github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-07-17 13:47:08 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[GH-ISSUE #14960] [Bug]: RTX 5070 Ti not detected (Total VRAM: 0 B) - Compute Capability 12.0 / Blackwell Architecture #56133

New Issue
Closed
opened 2026-04-29 10:18:19 -05:00 by GiteaMirror · 11 comments
GiteaMirror commented 2026-04-29 10:18:19 -05:00
Owner
Copy Link

Originally created by @Noct2vis on GitHub (Mar 19, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14960

Description

Ollama version 0.18.1 fails to detect the newly released NVIDIA GeForce RTX 5070 Ti (12GB) on Windows. Even with CUDA Toolkit 13.2 installed and proper environment variables set, the server defaults to CPU and reports total_vram="0 B".

Environment

  • Ollama Version: 0.18.1 (Windows Native)
  • GPU: NVIDIA GeForce RTX 5070 Ti (12GB)
  • Architecture: Blackwell (Compute Capability 12.0)
  • OS: Windows 11
  • Driver Version: 58x.xx (Latest available for 50-series)
  • CUDA Toolkit: v13.2

Steps Taken to Troubleshoot

  1. Verified nvidia-smi correctly displays the GPU and CUDA version.
  2. Set OLLAMA_DEBUG=1 to capture detailed logs.
  3. Attempted to manually link CUDA 13.2 DLLs (renamed to *_12.dll) into the Ollama program directory to resolve library version mismatches.
  4. Set CUDA_VISIBLE_DEVICES=0 and OLLAMA_VULKAN=1, but the discovery still resulted in 0 B VRAM.
  5. Killed all background processes (ollama.exe, ollama_llama_server.exe) to ensure a clean start with ollama serve.

Logs

time=2026-03-19T16:36:09.609Z level=INFO source=runner.go:67 msg="discovering available GPUs..."
time=2026-03-19T16:36:09.634Z level=DEBUG source=server.go:431 msg=subprocess CUDA_PATH="C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v13.2" OLLAMA_CUDA_VERSION=12 OLLAMA_DEBUG=1 OLLAMA_GPU_LAYERS=cuda OLLAMA_LLM_LIBRARY=cuda
time=2026-03-19T16:36:09.718Z level=DEBUG source=runner.go:437 msg="bootstrap discovery took" duration=102.1248ms
time=2026-03-19T16:36:09.718Z level=DEBUG source=runner.go:124 msg="evaluating which, if any, devices to filter out" initial_count=0
time=2026-03-19T16:36:09.718Z level=DEBUG source=runner.go:40 msg="GPU bootstrap discovery took" duration=110.4689ms
time=2026-03-19T16:36:09.718Z level=INFO source=types.go:60 msg="inference compute" id=cpu library=cpu compute="" name=cpu
time=2026-03-19T16:36:09.718Z level=INFO source=routes.go:1832 msg="vram-based default context" total_vram="0 B" default_num_ctx=4096

### Relevant log output

```shell

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.18.1

Originally created by @Noct2vis on GitHub (Mar 19, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14960 ### Description Ollama version 0.18.1 fails to detect the newly released **NVIDIA GeForce RTX 5070 Ti (12GB)** on Windows. Even with CUDA Toolkit 13.2 installed and proper environment variables set, the server defaults to CPU and reports `total_vram="0 B"`. ### Environment - **Ollama Version:** 0.18.1 (Windows Native) - **GPU:** NVIDIA GeForce RTX 5070 Ti (12GB) - **Architecture:** Blackwell (Compute Capability 12.0) - **OS:** Windows 11 - **Driver Version:** 58x.xx (Latest available for 50-series) - **CUDA Toolkit:** v13.2 ### Steps Taken to Troubleshoot 1. Verified `nvidia-smi` correctly displays the GPU and CUDA version. 2. Set `OLLAMA_DEBUG=1` to capture detailed logs. 3. Attempted to manually link CUDA 13.2 DLLs (renamed to `*_12.dll`) into the Ollama program directory to resolve library version mismatches. 4. Set `CUDA_VISIBLE_DEVICES=0` and `OLLAMA_VULKAN=1`, but the discovery still resulted in 0 B VRAM. 5. Killed all background processes (`ollama.exe`, `ollama_llama_server.exe`) to ensure a clean start with `ollama serve`. ### Logs ```text time=2026-03-19T16:36:09.609Z level=INFO source=runner.go:67 msg="discovering available GPUs..." time=2026-03-19T16:36:09.634Z level=DEBUG source=server.go:431 msg=subprocess CUDA_PATH="C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v13.2" OLLAMA_CUDA_VERSION=12 OLLAMA_DEBUG=1 OLLAMA_GPU_LAYERS=cuda OLLAMA_LLM_LIBRARY=cuda time=2026-03-19T16:36:09.718Z level=DEBUG source=runner.go:437 msg="bootstrap discovery took" duration=102.1248ms time=2026-03-19T16:36:09.718Z level=DEBUG source=runner.go:124 msg="evaluating which, if any, devices to filter out" initial_count=0 time=2026-03-19T16:36:09.718Z level=DEBUG source=runner.go:40 msg="GPU bootstrap discovery took" duration=110.4689ms time=2026-03-19T16:36:09.718Z level=INFO source=types.go:60 msg="inference compute" id=cpu library=cpu compute="" name=cpu time=2026-03-19T16:36:09.718Z level=INFO source=routes.go:1832 msg="vram-based default context" total_vram="0 B" default_num_ctx=4096 ### Relevant log output ```shell ``` ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version 0.18.1
GiteaMirror added the bug label 2026-04-29 10:18:19 -05:00
GiteaMirror commented 2026-04-29 10:18:22 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 19, 2026):

Set OLLAMA_DEBUG=2 and post the complete log (ie no truncated lines) up to and including the line with inference compute.

OLLAMA_CUDA_VERSION and OLLAMA_GPU_LAYERS are not ollama configuration variables.

<!-- gh-comment-id:4091922073 --> @rick-github commented on GitHub (Mar 19, 2026): Set `OLLAMA_DEBUG=2` and post the complete log (ie no truncated lines) up to and including the line with `inference compute`. `OLLAMA_CUDA_VERSION` and `OLLAMA_GPU_LAYERS` are not ollama configuration variables.
GiteaMirror commented 2026-04-29 10:18:22 -05:00
Author
Owner
Copy Link

@viba1 commented on GitHub (Mar 19, 2026):

Same problem here (RTX 3090).
Do you perform regression/performance tests before deploying new versions, considering different environmental variables?

<!-- gh-comment-id:4093274770 --> @viba1 commented on GitHub (Mar 19, 2026): Same problem here (RTX 3090). Do you perform regression/performance tests before deploying new versions, considering different environmental variables?
GiteaMirror commented 2026-04-29 10:18:22 -05:00
Author
Owner
Copy Link

@viba1 commented on GitHub (Mar 19, 2026):

Removing OLLAMA_CUDA_VERSION and OLLAMA_GPU_LAYERS does'nt change anything.
See logs (DEBUG=2) in associated file

ollama.log

<!-- gh-comment-id:4093352131 --> @viba1 commented on GitHub (Mar 19, 2026): Removing OLLAMA_CUDA_VERSION and OLLAMA_GPU_LAYERS does'nt change anything. See logs (DEBUG=2) in associated file [ollama.log](https://github.com/user-attachments/files/26126854/ollama.log)
GiteaMirror commented 2026-04-29 10:18:23 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 19, 2026):

time=2026-03-19T22:12:23.525+01:00 level=DEBUG source=ggml.go:94 msg="ggml backend load all from path" path=/usr/local/lib/ollama
time=2026-03-19T22:12:23.525+01:00 level=INFO source=ggml.go:104 msg=system CPU.0.LLAMAFILE=1 compiler=cgo(gcc)

No accelerator backends found. How was ollama installed, what variant of Linux is used, what's the output of ls -l /usr/local/lib/ollama.

<!-- gh-comment-id:4093439547 --> @rick-github commented on GitHub (Mar 19, 2026): ``` time=2026-03-19T22:12:23.525+01:00 level=DEBUG source=ggml.go:94 msg="ggml backend load all from path" path=/usr/local/lib/ollama time=2026-03-19T22:12:23.525+01:00 level=INFO source=ggml.go:104 msg=system CPU.0.LLAMAFILE=1 compiler=cgo(gcc) ``` No accelerator backends found. How was ollama installed, what variant of Linux is used, what's the output of `ls -l /usr/local/lib/ollama`.
GiteaMirror commented 2026-04-29 10:18:23 -05:00
Author
Owner
Copy Link

@viba1 commented on GitHub (Mar 19, 2026):

It was installed using the script : curl -fsSL https://ollama.com/install.sh | sh
output of "ls -l /usr/local/lib/ollama" is: "drwx------ 2 root root 4096 19 mars 22:22 cuda_v12"

<!-- gh-comment-id:4093476846 --> @viba1 commented on GitHub (Mar 19, 2026): It was installed using the script : curl -fsSL https://ollama.com/install.sh | sh output of "ls -l /usr/local/lib/ollama" is: "drwx------ 2 root root 4096 19 mars 22:22 cuda_v12"
GiteaMirror commented 2026-04-29 10:18:24 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 19, 2026):

output of "ls -l /usr/local/lib/ollama" is: "drwx------ 2 root root 4096 19 mars 22:22 cuda_v12"

Your installation is incomplete. It should look like:

total 6484
drwxr-xr-x 2 root root    4096 Mär 18 21:53 cuda_v12
drwxr-xr-x 2 root root    4096 Mär 18 21:41 cuda_v13
lrwxrwxrwx 1 root root      17 Mär 18 21:30 libggml-base.so -> libggml-base.so.0
lrwxrwxrwx 1 root root      21 Mär 18 21:30 libggml-base.so.0 -> libggml-base.so.0.0.0
-rwxr-xr-x 1 root root  748152 Mär 18 21:30 libggml-base.so.0.0.0
-rwxr-xr-x 1 root root  873912 Mär 18 21:30 libggml-cpu-alderlake.so
-rwxr-xr-x 1 root root  873912 Mär 18 21:30 libggml-cpu-haswell.so
-rwxr-xr-x 1 root root 1009080 Mär 18 21:30 libggml-cpu-icelake.so
-rwxr-xr-x 1 root root  820728 Mär 18 21:30 libggml-cpu-sandybridge.so
-rwxr-xr-x 1 root root 1009080 Mär 18 21:30 libggml-cpu-skylakex.so
-rwxr-xr-x 1 root root  636536 Mär 18 21:30 libggml-cpu-sse42.so
-rwxr-xr-x 1 root root  632472 Mär 18 21:30 libggml-cpu-x64.so
drwxr-xr-x 3 root root    4096 Mär 18 21:54 mlx_cuda_v13
drwxr-xr-x 2 root root    4096 Mär 18 21:36 vulkan
<!-- gh-comment-id:4093490066 --> @rick-github commented on GitHub (Mar 19, 2026): > output of "ls -l /usr/local/lib/ollama" is: "drwx------ 2 root root 4096 19 mars 22:22 cuda_v12" Your installation is incomplete. It should look like: ``` total 6484 drwxr-xr-x 2 root root 4096 Mär 18 21:53 cuda_v12 drwxr-xr-x 2 root root 4096 Mär 18 21:41 cuda_v13 lrwxrwxrwx 1 root root 17 Mär 18 21:30 libggml-base.so -> libggml-base.so.0 lrwxrwxrwx 1 root root 21 Mär 18 21:30 libggml-base.so.0 -> libggml-base.so.0.0.0 -rwxr-xr-x 1 root root 748152 Mär 18 21:30 libggml-base.so.0.0.0 -rwxr-xr-x 1 root root 873912 Mär 18 21:30 libggml-cpu-alderlake.so -rwxr-xr-x 1 root root 873912 Mär 18 21:30 libggml-cpu-haswell.so -rwxr-xr-x 1 root root 1009080 Mär 18 21:30 libggml-cpu-icelake.so -rwxr-xr-x 1 root root 820728 Mär 18 21:30 libggml-cpu-sandybridge.so -rwxr-xr-x 1 root root 1009080 Mär 18 21:30 libggml-cpu-skylakex.so -rwxr-xr-x 1 root root 636536 Mär 18 21:30 libggml-cpu-sse42.so -rwxr-xr-x 1 root root 632472 Mär 18 21:30 libggml-cpu-x64.so drwxr-xr-x 3 root root 4096 Mär 18 21:54 mlx_cuda_v13 drwxr-xr-x 2 root root 4096 Mär 18 21:36 vulkan ```
GiteaMirror commented 2026-04-29 10:18:24 -05:00
Author
Owner
Copy Link

@viba1 commented on GitHub (Mar 19, 2026):

Thanks,
On my side, it's working after a complete reinstall.
I still don't understand the initial cause. I just restarted the machine. I hope it's not a disk issue.

<!-- gh-comment-id:4093995784 --> @viba1 commented on GitHub (Mar 19, 2026): Thanks, On my side, it's working after a complete reinstall. I still don't understand the initial cause. I just restarted the machine. I hope it's not a disk issue.
GiteaMirror commented 2026-04-29 10:18:25 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 19, 2026):

Could have been a network issue, the installer streams the package from the server and unpacks it as bytes arrive. If the network connection is interrupted, the installation will not be complete. Would have been visible during running the install script, though.

<!-- gh-comment-id:4094019695 --> @rick-github commented on GitHub (Mar 19, 2026): Could have been a network issue, the installer streams the package from the server and unpacks it as bytes arrive. If the network connection is interrupted, the installation will not be complete. Would have been visible during running the install script, though.
GiteaMirror commented 2026-04-29 10:18:25 -05:00
Author
Owner
Copy Link

@Noct2vis commented on GitHub (Mar 19, 2026):

I dont really think that works. Cause my cmd window told me my cuda version is 13.1 after nvidia-smi. However, all the checks from ollama are cuda 12.4 or lower. So it always told me there is no GPU there.

<!-- gh-comment-id:4094051815 --> @Noct2vis commented on GitHub (Mar 19, 2026): I dont really think that works. Cause my cmd window told me my cuda version is 13.1 after nvidia-smi. However, all the checks from ollama are cuda 12.4 or lower. So it always told me there is no GPU there.
GiteaMirror commented 2026-04-29 10:18:26 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 19, 2026):

ollama supports cuda12 and cuda13.

<!-- gh-comment-id:4094063245 --> @rick-github commented on GitHub (Mar 19, 2026): ollama supports cuda12 and cuda13.
GiteaMirror commented 2026-04-29 10:18:26 -05:00
Author
Owner
Copy Link

@Noct2vis commented on GitHub (Mar 19, 2026):

It works after I reinstall it.

<!-- gh-comment-id:4094182905 --> @Noct2vis commented on GitHub (Mar 19, 2026): It works after I reinstall it.
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
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#56133
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?