XRT-SMI Utility#
The xrt-smi utility helps you to verify AIE readiness, DDR connectivity, and kernel execution during the initial board boot. It provides a quick, command-line way to confirm that the hardware, drivers, and runtime stack are correctly installed and operational before deploying AI workloads, making it an essential first step for board bring-up and ongoing system health checks.
Overview#
The XRT System Management Interface (xrt-smi) is a command-line utility included with the XRT runtime package. It provides device configuration, examination, and validation capabilities.
On the AI Engine-ML v2 Board (AMD Versal VEK385), the system exposes two XRT devices:
BDF |
Name |
Description |
|---|---|---|
|
VE2 |
AIE2-PS NPU device (aie2ps, 7x36) |
|
edge |
ZOCL edge device (PL/PS interface) |
Important
All AI Engine-ML v2 specific xrt-smi commands must target device 0000:00:00.0 using the -d flag. The edge device (0000:00:00.1) does not support standard validation tests.
Prerequisites#
Before running xrt-smi, the PL + AIE overlay must be loaded:
fpgautil -b /overlay/x_plus_ml.pdi -o /overlay/x_plus_ml.dtbo
General Syntax#
xrt-smi [--help] [--version] [--verbose] [--batch] [--force] [command [commandArgs]]
Available top-level commands:
Command |
Description |
|---|---|
|
Device and host configuration |
|
Status of the system and device |
|
Download the acceleration program to a given device |
|
Resets the given device |
|
Validates the basic device acceleration functionality |
xrt-smi examine#
Description#
Displays comprehensive system information, XRT software version details, driver versions, and a summary of all discovered XRT devices.
Syntax#
xrt-smi examine
Output Sections#
System Configuration — host-level information about the operating environment.
Field |
Description |
Example Value |
|---|---|---|
OS Name |
Operating system |
Linux |
Release |
Kernel version |
6.12.40-xilinx |
Machine |
CPU architecture |
aarch64 |
CPU Cores |
Number of available CPU cores |
8 |
Memory |
Total system RAM (MB) |
19871 MB |
Distribution |
Linux distribution name and build |
AMD Embedded Development Framework … |
GLIBC |
GNU C Library version |
2.39 |
Model |
Hardware board model |
AMD Versal VEK385 revB |
XRT — XRT runtime and driver version information.
Field |
Description |
Example Value |
|---|---|---|
Version |
XRT software version |
2.23.0 |
Branch |
Git branch of the XRT build |
HEAD |
Hash |
Git commit hash of the XRT build |
eb7059812aff4e68177b143b… |
Hash Date |
Timestamp of the XRT build commit |
Thu, 23 Apr 2026 17:00:35 -0700 |
amdxdna Version |
NPU kernel driver version and commit |
2.23.0_20260424, 9a27eba… |
zocl Version |
ZOCL edge driver version and commit |
2.23.0, eb70598… |
CERT Firmware Version |
Microcontroller CERT firmware version |
1.1.0.29 |
Device(s) Present — all discovered XRT devices.
Field |
Description |
|---|---|
BDF |
Bus:Device.Function identifier |
Shell |
Device shell type ( |
Logic UUID |
UUID of the loaded logic/bitstream |
Device ID |
Device instance identifier |
Device Ready |
Whether the device is ready for use |
Name |
Platform name (VE2 for the NPU device) |
Architecture |
AIE architecture type (aie2ps for AI Engine-ML v2) |
Topology |
AIE array dimensions (7x36 for AI Engine-ML v2) |
Example#
root@amd-edf:~# xrt-smi examine
System Configuration
OS Name : Linux
Release : 6.12.40-xilinx
Machine : aarch64
CPU Cores : 8
Memory : 19871 MB
Distribution : AMD Embedded Development Framework Linux distribution ...
GLIBC : 2.39
Model : AMD Versal VEK385 revB
XRT
Version : 2.23.0
Branch : HEAD
Hash : eb7059812aff4e68177b143b183bf54c141e0782
Hash Date : Thu, 23 Apr 2026 17:00:35 -0700
amdxdna Version : 2.23.0_20260424, 9a27eba307eea1e397e06419ec29f3158982cf0c
zocl Version : 2.23.0, eb7059812aff4e68177b143b183bf54c141e0782
CERT Firmware Version: 1.1.0.29
Device(s) Present
|BDF |Shell |Logic UUID |Device ID |Device Ready* |
|----------------|-------|------------|--------------|---------------|
|[0000:00:00.1] |edge |0x0 |user(inst=0) |Yes |
|BDF |Name |Architecture |Topology |
|----------------|------|--------------|----------|
|[0000:00:00.0] |VE2 |aie2ps |7x36 |
xrt-smi examine –report#
Description#
Generates specific device reports for the AI Engine-ML v2 NPU. The --report flag accepts a report type to produce targeted diagnostic information about the device.
Syntax#
xrt-smi examine -d <BDF> --report <report_type>
Parameters#
Parameter |
Required |
Description |
|---|---|---|
|
Yes |
Target device BDF (use |
|
Yes |
Report type to produce (see the following table) |
|
No |
Output format: |
|
No |
Write output to the specified file |
Available Report Types for AI Engine-ML v2#
To see all available report types, run:
xrt-smi examine -d 0000:00:00.0 --report --help
Report Type |
Description |
|---|---|
|
All known reports are produced |
|
AIE metadata in xclbin |
|
AIE memory tile information |
|
AIE shim tile status |
|
AIE partition information |
|
Host information |
|
Platforms flashed on the device |
–report platform#
Reports platform-specific hardware details including the platform name, power mode support, and AIE array topology.
Output Fields
Field |
Description |
Example Value |
|---|---|---|
Name |
Platform name |
VE2 |
Power Mode |
Current power mode |
not supported |
Total Columns |
Total number of AIE columns available |
36 |
Estimated Power |
Estimated device power consumption |
N/A |
Temperature |
Current device temperature (Celsius) |
N/A |
Example
root@amd-edf:~# xrt-smi examine -d 0000:00:00.0 --report platform
----------------------
[0000:00:00.0] : VE2
----------------------
Platform
Name : VE2
Power Mode : not supported
Total Columns : 36
Estimated Power : N/A
Temperature (C) : N/A
Note
Power estimation and temperature monitoring are not currently supported on AI Engine-ML v2. These fields report N/A.
root@amd-edf:~# xrt-smi examine --device 0 --report aie-partitions
---------------------
[0000:00:00.0] : VE2
---------------------
AIE Partitions
Total Memory Usage: N/A
Partition Index : 0
Columns: [0, 1, 2, 3]
HW Contexts:
|PID |Ctx ID |Submissions |Migrations |Err |Priority |
|Process Name |Status |Completions |Suspensions | |GOPS |
|Memory Usage |Instr BO | | | |FPS |
| | | | | |Latency |
|====================|===========|============|============|=====|=========|
|1222 |1 |5632 |0 |0 |Normal |
|N/A |Active |5632 |0 | |1 |
|105 MB |N/A | | | |1 |
| | | | | |2000 |
|--------------------|-----------|------------|------------|-----|---------|
Partition Index : 1
Columns: [4, 5, 6, 7]
HW Contexts:
|PID |Ctx ID |Submissions |Migrations |Err |Priority |
|Process Name |Status |Completions |Suspensions | |GOPS |
|Memory Usage |Instr BO | | | |FPS |
| | | | | |Latency |
|====================|===========|============|============|=====|=========|
|1222 |2 |5686 |0 |0 |Normal |
|N/A |Active |5686 |0 | |1 |
|105 MB |N/A | | | |1 |
| | | | | |2000 |
|--------------------|-----------|------------|------------|-----|---------|
Partition Index : 2
Columns: [8, 9, 10, 11]
HW Contexts:
|PID |Ctx ID |Submissions |Migrations |Err |Priority |
|Process Name |Status |Completions |Suspensions | |GOPS |
|Memory Usage |Instr BO | | | |FPS |
| | | | | |Latency |
|====================|===========|============|============|=====|=========|
|1222 |3 |5768 |0 |0 |Normal |
|N/A |Active |5768 |0 | |1 |
|105 MB |N/A | | | |1 |
| | | | | |2000 |
|--------------------|-----------|------------|------------|-----|---------|
Partition Index : 3
Columns: [12, 13, 14, 15]
HW Contexts:
|PID |Ctx ID |Submissions |Migrations |Err |Priority |
|Process Name |Status |Completions |Suspensions | |GOPS |
|Memory Usage |Instr BO | | | |FPS |
| | | | | |Latency |
|====================|===========|============|============|=====|=========|
|1222 |4 |5802 |0 |0 |Normal |
|N/A |Idle |5801 |0 | |1 |
|105 MB |N/A | | | |1 |
| | | | | |2000 |
|--------------------|-----------|------------|------------|-----|---------|
–report all#
Produces all available reports in a single command. This combines the System Configuration, XRT version, device list, and all device-specific reports (AIE core tiles, AIE memory tiles, AIE shim tiles, AIE partitions, and platform).
Output Sections
Section |
Description |
|---|---|
System Configuration |
Host OS, kernel, CPU, memory, board model (same as |
XRT |
XRT version, driver versions, CERT firmware version |
Device(s) Present |
All discovered devices with BDF, shell, architecture |
AIE Core Tiles |
Status of AIE core tile columns (active or idle) |
AIE Mem Tiles |
Status of AIE memory tile columns (active or idle) |
AIE Shim Tiles |
Status of AIE shim tile columns (active or idle) |
AIE Partitions |
Active hardware contexts running on the device |
Platform |
Platform name, power mode, total columns, power, temperature |
Note
When no AIE workload is running, the AIE tile sections report “No AIE columns are active on the device” and AIE Partitions reports “No hardware contexts running on device”. These sections become populated when an xclbin is loaded and a hardware context is active.
Example
root@amd-edf:~# xrt-smi examine -d 0000:00:00.0 --report all
System Configuration
OS Name : Linux
Release : 6.12.40-xilinx
Machine : aarch64
CPU Cores : 8
Memory : 19871 MB
Distribution : AMD Embedded Development Framework Linux distribution ...
GLIBC : 2.39
Model : AMD Versal VEK385 revB
XRT
Version : 2.23.0
Branch : HEAD
Hash : eb7059812aff4e68177b143b183bf54c141e0782
Hash Date : Thu, 23 Apr 2026 17:00:35 -0700
amdxdna Version : 2.23.0_20260424, 9a27eba307eea1e397e06419ec29f3158982cf0c
zocl Version : 2.23.0, eb7059812aff4e68177b143b183bf54c141e0782
CERT Firmware Version: 1.1.0.29
Device(s) Present
|BDF |Shell |Logic UUID |Device ID |Device Ready* |
|----------------|-------|------------|--------------|---------------|
|[0000:00:00.1] |edge |0x0 |user(inst=0) |Yes |
|BDF |Name |Architecture |Topology |
|----------------|------|--------------|----------|
|[0000:00:00.0] |VE2 |aie2ps |7x36 |
---------------------
[0000:00:00.0] : VE2
---------------------
AIE Core Tiles
No AIE columns are active on the device
AIE Mem Tiles
No AIE columns are active on the device
AIE Shim Tiles
No AIE columns are active on the device
AIE Partitions
No hardware contexts running on device
Platform
Name : VE2
Power Mode : not supported
Total Columns : 36
Estimated Power : N/A
Temperature (C) : N/A
xrt-smi validate#
Description#
Runs the complete platform validation test suite on the specified device. For AI Engine-ML v2, this executes two tests sequentially: the latency test followed by the throughput test. Both tests use a no-operation (NOP) AIE kernel to measure the XRT command submission infrastructure overhead, independent of any actual compute workload.
Syntax#
xrt-smi validate -d <BDF> [--verbose] [--batch] [--force]
Parameters#
Parameter |
Required |
Description |
|---|---|---|
|
Yes |
Target device BDF (use |
|
No |
Run a subset of the test suite (default |
|
No |
Display additional test details (for example, descriptions) |
|
No |
Output format: |
|
No |
Write output to the specified file |
Available Tests on AI Engine-ML v2#
Test Name |
Description |
|---|---|
|
Run end-to-end latency test |
|
Run end-to-end throughput test |
Note
The generic --help output lists many additional test names (aie, dma, pcie-link, etc.) that are applicable to PCIe-based datacenter accelerators. On the AI Engine-ML v2 platform, only latency and throughput are valid. Attempting to run an unsupported test returns ERROR: Invalid test name.
Test Details#
Latency Test
Measures the round-trip latency for a single AIE kernel dispatch-to-completion cycle.
What it measures: End-to-end time from the host issuing a single kernel execution command through the amdxdna driver to the AIE array, executing a NOP instruction, and returning completion status back to the host.
Method: Executes 10,000 iterations of a single NOP kernel run and computes the average latency per iteration.
Pass criteria: The test completes successfully and reports a latency value.
Typical result: ~29-31 microseconds.
What it validates: XRT/amdxdna driver command path, hardware context creation, kernel dispatch mechanism, AIE mailbox/doorbell signaling, and completion interrupt handling.
Throughput Test
Measures the sustained kernel dispatch rate (operations per second).
What it measures: The maximum number of NOP kernel executions that can be dispatched and completed per second.
Method: Executes 10,000 iterations of a single NOP kernel run and computes the operations per second based on total elapsed time.
Pass criteria: The test completes successfully and reports a throughput value.
Typical result: ~100,000-101,000 ops/s.
What it validates: XRT/amdxdna driver sustained dispatch performance, command queue efficiency, and overhead of the full host-to-AIE-to-host round trip.
Example#
root@amd-edf:~# xrt-smi validate -d 0000:00:00.0
Validate Device : [0000:00:00.0]
Platform : VE2
Power Mode : default
-------------------------------------------------------------------------------
Test 1 [0000:00:00.0] : latency
Details : Average latency: 31.0 us
Test Status : [PASSED]
-------------------------------------------------------------------------------
Test 2 [0000:00:00.0] : throughput
Details : Average throughput: 100929.0 op/s
Test Status : [PASSED]
-------------------------------------------------------------------------------
Validation completed
Quick Reference#
Device Selection#
When multiple devices are present, xrt-smi requires explicit device selection. Without -d, it prompts:
NOTE: Multiple devices found. Showing help for 0000:00:00.1 device
ERROR: Please specify a device using --device option
Available devices:
[0000:00:00.0] : VE2
[0000:00:00.1] : edge
Always use -d 0000:00:00.0 for AI Engine-ML v2 NPU operations.
Command Summary#
Command |
Purpose |
|---|---|
|
System info, XRT version, device list |
|
List available report types for AI Engine-ML v2 |
|
AI Engine-ML v2 platform details (columns, etc.) |
|
All reports combined (AIE tiles, platform) |
|
Run all validation tests |
|
Run latency test only |
|
Run throughput test only |
|
Run all tests with descriptions |
Interpreting Results#
Metric |
Typical Value |
What It Means |
|---|---|---|
Average latency |
~30 us |
Single kernel dispatch round-trip time |
Average throughput |
~100,000 ops/s |
Sustained kernel dispatch rate |
Troubleshooting#
Issue |
Cause |
Resolution |
|---|---|---|
|
Test not supported on AI Engine-ML v2 |
Use |
|
Multiple devices, no |
Add |
|
Edge device does not support this validate |
Use |
Validation tests not found |
PL overlay not loaded |
Run |