Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[GPU-Plugin] Make node_idx type 32 bit for hist algo. Set default n_gpus to 1. (#2445)

Browse Source
This commit is contained in:
Rory Mitchell 2017-06-23 18:26:45 +12:00 committed by GitHub
parent 169c983b5f
commit 0e48f87529
3 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
plugin/updater_gpu/README.md
View File

@ -21,7 +21,7 @@ n_gpus | ✖ | ✔ |
The device ordinal can be selected using the 'gpu_id' parameter, which defaults to 0. The device ordinal can be selected using the 'gpu_id' parameter, which defaults to 0.
Multiple GPUs can be used with the grow_gpu_hist parameter using the n_gpus parameter, which defaults to -1 (indicating use all visible GPUs). If gpu_id is specified as non-zero, the gpu device order is mod(gpu_id + i) % n_visible_devices for i=0 to n_gpus-1. As with GPU vs. CPU, multi-GPU will not always be faster than a single GPU due to PCI bus bandwidth that can limit performance. For example, when n_features * n_bins * 2^depth divided by time of each round/iteration becomes comparable to the real PCI 16x bus bandwidth of order 4GB/s to 10GB/s, then AllReduce will dominant code speed and multiple GPUs become ineffective at increasing performance. Also, CPU overhead between GPU calls can limit usefulness of multiple GPUs. Multiple GPUs can be used with the grow_gpu_hist parameter using the n_gpus parameter. which defaults to 1. If this is set to -1 all available GPUs will be used. If gpu_id is specified as non-zero, the gpu device order is mod(gpu_id + i) % n_visible_devices for i=0 to n_gpus-1. As with GPU vs. CPU, multi-GPU will not always be faster than a single GPU due to PCI bus bandwidth that can limit performance. For example, when n_features * n_bins * 2^depth divided by time of each round/iteration becomes comparable to the real PCI 16x bus bandwidth of order 4GB/s to 10GB/s, then AllReduce will dominant code speed and multiple GPUs become ineffective at increasing performance. Also, CPU overhead between GPU calls can limit usefulness of multiple GPUs.
This plugin currently works with the CLI version and python version. This plugin currently works with the CLI version and python version.

4
plugin/updater_gpu/src/gpu_hist_builder.cuh
View File

@ -121,8 +121,8 @@ class GPUHistBuilder {
std::vector<dh::dvec<float>> fidx_min_map; std::vector<dh::dvec<float>> fidx_min_map;
std::vector<dh::dvec<int>> feature_segments; std::vector<dh::dvec<int>> feature_segments;
std::vector<dh::dvec<bst_float>> prediction_cache; std::vector<dh::dvec<bst_float>> prediction_cache;
std::vector<dh::dvec<NodeIdT>> position; std::vector<dh::dvec<int>> position;
std::vector<dh::dvec<NodeIdT>> position_tmp; std::vector<dh::dvec<int>> position_tmp;
std::vector<DeviceGMat> device_matrix; std::vector<DeviceGMat> device_matrix;
std::vector<dh::dvec<gpu_gpair>> device_gpair; std::vector<dh::dvec<gpu_gpair>> device_gpair;
std::vector<dh::dvec<int>> gidx_feature_map; std::vector<dh::dvec<int>> gidx_feature_map;

2
src/tree/param.h
View File

@ -196,7 +196,7 @@ struct TrainParam : public dmlc::Parameter<TrainParam> {
.describe("gpu to use for single gpu algorithms"); .describe("gpu to use for single gpu algorithms");
DMLC_DECLARE_FIELD(n_gpus) DMLC_DECLARE_FIELD(n_gpus)
.set_lower_bound(-1) .set_lower_bound(-1)
.set_default(-1) .set_default(1)
.describe("Number of GPUs to use for multi-gpu algorithms: -1=use all GPUs"); .describe("Number of GPUs to use for multi-gpu algorithms: -1=use all GPUs");
// add alias of parameters // add alias of parameters
DMLC_DECLARE_ALIAS(reg_lambda, lambda); DMLC_DECLARE_ALIAS(reg_lambda, lambda);