github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-08 08:59:13 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #1204] GPU VRAM size calculations seem incorrect #26376

New Issue
Closed
opened 2026-04-22 02:37:47 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-22 02:37:47 -05:00
Owner
Copy Link

Originally created by @FairyTail2000 on GitHub (Nov 20, 2023).
Original GitHub issue: https://github.com/ollama/ollama/issues/1204

[GIN] 2023/11/20 - 09:01:35 | 204 | 14.066µs | 127.0.0.1 | OPTIONS "/api/generate"
2023/11/20 09:01:36 llama.go:290: 3849 MB VRAM available, loading up to 25 GPU layers
2023/11/20 09:01:36 llama.go:415: starting llama runner
2023/11/20 09:01:36 llama.go:473: waiting for llama runner to start responding
ggml_init_cublas: found 1 CUDA devices:
Device 0: NVIDIA GeForce GTX 1650, compute capability 7.5
{"timestamp":1700467297,"level":"INFO","function":"main","line":1190,"message":"build info","build":1009,"commit":"9e232f0"}
{"timestamp":1700467297,"level":"INFO","function":"main","line":1192,"message":"system info","n_threads":8,"total_threads":16,"system_info":"AVX = 1 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | VSX = 0 | "}
llama.cpp: loading model from /home/rafaels/.ollama/models/blobs/sha256:8daa9615cce30c259a9555b1cc250d461d1bc69980a274b44d7eda0be78076d8
llama_model_load_internal: format = ggjt v3 (latest)
llama_model_load_internal: n_vocab = 32000
llama_model_load_internal: n_ctx = 2048
llama_model_load_internal: n_embd = 4096
llama_model_load_internal: n_mult = 256
llama_model_load_internal: n_head = 32
llama_model_load_internal: n_head_kv = 32
llama_model_load_internal: n_layer = 32
llama_model_load_internal: n_rot = 128
llama_model_load_internal: n_gqa = 1
llama_model_load_internal: rnorm_eps = 5.0e-06
llama_model_load_internal: n_ff = 11008
llama_model_load_internal: freq_base = 10000.0
llama_model_load_internal: freq_scale = 1
llama_model_load_internal: ftype = 2 (mostly Q4_0)
llama_model_load_internal: model size = 7B
llama_model_load_internal: ggml ctx size = 0.08 MB
llama_model_load_internal: using CUDA for GPU acceleration
llama_model_load_internal: mem required = 1298.89 MB (+ 1024.00 MB per state)
llama_model_load_internal: allocating batch_size x (512 kB + n_ctx x 128 B) = 384 MB VRAM for the scratch buffer
llama_model_load_internal: offloading 25 repeating layers to GPU
llama_model_load_internal: offloaded 25/35 layers to GPU
llama_model_load_internal: total VRAM used: 3099 MB
llama_new_context_with_model: kv self size = 1024.00 MB
llama server listening at http://127.0.0.1:59278
{"timestamp":1700467302,"level":"INFO","function":"main","line":1443,"message":"HTTP server listening","hostname":"127.0.0.1","port":59278}
{"timestamp":1700467302,"level":"INFO","function":"log_server_request","line":1157,"message":"request","remote_addr":"127.0.0.1","remote_port":58042,"status":200,"method":"HEAD","path":"/","params":{}}
2023/11/20 09:01:42 llama.go:487: llama runner started in 6.203332 seconds
CUDA error 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/ggml/ggml-cuda.cu:4856: out of memory
2023/11/20 09:01:46 llama.go:430: 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/ggml/ggml-cuda.cu:4856: out of memory
2023/11/20 09:01:46 llama.go:504: llama runner stopped successfully
[GIN] 2023/11/20 - 09:01:46 | 200 | 11.036819193s | 127.0.0.1 | POST "/api/generate"

As visible it seems like ollama correctly calculated that not all layers fit into the gpu. However other factors seem to also play a role, espacially mem required

I'm running a GTX 1650 Laptop 4GB VRAM and 32 GB Main Memory. The current load before/after ollama on the VRAM are 54 MiB allocated by the system. Ollama Version: 0.1.10

Originally created by @FairyTail2000 on GitHub (Nov 20, 2023). Original GitHub issue: https://github.com/ollama/ollama/issues/1204 > [GIN] 2023/11/20 - 09:01:35 | 204 | 14.066µs | 127.0.0.1 | OPTIONS "/api/generate" 2023/11/20 09:01:36 llama.go:290: 3849 MB VRAM available, loading up to 25 GPU layers 2023/11/20 09:01:36 llama.go:415: starting llama runner 2023/11/20 09:01:36 llama.go:473: waiting for llama runner to start responding ggml_init_cublas: found 1 CUDA devices: Device 0: NVIDIA GeForce GTX 1650, compute capability 7.5 {"timestamp":1700467297,"level":"INFO","function":"main","line":1190,"message":"build info","build":1009,"commit":"9e232f0"} {"timestamp":1700467297,"level":"INFO","function":"main","line":1192,"message":"system info","n_threads":8,"total_threads":16,"system_info":"AVX = 1 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | VSX = 0 | "} llama.cpp: loading model from /home/rafaels/.ollama/models/blobs/sha256:8daa9615cce30c259a9555b1cc250d461d1bc69980a274b44d7eda0be78076d8 llama_model_load_internal: format = ggjt v3 (latest) llama_model_load_internal: n_vocab = 32000 llama_model_load_internal: n_ctx = 2048 llama_model_load_internal: n_embd = 4096 llama_model_load_internal: n_mult = 256 llama_model_load_internal: n_head = 32 llama_model_load_internal: n_head_kv = 32 llama_model_load_internal: n_layer = 32 llama_model_load_internal: n_rot = 128 llama_model_load_internal: n_gqa = 1 llama_model_load_internal: rnorm_eps = 5.0e-06 llama_model_load_internal: n_ff = 11008 llama_model_load_internal: freq_base = 10000.0 llama_model_load_internal: freq_scale = 1 llama_model_load_internal: ftype = 2 (mostly Q4_0) llama_model_load_internal: model size = 7B llama_model_load_internal: ggml ctx size = 0.08 MB llama_model_load_internal: using CUDA for GPU acceleration llama_model_load_internal: mem required = 1298.89 MB (+ 1024.00 MB per state) llama_model_load_internal: allocating batch_size x (512 kB + n_ctx x 128 B) = 384 MB VRAM for the scratch buffer llama_model_load_internal: offloading 25 repeating layers to GPU llama_model_load_internal: offloaded 25/35 layers to GPU llama_model_load_internal: total VRAM used: 3099 MB llama_new_context_with_model: kv self size = 1024.00 MB llama server listening at http://127.0.0.1:59278 {"timestamp":1700467302,"level":"INFO","function":"main","line":1443,"message":"HTTP server listening","hostname":"127.0.0.1","port":59278} {"timestamp":1700467302,"level":"INFO","function":"log_server_request","line":1157,"message":"request","remote_addr":"127.0.0.1","remote_port":58042,"status":200,"method":"HEAD","path":"/","params":{}} 2023/11/20 09:01:42 llama.go:487: llama runner started in 6.203332 seconds CUDA error 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/ggml/ggml-cuda.cu:4856: out of memory 2023/11/20 09:01:46 llama.go:430: 2 at /go/src/github.com/jmorganca/ollama/llm/llama.cpp/ggml/ggml-cuda.cu:4856: out of memory 2023/11/20 09:01:46 llama.go:504: llama runner stopped successfully [GIN] 2023/11/20 - 09:01:46 | 200 | 11.036819193s | 127.0.0.1 | POST "/api/generate" As visible it seems like ollama correctly calculated that not all layers fit into the gpu. However other factors seem to also play a role, espacially mem required I'm running a GTX 1650 Laptop 4GB VRAM and 32 GB Main Memory. The current load before/after ollama on the VRAM are 54 MiB allocated by the system. Ollama Version: 0.1.10
GiteaMirror commented 2026-04-22 02:37:49 -05:00
Author
Owner
Copy Link

@FairyTail2000 commented on GitHub (Nov 20, 2023):

Probably should have added, I'm using llama2:7b

<!-- gh-comment-id:1818419334 --> @FairyTail2000 commented on GitHub (Nov 20, 2023): Probably should have added, I'm using llama2:7b
GiteaMirror commented 2026-04-22 02:37:49 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Jan 25, 2024):

@jmorganca has been doing a bunch of changes with this, so it should be better. @FairyTail2000 are you still seeing issues?

<!-- gh-comment-id:1911066512 --> @pdevine commented on GitHub (Jan 25, 2024): @jmorganca has been doing a bunch of changes with this, so it should be better. @FairyTail2000 are you still seeing issues?
GiteaMirror commented 2026-04-22 02:37:50 -05:00
Author
Owner
Copy Link

@FairyTail2000 commented on GitHub (Jan 26, 2024):

Looks good. Although I'm now using the chat api which might be different in handling the calculation but since I don't known, solved

<!-- gh-comment-id:1911622147 --> @FairyTail2000 commented on GitHub (Jan 26, 2024): Looks good. Although I'm now using the chat api which might be different in handling the calculation but since I don't known, solved
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
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#26376
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?