[Bug] Slow on gfx1150 with both Vulkan and ROCm builds #1049
Description
Git commit
Operating System & Version
Arch
GGML backends
Vulkan, HIP
Command-line arguments used
./bin/sd --diffusion-model /data/comfyui/models/diffusion_models/z_image_turbo_bf16.safetensors --vae /data/comfyui/models/vae/ae.safetensors --llm /data/comfyui/models/text_encoders/qwen_3_4b.safetensors -p "A cinematic, melancholic photograph of a solitary hooded figure walking through a sprawling, rain-slicked metropolis at night. The city lights are a chaotic blur of neon orange and cool blue, reflecting on the wet asphalt. The scene evokes a sense of being a single component in a vast machine. Superimposed over the image in a sleek, modern, slightly glitched font is the philosophical quote: 'THE CITY IS A CIRCUIT BOARD, AND I AM A BROKEN TRANSISTOR.' -- moody, atmospheric, profound, dark academic" --cfg-scale 1.0 -v --offload-to-cpu -H 512 -W 512 --rng cpu --steps 5 --rng cpu --seed 1061061743296960 --scheduler simple
Steps to reproduce
Run the commands with either the Vulkan or ROCm build of sd on gfx1150
What you expected to happen
Get a little over 2s/it like in ComfyUI (comfyui is using ROCm)
What actually happened
Get 12s/it with Vulkan and 14s/it with ROCm
Logs / error messages / stack trace
System Info:
SSE3 = 1 | AVX = 1 | AVX2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | VSX = 0 | SDCliParams {
mode: img_gen,
output_path: "output.png",
verbose: true,
color: false,
canny_preprocess: false,
preview_method: none,
preview_interval: 1,
preview_path: "preview.png",
preview_fps: 16,
taesd_preview: false,
preview_noisy: false
}
SDContextParams {
n_threads: 12,
model_path: "",
clip_l_path: "",
clip_g_path: "",
clip_vision_path: "",
t5xxl_path: "",
llm_path: "/data/comfyui/models/text_encoders/qwen_3_4b.safetensors",
llm_vision_path: "",
diffusion_model_path: "/data/comfyui/models/diffusion_models/z_image_turbo_bf16.safetensors",
high_noise_diffusion_model_path: "",
vae_path: "/data/comfyui/models/vae/ae.safetensors",
taesd_path: "",
esrgan_path: "",
control_net_path: "",
embedding_dir: "",
wtype: NONE,
tensor_type_rules: "",
lora_model_dir: "",
photo_maker_path: "",
rng_type: cpu,
sampler_rng_type: NONE,
flow_shift: INF
offload_params_to_cpu: true,
control_net_cpu: false,
clip_on_cpu: false,
vae_on_cpu: false,
diffusion_flash_attn: false,
diffusion_conv_direct: false,
vae_conv_direct: false,
chroma_use_dit_mask: true,
chroma_use_t5_mask: false,
chroma_t5_mask_pad: 1,
prediction: NONE,
lora_apply_mode: auto,
vae_tiling_params: { 0, 0, 0, 0.5, 0, 0 },
force_sdxl_vae_conv_scale: false
}
SDGenerationParams {
prompt: "A cinematic, melancholic photograph of a solitary hooded figure walking through a sprawling, rain-slicked metropolis at night. The city lights are a chaotic blur of neon orange and cool blue, reflecting on the wet asphalt. The scene evokes a sense of being a single component in a vast machine. Superimposed over the image in a sleek, modern, slightly glitched font is the philosophical quote: 'THE CITY IS A CIRCUIT BOARD, AND I AM A BROKEN TRANSISTOR.' -- moody, atmospheric, profound, dark academic",
negative_prompt: "",
clip_skip: -1,
width: 512,
height: 512,
batch_count: 1,
init_image_path: "",
end_image_path: "",
mask_image_path: "",
control_image_path: "",
ref_image_paths: [],
control_video_path: "",
auto_resize_ref_image: true,
increase_ref_index: false,
pm_id_images_dir: "",
pm_id_embed_path: "",
pm_style_strength: 20,
skip_layers: [7, 8, 9],
sample_params: (txt_cfg: 1.00, img_cfg: 1.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: simple, sample_method: NONE, sample_steps: 5, eta: 0.00, shifted_timestep: 0),
high_noise_skip_layers: [7, 8, 9],
high_noise_sample_params: (txt_cfg: 7.00, img_cfg: 7.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: 0.00, shifted_timestep: 0),
easycache_option: "",
easycache: disabled (threshold=0, start=0, end=0),
moe_boundary: 0.875,
video_frames: 1,
fps: 16,
vace_strength: 1,
strength: 0.75,
control_strength: 0.9,
seed: 1061061743296960,
upscale_repeats: 1,
}
[DEBUG] stable-diffusion.cpp:167 - Using Vulkan backend
[DEBUG] ggml_extend.hpp:66 - ggml_vulkan: Found 1 Vulkan devices:
[DEBUG] ggml_extend.hpp:66 - ggml_vulkan: 0 = AMD Radeon 890M Graphics (RADV GFX1150) (radv) | uma: 1 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat
[INFO ] stable-diffusion.cpp:234 - loading diffusion model from '/data/comfyui/models/diffusion_models/z_image_turbo_bf16.safetensors'
[INFO ] model.cpp:385 - load /data/comfyui/models/diffusion_models/z_image_turbo_bf16.safetensors using safetensors format
[DEBUG] model.cpp:515 - init from '/data/comfyui/models/diffusion_models/z_image_turbo_bf16.safetensors', prefix = 'model.diffusion_model.'
[INFO ] stable-diffusion.cpp:281 - loading llm from '/data/comfyui/models/text_encoders/qwen_3_4b.safetensors'
[INFO ] model.cpp:385 - load /data/comfyui/models/text_encoders/qwen_3_4b.safetensors using safetensors format
[DEBUG] model.cpp:515 - init from '/data/comfyui/models/text_encoders/qwen_3_4b.safetensors', prefix = 'text_encoders.llm.'
[INFO ] stable-diffusion.cpp:295 - loading vae from '/data/comfyui/models/vae/ae.safetensors'
[INFO ] model.cpp:385 - load /data/comfyui/models/vae/ae.safetensors using safetensors format
[DEBUG] model.cpp:515 - init from '/data/comfyui/models/vae/ae.safetensors', prefix = 'vae.'
[INFO ] stable-diffusion.cpp:318 - Version: Z-Image
[INFO ] stable-diffusion.cpp:346 - Weight type stat: f32: 1095
[INFO ] stable-diffusion.cpp:347 - Conditioner weight type stat: f32: 398
[INFO ] stable-diffusion.cpp:348 - Diffusion model weight type stat: f32: 453
[INFO ] stable-diffusion.cpp:349 - VAE weight type stat: f32: 244
[DEBUG] stable-diffusion.cpp:351 - ggml tensor size = 400 bytes
[DEBUG] llm.hpp:285 - merges size 151387
[DEBUG] llm.hpp:317 - vocab size: 151665
[DEBUG] ggml_extend.hpp:1873 - qwen3 params backend buffer size = 7672.62 MB(RAM) (398 tensors)
[DEBUG] ggml_extend.hpp:1873 - z_image params backend buffer size = 23479.11 MB(RAM) (453 tensors)
[DEBUG] ggml_extend.hpp:1873 - vae params backend buffer size = 94.57 MB(RAM) (138 tensors)
[DEBUG] stable-diffusion.cpp:683 - loading weights
[DEBUG] model.cpp:1363 - using 12 threads for model loading
[DEBUG] model.cpp:1385 - loading tensors from /data/comfyui/models/diffusion_models/z_image_turbo_bf16.safetensors
|====================> | 453/1095 - 109.84it/s
[DEBUG] model.cpp:1385 - loading tensors from /data/comfyui/models/text_encoders/qwen_3_4b.safetensors
|======================================> | 851/1095 - 116.51it/s
[DEBUG] model.cpp:1385 - loading tensors from /data/comfyui/models/vae/ae.safetensors
|==================================================| 1095/1095 - 145.92it/s
[INFO ] model.cpp:1588 - loading tensors completed, taking 7.50s (process: 0.00s, read: 5.71s, memcpy: 0.00s, convert: 1.05s, copy_to_backend: 0.00s)
[DEBUG] stable-diffusion.cpp:710 - finished loaded file
[INFO ] stable-diffusion.cpp:767 - total params memory size = 31246.31MB (VRAM 31246.31MB, RAM 0.00MB): text_encoders 7672.62MB(VRAM), diffusion_model 23479.11MB(VRAM), vae 94.57MB(VRAM), controlnet 0.00MB(VRAM), pmid 0.00MB(VRAM)
[INFO ] stable-diffusion.cpp:850 - running in FLOW mode
[DEBUG] stable-diffusion.cpp:3146 - generate_image 512x512
[INFO ] stable-diffusion.cpp:3177 - sampling using Euler method
[INFO ] denoiser.hpp:388 - get_sigmas with Simple scheduler
[INFO ] stable-diffusion.cpp:3290 - TXT2IMG
[DEBUG] conditioner.hpp:1701 - parse '<|im_start|>user
A cinematic, melancholic photograph of a solitary hooded figure walking through a sprawling, rain-slicked metropolis at night. The city lights are a chaotic blur of neon orange and cool blue, reflecting on the wet asphalt. The scene evokes a sense of being a single component in a vast machine. Superimposed over the image in a sleek, modern, slightly glitched font is the philosophical quote: 'THE CITY IS A CIRCUIT BOARD, AND I AM A BROKEN TRANSISTOR.' -- moody, atmospheric, profound, dark academic<|im_end|>
<|im_start|>assistant
' to [['<|im_start|>user
', 1], ['A cinematic, melancholic photograph of a solitary hooded figure walking through a sprawling, rain-slicked metropolis at night. The city lights are a chaotic blur of neon orange and cool blue, reflecting on the wet asphalt. The scene evokes a sense of being a single component in a vast machine. Superimposed over the image in a sleek, modern, slightly glitched font is the philosophical quote: 'THE CITY IS A CIRCUIT BOARD, AND I AM A BROKEN TRANSISTOR.' -- moody, atmospheric, profound, dark academic', 1], ['<|im_end|>
<|im_start|>assistant
', 1], ]
[DEBUG] llm.hpp:259 - split prompt "<|im_start|>user
" to tokens ["<|im_start|>", "user", "Ċ", ]
[DEBUG] llm.hpp:259 - split prompt "A cinematic, melancholic photograph of a solitary hooded figure walking through a sprawling, rain-slicked metropolis at night. The city lights are a chaotic blur of neon orange and cool blue, reflecting on the wet asphalt. The scene evokes a sense of being a single component in a vast machine. Superimposed over the image in a sleek, modern, slightly glitched font is the philosophical quote: 'THE CITY IS A CIRCUIT BOARD, AND I AM A BROKEN TRANSISTOR.' -- moody, atmospheric, profound, dark academic" to tokens ["A", "Ġcinematic", ",", "Ġmelanch", "olic", "Ġphotograph", "Ġof", "Ġa", "Ġsolitary", "Ġhood", "ed", "Ġfigure", "Ġwalking", "Ġthrough", "Ġa", "Ġsprawling", ",", "Ġrain", "-s", "lick", "ed", "Ġmet", "ropolis", "Ġat", "Ġnight", ".", "ĠThe", "Ġcity", "Ġlights", "Ġare", "Ġa", "Ġchaotic", "Ġblur", "Ġof", "Ġneon", "Ġorange", "Ġand", "Ġcool", "Ġblue", ",", "Ġreflecting", "Ġon", "Ġthe", "Ġwet", "Ġasphalt", ".", "ĠThe", "Ġscene", "Ġev", "okes", "Ġa", "Ġsense", "Ġof", "Ġbeing", "Ġa", "Ġsingle", "Ġcomponent", "Ġin", "Ġa", "Ġvast", "Ġmachine", ".", "ĠSuper", "im", "posed", "Ġover", "Ġthe", "Ġimage", "Ġin", "Ġa", "Ġsleek", ",", "Ġmodern", ",", "Ġslightly", "Ġglitch", "ed", "Ġfont", "Ġis", "Ġthe", "Ġphilosophical", "Ġquote", ":", "Ġ'", "THE", "ĠCITY", "ĠIS", "ĠA", "ĠC", "IR", "CU", "IT", "ĠBOARD", ",", "ĠAND", "ĠI", "ĠAM", "ĠA", "ĠBRO", "KEN", "ĠTRANS", "IST", "OR", ".'", "Ġ--", "Ġmo", "ody", ",", "Ġatmospheric", ",", "Ġprofound", ",", "Ġdark", "Ġacademic", ]
[DEBUG] llm.hpp:259 - split prompt "<|im_end|>
<|im_start|>assistant
" to tokens ["<|im_end|>", "Ċ", "<|im_start|>", "assistant", "Ċ", ]
[INFO ] ggml_extend.hpp:1786 - qwen3 offload params (7672.62 MB, 398 tensors) to runtime backend (Vulkan0), taking 1.07s
[DEBUG] ggml_extend.hpp:1688 - qwen3 compute buffer size: 13.34 MB(VRAM)
[DEBUG] conditioner.hpp:1896 - computing condition graph completed, taking 1638 ms
[INFO ] stable-diffusion.cpp:2921 - get_learned_condition completed, taking 1640 ms
[INFO ] stable-diffusion.cpp:3032 - generating image: 1/1 - seed 1061061743296960
[INFO ] ggml_extend.hpp:1786 - z_image offload params (23479.11 MB, 453 tensors) to runtime backend (Vulkan0), taking 2.67s
[DEBUG] ggml_extend.hpp:1688 - z_image compute buffer size: 255.60 MB(VRAM)
|==================================================| 5/5 - 12.12s/it
[INFO ] stable-diffusion.cpp:3074 - sampling completed, taking 60.58s
[INFO ] stable-diffusion.cpp:3085 - generating 1 latent images completed, taking 60.66s
[INFO ] stable-diffusion.cpp:3088 - decoding 1 latents
[INFO ] ggml_extend.hpp:1786 - vae offload params ( 94.57 MB, 138 tensors) to runtime backend (Vulkan0), taking 0.02s
[DEBUG] ggml_extend.hpp:1688 - vae compute buffer size: 2112.25 MB(VRAM)
[DEBUG] stable-diffusion.cpp:2291 - computing vae decode graph completed, taking 9.97s
[INFO ] stable-diffusion.cpp:3098 - latent 1 decoded, taking 10.00s
[INFO ] stable-diffusion.cpp:3102 - decode_first_stage completed, taking 10.00s
[INFO ] stable-diffusion.cpp:3398 - generate_image completed in 72.30s
save result PNG image to 'output.png' (success)
Additional context / environment details
No response