Be notified of new releases
Create your free GitHub account today to subscribe to this repository for new releases and build software alongside 28 million developers.Sign up
Continuous monitoring function (fixed 1 second interval, 5 minutes ) #32
Stopping the monitor will allow you to save the data
Enable value/utilization inspector during active profiling or monitoring
Bug fix version of v1.06, no new features added.
Corrects the following bugs:
#89 : Process utilization not shown during profiling or monitoring
#90 : Application exception on launch when running on Windows Server 2016
#91 : Exception error while using data inspector within areas not yet containing profile data
#92 : Monitor mode values not correct during scrolling, utilization values are correct
Bug fix release
- Exception error when loading saved GPD file via explorer file association #68
- Incorrect header in CSV output #69
- Automatic timed series profiling works only on first loop #72
- Automatic timed series profiling, file path handling improvement #73
- Fallback to NVAPI where NVML functionality is limited and restricts accurate utilization metrics (frame buffer reported to be 100% utilized) #74
- Suppress warning dialog(s) when GPUProfiler is run before the driver is loaded after installation (reboot requirement not met) #75
- Process utilization header "FRB" -> "MC" the values returned from NVML include the memory controller utilization and not the Framebuffer utilization #76
- Show the total memory controller utilization in the "_TOTAL" line #76
The latest release of GPUProfiler v1.05 is finally available.
Dark UI theme (light theme can be enabled in options) #50
Automated timed series profiling via command line options #55
|-d||int||3660||Profile duration (in minutes)|
|-r||int||500||Repeat (number of times)|
|-t||Append time stamp|
|-m||Minimize during profiling|
|-x||Exit on completion|
GPUProfiler.exe -i 1 -d 30 -r 10 -f c:\tmp\profile.gpd -t -m -x
- Default interval, duration, plot bold level as well as the application standard and dash-view window sizes /positions are also saved.
GPU WDDM or TCC mode detection #43
New Monitoring Capabilities
- Per-Process GPU utilization monitoring (currently not serialized in the GPD file, no graph support -yet)
- Improved DPI scaling support, 100%, 125%, 200% #40
- Default profile view displays the first 60 samples and scrolls from then on #57
- Improved display performance for longer profile runs.
- Two step bolding of the individual plot lines, (third step is to hide it) hotkey ‘g’ for GPU, etc. #22
- Three step bolding of the plot lines (hotkey ‘b’) #41
- Legend display toggle (hotkey 'l') #42
- Dash-view automatically hides legend and Axis labels #42
- Resource utilization histogram update when visible data range changes
- Arrow keys translate within the plot view
- Ctrl+up (zoom in), Ctrl+down (zoom out)
- 4 state 'Home' and 'End' functions
- Zoom all
- Show first / last 50% of total
- Show first / last 25% of total
- Show first / last 13% of total.
- 'Page Up' / 'Page Down' translate along the X-Axis
There are other small changes from v1.04 that I have made, and as with all new releases, there are corner cases that I have not completely tested and would appreciate any discovered bugs to be added to the page.
New feature requests can be added and tracked there as well.
A short list of changes from v1.03 -> 1.04
- In-graph display of utilization data in percentage or actual value for loaded files and for post collection results
- Scroll, pan and zoom support via mouse wheel.
- Re-sizable windows, size retained for each mode (normal/dashboard)
- Dashboard mode via double clicking on the graph (revert to prior view via same method)
- Resource histogram analysis analyzes the current visible time range (zoom in to examine a particular operation’s resource utilization)
- Improvement in histogram calculation and handling of 100% and 0% values.
- Histogram width increased to give better visibility
- Selective display of histogram data based on current data display options
- Double clicking each histogram toggles between auto-scale and fixed scale mode
-- Fixed scale
-- Automatic scale
- Keyboard accelerators to show/hide the graph data in the plot, data tooltip and the histogram:
-- 'C' - CPU
-- 'R' - RAM
-- 'G' - GPU
-- 'F' - Framebuffer
-- 'E' - Video Encoder
-- 'D' - Video Decoder
- The Analysis (histogram) can be shown or hidden with
-- 'A' - Analysis
- When data is loaded these hotkey control the display of utilization information when the mouse is within the graph results:
-- 'U' - Utilization
-- 'V' - Values (and utilization where appropriate)
- Dependencies have been statically linked to remove the prior requirement to install Visual Studio redistributables
- ‘GPD’ file associations are set on startup based on current executable path, GPD file icon added as well.
- Zoom full to profile overview when profile collection has been completed
- Removed checks that prevented the application from running in environments without NVML or an NVIDIA GPU
- Improved scrolling/panning during profiling when in a zoomed view state
- Multi GPU collection and support in the GPD data file but no current display in the UI
- Added a disclaimer for my employer (thank you Rachel) in the about box
- Font improvements to handle large DPI display users
- Resource utilization histograms
- Resizable window, minimize/maximize
- NVML and x64 support only (x86 users should continue to use 1.01 version)
- Video Encode/Decode utilization
- vGPU GPU utilization
- Add always on top option
- Add minimize on profiling start option
- Switched to NVML to use advanced monitoring features unavailable in NVAPI
- Support for x86 ends due to lack of NVML support for x86 platforms
- Many bugs fixed
- Redirect to Github releases page from "About..." dialog
- Additional sample interval and duration input data checks
Initial public release.
Requires prior installation of VS2010 redistributables.