
"型号": "H100 PCIe",
"用途": "Training/Inference",
"照片": "vendor_zoo/NVIDIA/image_5.png"
},
"硬件参数": {
"制程(NM)": 4,
"尺寸": "FHFL, Dual Slot Card",
"接口": "PCIe 5.0x16",
"功耗(W/TDP)": 350
},
"内存参数": {
"内存层次架构图": null,
"内存": {
"内存类型": "HBM3",
"内存容量(GB)": 80,
"内存带宽(GB/s)": 2039
},
"一级缓存": {
"缓存类型": "Cache(x114)",
"缓存容量(MB)": 29.184,
"缓存带宽(TB/s)": null
},
"二级缓存": {
"缓存类型": null,
"缓存容量(MB)": 50,
"缓存带宽(TB/s)": null
}
},
"算力参数": {
"PE层次架构图": null,
"PE参数": {
"算力架构": "同构众核",
"并行方式": "SIMT",
"通信带宽(GB/s)": null
},
"标量参数": {
"标量精度": null,
"INT8标量算力(TOPS)": null,
"FP16标量算力(TFLOPS)": null,
"FP32标量算力(TFLOPS)": null
},
"向量参数": {
"向量精度": "FP64, FP32",
"INT8量算力(TOPS)": null,
"FP16向量算力(TFLOPS)": null,
"FP32向量算力(TFLOPS)": 51.0,
"FP64向量算力(TFLOPS)": 26.0
},
"张量参数": {
"张量精度": "FP64, TF32, FP16, BF16, FP8, INT8",
"INT8张量算力(TOPS)": 3026.0,
"FP8张量算力(TFLOPS)": 3026.0,
"BF16张量算力(TFLOPS)": 1513.0,
"FP16张量算力(TFLOPS)": 1513.0,
"TF32张量算力(TFLOPS)": 756.0,
"FP64张量算力(TFLOPS)": 51.0
}
},
"卡间通信参数": {
"通信方式": "NV-Link",
"端口数量": null,
"RDMA协议": null,
"下行带宽(GB/s)": null,
"上行带宽(GB/s)": null
},
"数据源": [
"https://resources.nvidia.com/en-us-tensor-core/nvidia-tensor-core-gpu-datasheet?ncid=no-ncid",
"https://www.techpowerup.com/gpu-specs/h100-pcie-80-gb.c3899"
]
}
\ No newline at end of file
ByteMLPerf/vendor_zoo/NVIDIA/H100-SXM5.json 0 → 100644
View file @ 24b257f1
{
"SKU参数": {
"厂商": "NVIDIA",
"型号": "H100 SXM5",
"用途": "Training/Inference",
"照片": "vendor_zoo/NVIDIA/image_6.png"
},
"硬件参数": {
"制程(NM)": 4,
"尺寸": "N/A",
"接口": "SXM",
"功耗(W/TDP)": 700
},
"内存参数": {
"内存层次架构图": null,
"内存": {
"内存类型": "HBM3",
"内存容量(GB)": 80,
"内存带宽(GB/s)": 3350
},
"一级缓存": {
"缓存类型": "Cache(x132)",
"缓存容量(MB)": 33.792,
"缓存带宽(TB/s)": null
},
"二级缓存": {
"缓存类型": null,
"缓存容量(MB)": 50,
"缓存带宽(TB/s)": null
}
},
"算力参数": {
"PE层次架构图": null,
"PE参数": {
"算力架构": "同构众核",
"并行方式": "SIMT",
"通信带宽(GB/s)": null
},
"标量参数": {
"标量精度": null,
"INT8标量算力(TOPS)": null,
"FP16标量算力(TFLOPS)": null,
"FP32标量算力(TFLOPS)": null
},
"向量参数": {
"向量精度": "FP64, FP32",
"INT8量算力(TOPS)": null,
"FP16向量算力(TFLOPS)": null,
"FP32向量算力(TFLOPS)": 67.0,
"FP64向量算力(TFLOPS)": 34.0
},
"张量参数": {
"张量精度": "FP64, TF32, FP16, BF16, FP8, INT8",
"INT8张量算力(TOPS)": 3958.0,
"FP8张量算力(TFLOPS)": 3958.0,
"BF16张量算力(TFLOPS)": 1979.0,
"FP16张量算力(TFLOPS)": 1979.0,
"TF32张量算力(TFLOPS)": 989.0,
"FP64张量算力(TFLOPS)": 67.0
}
},
"卡间通信参数": {
"通信方式": "NV-Link",
"端口数量": null,
"RDMA协议": null,
"下行带宽(GB/s)": null,
"上行带宽(GB/s)": null
},
"数据源": [
"https://resources.nvidia.com/en-us-tensor-core/nvidia-tensor-core-gpu-datasheet?ncid=no-ncid",
"https://www.techpowerup.com/gpu-specs/h100-sxm5-80-gb.c3900"
]
}
\ No newline at end of file
ByteMLPerf/vendor_zoo/NVIDIA/T4.json 0 → 100644
View file @ 24b257f1
{
"SKU参数": {
"厂商": "NVIDIA",
"型号": "T4",
"用途": "Training/Inference",
"照片": "vendor_zoo/NVIDIA/image_0.png"
},
"硬件参数": {
"制程(NM)": 12,
"尺寸": "HHHL, Single Slot Card",
"接口": "PCIe 3.0x16",
"功耗(W/TDP)": 70
},
"内存参数": {
"内存层次架构图": null,
"内存": {
"内存类型": "GDDR6",
"内存容量(GB)": 16,
"内存带宽(GB/s)": 320
},
"一级缓存": {
"缓存类型": "Cache(x40)",
"缓存容量(MB)": 2.56,
"缓存带宽(TB/s)": null
},
"二级缓存": {
"缓存类型": null,
"缓存容量(MB)": 4,
"缓存带宽(TB/s)": null
}
},
"算力参数": {
"PE层次架构图": null,
"PE参数": {
"算力架构": "同构众核",
"并行方式": "SIMT",
"通信带宽(GB/s)": null
},
"标量参数": {
"标量精度": null,
"INT8标量算力(TOPS)": null,
"FP16标量算力(TFLOPS)": null,
"FP32标量算力(TFLOPS)": null
},
"向量参数": {
"向量精度": "FP32",
"INT8量算力(TOPS)": null,
"FP16向量算力(TFLOPS)": null,
"FP32向量算力(TFLOPS)": 8.1
},
"张量参数": {
"张量精度": "FP16, INT8, INT4",
"INT4张量算力(TOPS)": 260.0,
"INT8张量算力(TOPS)": 130.0,
"FP16张量算力(TFLOPS)": 65.0,
"FP32张量算力(TFLOPS)": null
}
},
"卡间通信参数": {
"通信方式": null,
"端口数量": null,
"RDMA协议": null,
"下行带宽(GB/s)": null,
"上行带宽(GB/s)": null
},
"数据源": [
"https://www.nvidia.com/en-us/data-center/tesla-t4/",
"https://www.techpowerup.com/gpu-specs/tesla-t4.c3316"
]
}
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment