github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #5022] GPU VRAM estimate not accounting for flash attetion #49690

New Issue
Closed
opened 2026-04-28 12:43:07 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-04-28 12:43:07 -05:00
Owner
Copy Link

Originally created by @theasp on GitHub (Jun 13, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/5022

What is the issue?

Hi,

I'm using a q6_K quant of of codestral-22 with a 18k context and flash attention enabled. I'm trying to get a higher context configured, but I always have VRAM left. It appears that the estimate does not account for the use of flash attention as I still have 2882 GB left.

NAME                            ID              SIZE    PROCESSOR       UNTIL
DEFAULT/codestral-22b:latest    cd78ecba62ae    25 GB   100% GPU        Forever

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 555.42.02              Driver Version: 555.42.02      CUDA Version: 12.5     |
|-----------------------------------------+------------------------+----------------------+
| 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 GeForce RTX 3090        Off |   00000000:01:00.0 Off |                  N/A |
|  0%   40C    P8             34W /  420W |   21694MiB /  24576MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A   3659350      C   ...unners/cuda_v11/ollama_llama_server      21684MiB |
+-----------------------------------------------------------------------------------------+

ollama-1  | time=2024-06-13T13:02:05.501Z level=INFO source=memory.go:133 msg="offload to gpu" layers.requested=-1 layers.real=57 memory.available="23.3 GiB" memory.required.full="23.3 GiB" memory.required.partial="23.3 GiB" memory.required.kv="3.9 GiB" memory.weights.total="16.8 GiB" memory.weights.repeating="16.7 GiB" memory.weights.nonrepeating="157.5 MiB" memory.graph.full="1.8 GiB" memory.graph.partial="1.8 GiB"
ollama-1  | time=2024-06-13T13:02:05.501Z level=INFO source=server.go:341 msg="starting llama server" cmd="/tmp/ollama151144087/runners/cuda_v11/ollama_llama_server --model
 /root/.ollama/models/blobs/sha256-83d371fdab7d62c12eb780a034bf9b5ea89403e4d69e46d332d9bdaeff765c31 --ctx-size 18432 --batch-size 512 --embedding --log-disable --n-gpu-layers 57 --flash-attn --parallel 1 --port 35209"
[...]
time=2024-06-13T13:02:05.953Z level=INFO source=server.go:567 msg="waiting for server to become available" status="llm server not responding"
llm_load_tensors: offloading 56 repeating layers to GPU
llm_load_tensors: offloading non-repeating layers to GPU
llm_load_tensors: offloaded 57/57 layers to GPU
llm_load_tensors:        CPU buffer size =   157.50 MiB
llm_load_tensors:      CUDA0 buffer size = 17248.90 MiB
time=2024-06-13T13:02:06.656Z level=INFO source=server.go:567 msg="waiting for server to become available" status="llm server loading model"
llama_new_context_with_model: n_ctx      = 18432
llama_new_context_with_model: n_batch    = 512
llama_new_context_with_model: n_ubatch   = 512
llama_new_context_with_model: flash_attn = 1
llama_new_context_with_model: freq_base  = 1000000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init:      CUDA0 KV buffer size =  4032.00 MiB
llama_new_context_with_model: KV self size  = 4032.00 MiB, K (f16): 2016.00 MiB, V (f16): 2016.00 MiB
llama_new_context_with_model:  CUDA_Host  output buffer size =     0.15 MiB
llama_new_context_with_model:      CUDA0 compute buffer size =   130.00 MiB
llama_new_context_with_model:  CUDA_Host compute buffer size =    48.01 MiB
llama_new_context_with_model: graph nodes  = 1575
llama_new_context_with_model: graph splits = 2

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

0.1.43

Originally created by @theasp on GitHub (Jun 13, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/5022 ### What is the issue? Hi, I'm using a q6_K quant of of codestral-22 with a 18k context and flash attention enabled. I'm trying to get a higher context configured, but I always have VRAM left. It appears that the estimate does not account for the use of flash attention as I still have 2882 GB left. ``` NAME ID SIZE PROCESSOR UNTIL DEFAULT/codestral-22b:latest cd78ecba62ae 25 GB 100% GPU Forever ``` ``` +-----------------------------------------------------------------------------------------+ | NVIDIA-SMI 555.42.02 Driver Version: 555.42.02 CUDA Version: 12.5 | |-----------------------------------------+------------------------+----------------------+ | 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 GeForce RTX 3090 Off | 00000000:01:00.0 Off | N/A | | 0% 40C P8 34W / 420W | 21694MiB / 24576MiB | 0% Default | | | | N/A | +-----------------------------------------+------------------------+----------------------+ +-----------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=========================================================================================| | 0 N/A N/A 3659350 C ...unners/cuda_v11/ollama_llama_server 21684MiB | +-----------------------------------------------------------------------------------------+ ``` ``` ollama-1 | time=2024-06-13T13:02:05.501Z level=INFO source=memory.go:133 msg="offload to gpu" layers.requested=-1 layers.real=57 memory.available="23.3 GiB" memory.required.full="23.3 GiB" memory.required.partial="23.3 GiB" memory.required.kv="3.9 GiB" memory.weights.total="16.8 GiB" memory.weights.repeating="16.7 GiB" memory.weights.nonrepeating="157.5 MiB" memory.graph.full="1.8 GiB" memory.graph.partial="1.8 GiB" ollama-1 | time=2024-06-13T13:02:05.501Z level=INFO source=server.go:341 msg="starting llama server" cmd="/tmp/ollama151144087/runners/cuda_v11/ollama_llama_server --model /root/.ollama/models/blobs/sha256-83d371fdab7d62c12eb780a034bf9b5ea89403e4d69e46d332d9bdaeff765c31 --ctx-size 18432 --batch-size 512 --embedding --log-disable --n-gpu-layers 57 --flash-attn --parallel 1 --port 35209" [...] time=2024-06-13T13:02:05.953Z level=INFO source=server.go:567 msg="waiting for server to become available" status="llm server not responding" llm_load_tensors: offloading 56 repeating layers to GPU llm_load_tensors: offloading non-repeating layers to GPU llm_load_tensors: offloaded 57/57 layers to GPU llm_load_tensors: CPU buffer size = 157.50 MiB llm_load_tensors: CUDA0 buffer size = 17248.90 MiB time=2024-06-13T13:02:06.656Z level=INFO source=server.go:567 msg="waiting for server to become available" status="llm server loading model" llama_new_context_with_model: n_ctx = 18432 llama_new_context_with_model: n_batch = 512 llama_new_context_with_model: n_ubatch = 512 llama_new_context_with_model: flash_attn = 1 llama_new_context_with_model: freq_base = 1000000.0 llama_new_context_with_model: freq_scale = 1 llama_kv_cache_init: CUDA0 KV buffer size = 4032.00 MiB llama_new_context_with_model: KV self size = 4032.00 MiB, K (f16): 2016.00 MiB, V (f16): 2016.00 MiB llama_new_context_with_model: CUDA_Host output buffer size = 0.15 MiB llama_new_context_with_model: CUDA0 compute buffer size = 130.00 MiB llama_new_context_with_model: CUDA_Host compute buffer size = 48.01 MiB llama_new_context_with_model: graph nodes = 1575 llama_new_context_with_model: graph splits = 2 ``` ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version 0.1.43
GiteaMirror added the bug label 2026-04-28 12:43:07 -05:00
GiteaMirror commented 2026-04-28 12:43:13 -05:00
Author
Owner
Copy Link

@hdnh2006 commented on GitHub (Oct 18, 2024):

I am facing a very similar problem but with another model (QWEN).

I have dropped a comment here: https://github.com/ollama/ollama/issues/3078#issuecomment-2421960239.

Any help to set ollama use the 100% of GPU would be fantastic.

<!-- gh-comment-id:2421991455 --> @hdnh2006 commented on GitHub (Oct 18, 2024): I am facing a very similar problem but with another model (QWEN). I have dropped a comment here: https://github.com/ollama/ollama/issues/3078#issuecomment-2421960239. Any help to set ollama use the 100% of GPU would be fantastic.
GiteaMirror commented 2026-04-28 12:43:18 -05:00
Author
Owner
Copy Link

@jessegross commented on GitHub (Sep 24, 2025):

I'm going to go ahead and close this now that the new memory management logic is on by default. If you continue to see problems, please file a new issue.

<!-- gh-comment-id:3330114600 --> @jessegross commented on GitHub (Sep 24, 2025): I'm going to go ahead and close this now that the new memory management logic is on by default. If you continue to see problems, please file a new issue.
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#49690
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?