Low GPU utilization on windows

[FredericoPerimLopes]

Hello,
I've build llama.cpp with CUDA and it built fine. The issue is when I run inference I see GPU utilization close to 0 but I can see memory increasing, so what could be the issue?

Log start
main: build = 1999 (d2f650c)
main: built with MSVC 19.33.31629.0 for x64
main: seed = 1706478765
ggml_init_cublas: GGML_CUDA_FORCE_MMQ: no
ggml_init_cublas: CUDA_USE_TENSOR_CORES: yes
ggml_init_cublas: found 1 CUDA devices:
Device 0: NVIDIA GeForce MX230, compute capability 6.1, VMM: yes
llama_model_loader: loaded meta data with 22 key-value pairs and 291 tensors from E:\llama.cpp\models\deepseek-coder-6.7b-instruct.Q5_K_M.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv 0: general.architecture str = llama
llama_model_loader: - kv 1: general.name str = deepseek-ai_deepseek-coder-6.7b-instruct
llama_model_loader: - kv 2: llama.context_length u32 = 16384
llama_model_loader: - kv 3: llama.embedding_length u32 = 4096
llama_model_loader: - kv 4: llama.block_count u32 = 32
llama_model_loader: - kv 5: llama.feed_forward_length u32 = 11008
llama_model_loader: - kv 6: llama.rope.dimension_count u32 = 128
llama_model_loader: - kv 7: llama.attention.head_count u32 = 32
llama_model_loader: - kv 8: llama.attention.head_count_kv u32 = 32
llama_model_loader: - kv 9: llama.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 10: llama.rope.freq_base f32 = 100000.000000
llama_model_loader: - kv 11: llama.rope.scale_linear f32 = 4.000000
llama_model_loader: - kv 12: general.file_type u32 = 17
llama_model_loader: - kv 13: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 14: tokenizer.ggml.tokens arr[str,32256] = ["!", """, "#", "$", "%", "&", "'", ...
llama_model_loader: - kv 15: tokenizer.ggml.scores arr[f32,32256] = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv 16: tokenizer.ggml.token_type arr[i32,32256] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 17: tokenizer.ggml.merges arr[str,31757] = ["Ġ Ġ", "Ġ t", "Ġ a", "i n", "h e...
llama_model_loader: - kv 18: tokenizer.ggml.bos_token_id u32 = 32013
llama_model_loader: - kv 19: tokenizer.ggml.eos_token_id u32 = 32021
llama_model_loader: - kv 20: tokenizer.ggml.padding_token_id u32 = 32014
llama_model_loader: - kv 21: general.quantization_version u32 = 2
llama_model_loader: - type f32: 65 tensors
llama_model_loader: - type q5_K: 193 tensors
llama_model_loader: - type q6_K: 33 tensors
llm_load_vocab: mismatch in special tokens definition ( 243/32256 vs 237/32256 ).
llm_load_print_meta: format = GGUF V3 (latest)
llm_load_print_meta: arch = llama
llm_load_print_meta: vocab type = BPE
llm_load_print_meta: n_vocab = 32256
llm_load_print_meta: n_merges = 31757
llm_load_print_meta: n_ctx_train = 16384
llm_load_print_meta: n_embd = 4096
llm_load_print_meta: n_head = 32
llm_load_print_meta: n_head_kv = 32
llm_load_print_meta: n_layer = 32
llm_load_print_meta: n_rot = 128
llm_load_print_meta: n_embd_head_k = 128
llm_load_print_meta: n_embd_head_v = 128
llm_load_print_meta: n_gqa = 1
llm_load_print_meta: n_embd_k_gqa = 4096
llm_load_print_meta: n_embd_v_gqa = 4096
llm_load_print_meta: f_norm_eps = 0.0e+00
llm_load_print_meta: f_norm_rms_eps = 1.0e-06
llm_load_print_meta: f_clamp_kqv = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: n_ff = 11008
llm_load_print_meta: n_expert = 0
llm_load_print_meta: n_expert_used = 0
llm_load_print_meta: rope scaling = linear
llm_load_print_meta: freq_base_train = 100000.0
llm_load_print_meta: freq_scale_train = 0.25
llm_load_print_meta: n_yarn_orig_ctx = 16384
llm_load_print_meta: rope_finetuned = unknown
llm_load_print_meta: model type = 7B
llm_load_print_meta: model ftype = Q5_K - Medium
llm_load_print_meta: model params = 6.74 B
llm_load_print_meta: model size = 4.46 GiB (5.68 BPW)
llm_load_print_meta: general.name = deepseek-ai_deepseek-coder-6.7b-instruct
llm_load_print_meta: BOS token = 32013 '<｜begin▁of▁sentence｜>'
llm_load_print_meta: EOS token = 32021 '<|EOT|>'
llm_load_print_meta: PAD token = 32014 '<｜end▁of▁sentence｜>'
llm_load_print_meta: LF token = 30 '?'
llm_load_tensors: ggml ctx size = 0.22 MiB
llm_load_tensors: offloading 10 repeating layers to GPU
llm_load_tensors: offloaded 10/33 layers to GPU
llm_load_tensors: CPU buffer size = 4562.38 MiB
llm_load_tensors: CUDA0 buffer size = 1374.21 MiB
...................................................................................................
llama_new_context_with_model: n_ctx = 2048
llama_new_context_with_model: freq_base = 100000.0
llama_new_context_with_model: freq_scale = 0.25
llama_kv_cache_init: CUDA_Host KV buffer size = 704.00 MiB
llama_kv_cache_init: CUDA0 KV buffer size = 320.00 MiB
llama_new_context_with_model: KV self size = 1024.00 MiB, K (f16): 512.00 MiB, V (f16): 512.00 MiB
llama_new_context_with_model: CUDA_Host input buffer size = 12.01 MiB
llama_new_context_with_model: CUDA0 compute buffer size = 171.60 MiB
llama_new_context_with_model: CUDA_Host compute buffer size = 167.20 MiB
llama_new_context_with_model: graph splits (measure): 5

system_info: n_threads = 4 / 8 | AVX = 1 | AVX_VNNI = 0 | 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 | SSSE3 = 0 | VSX = 0 |
sampling:
repeat_last_n = 64, repeat_penalty = 1.100, frequency_penalty = 0.000, presence_penalty = 0.000
top_k = 40, tfs_z = 1.000, top_p = 0.950, min_p = 0.050, typical_p = 1.000, temp = 0.700
mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000
sampling order:
CFG -> Penalties -> top_k -> tfs_z -> typical_p -> top_p -> min_p -> temp
generate: n_ctx = 2048, n_batch = 512, n_predict = -1, n_keep = 0

[divinity76]

llm_load_tensors: offloaded 10/33 layers to GPU

i guess you have -ngl 10 ? try changing it to -ngl 33

edit:

Device 0: NVIDIA GeForce MX230

can also guess that MX230 don't have enough graphics RAM to offload enough to the GPU to actually utilize much of it

[FredericoPerimLopes]

@divinity76 yes it’s a mobile gpu, quite old. I noticed that if the model can fit on vram it works great, but if
I try to offload with larger models it runs on cpu only.

[JohannesGaessler]

It's not running on CPU only, the GPU just has to wait for the much slower CPU to resume work. Consider this plot:

[divinity76]

@FredericoPerimLopes in conclusion, your problem can be solved by upgrading to Nvidia RTX 4090 and AMD Ryzen 9 7950x.

[FredericoPerimLopes]

@JohannesGaessler @divinity76 thanks for the support. I was hoping that although my gpu is quite old and has only 2gb of vram it could at least offload some layers to it. I noticed that sometimes I see some activity on the gpu but most of the time it seems to be idle with only vram increasing which seems to be odd.