为什么 Tensorflow 显示可用内存低于 GPU 规格？

Question

此命令为 GPU 提供 4614 MB 内存。

但是RTX2060有6GB内存。为什么只显示 4GB 多一点？

>>> tf.test.is_built_with_cuda()
True
>>> tf.test.is_gpu_available(cuda_only=False,min_cuda_compute_capability=None)
2019-10-29 17:02:40.062465: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2019-10-29 17:02:40.072455: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library nvcuda.dll
2019-10-29 17:02:40.105489: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties:
name: GeForce RTX 2060 major: 7 minor: 5 memoryClockRate(GHz): 1.2
pciBusID: 0000:01:00.0
2019-10-29 17:02:40.111138: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
2019-10-29 17:02:40.117217: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1763] Adding visible gpu devices: 0
2019-10-29 17:02:44.865862: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1181] Device interconnect StreamExecutor with strength 1 edge matrix:
2019-10-29 17:02:44.870341: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1187]      0
2019-10-29 17:02:44.872351: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1200] 0:   N
2019-10-29 17:02:44.876727: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1326] Created TensorFlow device (/device:GPU:0 with 4614 MB memory) -> physical GPU (device: 0, name: GeForce RTX 2060, pci bus id: 0000:01:00.0, compute capability: 7.5)
True

Answer 1

Tensorflow 报告它能够在设备上使用的内存。因此，如果您有任何其他使用一些 GPU 内存的东西，TensorFlow 报告的内存将低于总可用内存。

此外，一些 OS 只允许应用程序使用特定部分的内存，以确保它仍然能够使用设备（参见 TensorFlow GitHub 上的这个问题：https://github.com/tensorflow/tensorflow/issues/22623)

为什么 Tensorflow 显示可用内存低于 GPU 规格？

Why does Tensorflow show memory available less than GPU specs?

nvidia

tensorflow