I want to reduce the execution time using multiple Alveo U55C cards.
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 35/100
- Issue 类型
- 缺陷
- 描述清晰度
- 需要澄清
- 活跃度
- 停滞
调研方向
Start with the linked sys_opt/multiple_devices example and reproduce the reported timings on one, two, and four Alveo U55C cards. Compare the execution logs and determine what prevents proportional scaling; done means the multi-device run achieves the expected speedup or the limiting behavior is clearly documented.
由索引模型根据 Issue 内容生成。
描述
"I tried to implement this example project: https://github.com/Xilinx/Vitis_Accel_Examples/tree/main/sys_opt/multiple_devices. It reduces the execution time as the number of devices increases. However, when I use four Alveo U55C cards, the implementation does not reduce the execution time proportionally; it only improves performance by a factor of four.
"INFO: Reading ./build_dir.hw.xilinx_u55c_gen3x16_xdma_3_202210_1/vector_addition.xclbin
Loading: './build_dir.hw.xilinx_u55c_gen3x16_xdma_3_202210_1/vector_addition.xclbin'
Creating Buffers[0]...
Creating Context[1]...
INFO: Reading ./build_dir.hw.xilinx_u55c_gen3x16_xdma_3_202210_1/vector_addition.xclbin
Loading: './build_dir.hw.xilinx_u55c_gen3x16_xdma_3_202210_1/vector_addition.xclbin'
Creating Buffers[1]...
Creating Context[2]...
INFO: Reading ./build_dir.hw.xilinx_u55c_gen3x16_xdma_3_202210_1/vector_addition.xclbin
Loading: './build_dir.hw.xilinx_u55c_gen3x16_xdma_3_202210_1/vector_addition.xclbin'
Creating Buffers[2]...
Creating Context[3]...
INFO: Reading ./build_dir.hw.xilinx_u55c_gen3x16_xdma_3_202210_1/vector_addition.xclbin
Loading: './build_dir.hw.xilinx_u55c_gen3x16_xdma_3_202210_1/vector_addition.xclbin'
Creating Buffers[3]...
Copying data...
Launching Kernel...
Getting Results...
Copying data...
Launching Kernel...
Getting Results...
Copying data...
Launching Kernel...
Getting Results...
Copying data...
Launching Kernel...
Getting Results...
Waiting for work to finish on device 0
Waiting for work to finish on device 1
Waiting for work to finish on device 2
Waiting for work to finish on device 3
Total Size : 480.00 GB
Time Taken : 42.5177sec
TEST PASSED
" And for Single device u55c output is "INFO: Reading ./build_dir.hw.xilinx_u55c_gen3x16_xdma_3_202210_1/vector_addition.xclbin
Loading: './build_dir.hw.xilinx_u55c_gen3x16_xdma_3_202210_1/vector_addition.xclbin'
Creating Buffers[0]...
Copying data...
Launching Kernel...
Getting Results...
Waiting for work to finish on device 0
Total Size : 120.00 GB
Time Taken : 42.5148sec
TEST PASSED
"
,I expected this output. "
1 | 47.41
2 | 23.71
Speedup | 1.99 This is for two devices
- 主要语言
- Makefile
- 星标
- 609
- 派生
- 233
- PR 合并指标
- 30 天内没有已合并 PR
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
Xilinx/Vitis_Accel_Examples 的其他 Issue
-
难度 1/5 1 小时以内 新手友好度 35/100
-
hw_emu hangs未关闭
难度 4/5 3-5 天 新手友好度 35/100
-
难度 4/5 3-5 天 新手友好度 25/100
-
难度 4/5 3-5 天 新手友好度 30/100
-
难度 5/5 一周以上 新手友好度 25/100
查看 Xilinx/Vitis_Accel_Examples 的全部 Issue
相似的 Issue
-
Arch: xtensa Area: Board support Area: Build system OS: Linux Type: Bug
难度 2/5 1-3 小时 新手友好度 72/100
维护者通常 1 天内回复
-
device support request
难度 2/5 1-3 小时 新手友好度 68/100
zigpy/zha-device-handlers#5461 ·
维护者通常 4 天内回复
-
external converter
难度 1/5 1 小时以内 新手友好度 62/100
Koenkk/zigbee2mqtt#33396 · 1 条评论 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 70/100
维护者通常 1 天内回复
-
enhancement good first issue
难度 2/5 3-5 天 新手友好度 82/100
维护者通常 1 天内回复