github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 08:02:14 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #4814] Only Detecting One MIG Instance #3040

New Issue
Open
opened 2026-04-12 13:27:52 -05:00 by GiteaMirror · 6 comments
GiteaMirror commented 2026-04-12 13:27:52 -05:00
Owner
Copy Link

Originally created by @Magitoneu on GitHub (Jun 4, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4814

Originally assigned to: @dhiltgen on GitHub.

What is the issue?

I am running Ollama as a service on a server with 2xA100 GPUs, both split into 4 MIG instances. I want Ollama to keep two models loaded, each in a different MIG instance. However, Ollama is detecting only one of the MIG instances that I have assigned, so it does not load both models but swaps them.

Environment vars:

Environment="CUDA_VISIBLE_DEVICES=MIG-d001b894-6dc3-5220-a1bf-fb944eb1b13b,MIG-0665cd24-2540-5c33-a995-636c4fcab1cf"
Environment="OLLAMA_MAX_LOADED_MODELS=2"

Nvidia-smi

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 510.108.03   Driver Version: 510.108.03   CUDA Version: 11.6     |
|-------------------------------+----------------------+----------------------+
| 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 A100 80G...  Off  | 00000000:05:00.0 Off |                   On |
| N/A   44C    P0    66W / 300W |     48MiB / 81920MiB |     N/A      Default |
|                               |                      |              Enabled |
+-------------------------------+----------------------+----------------------+
|   1  NVIDIA A100 80G...  Off  | 00000000:86:00.0 Off |                   On |
| N/A   34C    P0    67W / 300W |     45MiB / 81920MiB |     N/A      Default |
|                               |                      |              Enabled |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| MIG devices:                                                                |
+------------------+----------------------+-----------+-----------------------+
| GPU  GI  CI  MIG |         Memory-Usage |        Vol|         Shared        |
|      ID  ID  Dev |           BAR1-Usage | SM     Unc| CE  ENC  DEC  OFA  JPG|
|                  |                      |        ECC|                       |
|==================+======================+===========+=======================|
|  0    2   0   0  |     19MiB / 40192MiB | 42      0 |  3   0    2    0    0 |
|                  |      0MiB / 65535MiB |           |                       |
+------------------+----------------------+-----------+-----------------------+
|  0    7   0   1  |      8MiB /  9728MiB | 14      0 |  1   0    0    0    0 |
|                  |      0MiB / 16383MiB |           |                       |
+------------------+----------------------+-----------+-----------------------+
|  0    8   0   2  |      6MiB /  9728MiB | 14      0 |  1   0    0    0    0 |
|                  |      0MiB / 16383MiB |           |                       |
+------------------+----------------------+-----------+-----------------------+
|  0    9   0   3  |      6MiB /  9728MiB | 14      0 |  1   0    0    0    0 |
|                  |      0MiB / 16383MiB |           |                       |
+------------------+----------------------+-----------+-----------------------+
|  0   10   0   4  |      6MiB /  9728MiB | 14      0 |  1   0    0    0    0 |
|                  |      0MiB / 16383MiB |           |                       |
+------------------+----------------------+-----------+-----------------------+
|  1    2   0   0  |     19MiB / 40192MiB | 42      0 |  3   0    2    0    0 |
|                  |      0MiB / 65535MiB |           |                       |
+------------------+----------------------+-----------+-----------------------+
|  1    7   0   1  |      6MiB /  9728MiB | 14      0 |  1   0    0    0    0 |
|                  |      0MiB / 16383MiB |           |                       |
+------------------+----------------------+-----------+-----------------------+
|  1    8   0   2  |      6MiB /  9728MiB | 14      0 |  1   0    0    0    0 |
|                  |      0MiB / 16383MiB |           |                       |
+------------------+----------------------+-----------+-----------------------+
|  1    9   0   3  |      6MiB /  9728MiB | 14      0 |  1   0    0    0    0 |
|                  |      0MiB / 16383MiB |           |                       |
+------------------+----------------------+-----------+-----------------------+
|  1   10   0   4  |      6MiB /  9728MiB | 14      0 |  1   0    0    0    0 |
|                  |      0MiB / 16383MiB |           |                       |
+------------------+----------------------+-----------+-----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

GPU 0: NVIDIA A100 80GB PCIe (UUID: GPU-fce143c8-f3e2-5db9-a955-1541d6ae4ed6)
  MIG 3g.40gb     Device  0: (UUID: MIG-cb4aa05b-5bb3-5f35-8028-590348715f02)
  MIG 1g.10gb     Device  1: (UUID: MIG-d001b894-6dc3-5220-a1bf-fb944eb1b13b)
  MIG 1g.10gb     Device  2: (UUID: MIG-0665cd24-2540-5c33-a995-636c4fcab1cf)
  MIG 1g.10gb     Device  3: (UUID: MIG-3a059506-610c-5274-984f-065c39393bdc)
  MIG 1g.10gb     Device  4: (UUID: MIG-91c89459-4898-5209-90b7-b0ede2e5ec35)
GPU 1: NVIDIA A100 80GB PCIe (UUID: GPU-14122cde-73d9-d15c-5de8-4ce2da4ff2a2)
  MIG 3g.40gb     Device  0: (UUID: MIG-4770598e-9eea-56f1-a483-dd22f5725c58)
  MIG 1g.10gb     Device  1: (UUID: MIG-94bf4e63-eeda-55e0-a038-ca15deeb6a98)
  MIG 1g.10gb     Device  2: (UUID: MIG-e15bc69a-fdaa-5b4a-bbdd-e2c836785d91)
  MIG 1g.10gb     Device  3: (UUID: MIG-dc0d7d2f-15be-5eb7-bfbb-800b1c8f0368)
  MIG 1g.10gb     Device  4: (UUID: MIG-6c1c7cef-7926-5217-a5bd-b2298a3771a9)

Ollama Logs

Jun 04 15:13:42 rack-ai-0 systemd[1]: Started Ollama Service.
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: 2024/06/04 15:13:42 routes.go:1007: INFO server config env="map[OLLAMA_DEBUG:true OLLAMA_FLASH_ATTENTION:false OLLAMA_HOST: OLLAMA_KEEP_ALIVE: OLLAMA_LLM_LIBRARY: OLLAMA_MAX_LOADED_MODELS:2 OLLAMA_MAX_QUEUE:512>
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.076+02:00 level=INFO source=images.go:729 msg="total blobs: 44"
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.076+02:00 level=INFO source=images.go:736 msg="total unused blobs removed: 0"
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.076+02:00 level=INFO source=routes.go:1053 msg="Listening on [::]:11434 (version 0.1.41)"
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=INFO source=payload.go:30 msg="extracting embedded files" dir=/tmp/ollama1162914854/runners
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cpu file=build/linux/x86_64/cpu/bin/ollama_llama_server.gz
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cpu_avx file=build/linux/x86_64/cpu_avx/bin/ollama_llama_server.gz
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cpu_avx2 file=build/linux/x86_64/cpu_avx2/bin/ollama_llama_server.gz
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cuda_v11 file=build/linux/x86_64/cuda_v11/bin/libcublas.so.11.gz
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cuda_v11 file=build/linux/x86_64/cuda_v11/bin/libcublasLt.so.11.gz
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cuda_v11 file=build/linux/x86_64/cuda_v11/bin/libcudart.so.11.0.gz
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cuda_v11 file=build/linux/x86_64/cuda_v11/bin/ollama_llama_server.gz
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=rocm_v60002 file=build/linux/x86_64/rocm_v60002/bin/deps.txt.gz
Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=rocm_v60002 file=build/linux/x86_64/rocm_v60002/bin/ollama_llama_server.gz
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:71 msg="availableServers : found" file=/tmp/ollama1162914854/runners/cpu
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:71 msg="availableServers : found" file=/tmp/ollama1162914854/runners/cpu_avx
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:71 msg="availableServers : found" file=/tmp/ollama1162914854/runners/cpu_avx2
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:71 msg="availableServers : found" file=/tmp/ollama1162914854/runners/cuda_v11
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:71 msg="availableServers : found" file=/tmp/ollama1162914854/runners/rocm_v60002
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=INFO source=payload.go:44 msg="Dynamic LLM libraries [cpu cpu_avx cpu_avx2 cuda_v11 rocm_v60002]"
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:45 msg="Override detection logic by setting OLLAMA_LLM_LIBRARY"
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=sched.go:90 msg="starting llm scheduler"
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=gpu.go:132 msg="Detecting GPUs"
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=gpu.go:274 msg="Searching for GPU library" name=libcuda.so*
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=gpu.go:293 msg="gpu library search" globs="[/libcuda.so** /usr/local/cuda*/targets/*/lib/libcuda.so* /usr/lib/*-linux-gnu/nvidia/current/libcuda.so* /usr/li>
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.660+02:00 level=DEBUG source=gpu.go:326 msg="discovered GPU libraries" paths=[/usr/lib/x86_64-linux-gnu/libcuda.so.510.108.03]
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: CUDA driver version: 11.6
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.749+02:00 level=DEBUG source=gpu.go:137 msg="detected GPUs" count=1 library=/usr/lib/x86_64-linux-gnu/libcuda.so.510.108.03
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.749+02:00 level=DEBUG source=cpu_common.go:11 msg="CPU has AVX2"
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: [GPU-fce143c8-f3e2-5db9-a955-1541d6ae4ed6] CUDA totalMem 9728 mb
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: [GPU-fce143c8-f3e2-5db9-a955-1541d6ae4ed6] CUDA freeMem 9645 mb
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: [GPU-fce143c8-f3e2-5db9-a955-1541d6ae4ed6] Compute Capability 8.0
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.888+02:00 level=DEBUG source=amd_linux.go:322 msg="amdgpu driver not detected /sys/module/amdgpu"
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: releasing nvcuda library
Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.888+02:00 level=INFO source=types.go:71 msg="inference compute" id=GPU-fce143c8-f3e2-5db9-a955-1541d6ae4ed6 library=cuda compute=8.0 driver=11.6 name="NVIDIA A100 80GB PCIe MIG 1g.10gb"

OS

Linux

GPU

Nvidia

CPU

AMD

Ollama version

0.1.41

Originally created by @Magitoneu on GitHub (Jun 4, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4814 Originally assigned to: @dhiltgen on GitHub. ### What is the issue? I am running Ollama as a service on a server with 2xA100 GPUs, both split into 4 MIG instances. I want Ollama to keep two models loaded, each in a different MIG instance. However, Ollama is detecting only one of the MIG instances that I have assigned, so it does not load both models but swaps them. Environment vars: ``` Environment="CUDA_VISIBLE_DEVICES=MIG-d001b894-6dc3-5220-a1bf-fb944eb1b13b,MIG-0665cd24-2540-5c33-a995-636c4fcab1cf" Environment="OLLAMA_MAX_LOADED_MODELS=2" ``` Nvidia-smi ``` +-----------------------------------------------------------------------------+ | NVIDIA-SMI 510.108.03 Driver Version: 510.108.03 CUDA Version: 11.6 | |-------------------------------+----------------------+----------------------+ | 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 A100 80G... Off | 00000000:05:00.0 Off | On | | N/A 44C P0 66W / 300W | 48MiB / 81920MiB | N/A Default | | | | Enabled | +-------------------------------+----------------------+----------------------+ | 1 NVIDIA A100 80G... Off | 00000000:86:00.0 Off | On | | N/A 34C P0 67W / 300W | 45MiB / 81920MiB | N/A Default | | | | Enabled | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | MIG devices: | +------------------+----------------------+-----------+-----------------------+ | GPU GI CI MIG | Memory-Usage | Vol| Shared | | ID ID Dev | BAR1-Usage | SM Unc| CE ENC DEC OFA JPG| | | | ECC| | |==================+======================+===========+=======================| | 0 2 0 0 | 19MiB / 40192MiB | 42 0 | 3 0 2 0 0 | | | 0MiB / 65535MiB | | | +------------------+----------------------+-----------+-----------------------+ | 0 7 0 1 | 8MiB / 9728MiB | 14 0 | 1 0 0 0 0 | | | 0MiB / 16383MiB | | | +------------------+----------------------+-----------+-----------------------+ | 0 8 0 2 | 6MiB / 9728MiB | 14 0 | 1 0 0 0 0 | | | 0MiB / 16383MiB | | | +------------------+----------------------+-----------+-----------------------+ | 0 9 0 3 | 6MiB / 9728MiB | 14 0 | 1 0 0 0 0 | | | 0MiB / 16383MiB | | | +------------------+----------------------+-----------+-----------------------+ | 0 10 0 4 | 6MiB / 9728MiB | 14 0 | 1 0 0 0 0 | | | 0MiB / 16383MiB | | | +------------------+----------------------+-----------+-----------------------+ | 1 2 0 0 | 19MiB / 40192MiB | 42 0 | 3 0 2 0 0 | | | 0MiB / 65535MiB | | | +------------------+----------------------+-----------+-----------------------+ | 1 7 0 1 | 6MiB / 9728MiB | 14 0 | 1 0 0 0 0 | | | 0MiB / 16383MiB | | | +------------------+----------------------+-----------+-----------------------+ | 1 8 0 2 | 6MiB / 9728MiB | 14 0 | 1 0 0 0 0 | | | 0MiB / 16383MiB | | | +------------------+----------------------+-----------+-----------------------+ | 1 9 0 3 | 6MiB / 9728MiB | 14 0 | 1 0 0 0 0 | | | 0MiB / 16383MiB | | | +------------------+----------------------+-----------+-----------------------+ | 1 10 0 4 | 6MiB / 9728MiB | 14 0 | 1 0 0 0 0 | | | 0MiB / 16383MiB | | | +------------------+----------------------+-----------+-----------------------+ +-----------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=============================================================================| | No running processes found | +-----------------------------------------------------------------------------+ GPU 0: NVIDIA A100 80GB PCIe (UUID: GPU-fce143c8-f3e2-5db9-a955-1541d6ae4ed6) MIG 3g.40gb Device 0: (UUID: MIG-cb4aa05b-5bb3-5f35-8028-590348715f02) MIG 1g.10gb Device 1: (UUID: MIG-d001b894-6dc3-5220-a1bf-fb944eb1b13b) MIG 1g.10gb Device 2: (UUID: MIG-0665cd24-2540-5c33-a995-636c4fcab1cf) MIG 1g.10gb Device 3: (UUID: MIG-3a059506-610c-5274-984f-065c39393bdc) MIG 1g.10gb Device 4: (UUID: MIG-91c89459-4898-5209-90b7-b0ede2e5ec35) GPU 1: NVIDIA A100 80GB PCIe (UUID: GPU-14122cde-73d9-d15c-5de8-4ce2da4ff2a2) MIG 3g.40gb Device 0: (UUID: MIG-4770598e-9eea-56f1-a483-dd22f5725c58) MIG 1g.10gb Device 1: (UUID: MIG-94bf4e63-eeda-55e0-a038-ca15deeb6a98) MIG 1g.10gb Device 2: (UUID: MIG-e15bc69a-fdaa-5b4a-bbdd-e2c836785d91) MIG 1g.10gb Device 3: (UUID: MIG-dc0d7d2f-15be-5eb7-bfbb-800b1c8f0368) MIG 1g.10gb Device 4: (UUID: MIG-6c1c7cef-7926-5217-a5bd-b2298a3771a9) ``` Ollama Logs ``` Jun 04 15:13:42 rack-ai-0 systemd[1]: Started Ollama Service. Jun 04 15:13:42 rack-ai-0 ollama[3755267]: 2024/06/04 15:13:42 routes.go:1007: INFO server config env="map[OLLAMA_DEBUG:true OLLAMA_FLASH_ATTENTION:false OLLAMA_HOST: OLLAMA_KEEP_ALIVE: OLLAMA_LLM_LIBRARY: OLLAMA_MAX_LOADED_MODELS:2 OLLAMA_MAX_QUEUE:512> Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.076+02:00 level=INFO source=images.go:729 msg="total blobs: 44" Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.076+02:00 level=INFO source=images.go:736 msg="total unused blobs removed: 0" Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.076+02:00 level=INFO source=routes.go:1053 msg="Listening on [::]:11434 (version 0.1.41)" Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=INFO source=payload.go:30 msg="extracting embedded files" dir=/tmp/ollama1162914854/runners Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cpu file=build/linux/x86_64/cpu/bin/ollama_llama_server.gz Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cpu_avx file=build/linux/x86_64/cpu_avx/bin/ollama_llama_server.gz Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cpu_avx2 file=build/linux/x86_64/cpu_avx2/bin/ollama_llama_server.gz Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cuda_v11 file=build/linux/x86_64/cuda_v11/bin/libcublas.so.11.gz Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cuda_v11 file=build/linux/x86_64/cuda_v11/bin/libcublasLt.so.11.gz Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cuda_v11 file=build/linux/x86_64/cuda_v11/bin/libcudart.so.11.0.gz Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=cuda_v11 file=build/linux/x86_64/cuda_v11/bin/ollama_llama_server.gz Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=rocm_v60002 file=build/linux/x86_64/rocm_v60002/bin/deps.txt.gz Jun 04 15:13:42 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:42.077+02:00 level=DEBUG source=payload.go:180 msg=extracting variant=rocm_v60002 file=build/linux/x86_64/rocm_v60002/bin/ollama_llama_server.gz Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:71 msg="availableServers : found" file=/tmp/ollama1162914854/runners/cpu Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:71 msg="availableServers : found" file=/tmp/ollama1162914854/runners/cpu_avx Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:71 msg="availableServers : found" file=/tmp/ollama1162914854/runners/cpu_avx2 Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:71 msg="availableServers : found" file=/tmp/ollama1162914854/runners/cuda_v11 Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:71 msg="availableServers : found" file=/tmp/ollama1162914854/runners/rocm_v60002 Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=INFO source=payload.go:44 msg="Dynamic LLM libraries [cpu cpu_avx cpu_avx2 cuda_v11 rocm_v60002]" Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=payload.go:45 msg="Override detection logic by setting OLLAMA_LLM_LIBRARY" Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=sched.go:90 msg="starting llm scheduler" Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=gpu.go:132 msg="Detecting GPUs" Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=gpu.go:274 msg="Searching for GPU library" name=libcuda.so* Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.658+02:00 level=DEBUG source=gpu.go:293 msg="gpu library search" globs="[/libcuda.so** /usr/local/cuda*/targets/*/lib/libcuda.so* /usr/lib/*-linux-gnu/nvidia/current/libcuda.so* /usr/li> Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.660+02:00 level=DEBUG source=gpu.go:326 msg="discovered GPU libraries" paths=[/usr/lib/x86_64-linux-gnu/libcuda.so.510.108.03] Jun 04 15:13:44 rack-ai-0 ollama[3755267]: CUDA driver version: 11.6 Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.749+02:00 level=DEBUG source=gpu.go:137 msg="detected GPUs" count=1 library=/usr/lib/x86_64-linux-gnu/libcuda.so.510.108.03 Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.749+02:00 level=DEBUG source=cpu_common.go:11 msg="CPU has AVX2" Jun 04 15:13:44 rack-ai-0 ollama[3755267]: [GPU-fce143c8-f3e2-5db9-a955-1541d6ae4ed6] CUDA totalMem 9728 mb Jun 04 15:13:44 rack-ai-0 ollama[3755267]: [GPU-fce143c8-f3e2-5db9-a955-1541d6ae4ed6] CUDA freeMem 9645 mb Jun 04 15:13:44 rack-ai-0 ollama[3755267]: [GPU-fce143c8-f3e2-5db9-a955-1541d6ae4ed6] Compute Capability 8.0 Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.888+02:00 level=DEBUG source=amd_linux.go:322 msg="amdgpu driver not detected /sys/module/amdgpu" Jun 04 15:13:44 rack-ai-0 ollama[3755267]: releasing nvcuda library Jun 04 15:13:44 rack-ai-0 ollama[3755267]: time=2024-06-04T15:13:44.888+02:00 level=INFO source=types.go:71 msg="inference compute" id=GPU-fce143c8-f3e2-5db9-a955-1541d6ae4ed6 library=cuda compute=8.0 driver=11.6 name="NVIDIA A100 80GB PCIe MIG 1g.10gb" ``` ### OS Linux ### GPU Nvidia ### CPU AMD ### Ollama version 0.1.41
GiteaMirror added the nvidiabug labels 2026-04-12 13:27:52 -05:00
GiteaMirror commented 2026-04-12 13:27:54 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Jun 18, 2024):

Based on the log output it seems like the driver API may not be MIG aware. Maybe we were premature in closing #1500

<!-- gh-comment-id:2177093239 --> @dhiltgen commented on GitHub (Jun 18, 2024): Based on the log output it seems like the driver API may not be MIG aware. Maybe we were premature in closing #1500
GiteaMirror commented 2026-04-12 13:27:54 -05:00
Author
Owner
Copy Link

@mhoehl05 commented on GitHub (Aug 1, 2024):

As already mentioned in #1500

been testing the same setup with 1x h100 and 20gb slices for a proof of concept, running into the same issue. Ollama only utilizes 1 of 3 passed Migs:

image

used llama3.1:70b and noticed a massive performance loss compared to our "no-mig" setup

<!-- gh-comment-id:2263029059 --> @mhoehl05 commented on GitHub (Aug 1, 2024): As already mentioned in #1500 been testing the same setup with 1x h100 and 20gb slices for a proof of concept, running into the same issue. Ollama only utilizes 1 of 3 passed Migs: ![image](https://github.com/user-attachments/assets/89ac70fe-dd29-45f8-9f50-64641fb499d6) used llama3.1:70b and noticed a massive performance loss compared to our "no-mig" setup
GiteaMirror commented 2026-04-12 13:27:55 -05:00
Author
Owner
Copy Link

@waTeim commented on GitHub (Aug 1, 2024):

Hmm, ok what's the deal here? -- we are probably also being stung by this.

<!-- gh-comment-id:2263392095 --> @waTeim commented on GitHub (Aug 1, 2024): Hmm, ok what's the deal here? -- we are probably also being stung by this.
GiteaMirror commented 2026-04-12 13:27:56 -05:00
Author
Owner
Copy Link

@Magitoneu commented on GitHub (Nov 8, 2024):

Is there any update on this topic? @dhiltgen

<!-- gh-comment-id:2464283077 --> @Magitoneu commented on GitHub (Nov 8, 2024): Is there any update on this topic? @dhiltgen
GiteaMirror commented 2026-04-12 13:27:57 -05:00
Author
Owner
Copy Link

@Eridoc commented on GitHub (Feb 24, 2025):

MIG Support would be very beneficial.

<!-- gh-comment-id:2678350963 --> @Eridoc commented on GitHub (Feb 24, 2025): MIG Support would be very beneficial.
GiteaMirror commented 2026-04-12 13:27:57 -05:00
Author
Owner
Copy Link

@Eridoc commented on GitHub (Mar 12, 2025):

Alright I think there is a misconception here.

MIG is indeed supported, that's why we are able to run Ollama in a single MIG instance.
However, MIG instances are inherently isolated from one another and do not share SMs and Memory, which is why they do not "support" multiple MIG instances.
https://forums.developer.nvidia.com/t/a100-mig-inter-instance-communication/190424

If I am not mistaken, best practice would be to have bigger MIG instances if the model does not fit into the smaller MIG instance, or enable MIG for only a few GPUs, and run Ollama on MIG disabled GPUs, so that they can share SMs and Memory.

<!-- gh-comment-id:2718363834 --> @Eridoc commented on GitHub (Mar 12, 2025): Alright I think there is a misconception here. MIG is indeed supported, that's why we are able to run Ollama in a single MIG instance. However, MIG instances are inherently isolated from one another and do not share SMs and Memory, which is why they do not "support" multiple MIG instances. https://forums.developer.nvidia.com/t/a100-mig-inter-instance-communication/190424 If I am not mistaken, best practice would be to have bigger MIG instances if the model does not fit into the smaller MIG instance, or enable MIG for only a few GPUs, and run Ollama on MIG disabled GPUs, so that they can share SMs and Memory.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug nvidia
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#3040
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?