Skip to content

GitLab

Unverified Commit e083a598 authored by Yuting Jiang's avatar Yuting Jiang Committed by GitHub
Browse files

Benchmarks: Add Benchmark - Add NCCL performance benchmark (#113) 

**Description**
Add NCCL performance microbenchmark.

**Major Revision**
- Add microbenchmark, example, test, config for NCCL
parent b0c5addc
Show whitespace changes
Inline Side-by-side
Showing with 675 additions and 1 deletion
examples/benchmarks/cuda_nccl_bw_performance.py 0 → 100644
View file @ e083a598
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT license.
"""Micro benchmark example for NCCL Bus bandwidth performance.
Commands to run:
python3 examples/benchmarks/nccl_bw_performance.py
"""
from superbench.benchmarks import BenchmarkRegistry, Platform
from superbench.common.utils import logger
if __name__ == '__main__':
context = BenchmarkRegistry.create_benchmark_context(
'nccl-bw', platform=Platform.CUDA, parameters='--operations allreduce'
)
benchmark = BenchmarkRegistry.launch_benchmark(context)
if benchmark:
logger.info(
'benchmark: {}, return code: {}, result: {}'.format(
benchmark.name, benchmark.return_code, benchmark.result
)
)
superbench/benchmarks/micro_benchmarks/__init__.py
View file @ e083a598
......@@ -13,8 +13,10 @@
from superbench.benchmarks.micro_benchmarks.cuda_memory_bw_performance import CudaMemBwBenchmark
from superbench.benchmarks.micro_benchmarks.disk_performance import DiskBenchmark
from superbench.benchmarks.micro_benchmarks.ib_loopback_performance import IBLoopbackBenchmark
from superbench.benchmarks.micro_benchmarks.cuda_nccl_bw_performance import CudaNcclBwBenchmark
__all__ = [
'MicroBenchmark', 'MicroBenchmarkWithInvoke', 'ShardingMatmul', 'ComputationCommunicationOverlap', 'KernelLaunch',
'CublasBenchmark', 'CudnnBenchmark', 'GemmFlopsCuda', 'CudaMemBwBenchmark', 'DiskBenchmark', 'IBLoopbackBenchmark'
'CublasBenchmark', 'CudnnBenchmark', 'GemmFlopsCuda', 'CudaMemBwBenchmark', 'DiskBenchmark', 'IBLoopbackBenchmark',
'CudaNcclBwBenchmark'
]
superbench/benchmarks/micro_benchmarks/cuda_nccl_bw_performance.py 0 → 100644
View file @ e083a598
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT license.
"""Module of the NCCL performance benchmarks."""
import os
import re
from superbench.common.utils import logger
from superbench.benchmarks import BenchmarkRegistry, Platform, ReturnCode
from superbench.benchmarks.micro_benchmarks import MicroBenchmarkWithInvoke
class CudaNcclBwBenchmark(MicroBenchmarkWithInvoke):
"""The NCCL bus bandwidth performance benchmark class."""
def __init__(self, name, parameters=''):
"""Constructor.
Args:
name (str): benchmark name.
parameters (str): benchmark parameters.
"""
super().__init__(name, parameters)
self._bin_name = 'all_reduce_perf'
self.__operations = {
'allreduce': 'all_reduce_perf',
'allgather': 'all_gather_perf',
'broadcast': 'broadcast_perf',
'reduce': 'reduce_perf',
'reducescatter': 'reduce_scatter_perf',
'alltoall': 'alltoall_perf'
}
def add_parser_arguments(self):
"""Add the specified arguments."""
super().add_parser_arguments()
self._parser.add_argument(
'--operations',
type=str,
nargs='+',
default=list(self.__operations.keys()),
help='Nccl operations to benchmark, e.g., {}.'.format(' '.join(list(self.__operations.keys()))),
)
self._parser.add_argument(
'--ngpus',
type=int,
default=8,
help='Number of gpus per thread to run the nccl test.',
)
self._parser.add_argument(
'--maxbytes',
type=str,
default='8G',
help='Max size in bytes to run the nccl test. E.g. 8G.',
)
self._parser.add_argument(
'--minbytes',
type=str,
default='8',
help='Min size in bytes to run the nccl test. E.g. 1.',
)
self._parser.add_argument(
'--stepfactor',
type=int,
default=2,
help='Increment factor, multiplication factor between sizes. E.g. 2.',
)
self._parser.add_argument(
'--check',
type=int,
default=0,
help='Check correctness of results. This can be quite slow on large numbers of GPUs. E.g. 0 or 1.',
)
self._parser.add_argument(
'--iters',
type=int,
default=20,
help='Number of iterations. Default: 20.',
)
self._parser.add_argument(
'--warmup_iters',
type=int,
default=5,
help='Number of warmup iterations. Default: 5.',
)
def _preprocess(self):
"""Preprocess/preparation operations before the benchmarking.
Return:
True if _preprocess() succeed.
"""
if not super()._preprocess():
return False
# Format the arguments
self._args.operations = [p.lower() for p in self._args.operations]
# Check the arguments and generate the commands
for op in self._args.operations:
if op not in self.__operations:
self._result.set_return_code(ReturnCode.INVALID_ARGUMENT)
logger.error(
'Unsupported operation of NCCL test - benchmark: {}, operation: {}, expected: {}.'.format(
self._name, op, ' '.join(list(self.__operations.keys()))
)
)
return False
else:
self._bin_name = self.__operations[op]
if not self._set_binary_path():
return False
command = os.path.join(self._args.bin_dir, self._bin_name)
command += ' -b {} -e {} -f {} -g {} -c {} -n {} -w {}'.format(
self._args.minbytes, self._args.maxbytes, str(self._args.stepfactor), str(self._args.ngpus),
str(self._args.check), str(self._args.iters), str(self._args.warmup_iters)
)
self._commands.append(command)
return True
def _process_raw_result(self, cmd_idx, raw_output): # noqa: C901
"""Function to parse raw results and save the summarized results.
self._result.add_raw_data() and self._result.add_result() need to be called to save the results.
Args:
cmd_idx (int): the index of command corresponding with the raw_output.
raw_output (str): raw output string of the micro-benchmark.
Return:
True if the raw output string is valid and result can be extracted.
"""
# If it's invoked by MPI and rank is not 0, empty content is expected
if os.getenv('OMPI_COMM_WORLD_RANK'):
rank = int(os.getenv('OMPI_COMM_WORLD_RANK'))
if rank > 0:
return True
self._result.add_raw_data('raw_output_' + self._args.operations[cmd_idx], raw_output)
content = raw_output.splitlines()
size = -1
busbw_out = -1
time_out = -1
algbw_out = -1
try:
# Filter useless output
out_of_place_index = -1
out_of_bound_index = -1
for index, line in enumerate(content):
if 'out-of-place' in line:
out_of_place_index = index
if 'Out of bounds values' in line:
out_of_bound_index = index
content = content[out_of_place_index + 1:out_of_bound_index]
# Parse max out of bound bus bw as the result
size_index = -1
time_index = -1
busbw_index = -1
algbw_index = -1
for line in content:
if 'time' in line and 'busbw' in line:
# Get index of selected column
line = line[1:].strip(' ')
line = re.sub(r' +', ' ', line).split(' ')
# Get first index of condition in list, if it not existing, raise exception
size_index = line.index('size')
time_index = line.index('time') - len(line)
busbw_index = line.index('busbw') - len(line)
algbw_index = line.index('algbw') - len(line)
break
if size_index != -1 and busbw_index != -1 and time_index != -1 and algbw_index != -1:
for line in content:
line = line.strip(' ')
line = re.sub(r' +', ' ', line).split(' ')
# Filter line not started with number
if len(line) == 0 or not re.match(r'\d+', line[0]):
continue
size = int(line[size_index])
if size != 0:
busbw_out = float(line[busbw_index])
time_out = float(line[time_index])
algbw_out = float(line[algbw_index])
self._result.add_result(
'NCCL_' + self._args.operations[cmd_idx] + '_' + str(size) + '_busbw', busbw_out
)
self._result.add_result(
'NCCL_' + self._args.operations[cmd_idx] + '_' + str(size) + '_algbw', algbw_out
)
self._result.add_result(
'NCCL_' + self._args.operations[cmd_idx] + '_' + str(size) + '_time', time_out
)
except BaseException as e:
logger.error(
'The result format is invalid - round: {}, benchmark: {}, raw output: {}, message: {}.'.format(
self._curr_run_index, self._name, raw_output, str(e)
)
)
return False
if out_of_place_index == -1 or out_of_bound_index == -1 or busbw_out == -1:
logger.error(
'The result format is invalid - round: {}, benchmark: {}, raw output: {}.'.format(
self._curr_run_index, self._name, raw_output
)
)
return False
return True
BenchmarkRegistry.register_benchmark('nccl-bw', CudaNcclBwBenchmark, platform=Platform.CUDA)
superbench/config/default.yaml
View file @ e083a598
......@@ -28,6 +28,11 @@ superbench:
model_action:
- train
benchmarks:
nccl-bw:
enable: true
modes:
- name: local
prefix: NCCL_DEBUG=INFO NCCL_IB_DISABLE=1
ib-loopback:
enable: true
modes:
......
tests/benchmarks/micro_benchmarks/test_cuda_nccl_bw_performance.py 0 → 100644
View file @ e083a598
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT License.
"""Tests for nccl-bw benchmark."""
import os
import numbers
import unittest
from pathlib import Path
from superbench.benchmarks import BenchmarkRegistry, BenchmarkType, ReturnCode, Platform
class CudaNcclBwBenchmarkTest(unittest.TestCase):
"""Tests for CudaNcclBwBenchmark benchmark."""
def setUp(self):
"""Method called to prepare the test fixture."""
# Create fake binary file just for testing.
os.environ['SB_MICRO_PATH'] = '/tmp/superbench/'
binary_path = os.path.join(os.getenv('SB_MICRO_PATH'), 'bin')
Path(binary_path).mkdir(parents=True, exist_ok=True)
self.__binary_files = []
for bin_name in [
'all_reduce_perf', 'all_gather_perf', 'broadcast_perf', 'reduce_perf', 'reduce_scatter_perf',
'alltoall_perf'
]:
self.__binary_files.append(Path(binary_path, bin_name))
Path(binary_path, bin_name).touch(mode=0o755, exist_ok=True)
def tearDown(self):
"""Method called after the test method has been called and the result recorded."""
for binary_file in self.__binary_files:
binary_file.unlink()
def test_nccl_bw_performance(self):
"""Test nccl-bw benchmark."""
benchmark_name = 'nccl-bw'
(benchmark_class,
predefine_params) = BenchmarkRegistry._BenchmarkRegistry__select_benchmark(benchmark_name, Platform.CUDA)
assert (benchmark_class)
benchmark = benchmark_class(benchmark_name)
ret = benchmark._preprocess()
assert (ret is True)
assert (benchmark.return_code == ReturnCode.SUCCESS)
# Check basic information.
assert (benchmark)
assert (benchmark.name == 'nccl-bw')
assert (benchmark.type == BenchmarkType.MICRO)
# Check parameters specified in BenchmarkContext.
assert (
benchmark._args.operations == [
'allreduce', 'allgather', 'broadcast', 'reduce', 'reducescatter', 'alltoall'
]
)
assert (benchmark._args.ngpus == 8)
assert (benchmark._args.minbytes == '8')
assert (benchmark._args.maxbytes == '8G')
assert (benchmark._args.stepfactor == 2)
assert (benchmark._args.check == 0)
assert (benchmark._args.iters == 20)
assert (benchmark._args.warmup_iters == 5)
# Check command list
bin_names = [
'all_reduce_perf', 'all_gather_perf', 'broadcast_perf', 'reduce_perf', 'reduce_scatter_perf',
'alltoall_perf'
]
for i in range(len(benchmark._args.operations)):
commnad = bin_names[i] + benchmark._commands[i].split(bin_names[i])[1]
expected_command = '{} -b 8 -e 8G -f 2 -g 8 -c 0 -n 20 -w 5'.format(bin_names[i])
assert (commnad == expected_command)
# Check results and metrics.
# Case with no raw_output
assert (benchmark._process_raw_result(0, '') is False)
# Case with valid raw_output
raw_output = {}
raw_output['allgather'] = """
# nThread 1 nGpus 8 minBytes 1 maxBytes 8589934592 step: 2(factor) warmup iters: 5 iters: 20 validation: 0
#
# Using devices
# Rank 0 Pid 112372 on localhost device 0 [0x00] A100-SXM4-40GB
# Rank 1 Pid 112372 on localhost device 1 [0x00] A100-SXM4-40GB
# Rank 2 Pid 112372 on localhost device 2 [0x00] A100-SXM4-40GB
# Rank 3 Pid 112372 on localhost device 3 [0x00] A100-SXM4-40GB
# Rank 4 Pid 112372 on localhost device 4 [0x00] A100-SXM4-40GB
# Rank 5 Pid 112372 on localhost device 5 [0x00] A100-SXM4-40GB
# Rank 6 Pid 112372 on localhost device 6 [0x00] A100-SXM4-40GB
# Rank 7 Pid 112372 on localhost device 7 [0x00] A100-SXM4-40GB
#
# out-of-place in-place
# size count type time algbw busbw error time algbw busbw error
# (B) (elements) (us) (GB/s) (GB/s) (us) (GB/s) (GB/s)
hostname:3442:3442 [0] NCCL INFO Launch mode Parallel
0 0 float 34.27 0.00 0.00 N/A 33.57 0.00 0.00 N/A
0 0 float 33.41 0.00 0.00 N/A 33.62 0.00 0.00 N/A
0 0 float 33.94 0.00 0.00 N/A 33.48 0.00 0.00 N/A
0 0 float 33.83 0.00 0.00 N/A 33.62 0.00 0.00 N/A
0 0 float 33.82 0.00 0.00 N/A 33.57 0.00 0.00 N/A
32 1 float 35.03 0.00 0.00 N/A 34.15 0.00 0.00 N/A
64 2 float 34.36 0.00 0.00 N/A 33.83 0.00 0.00 N/A
128 4 float 33.94 0.00 0.00 N/A 35.22 0.00 0.00 N/A
256 8 float 34.44 0.01 0.01 N/A 34.82 0.01 0.01 N/A
512 16 float 34.84 0.01 0.01 N/A 34.76 0.01 0.01 N/A
1024 32 float 35.38 0.03 0.03 N/A 34.53 0.03 0.03 N/A
2048 64 float 34.67 0.06 0.05 N/A 34.91 0.06 0.05 N/A
4096 128 float 34.62 0.12 0.10 N/A 34.81 0.12 0.10 N/A
8192 256 float 34.76 0.24 0.21 N/A 35.03 0.23 0.20 N/A
16384 512 float 34.80 0.47 0.41 N/A 34.90 0.47 0.41 N/A
32768 1024 float 34.54 0.95 0.83 N/A 35.23 0.93 0.81 N/A
65536 2048 float 36.34 1.80 1.58 N/A 36.01 1.82 1.59 N/A
131072 4096 float 40.18 3.26 2.85 N/A 39.43 3.32 2.91 N/A
262144 8192 float 46.45 5.64 4.94 N/A 46.27 5.67 4.96 N/A
524288 16384 float 58.48 8.96 7.84 N/A 60.40 8.68 7.60 N/A
1048576 32768 float 72.95 14.37 12.58 N/A 73.07 14.35 12.56 N/A
2097152 65536 float 77.28 27.14 23.75 N/A 75.84 27.65 24.20 N/A
4194304 131072 float 100.7 41.64 36.43 N/A 99.56 42.13 36.86 N/A
8388608 262144 float 123.5 67.94 59.44 N/A 120.7 69.51 60.82 N/A
16777216 524288 float 167.7 100.03 87.52 N/A 164.6 101.94 89.20 N/A
33554432 1048576 float 265.8 126.24 110.46 N/A 257.5 130.33 114.04 N/A
67108864 2097152 float 379.7 176.74 154.65 N/A 367.6 182.57 159.75 N/A
134217728 4194304 float 698.6 192.13 168.12 N/A 657.3 204.20 178.67 N/A
268435456 8388608 float 1192.2 225.16 197.01 N/A 1136.0 236.29 206.76 N/A
536870912 16777216 float 2304.1 233.01 203.88 N/A 2227.9 240.98 210.85 N/A
1073741824 33554432 float 4413.4 243.29 212.88 N/A 4258.8 252.12 220.61 N/A
2147483648 67108864 float 8658.8 248.01 217.01 N/A 8389.4 255.98 223.98 N/A
4294967296 134217728 float 17016 252.40 220.85 N/A 16474 260.71 228.12 N/A
8589934592 268435456 float 33646 255.31 223.39 N/A 32669 262.94 230.07 N/A
# Out of bounds values : 0 OK
# Avg bus bandwidth : 58.2651
#
"""
raw_output['allreduce'] = """
# nThread 1 nGpus 8 minBytes 1 maxBytes 8589934592 step: 2(factor) warmup iters: 5 iters: 20 validation: 0
#
# Using devices
# Rank 0 Pid 112424 on localhost device 0 [0x00] A100-SXM4-40GB
# Rank 1 Pid 112424 on localhost device 1 [0x00] A100-SXM4-40GB
# Rank 2 Pid 112424 on localhost device 2 [0x00] A100-SXM4-40GB
# Rank 3 Pid 112424 on localhost device 3 [0x00] A100-SXM4-40GB
# Rank 4 Pid 112424 on localhost device 4 [0x00] A100-SXM4-40GB
# Rank 5 Pid 112424 on localhost device 5 [0x00] A100-SXM4-40GB
# Rank 6 Pid 112424 on localhost device 6 [0x00] A100-SXM4-40GB
# Rank 7 Pid 112424 on localhost device 7 [0x00] A100-SXM4-40GB
#
# out-of-place in-place
# size count type redop time algbw busbw error time algbw busbw error
# (B) (elements) (us) (GB/s) (GB/s) (us) (GB/s) (GB/s)
hostname:3442:3442 [0] NCCL INFO Launch mode Parallel
0 0 float sum 35.20 0.00 0.00 N/A 34.05 0.00 0.00 N/A
0 0 float sum 34.18 0.00 0.00 N/A 33.50 0.00 0.00 N/A
4 1 float sum 34.73 0.00 0.00 N/A 35.30 0.00 0.00 N/A
8 2 float sum 34.66 0.00 0.00 N/A 34.84 0.00 0.00 N/A
16 4 float sum 35.00 0.00 0.00 N/A 35.61 0.00 0.00 N/A
32 8 float sum 35.60 0.00 0.00 N/A 35.27 0.00 0.00 N/A
64 16 float sum 34.83 0.00 0.00 N/A 34.61 0.00 0.00 N/A
128 32 float sum 34.53 0.00 0.01 N/A 43.78 0.00 0.01 N/A
256 64 float sum 34.56 0.01 0.01 N/A 34.95 0.01 0.01 N/A
512 128 float sum 34.94 0.01 0.03 N/A 35.20 0.01 0.03 N/A
1024 256 float sum 36.07 0.03 0.05 N/A 35.77 0.03 0.05 N/A
2048 512 float sum 35.42 0.06 0.10 N/A 35.89 0.06 0.10 N/A
4096 1024 float sum 35.92 0.11 0.20 N/A 36.11 0.11 0.20 N/A
8192 2048 float sum 35.91 0.23 0.40 N/A 36.07 0.23 0.40 N/A
16384 4096 float sum 36.18 0.45 0.79 N/A 35.87 0.46 0.80 N/A
32768 8192 float sum 36.65 0.89 1.56 N/A 35.73 0.92 1.60 N/A
65536 16384 float sum 37.82 1.73 3.03 N/A 37.25 1.76 3.08 N/A
131072 32768 float sum 41.19 3.18 5.57 N/A 41.11 3.19 5.58 N/A
262144 65536 float sum 47.53 5.52 9.65 N/A 47.94 5.47 9.57 N/A
524288 131072 float sum 60.32 8.69 15.21 N/A 60.52 8.66 15.16 N/A
1048576 262144 float sum 74.78 14.02 24.54 N/A 76.17 13.77 24.09 N/A
2097152 524288 float sum 93.48 22.43 39.26 N/A 96.10 21.82 38.19 N/A
4194304 1048576 float sum 112.0 37.44 65.52 N/A 110.2 38.06 66.60 N/A
8388608 2097152 float sum 162.0 51.79 90.63 N/A 160.0 52.44 91.77 N/A
16777216 4194304 float sum 226.0 74.23 129.90 N/A 225.0 74.57 130.49 N/A
33554432 8388608 float sum 374.3 89.65 156.89 N/A 372.8 90.00 157.50 N/A
67108864 16777216 float sum 584.5 114.81 200.91 N/A 581.9 115.33 201.82 N/A
134217728 33554432 float sum 1162.2 115.49 202.11 N/A 1162.5 115.46 202.05 N/A
268435456 67108864 float sum 2112.2 127.09 222.40 N/A 2111.8 127.11 222.45 N/A
536870912 134217728 float sum 4200.3 127.82 223.68 N/A 4184.0 128.32 224.55 N/A
1073741824 268435456 float sum 8159.5 131.59 230.29 N/A 8176.5 131.32 229.81 N/A
2147483648 536870912 float sum 16215 132.44 231.76 N/A 16203 132.53 231.93 N/A
4294967296 1073741824 float sum 32070 133.92 234.37 N/A 32052 134.00 234.50 N/A
8589934592 2147483648 float sum 63896 134.44 235.26 N/A 63959 134.30 235.03 N/A
# Out of bounds values : 0 OK
# Avg bus bandwidth : 68.4048
#
"""
raw_output['reduce'] = """
# nThread 1 nGpus 8 minBytes 1 maxBytes 8589934592 step: 2(factor) warmup iters: 5 iters: 20 validation: 0
#
# Using devices
# Rank 0 Pid 112476 on localhost device 0 [0x00] A100-SXM4-40GB
# Rank 1 Pid 112476 on localhost device 1 [0x00] A100-SXM4-40GB
# Rank 2 Pid 112476 on localhost device 2 [0x00] A100-SXM4-40GB
# Rank 3 Pid 112476 on localhost device 3 [0x00] A100-SXM4-40GB
# Rank 4 Pid 112476 on localhost device 4 [0x00] A100-SXM4-40GB
# Rank 5 Pid 112476 on localhost device 5 [0x00] A100-SXM4-40GB
# Rank 6 Pid 112476 on localhost device 6 [0x00] A100-SXM4-40GB
# Rank 7 Pid 112476 on localhost device 7 [0x00] A100-SXM4-40GB
#
# out-of-place in-place
# size count type redop root time algbw busbw error time algbw busbw error
# (B) (elements) (us) (GB/s) (GB/s) (us) (GB/s) (GB/s)
hostname:3442:3442 [0] NCCL INFO Launch mode Parallel
0 0 float sum 0 36.90 0.00 0.00 N/A 36.47 0.00 0.00 N/A
0 0 float sum 0 34.18 0.00 0.00 N/A 35.70 0.00 0.00 N/A
4 1 float sum 0 35.40 0.00 0.00 N/A 35.59 0.00 0.00 N/A
8 2 float sum 0 36.35 0.00 0.00 N/A 35.74 0.00 0.00 N/A
16 4 float sum 0 35.47 0.00 0.00 N/A 34.27 0.00 0.00 N/A
32 8 float sum 0 36.16 0.00 0.00 N/A 36.19 0.00 0.00 N/A
64 16 float sum 0 35.61 0.00 0.00 N/A 35.45 0.00 0.00 N/A
128 32 float sum 0 34.78 0.00 0.00 N/A 35.80 0.00 0.00 N/A
256 64 float sum 0 35.37 0.01 0.01 N/A 35.89 0.01 0.01 N/A
512 128 float sum 0 35.49 0.01 0.01 N/A 35.53 0.01 0.01 N/A
1024 256 float sum 0 35.38 0.03 0.03 N/A 35.52 0.03 0.03 N/A
2048 512 float sum 0 35.97 0.06 0.06 N/A 35.13 0.06 0.06 N/A
4096 1024 float sum 0 36.03 0.11 0.11 N/A 35.82 0.11 0.11 N/A
8192 2048 float sum 0 36.80 0.22 0.22 N/A 36.71 0.22 0.22 N/A
16384 4096 float sum 0 35.37 0.46 0.46 N/A 36.79 0.45 0.45 N/A
32768 8192 float sum 0 35.16 0.93 0.93 N/A 35.72 0.92 0.92 N/A
65536 16384 float sum 0 38.08 1.72 1.72 N/A 37.74 1.74 1.74 N/A
131072 32768 float sum 0 43.07 3.04 3.04 N/A 41.59 3.15 3.15 N/A
262144 65536 float sum 0 52.16 5.03 5.03 N/A 50.49 5.19 5.19 N/A
524288 131072 float sum 0 67.58 7.76 7.76 N/A 66.57 7.88 7.88 N/A
1048576 262144 float sum 0 76.74 13.66 13.66 N/A 80.47 13.03 13.03 N/A
2097152 524288 float sum 0 78.51 26.71 26.71 N/A 78.76 26.63 26.63 N/A
4194304 1048576 float sum 0 81.47 51.48 51.48 N/A 80.30 52.23 52.23 N/A
8388608 2097152 float sum 0 94.72 88.57 88.57 N/A 94.06 89.19 89.19 N/A
16777216 4194304 float sum 0 137.7 121.83 121.83 N/A 139.6 120.17 120.17 N/A
33554432 8388608 float sum 0 218.3 153.70 153.70 N/A 218.1 153.83 153.83 N/A
67108864 16777216 float sum 0 370.8 180.96 180.96 N/A 369.8 181.49 181.49 N/A
134217728 33554432 float sum 0 661.0 203.06 203.06 N/A 659.9 203.39 203.39 N/A
268435456 67108864 float sum 0 1251.4 214.52 214.52 N/A 1268.1 211.68 211.68 N/A
536870912 134217728 float sum 0 2421.6 221.70 221.70 N/A 2413.4 222.45 222.45 N/A
1073741824 268435456 float sum 0 4736.0 226.72 226.72 N/A 4757.9 225.68 225.68 N/A
2147483648 536870912 float sum 0 9323.5 230.33 230.33 N/A 9354.0 229.58 229.58 N/A
4294967296 1073741824 float sum 0 18594 230.99 230.99 N/A 18570 231.28 231.28 N/A
8589934592 2147483648 float sum 0 37613 228.38 228.38 N/A 37539 228.83 228.83 N/A
# Out of bounds values : 0 OK
# Avg bus bandwidth : 65.018
#
"""
raw_output['broadcast'] = """
# nThread 1 nGpus 8 minBytes 1 maxBytes 8589934592 step: 2(factor) warmup iters: 5 iters: 20 validation: 0
#
# Using devices
# Rank 0 Pid 112528 on localhost device 0 [0x00] A100-SXM4-40GB
# Rank 1 Pid 112528 on localhost device 1 [0x00] A100-SXM4-40GB
# Rank 2 Pid 112528 on localhost device 2 [0x00] A100-SXM4-40GB
# Rank 3 Pid 112528 on localhost device 3 [0x00] A100-SXM4-40GB
# Rank 4 Pid 112528 on localhost device 4 [0x00] A100-SXM4-40GB
# Rank 5 Pid 112528 on localhost device 5 [0x00] A100-SXM4-40GB
# Rank 6 Pid 112528 on localhost device 6 [0x00] A100-SXM4-40GB
# Rank 7 Pid 112528 on localhost device 7 [0x00] A100-SXM4-40GB
#
# out-of-place in-place
# size count type root time algbw busbw error time algbw busbw error
# (B) (elements) (us) (GB/s) (GB/s) (us) (GB/s) (GB/s)
hostname:3442:3442 [0] NCCL INFO Launch mode Parallel
0 0 float 0 34.61 0.00 0.00 N/A 34.33 0.00 0.00 N/A
0 0 float 0 34.43 0.00 0.00 N/A 35.06 0.00 0.00 N/A
4 1 float 0 33.96 0.00 0.00 N/A 33.80 0.00 0.00 N/A
8 2 float 0 34.16 0.00 0.00 N/A 34.32 0.00 0.00 N/A
16 4 float 0 34.47 0.00 0.00 N/A 34.85 0.00 0.00 N/A
32 8 float 0 35.24 0.00 0.00 N/A 34.75 0.00 0.00 N/A
64 16 float 0 35.12 0.00 0.00 N/A 34.89 0.00 0.00 N/A
128 32 float 0 34.67 0.00 0.00 N/A 34.36 0.00 0.00 N/A
256 64 float 0 34.23 0.01 0.01 N/A 34.42 0.01 0.01 N/A
512 128 float 0 34.26 0.01 0.01 N/A 35.20 0.01 0.01 N/A
1024 256 float 0 34.87 0.03 0.03 N/A 34.80 0.03 0.03 N/A
2048 512 float 0 34.90 0.06 0.06 N/A 35.27 0.06 0.06 N/A
4096 1024 float 0 35.37 0.12 0.12 N/A 34.59 0.12 0.12 N/A
8192 2048 float 0 34.95 0.23 0.23 N/A 34.79 0.24 0.24 N/A
16384 4096 float 0 34.94 0.47 0.47 N/A 34.94 0.47 0.47 N/A
32768 8192 float 0 35.03 0.94 0.94 N/A 34.71 0.94 0.94 N/A
65536 16384 float 0 36.04 1.82 1.82 N/A 36.48 1.80 1.80 N/A
131072 32768 float 0 40.09 3.27 3.27 N/A 39.92 3.28 3.28 N/A
262144 65536 float 0 46.58 5.63 5.63 N/A 45.89 5.71 5.71 N/A
524288 131072 float 0 58.37 8.98 8.98 N/A 59.67 8.79 8.79 N/A
1048576 262144 float 0 76.02 13.79 13.79 N/A 78.43 13.37 13.37 N/A
2097152 524288 float 0 78.12 26.85 26.85 N/A 78.84 26.60 26.60 N/A
4194304 1048576 float 0 81.06 51.74 51.74 N/A 80.39 52.17 52.17 N/A
8388608 2097152 float 0 97.20 86.30 86.30 N/A 96.09 87.30 87.30 N/A
16777216 4194304 float 0 143.1 117.22 117.22 N/A 142.1 118.06 118.06 N/A
33554432 8388608 float 0 223.4 150.21 150.21 N/A 221.3 151.61 151.61 N/A
67108864 16777216 float 0 374.8 179.05 179.05 N/A 374.4 179.23 179.23 N/A
134217728 33554432 float 0 672.2 199.67 199.67 N/A 670.0 200.34 200.34 N/A
268435456 67108864 float 0 1271.5 211.11 211.11 N/A 1264.5 212.28 212.28 N/A
536870912 134217728 float 0 2436.3 220.37 220.37 N/A 2434.5 220.53 220.53 N/A
1073741824 268435456 float 0 4769.2 225.14 225.14 N/A 4697.5 228.58 228.58 N/A
2147483648 536870912 float 0 9314.2 230.56 230.56 N/A 9248.3 232.20 232.20 N/A
4294967296 1073741824 float 0 18487 232.33 232.33 N/A 18381 233.66 233.66 N/A
8589934592 2147483648 float 0 36896 232.81 232.81 N/A 36599 234.70 234.70 N/A
# Out of bounds values : 0 OK
# Avg bus bandwidth : 64.8653
#
"""
raw_output['reducescatter'] = """
# nThread 1 nGpus 8 minBytes 1 maxBytes 8589934592 step: 2(factor) warmup iters: 5 iters: 20 validation: 0
#
# Using devices
# Rank 0 Pid 112580 on localhost device 0 [0x00] A100-SXM4-40GB
# Rank 1 Pid 112580 on localhost device 1 [0x00] A100-SXM4-40GB
# Rank 2 Pid 112580 on localhost device 2 [0x00] A100-SXM4-40GB
# Rank 3 Pid 112580 on localhost device 3 [0x00] A100-SXM4-40GB
# Rank 4 Pid 112580 on localhost device 4 [0x00] A100-SXM4-40GB
# Rank 5 Pid 112580 on localhost device 5 [0x00] A100-SXM4-40GB
# Rank 6 Pid 112580 on localhost device 6 [0x00] A100-SXM4-40GB
# Rank 7 Pid 112580 on localhost device 7 [0x00] A100-SXM4-40GB
#
# out-of-place in-place
# size count type redop time algbw busbw error time algbw busbw error
# (B) (elements) (us) (GB/s) (GB/s) (us) (GB/s) (GB/s)
hostname:3442:3442 [0] NCCL INFO Launch mode Parallel
0 0 float sum 34.88 0.00 0.00 N/A 33.65 0.00 0.00 N/A
0 0 float sum 33.54 0.00 0.00 N/A 33.72 0.00 0.00 N/A
0 0 float sum 33.45 0.00 0.00 N/A 33.44 0.00 0.00 N/A
0 0 float sum 34.07 0.00 0.00 N/A 33.44 0.00 0.00 N/A
0 0 float sum 33.55 0.00 0.00 N/A 33.43 0.00 0.00 N/A
32 1 float sum 35.06 0.00 0.00 N/A 35.14 0.00 0.00 N/A
64 2 float sum 34.82 0.00 0.00 N/A 34.76 0.00 0.00 N/A
128 4 float sum 34.38 0.00 0.00 N/A 34.52 0.00 0.00 N/A
256 8 float sum 34.75 0.01 0.01 N/A 34.32 0.01 0.01 N/A
512 16 float sum 34.71 0.01 0.01 N/A 35.43 0.01 0.01 N/A
1024 32 float sum 35.16 0.03 0.03 N/A 34.75 0.03 0.03 N/A
2048 64 float sum 35.43 0.06 0.05 N/A 35.29 0.06 0.05 N/A
4096 128 float sum 35.49 0.12 0.10 N/A 35.17 0.12 0.10 N/A
8192 256 float sum 35.18 0.23 0.20 N/A 35.77 0.23 0.20 N/A
16384 512 float sum 35.27 0.46 0.41 N/A 35.49 0.46 0.40 N/A
32768 1024 float sum 35.00 0.94 0.82 N/A 35.09 0.93 0.82 N/A
65536 2048 float sum 36.78 1.78 1.56 N/A 36.92 1.77 1.55 N/A
131072 4096 float sum 40.71 3.22 2.82 N/A 39.78 3.29 2.88 N/A
262144 8192 float sum 48.12 5.45 4.77 N/A 46.65 5.62 4.92 N/A
524288 16384 float sum 59.81 8.77 7.67 N/A 58.88 8.90 7.79 N/A
1048576 32768 float sum 72.37 14.49 12.68 N/A 74.95 13.99 12.24 N/A
2097152 65536 float sum 80.64 26.01 22.76 N/A 79.62 26.34 23.05 N/A
4194304 131072 float sum 108.9 38.53 33.72 N/A 109.3 38.37 33.57 N/A
8388608 262144 float sum 147.3 56.96 49.84 N/A 166.8 50.28 44.00 N/A
16777216 524288 float sum 152.4 110.11 96.34 N/A 152.8 109.82 96.09 N/A
33554432 1048576 float sum 240.5 139.50 122.06 N/A 240.8 139.33 121.91 N/A
67108864 2097152 float sum 356.1 188.45 164.89 N/A 352.1 190.57 166.75 N/A
134217728 4194304 float sum 618.1 217.15 190.01 N/A 615.2 218.18 190.90 N/A
268435456 8388608 float sum 1108.7 242.11 211.84 N/A 1112.6 241.27 211.11 N/A
536870912 16777216 float sum 2169.0 247.52 216.58 N/A 2181.8 246.07 215.31 N/A
1073741824 33554432 float sum 4203.0 255.47 223.54 N/A 4206.3 255.27 223.36 N/A
2147483648 67108864 float sum 8356.9 256.97 224.85 N/A 8323.5 258.00 225.75 N/A
4294967296 134217728 float sum 16400 261.89 229.15 N/A 16402 261.86 229.13 N/A
8589934592 268435456 float sum 32464 264.60 231.52 N/A 32502 264.29 231.25 N/A
# Out of bounds values : 0 OK
# Avg bus bandwidth : 60.168
#
"""
raw_output['alltoall'] = """
# nThread 1 nGpus 8 minBytes 1 maxBytes 8589934592 step: 2(factor) warmup iters: 5 iters: 20 validation: 0
#
# Using devices
# Rank 0 Pid 167261 on localhost device 0 [0x00] A100-SXM4-40GB
# Rank 1 Pid 167261 on localhost device 1 [0x00] A100-SXM4-40GB
# Rank 2 Pid 167261 on localhost device 2 [0x00] A100-SXM4-40GB
# Rank 3 Pid 167261 on localhost device 3 [0x00] A100-SXM4-40GB
# Rank 4 Pid 167261 on localhost device 4 [0x00] A100-SXM4-40GB
# Rank 5 Pid 167261 on localhost device 5 [0x00] A100-SXM4-40GB
# Rank 6 Pid 167261 on localhost device 6 [0x00] A100-SXM4-40GB
# Rank 7 Pid 167261 on localhost device 7 [0x00] A100-SXM4-40GB
#
# out-of-place in-place
# size count type redop time algbw busbw error time algbw busbw error
# (B) (elements) (us) (GB/s) (GB/s) (us) (GB/s) (GB/s)
0 0 float 1.63 0.00 0.00 N/A 1.38 0.00 0.00 N/A
0 0 float 1.35 0.00 0.00 N/A 1.34 0.00 0.00 N/A
0 0 float 1.35 0.00 0.00 N/A 1.77 0.00 0.00 N/A
0 0 float 1.37 0.00 0.00 N/A 1.39 0.00 0.00 N/A
0 0 float 1.34 0.00 0.00 N/A 1.33 0.00 0.00 N/A
32 1 float 89.00 0.00 0.00 N/A 85.13 0.00 0.00 N/A
64 2 float 86.83 0.00 0.00 N/A 85.77 0.00 0.00 N/A
128 4 float 86.02 0.00 0.00 N/A 85.30 0.00 0.00 N/A
256 8 float 87.20 0.00 0.00 N/A 86.21 0.00 0.00 N/A
512 16 float 87.33 0.01 0.01 N/A 88.47 0.01 0.01 N/A
1024 32 float 88.17 0.01 0.01 N/A 88.98 0.01 0.01 N/A
2048 64 float 86.44 0.02 0.02 N/A 86.65 0.02 0.02 N/A
4096 128 float 86.75 0.05 0.04 N/A 86.68 0.05 0.04 N/A
8192 256 float 88.78 0.09 0.08 N/A 87.05 0.09 0.08 N/A
16384 512 float 87.71 0.19 0.16 N/A 86.76 0.19 0.17 N/A
32768 1024 float 86.26 0.38 0.33 N/A 88.92 0.37 0.32 N/A
65536 2048 float 87.67 0.75 0.65 N/A 89.16 0.74 0.64 N/A
131072 4096 float 87.35 1.50 1.31 N/A 86.76 1.51 1.32 N/A
262144 8192 float 87.02 3.01 2.64 N/A 87.98 2.98 2.61 N/A
524288 16384 float 86.58 6.06 5.30 N/A 89.33 5.87 5.14 N/A
1048576 32768 float 87.42 11.99 10.50 N/A 88.90 11.79 10.32 N/A
2097152 65536 float 89.61 23.40 20.48 N/A 90.10 23.27 20.37 N/A
4194304 131072 float 96.44 43.49 38.05 N/A 99.62 42.10 36.84 N/A
8388608 262144 float 121.1 69.28 60.62 N/A 120.6 69.56 60.87 N/A
16777216 524288 float 160.4 104.62 91.55 N/A 158.8 105.64 92.43 N/A
33554432 1048576 float 237.5 141.30 123.64 N/A 234.5 143.11 125.22 N/A
67108864 2097152 float 396.8 169.13 147.99 N/A 387.0 173.41 151.73 N/A
134217728 4194304 float 633.6 211.83 185.35 N/A 620.9 216.17 189.15 N/A
268435456 8388608 float 1189.1 225.75 197.53 N/A 1167.8 229.86 201.13 N/A
536870912 16777216 float 2236.6 240.04 210.03 N/A 2197.4 244.32 213.78 N/A
1073741824 33554432 float 4335.5 247.66 216.71 N/A 4274.2 251.22 219.81 N/A
2147483648 67108864 float 8510.4 252.34 220.79 N/A 8405.3 255.49 223.56 N/A
4294967296 134217728 float 16860 254.74 222.90 N/A 16678 257.53 225.34 N/A
8589934592 268435456 float 33508 256.36 224.31 N/A 33234 258.47 226.16 N/A
# Out of bounds values : 0 OK
# Avg bus bandwidth : 58.6481
"""
for i, op in enumerate(benchmark._args.operations):
assert (benchmark._process_raw_result(i, raw_output[op]))
for name in ['time', 'algbw', 'busbw']:
for size in ['8589934592', '4294967296', '2147483648', '1073741824', '536870912', '32']:
metric = 'NCCL_' + op + '_' + size + '_' + name
assert (metric in benchmark.result)
assert (len(benchmark.result[metric]) == 1)
assert (isinstance(benchmark.result[metric][0], numbers.Number))
assert (benchmark.result['NCCL_allreduce_8589934592_time'][0] == 63896.0)
assert (benchmark.result['NCCL_allreduce_8589934592_algbw'][0] == 134.44)
assert (benchmark.result['NCCL_allreduce_8589934592_busbw'][0] == 235.26)
assert (benchmark.result['NCCL_alltoall_8589934592_time'][0] == 33508.0)
assert (benchmark.result['NCCL_alltoall_8589934592_algbw'][0] == 256.36)
assert (benchmark.result['NCCL_alltoall_8589934592_busbw'][0] == 224.31)
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