NVIDIA/nvbench
1
0
Fork 0
mirror of https://github.com/NVIDIA/nvbench.git synced 2026-03-14 20:27:24 +00:00
Code Issues Packages Projects Releases Wiki Activity

Replace references to peak_sm_clock with default_sm_clock.

Browse Source 
The actual measured clock speed can exceed this value, so default is less confusing than peak.
This commit is contained in:
Allison Piper
2025-04-14 11:33:04 -04:00
parent 87dd03254f
commit 18926ced87
6 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
docs/cli_help.md
View File

@@ -90,7 +90,7 @@
before any `--benchmark` arguments.
* `--stopping-criterion <criterion>`
* After `--min-samples` is satisfied, use `<criterion>` to detect if enough
* After `--min-samples` is satisfied, use `<criterion>` to detect if enough
samples were collected.
* Only applies to Cold measurements.
* Default is stdrel (`--stopping-criterion stdrel`)
@@ -134,7 +134,7 @@
before any `--benchmark` arguments.
* `--throttle-threshold <value>`
* Set the GPU throttle threshold as percentage of the peak clock rate.
* Set the GPU throttle threshold as percentage of the device's default clock rate.
* Default is 75%.
* Applies to the most recent `--benchmark`, or all benchmarks if specified
before any `--benchmark` arguments.

2
nvbench/benchmark_base.cuh
View File

@@ -302,7 +302,7 @@ protected:
nvbench::float64_t m_skip_time{-1.};
nvbench::float64_t m_timeout{15.};
nvbench::float32_t m_throttle_threshold{0.75f}; // [% of peak SM clock rate]
nvbench::float32_t m_throttle_threshold{0.75f}; // [% of default SM clock rate]
nvbench::float32_t m_throttle_recovery_delay{0.05f}; // [seconds]
nvbench::criterion_params m_criterion_params;

2
nvbench/detail/gpu_frequency.cuh
View File

@@ -40,7 +40,7 @@ struct gpu_frequency
void stop(const nvbench::cuda_stream &stream) { m_stop.record(stream); }
[[nodiscard]] bool has_throttled(nvbench::float32_t peak_sm_clock_rate_hz,
[[nodiscard]] bool has_throttled(nvbench::float32_t default_sm_clock_rate_hz,
nvbench::float32_t throttle_threshold);
[[nodiscard]] nvbench::float32_t get_clock_frequency();

4
nvbench/detail/gpu_frequency.cxx
View File

@@ -31,10 +31,10 @@ nvbench::float32_t gpu_frequency::get_clock_frequency()
return clock_rate;
}
bool gpu_frequency::has_throttled(nvbench::float32_t peak_sm_clock_rate_hz,
bool gpu_frequency::has_throttled(nvbench::float32_t default_sm_clock_rate_hz,
nvbench::float32_t throttle_threshold)
{
float threshold = peak_sm_clock_rate_hz * throttle_threshold;
float threshold = default_sm_clock_rate_hz * throttle_threshold;
if (this->get_clock_frequency() < threshold)
{

2
nvbench/detail/measure_cold.cuh
View File

@@ -102,7 +102,7 @@ protected:
nvbench::float64_t m_skip_time{};
nvbench::float64_t m_timeout{};
nvbench::float32_t m_throttle_threshold; // [% of peak SM clock rate]
nvbench::float32_t m_throttle_threshold; // [% of default SM clock rate]
nvbench::float32_t m_throttle_recovery_delay; // [seconds]
nvbench::int64_t m_total_samples{};

2
nvbench/state.cuh
View File

@@ -331,7 +331,7 @@ private:
nvbench::float64_t m_skip_time;
nvbench::float64_t m_timeout;
nvbench::float32_t m_throttle_threshold; // [% of peak SM clock rate]
nvbench::float32_t m_throttle_threshold; // [% of default SM clock rate]
nvbench::float32_t m_throttle_recovery_delay; // [seconds]
// Deadlock protection. See blocking_kernel's class doc for details.