embeded/jetson2026. 7. 24. 11:39

헛다리를 짚은것 같은데 메모리가 안나오는건 공유 메모리를 써서 인가.. 싶기도 하고?

아무튼 thor iGPU에서는 최대 2개로 나누어 쓸 수 있나 본데

gfx는 외부 출력 포트를 쓸수 있냐 없냐의 차이인것 같고

[링크 : https://docs.nvidia.com/datacenter/tesla/mig-user-guide/supported-mig-profiles.html]

 

nvidia@localhost:~$ nvidia-smi -L
GPU 0: NVIDIA Thor (UUID: GPU-a7c66ad2-6dbb-0ab8-c1a2-37ba6dba3600)

nvidia@localhost:~$ nvidia-smi -i 0 -q

==============NVSMI LOG==============

Timestamp                                 : Fri Jul 24 11:33:07 2026
Driver Version                            : 580.00
CUDA Version                              : 13.0

Attached GPUs                             : 1
GPU 00000000:01:00.0
    Product Name                          : NVIDIA Thor
    Product Brand                         : GeForce
    Product Architecture                  : Blackwell
    Display Mode                          : Requested functionality has been deprecated
    Display Attached                      : No
    Display Active                        : Disabled
    Persistence Mode                      : Disabled
    Addressing Mode                       : ATS
    MIG Mode
        Current                           : Disabled
        Pending                           : Disabled
    Accounting Mode                       : Disabled
    Accounting Mode Buffer Size           : 4000
    Driver Model
        Current                           : N/A
        Pending                           : N/A
    Serial Number                         : N/A
    GPU UUID                              : GPU-a7c66ad2-6dbb-0ab8-c1a2-37ba6dba3600
    GPU PDI                               : 0x9257b02c0bd38a20
    Minor Number                          : 0
    VBIOS Version                         : N/A
    MultiGPU Board                        : No
    Board ID                              : 0x100
    Board Part Number                     : N/A
    GPU Part Number                       : 2B00--A1
    FRU Part Number                       : N/A
    Platform Info
        Chassis Serial Number             : N/A
        Slot Number                       : N/A
        Tray Index                        : N/A
        Host ID                           : N/A
        Peer Type                         : N/A
        Module Id                         : N/A
        GPU Fabric GUID                   : N/A
    Inforom Version
        Image Version                     : N/A
        OEM Object                        : N/A
        ECC Object                        : N/A
        Power Management Object           : N/A
    Inforom BBX Object Flush
        Latest Timestamp                  : N/A
        Latest Duration                   : N/A
    GPU Operation Mode
        Current                           : N/A
        Pending                           : N/A
    GPU C2C Mode                          : Disabled
    GPU Virtualization Mode
        Virtualization Mode               : None
        Host VGPU Mode                    : N/A
        vGPU Heterogeneous Mode           : N/A
    GPU Recovery Action                   : None
    GSP Firmware Version                  : 580.00
    IBMNPU
        Relaxed Ordering Mode             : N/A
    PCI
        Bus                               : 0x01
        Device                            : 0x00
        Domain                            : 0x0000
        Base Classcode                    : 0x3
        Sub Classcode                     : 0x2
        Device Id                         : 0x2B0010DE
        Bus Id                            : 00000000:01:00.0
        Sub System Id                     : 0x00000000
        GPU Link Info
            PCIe Generation
                Max                       : 1
                Current                   : 1
                Device Current            : 1
                Device Max                : 5
                Host Max                  : 1
            Link Width
                Max                       : 16x
                Current                   : 1x
        Bridge Chip
            Type                          : N/A
            Firmware                      : N/A
        Replays Since Reset               : 0
        Replay Number Rollovers           : 0
        Tx Throughput                     : N/A
        Rx Throughput                     : N/A
        Atomic Caps Outbound              : FETCHADD_32 FETCHADD_64 SWAP_32 SWAP_64 CAS_32 CAS_64 
        Atomic Caps Inbound               : N/A
    Fan Speed                             : N/A
    Performance State                     : N/A
    Clocks Event Reasons                  : N/A
    Clocks Event Reasons Counters
        SW Power Capping                  : N/A
        Sync Boost                        : N/A
        SW Thermal Slowdown               : N/A
        HW Thermal Slowdown               : N/A
        HW Power Braking                  : N/A
    Sparse Operation Mode                 : N/A
    FB Memory Usage
        Total                             : N/A
        Reserved                          : N/A
        Used                              : N/A
        Free                              : N/A
    BAR1 Memory Usage
        Total                             : N/A
        Used                              : N/A
        Free                              : N/A
    Conf Compute Protected Memory Usage
        Total                             : 0 MiB
        Used                              : 0 MiB
        Free                              : 0 MiB
    Compute Mode                          : Default
    Utilization
        GPU                               : 98 %
        Memory                            : 0 %
        Encoder                           : 0 %
        Decoder                           : 0 %
        JPEG                              : 0 %
        OFA                               : 0 %
    Encoder Stats
        Active Sessions                   : 0
        Average FPS                       : 0
        Average Latency                   : 0
    FBC Stats
        Active Sessions                   : 0
        Average FPS                       : 0
        Average Latency                   : 0
    DRAM Encryption Mode
        Current                           : N/A
        Pending                           : N/A
    ECC Mode
        Current                           : N/A
        Pending                           : N/A
    ECC Errors
        Volatile
            SRAM Correctable              : N/A
            SRAM Uncorrectable Parity     : N/A
            SRAM Uncorrectable SEC-DED    : N/A
            DRAM Correctable              : N/A
            DRAM Uncorrectable            : N/A
        Aggregate
            SRAM Correctable              : N/A
            SRAM Uncorrectable Parity     : N/A
            SRAM Uncorrectable SEC-DED    : N/A
            DRAM Correctable              : N/A
            DRAM Uncorrectable            : N/A
            SRAM Threshold Exceeded       : N/A
        Aggregate Uncorrectable SRAM Sources
            SRAM L2                       : N/A
            SRAM SM                       : N/A
            SRAM Microcontroller          : N/A
            SRAM PCIE                     : N/A
            SRAM Other                    : N/A
        Channel Repair Pending            : N/A
        TPC Repair Pending                : N/A
    Retired Pages
        Single Bit ECC                    : N/A
        Double Bit ECC                    : N/A
        Pending Page Blacklist            : N/A
    Remapped Rows                         : N/A
    Temperature
        GPU Current Temp                  : 60 C
        GPU T.Limit Temp                  : N/A
        GPU Shutdown Temp                 : N/A
        GPU Slowdown Temp                 : N/A
        GPU Max Operating Temp            : N/A
        GPU Target Temperature            : N/A
        Memory Current Temp               : N/A
        Memory Max Operating Temp         : N/A
    GPU Power Readings
        Average Power Draw                : N/A
        Instantaneous Power Draw          : 36.21 W
        Current Power Limit               : N/A
        Requested Power Limit             : N/A
        Default Power Limit               : N/A
        Min Power Limit                   : N/A
        Max Power Limit                   : N/A
    GPU Memory Power Readings 
        Average Power Draw                : N/A
        Instantaneous Power Draw          : N/A
    Module Power Readings
        Average Power Draw                : N/A
        Instantaneous Power Draw          : N/A
        Current Power Limit               : N/A
        Requested Power Limit             : N/A
        Default Power Limit               : N/A
        Min Power Limit                   : N/A
        Max Power Limit                   : N/A
    Power Smoothing                       : N/A
    Workload Power Profiles
        Requested Profiles                : N/A
        Enforced Profiles                 : N/A
    Clocks
        Graphics                          : 1575 MHz
        SM                                : N/A
        Memory                            : N/A
        Video                             : 315 MHz
    Applications Clocks
        Graphics                          : Requested functionality has been deprecated
        Memory                            : Requested functionality has been deprecated
    Default Applications Clocks
        Graphics                          : Requested functionality has been deprecated
        Memory                            : Requested functionality has been deprecated
    Deferred Clocks
        Memory                            : N/A
    Max Clocks
        Graphics                          : N/A
        SM                                : N/A
        Memory                            : N/A
        Video                             : N/A
    Max Customer Boost Clocks
        Graphics                          : N/A
    Clock Policy
        Auto Boost                        : N/A
        Auto Boost Default                : N/A
    Fabric
        State                             : Deprecated; See field Probe State instead
        Probe State                       : N/A
        Status                            : Deprecated; See field Probe Status instead
        Probe Status                      : N/A
        CliqueId                          : N/A
        ClusterUUID                       : N/A
        Health
            Summary                       : N/A
            Bandwidth                     : N/A
            Route Recovery in progress    : N/A
            Route Unhealthy               : N/A
            Access Timeout Recovery       : N/A
            Incorrect Configuration       : N/A
    Processes
        GPU instance ID                   : N/A
        Compute instance ID               : N/A
        Process ID                        : 3559
            Type                          : G
            Name                          : /usr/lib/xorg/Xorg
            Used GPU Memory               : 95 MiB
        GPU instance ID                   : N/A
        Compute instance ID               : N/A
        Process ID                        : 3775
            Type                          : G
            Name                          : /usr/bin/gnome-shell
            Used GPU Memory               : 58 MiB
        GPU instance ID                   : N/A
        Compute instance ID               : N/A
        Process ID                        : 8710
            Type                          : C+G
            Name                          : ./llama-b10099/llama-cli
            Used GPU Memory               : 33671 MiB
    Capabilities
        EGM                               : disabled

nvidia@localhost:~$ nvidia-smi mig -lgip
+-------------------------------------------------------------------------------+
| GPU instance profiles:                                                        |
| GPU   Name               ID    Instances   Memory     P2P    SM    DEC   ENC  |
|                                Free/Total   GiB              CE    JPEG  OFA  |
|===============================================================================|

[링크 : https://blog.naver.com/techtrip/224167049409]

[링크 : https://docs.nvidia.com/jetson/archives/r39.2/DeveloperGuide/SD/MiG.html#getting-started-with-mig-on-jetson-thor]

+

gdm만 죽이나 isolate 해서 죽이나 별반 차이 없는 것 같이 안된다 ㅠㅠ

nvidia@localhost:~$ nvidia-smi -i 0
Fri Jul 24 11:57:43 2026       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.00                 Driver Version: 580.00         CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA Thor                    Off |   00000000:01:00.0 Off |                  N/A |
| N/A   40C  N/A               2W /  N/A  | Not Supported          |      0%      Default |
|                                         |                        |             Disabled |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            3559      G   /usr/lib/xorg/Xorg                       95MiB |
|    0   N/A  N/A            3775      G   /usr/bin/gnome-shell                     58MiB |
+-----------------------------------------------------------------------------------------+
nvidia@localhost:~$ systemctl stop gdm
==== AUTHENTICATING FOR org.freedesktop.systemd1.manage-units ====
Authentication is required to stop 'gdm.service'.
Authenticating as: nvidia
Password: 
==== AUTHENTICATION COMPLETE ====
nvidia@localhost:~$ nvidia-smi -i 0
Fri Jul 24 11:58:10 2026       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.00                 Driver Version: 580.00         CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA Thor                    Off |   00000000:01:00.0 Off |                  N/A |
| N/A   40C  N/A               1W /  N/A  | Not Supported          |      0%      Default |
|                                         |                        |             Disabled |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+
nvidia@localhost:~$ nvidia-smi -L
GPU 0: NVIDIA Thor (UUID: GPU-a7c66ad2-6dbb-0ab8-c1a2-37ba6dba3600)
nvidia@localhost:~$ nvidia-smi -i 0 -mig 1
Unable to enable MIG Mode for GPU 00000000:01:00.0: Insufficient Permissions
Terminating early due to previous errors.
nvidia@localhost:~$ sudo nvidia-smi -i 0 -mig 1
Enabled MIG Mode for GPU 00000000:01:00.0

Warning: persistence mode is disabled on device 00000000:01:00.0. See the Known Issues section of the nvidia-smi(1) man page for more information. Run with [--help | -h] switch to get more information on how to enable persistence mode.
All done.
nvidia@localhost:~$ nvidia-smi -L
GPU 0: NVIDIA Thor (UUID: GPU-a7c66ad2-6dbb-0ab8-c1a2-37ba6dba3600)
nvidia@localhost:~$ sudo nvidia-smi -L
GPU 0: NVIDIA Thor (UUID: GPU-a7c66ad2-6dbb-0ab8-c1a2-37ba6dba3600)
nvidia@localhost:~$ nvidia-smi
Fri Jul 24 12:00:54 2026       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.00                 Driver Version: 580.00         CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA Thor                    Off |   00000000:01:00.0 Off |                  N/A |
| N/A   40C  N/A               3W /  N/A  | Not Supported          |     N/A      Default |
|                                         |                        |              Enabled |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| MIG devices:                                                                            |
+------------------+----------------------------------+-----------+-----------------------+
| GPU  GI  CI  MIG |              Shared Memory-Usage |        Vol|        Shared         |
|      ID  ID  Dev |                Shared BAR1-Usage | SM     Unc| CE ENC  DEC  OFA  JPG |
|                  |                                  |        ECC|                       |
|==================+==================================+===========+=======================|
|  No MIG devices found                                                                   |
+-----------------------------------------------------------------------------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+
nvidia@localhost:~$ nvidia-smi mig -lgip
+-------------------------------------------------------------------------------+
| GPU instance profiles:                                                        |
| GPU   Name               ID    Instances   Memory     P2P    SM    DEC   ENC  |
|                                Free/Total   GiB              CE    JPEG  OFA  |
|===============================================================================|
Failed to display GPU instance profiles: Unknown Error
nvidia@localhost:~$ sudo nvidia-smi mig -lgip
+-------------------------------------------------------------------------------+
| GPU instance profiles:                                                        |
| GPU   Name               ID    Instances   Memory     P2P    SM    DEC   ENC  |
|                                Free/Total   GiB              CE    JPEG  OFA  |
|===============================================================================|
Failed to display GPU instance profiles: Unknown Error







nvidia@localhost:~$ sudo nvidia-smi -pm 1
Enabled Legacy persistence mode for GPU 00000000:01:00.0.
All done.
nvidia@localhost:~$ sudo nvidia-smi -i 0 -mig 1
Enabled MIG Mode for GPU 00000000:01:00.0
All done.
nvidia@localhost:~$ sudo nvidia-smi mig -lgip
+-------------------------------------------------------------------------------+
| GPU instance profiles:                                                        |
| GPU   Name               ID    Instances   Memory     P2P    SM    DEC   ENC  |
|                                Free/Total   GiB              CE    JPEG  OFA  |
|===============================================================================|
Failed to display GPU instance profiles: Unknown Error

 

driver 버전은 이런데..

NVIDIA-SMI 580.00                 Driver Version: 580.00         CUDA Version: 13.0 

[링크 : https://forums.developer.nvidia.com/t/jetson-thor-mig-enabled-but-unable-to-create-gpu-instances-driver-580-00-l4t-r38-2/372073]

 

jetpack 버전은 38.4 같다. 일단 위에서 JetPack 7.2GA/R39.2 라고 했으니 업데이드 해봐야하나?

$ dpkg-query --show nvidia-l4t-core
nvidia-l4t-core 38.4.0-20251230160601

[링크 : https://yoo-hk.tistory.com/555]

 

 

'embeded > jetson' 카테고리의 다른 글

nvidia jetson 아니고 nvidia thor  (0) 2026.07.24
jetson nano + llama.cpp vulkan  (0) 2026.07.21
jetson nano jtop  (0) 2026.07.20
jetson nano 4GB / qwen3.5 0.8B  (0) 2026.07.20
jetson nano 과 llama.cpp (가속, 빌드 포기)  (0) 2026.07.20
Posted by 구차니