GPU Performance Summary

Home screen > Select Project > Server List > GPU Performance Summary

Monitor and manage the status and performance metrics of GPUs connected to the system in real time. You can view both physical GPUs and MIG instances together, and efficiently sort or filter each item as needed.

Note

Supported Agent Version

Only supported on Linux. Requires agent version 2.9.3 or higher.

Basic Screen Overview

Click Live to view or pause real-time data. You can search for GPU information using the search box.

Click on any column header in the data list to toggle between ascending and descending sort order.

• Total: Total number of GPU instances (including Physical + MIG)
• Physical: Number of actual physical GPU devices
• MIG: Number of MIG (Multi-Instance GPU) instances

GPU Information

Column	Description
Status	Current status of the GPU instance - N/A: GPU status is displayed only for Physical GPU targets. If the GPU Type is MIG, it is shown as N/A - Allocated: Same as the previous Active status - Unallocated: Same as the previous Inactive status - Active: Allocated and the 5-minute average GPU Utilization is 1% or higher - Idle: Allocated and the 5-minute average GPU Utilization is less than 1% - Effective: Allocated and the 5-minute average GPU Utilization exceeds 50%
HostName	Hostname of the server to which the GPU is connected
GPU Index	Unique index of the physical GPU or MIG instance (MIG is shown in a format like 0/6/0)
Model Name	GPU model name (e.g., NVIDIA A100-SXM4)
GPU Type	Type of the instance: either Physical or MIG

GPU Performance Metrics

Column	Description
GPU_Util (%)	GPU utilization rate (%) - Displayed only for physical GPUs; MIG instances are excluded
Encoder_Util (%)	Hardware encoder utilization rate (%) - Displayed only for physical GPUs
Decoder_Util (%)	Hardware decoder utilization rate (%) - Displayed only for physical GPUs
GR_Engine_Active_Util (%)	Ratio of time the graphic or compute engine on the GPU was active - Used to measure overall GPU utilization. In MIG environments, overhead between instances is reflected in physical GPU usage
SM_Active_Util (%)	Ratio of time at least one warp was executing in the SM - GPU threads run in groups of 32 called warps, not individually
SM_Occupancy (%)	Ratio of active warps to the maximum number of warps that can run on an SM - GPU threads run in groups of 32 called warps
Tensor_Core_Util (%)	Ratio of time the Tensor cores were active
Memory_Copy_Util (%)	Utilization of the GPU memory copy engine (%) - Measures memory transfer activity between the GPU and host or within the GPU - Displayed only for physical GPUs; not shown for MIG
DRAM_Active_Util (%)	DRAM read/write utilization - Displayed only for physical GPUs; not shown for MIG
FP64_Compute_Util (%)	Active time ratio of the FP64 (64-bit floating point) execution pipeline
FP32_Compute_Util (%)	Active time ratio of the FP32 (32-bit floating point) execution pipeline
FP16_Compute_Util (%)	Active time ratio of the FP16 (16-bit floating point) execution pipeline
BAR1_Total_Memory (Bytes)	Total BAR1 memory - BAR1 memory is used for data transfer between GPU and CPU
BAR1_Used_Memory (Bytes)	Amount of BAR1 memory currently in use - BAR1 memory is used for data transfer between GPU and CPU
BAR1_Free_Memory (Bytes)	Remaining BAR1 memory - BAR1 memory is used for data transfer between GPU and CPU
FB_Total_Memory (Bytes)	Total frame buffer memory size
FB_Free_Memory (Bytes)	Available frame buffer memory
FB_Used_Memory (Bytes)	Frame buffer memory currently in use
FB_Reserved_Memory (Bytes)	Reserved frame buffer memory
FB_Memory_Usage (%)	Frame buffer memory usage rate (%)
ECC_SBE_Total	Cumulative total of ECC SBE (Single Bit Errors)
ECC_DBE_Total	Cumulative total of ECC DBE (Double Bit Errors)
GPU_Temperature (°C)	Current GPU temperature
Power_Usage (W)	Current power usage by the GPU
Performance_State (P)	Current Performance State (P-State) of the GPU, represented as a number between 0 and 15 - P0 is the highest performance
Fan_Speed (%)	Current fan speed as a percentage
SM_Clock (MHz)	Current clock speed of the Streaming Multiprocessor (SM)
Memory_Clock (MHz)	Current memory clock speed
Video_Clock (MHz)	Clock speed for video processing
PCIE_TX (Bytes/s)	Amount of data transmitted by the GPU via the PCIe interface - Displayed only for physical GPUs; shown as N/A for MIG
PCIE_RX (Bytes/s)	Amount of data received by the GPU via the PCIe interface - Displayed only for physical GPUs; shown as N/A for MIG
NVLink_TX (Bytes/s)	Amount of data transmitted via NVLink
NVLink_RX (Bytes/s)	Amount of data received via NVLink

Download CSV

You can download the data as a CSV file.

Column Settings

Click the Column Settings button at the top right of the screen to open the Column Settings window.
In the Column Settings window, you can select which columns to display in the table.

• Select All: Select all columns

• Reset to Default Order: Reset the column order to default

• Reset to Default Selection: Restore to the default column selection

Note

The selected columns are saved in your browser cookies, so the selection persists even after refreshing the page.
If cookies are deleted or an error occurs, the default columns will be restored.

1. In the Column Settings window, select the columns to display under GPU Information and GPU Performance Metrics.

   • You can select columns by group or adjust their order.
   Caution

   The Status and HostName fields under GPU Information cannot be removed.

2. Click the Apply button to save your settings.

Add Filter

You can add filter conditions to quickly find the GPUs you need.

1. Click the Filter search bar to open the Add Filter window.

2. In the Add Filter window, select the desired GPU condition (filter key and condition), then click the Apply button.

Compare GPUs

When you select multiple GPUs to compare, the Compare panel appears below the GPU list, allowing you to visually compare metrics using charts.
You can track trends for the last 10 minutes and identify performance anomalies or bottlenecks.

• You can compare up to 20 items at a time.