github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #1105] Out of memory when using multiple GPUs #47064

New Issue
Closed
opened 2026-04-28 02:57:17 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-04-28 02:57:17 -05:00
Owner
Copy Link

Originally created by @BruceMacD on GitHub (Nov 12, 2023).
Original GitHub issue: https://github.com/ollama/ollama/issues/1105

Originally assigned to: @BruceMacD on GitHub.

When a system has multiple GPUs generation (ex: ollama run ...) may fail with an out of memory error.

Nov 05 22:41:50 example.com ollama[943528]: 2023/11/05 22:41:50 llama.go:259: 7197 MB VRAM available, loading up to 47 GPU layers
Nov 05 22:41:50 example.com ollama[943528]: 2023/11/05 22:41:50 llama.go:370: starting llama runner
Nov 05 22:41:50 example.com ollama[943528]: 2023/11/05 22:41:50 llama.go:428: waiting for llama runner to start responding
Nov 05 22:41:50 example.com ollama[943528]: ggml_init_cublas: found 2 CUDA devices:
Nov 05 22:41:50 example.com ollama[943528]:   Device 0: NVIDIA GeForce RTX 3060 Ti, compute capability 8.6
Nov 05 22:41:50 example.com ollama[943528]:   Device 1: NVIDIA GeForce RTX 3060, compute capability 8.6
Nov 05 22:41:52 example.com ollama[1418565]: {"timestamp":1699245712,"level":"INFO","function":"main","line":1323,"message":"build info","build":219,"commit":"9e70cc0"}
Nov 05 22:41:52 example.com ollama[1418565]: {"timestamp":1699245712,"level":"INFO","function":"main","line":1325,"message":"system info","n_threads":8,"n_threads_batch":-1,"total_threads":16,"system_info":"AVX = 1 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 0 | NEON = 0 | ARM_FMA = 0 | F16C = 0 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | "}
Nov 05 22:41:52 example.com ollama[943528]: llama_model_loader: loaded meta data with 19 key-value pairs and 291 tensors from /usr/share/ollama/.ollama/models/blobs/sha256:22f7f8ef5f4c791c1b03d7eb414399294764d7cc82c7e94aa81a1feb80a983a2 (version GGUF V2 (latest))
Nov 05 22:41:52 example.com ollama[943528]: llama_model_loader: - tensor    0:                token_embd.weight q4_0     [  4096, 32000,     1,     1 ]
...
Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: ggml ctx size =    0.10 MB
Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: using CUDA for GPU acceleration
Nov 05 22:41:52 example.com ollama[943528]: ggml_cuda_set_main_device: using device 0 (NVIDIA GeForce RTX 3060 Ti) as main device
Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: mem required  =   70.41 MB
Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: offloading 32 repeating layers to GPU
Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: offloading non-repeating layers to GPU
Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: offloaded 35/35 layers to GPU
Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: VRAM used: 3577.55 MB
Nov 05 22:41:53 example.com ollama[943528]: ....................................................................
Nov 05 22:41:53 example.com ollama[943528]: CUDA error 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:7233: out of memory
Nov 05 22:41:53 example.com ollama[943528]: current device: 0
Nov 05 22:41:53 example.com ollama[943528]: 2023/11/05 22:41:53 llama.go:385: 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:7233: out of memory
Nov 05 22:41:53 example.com ollama[943528]: current device: 0
Nov 05 22:41:53 example.com ollama[943528]: 2023/11/05 22:41:53 llama.go:393: error starting llama runner: llama runner process has terminated
Nov 05 22:41:53 example.com ollama[943528]: 2023/11/05 22:41:53 llama.go:459: llama runner stopped successfully

Possibly related:
https://github.com/ggerganov/llama.cpp/issues/1866
https://github.com/ggerganov/llama.cpp/issues/2432

Originally created by @BruceMacD on GitHub (Nov 12, 2023). Original GitHub issue: https://github.com/ollama/ollama/issues/1105 Originally assigned to: @BruceMacD on GitHub. When a system has multiple GPUs generation (ex: `ollama run ...`) may fail with an `out of memory` error. ``` Nov 05 22:41:50 example.com ollama[943528]: 2023/11/05 22:41:50 llama.go:259: 7197 MB VRAM available, loading up to 47 GPU layers Nov 05 22:41:50 example.com ollama[943528]: 2023/11/05 22:41:50 llama.go:370: starting llama runner Nov 05 22:41:50 example.com ollama[943528]: 2023/11/05 22:41:50 llama.go:428: waiting for llama runner to start responding Nov 05 22:41:50 example.com ollama[943528]: ggml_init_cublas: found 2 CUDA devices: Nov 05 22:41:50 example.com ollama[943528]: Device 0: NVIDIA GeForce RTX 3060 Ti, compute capability 8.6 Nov 05 22:41:50 example.com ollama[943528]: Device 1: NVIDIA GeForce RTX 3060, compute capability 8.6 Nov 05 22:41:52 example.com ollama[1418565]: {"timestamp":1699245712,"level":"INFO","function":"main","line":1323,"message":"build info","build":219,"commit":"9e70cc0"} Nov 05 22:41:52 example.com ollama[1418565]: {"timestamp":1699245712,"level":"INFO","function":"main","line":1325,"message":"system info","n_threads":8,"n_threads_batch":-1,"total_threads":16,"system_info":"AVX = 1 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 0 | NEON = 0 | ARM_FMA = 0 | F16C = 0 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | "} Nov 05 22:41:52 example.com ollama[943528]: llama_model_loader: loaded meta data with 19 key-value pairs and 291 tensors from /usr/share/ollama/.ollama/models/blobs/sha256:22f7f8ef5f4c791c1b03d7eb414399294764d7cc82c7e94aa81a1feb80a983a2 (version GGUF V2 (latest)) Nov 05 22:41:52 example.com ollama[943528]: llama_model_loader: - tensor 0: token_embd.weight q4_0 [ 4096, 32000, 1, 1 ] ... Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: ggml ctx size = 0.10 MB Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: using CUDA for GPU acceleration Nov 05 22:41:52 example.com ollama[943528]: ggml_cuda_set_main_device: using device 0 (NVIDIA GeForce RTX 3060 Ti) as main device Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: mem required = 70.41 MB Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: offloading 32 repeating layers to GPU Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: offloading non-repeating layers to GPU Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: offloaded 35/35 layers to GPU Nov 05 22:41:52 example.com ollama[943528]: llm_load_tensors: VRAM used: 3577.55 MB Nov 05 22:41:53 example.com ollama[943528]: .................................................................... Nov 05 22:41:53 example.com ollama[943528]: CUDA error 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:7233: out of memory Nov 05 22:41:53 example.com ollama[943528]: current device: 0 Nov 05 22:41:53 example.com ollama[943528]: 2023/11/05 22:41:53 llama.go:385: 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/gguf/ggml-cuda.cu:7233: out of memory Nov 05 22:41:53 example.com ollama[943528]: current device: 0 Nov 05 22:41:53 example.com ollama[943528]: 2023/11/05 22:41:53 llama.go:393: error starting llama runner: llama runner process has terminated Nov 05 22:41:53 example.com ollama[943528]: 2023/11/05 22:41:53 llama.go:459: llama runner stopped successfully ``` Possibly related: https://github.com/ggerganov/llama.cpp/issues/1866 https://github.com/ggerganov/llama.cpp/issues/2432
GiteaMirror added the bug label 2026-04-28 02:57:17 -05:00
GiteaMirror commented 2026-04-28 02:57:31 -05:00
Author
Owner
Copy Link

@phalexo commented on GitHub (Dec 4, 2023):

I get a similar error using multiple or a single GPU when the model is really too small for an OOM. The same models appear to work on the host. So if I set CUDA_VISIBLE_DEVICE='' it runs ok on the host.

<!-- gh-comment-id:1837663435 --> @phalexo commented on GitHub (Dec 4, 2023): I get a similar error using multiple or a single GPU when the model is really too small for an OOM. The same models appear to work on the host. So if I set CUDA_VISIBLE_DEVICE='' it runs ok on the host.
GiteaMirror commented 2026-04-28 02:57:32 -05:00
Author
Owner
Copy Link

@farhanhubble commented on GitHub (Dec 7, 2023):

I get a similar error using multiple or a single GPU when the model is really too small for an OOM. The same models appear to work on the host. So if I set CUDA_VISIBLE_DEVICE='' it runs ok on the host.

Nitpick: The environment variable should be CUDA_VISIBLE_DEVICES=''

<!-- gh-comment-id:1844684613 --> @farhanhubble commented on GitHub (Dec 7, 2023): > I get a similar error using multiple or a single GPU when the model is really too small for an OOM. The same models appear to work on the host. So if I set CUDA_VISIBLE_DEVICE='' it runs ok on the host. Nitpick: The environment variable should be ` CUDA_VISIBLE_DEVICES=''`
GiteaMirror commented 2026-04-28 02:57:34 -05:00
Author
Owner
Copy Link

@phalexo commented on GitHub (Dec 16, 2023):

git clone --recursive https://github.com/jmorganca/ollama.git
cd ollama/llm/llama.cpp
vi generate_linux.go

//go:generate cmake -S ggml -B ggml/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_CUDA_FORCE_MMQ=on
//go:generate cmake --build ggml/build/cuda --target server --config Release
//go:generate mv ggml/build/cuda/bin/server ggml/build/cuda/bin/ollama-runner
//go:generate cmake -S gguf -B gguf/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off -DLLAMA_CUDA_PEER_MAX_BATCH_SIZE=0 -DLLAMA_CUDA_FORCE_MMQ=on
//go:generate cmake --build gguf/build/cuda --target server --config Release
//go:generate mv gguf/build/cuda/bin/server gguf/build/cuda/bin/ollama-runner

cd ../..
go generate ./...
go build .
<!-- gh-comment-id:1858918180 --> @phalexo commented on GitHub (Dec 16, 2023): ```bash git clone --recursive https://github.com/jmorganca/ollama.git cd ollama/llm/llama.cpp vi generate_linux.go ``` ```go //go:generate cmake -S ggml -B ggml/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_CUDA_FORCE_MMQ=on //go:generate cmake --build ggml/build/cuda --target server --config Release //go:generate mv ggml/build/cuda/bin/server ggml/build/cuda/bin/ollama-runner //go:generate cmake -S gguf -B gguf/build/cuda -DLLAMA_CUBLAS=on -DLLAMA_ACCELERATE=on -DLLAMA_K_QUANTS=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off -DLLAMA_CUDA_PEER_MAX_BATCH_SIZE=0 -DLLAMA_CUDA_FORCE_MMQ=on //go:generate cmake --build gguf/build/cuda --target server --config Release //go:generate mv gguf/build/cuda/bin/server gguf/build/cuda/bin/ollama-runner ``` ```bash cd ../.. go generate ./... go build . ```
GiteaMirror commented 2026-04-28 02:57:39 -05:00
Author
Owner
Copy Link

@ex3ndr commented on GitHub (Dec 22, 2023):

I just finished 2x4090 build and getting same errors

<!-- gh-comment-id:1867326748 --> @ex3ndr commented on GitHub (Dec 22, 2023): I just finished 2x4090 build and getting same errors
GiteaMirror commented 2026-04-28 02:57:41 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Jan 10, 2024):

This should be fixed with https://github.com/jmorganca/ollama/pull/1850, but feel free to re-open an issue if not

<!-- gh-comment-id:1884881019 --> @jmorganca commented on GitHub (Jan 10, 2024): This should be fixed with https://github.com/jmorganca/ollama/pull/1850, but feel free to re-open an issue if not
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#47064
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?