[Guidance] Quantization for sdnext with Openvino

[ak-sangal]

I have windows 11 with intel irisx iGPU with 16 GB RAM. I am using sdnext with openvino Can u guide me which quantization I should use for GPU and how setup That quantization.

I want to use flux, z-image turbo and sdxl SDNQ quantization model will they work on igpu, if SDNQ not then which quantization work with Openvino , i did not get any guidance from site doc and discord, GUIDE me.

i am in DEV Branch of sdnext

[vladmandic]

yes, use sdnq
in settings: enable for model, te and llm, set type to either int8 (lower savings) or uint4 (higher savings)

[ak-sangal]

Will it work on on fly quantization

[ak-sangal]

If sdnq quantization enable will it use gpu or cpu

[vladmandic]

yes, it works with on-the-fly-quantization. it cant use quantization acceleration methods such as triton, but its still perfectly fine.
execution itself is not impacted, that will be on gpu regardless.

[ak-sangal]

I will try now thankyou for your valuable time

[ak-sangal]

shall i use Disty0/FLUX.2-klein-4B-SDNQ-4bit-dynamic
Disty0/Z-Image-Turbo-SDNQ-uint4-svd-r32
with my intel irisx igpu with 16 GB RAM as they are pre quantized model guide me

[vladmandic]

those are very different, all i can say - try them

[Disty0]

OpenVINO unfortunately doesn't support any quantization type. OpenVINO tries to be clever and converts the quantized weights to FP16 and keeps them at FP16. We don't have any control over this. Even their own NNCF quantization does this with OpenVINO.

SDNQ works as it doesn't error out and produces correct outputs but OpenVINO casts INT8 to FP16 and keeps the model as FP16.

[ak-sangal]

my setting is right as i want to SDNQ sdxl only with pre mode as i have 16 GB ram and intel irisx igpu

[Disty0]

SDXL doesn't support pre mode, set it to auto.

[ak-sangal]

ok

[ak-sangal]

shall i use pre SDQN model from your huggingface model Disty0/Z-Image-Turbo-SDNQ-uint4-svd-r32 in my 16 GB RAM intel irisx igpu

[Disty0]

If you want to use zimage turbo and don't want to deal with quantization process on model load, then yes.

[ak-sangal]

Disty0/Z-Image-Turbo-SDNQ-uint4-svd-r32 will this model work with my gpu

[ak-sangal]

Any guidance regarding zimage turbo in sdnext with openvino

[Disty0]

SDNQ works as it doesn't error out and produces correct outputs but OpenVINO casts INT8 to FP16 and keeps the model as FP16.

SDNQ will work but you won't see any memory reduction because of OpenVINO.

[ak-sangal]

Ok i try

[ak-sangal]

    i got black image when SDNQ with sdxl turbo mode-auto, type -int8

             'starlette.middleware.gzip.GZipMiddleware'>]

16:30:25-089320 DEBUG API initialize
16:30:25-248193 DEBUG Scripts setup: time=0.257 []
16:30:25-250590 INFO Model: autoload=True selected="model.safetensors"
16:30:25-252091 DEBUG Model requested: fn=threading.py:run:
16:30:25-253036 DEBUG Cache clear
16:30:25-254545 DEBUG Script init: ['system-info.py:app_started=0.06']
16:30:25-256047 INFO Startup time: total=35.36 launch=7.98 loader=7.83 installer=7.83 torch=4.42 libraries=2.38
gradio=1.26 ui-extensions=0.73 extensions=0.32 scans=0.32 ui-networks=0.31 diffusers=0.30
ui-defaults=0.29 ui-img2img=0.28 ui-control=0.19 ui-video=0.14 ui-txt2img=0.14
16:30:25-257046 DEBUG Save: file="F:\sdnext\config.json" json=11 bytes=460 time=0.003
16:30:25-262918 DEBUG Save: file="F:\sdnext\secrets.json" json=2 bytes=139 time=0.003
16:30:25-512720 INFO API user=None code=200 http/1.1 GET /sdapi/v1/progress 127.0.0.1 0.0032
16:30:29-499656 DEBUG UI: connected
16:30:46-722427 INFO API user=None code=200 http/1.1 GET / 127.0.0.1 0.5131
16:30:47-709565 INFO API user=None code=200 http/1.1 GET /sdapi/v1/motd 127.0.0.1 0.0425
16:30:52-800904 INFO API user=None code=200 http/1.1 GET /sdapi/v1/sd-models 127.0.0.1 0.0045
16:30:54-360119 INFO API user=None code=200 http/1.1 GET /sdapi/v1/browser/folders 127.0.0.1 0.0075
16:30:54-414293 INFO Browser session: user=None client=127.0.0.1 agent=Mozilla/5.0 (Windows NT 10.0; Win64; x64)
AppleWebKit/537.36 (KHTML, like Gecko) Chrome/146.0.0.0 Safari/537.36 Edg/146.0.0.0
16:30:54-422795 INFO API user=None code=200 http/1.1 GET /sdapi/v1/start 127.0.0.1 0.009
16:30:54-429292 INFO UI: ready time=7394
16:30:54-542392 DEBUG UI: connected
16:31:20-441274 INFO Settings: changed=2 ['models_not_to_quant', 'sdnq_quantize_weights']
16:31:20-460773 DEBUG Save: file="F:\sdnext\config.json" json=13 bytes=558 time=0.019
16:31:20-471776 DEBUG Save: file="F:\sdnext\secrets.json" json=2 bytes=139 time=0.004
16:31:27-817590 DEBUG Cache clear
16:31:36-329092 DEBUG Search model: name="Diffusers\stabilityai/sdxl-turbo [71153311d3]"
matched="models\Diffusers\models--stabilityai--sdxl-turbo\snapshots\71153311d3dbb46851df1931d3c
a6e939de83304" type=alias
16:31:36-332593 INFO Load model: select="Diffusers\stabilityai/sdxl-turbo [71153311d3]"
16:31:36-352092 INFO Autodetect model: detect="Stable Diffusion XL" class=StableDiffusionXLPipeline
file="models\Diffusers\models--stabilityai--sdxl-turbo\snapshots\71153311d3dbb46851df1931d3ca6e
939de83304"
16:31:36-387093 DEBUG Cache clear
Progress 8.36it/s ██████████████ 100% 1/1 00:00 00:00 Loading checkpoint shards
Loading weights: 100%|█████████████████████████████████████████████████████████████| 517/517 [00:00<00:00, 3259.12it/s]
Progress 54.03it/s ██████████████ 100% 1/1 00:00 00:00 Loading checkpoint shards | 359/517 [00:00<00:00, 3566.59it/s]
Loading weights: 100%|█████████████████████████████████████████████████████████████| 196/196 [00:00<00:00, 2613.35it/s]
Progress 4.08it/s █████████ 100% 7/7 00:01 00:00 Loading pipeline components... | 0/196 [00:00<?, ?it/s]
16:31:38-117957 INFO Network load: type=embeddings loaded=0 skipped=0 time=0.00
16:31:38-123455 DEBUG Setting model: component=vae {'slicing': True, 'tiling': False}
16:31:38-126455 DEBUG Setting model: attention="Scaled-Dot-Product"
16:31:38-128454 DEBUG Load model: post_quant=sdnq
16:31:38-129454 DEBUG Quantization: type=SDNQ modules=['Model', 'TE', 'LLM'] dtype=int8 dtype_te=Same as model
offload=none pre_forward=True
16:31:52-620108 INFO API user=None code=200 http/1.1 GET /sdapi/v1/network/desc 127.0.0.1 0.0045
16:31:56-104741 DEBUG GC: current={'gpu': 0, 'ram': 15.37, 'oom': 0} prev={'gpu': 0, 'ram': 15.67} load={'gpu': 0,
'ram': 98} gc={'gpu': 0, 'py': 762} fn=apply_function_to_model:sdnq_quantize_model why=sdnq
time=1.34
16:31:56-112241 DEBUG Quantization: module="unet" type=sdnq mode=post dtype=int8 matmul_dtype=auto matmul=False
svd=False dynamic=False:group=0:rank=32:steps=8:loss=-1e-08 matmul_conv=False quant_conv=False
quant_embedding=False fp32=True gpu=True device=cpu return=cpu map=default non_blocking=False
modules_skip=[] modules_dtype={}
16:31:57-103927 DEBUG GC: current={'gpu': 0, 'ram': 15.4, 'oom': 0} prev={'gpu': 0, 'ram': 15.4} load={'gpu': 0,
'ram': 98} gc={'gpu': 0, 'py': 0} fn=apply_function_to_model:sdnq_quantize_model why=sdnq
time=0.32
16:31:57-107431 DEBUG Quantization: module="text_encoder" type=sdnq mode=post dtype=int8 matmul_dtype=auto
matmul=False svd=False dynamic=False:group=0:rank=32:steps=8:loss=-1e-08 matmul_conv=False
quant_conv=False quant_embedding=False fp32=True gpu=True device=cpu return=cpu map=default
non_blocking=False modules_skip=[] modules_dtype={}
16:32:02-253301 DEBUG GC: current={'gpu': 0, 'ram': 15.32, 'oom': 0} prev={'gpu': 0, 'ram': 15.54} load={'gpu': 0,
'ram': 98} gc={'gpu': 0, 'py': 104} fn=apply_function_to_model:sdnq_quantize_model why=sdnq
time=0.29
16:32:02-255801 DEBUG Quantization: module="text_encoder_2" type=sdnq mode=post dtype=int8 matmul_dtype=auto
matmul=False svd=False dynamic=False:group=0:rank=32:steps=8:loss=-1e-08 matmul_conv=False
quant_conv=False quant_embedding=False fp32=True gpu=True device=cpu return=cpu map=default
non_blocking=False modules_skip=[] modules_dtype={}
16:32:02-259870 INFO Quantization: type=SDNQ time=24.13
16:32:02-262369 DEBUG Setting model: offload=none limit=0.0
16:32:02-268387 INFO Model compile: pipeline=StableDiffusionXLPipeline mode=default backend=openvino_fx options=[]
compile=['Model', 'VAE', 'Upscaler', 'Control']
16:32:02-275369 DEBUG Model compile: task=torch backends=['cudagraphs', 'inductor', 'openvino', 'openvino_fx',
'openxla', 'tvm']
16:32:02-385425 INFO Model compile: task=torch time=0.11
16:32:02-637539 DEBUG GC: current={'gpu': 0, 'ram': 15.34, 'oom': 0} prev={'gpu': 0, 'ram': 15.34} load={'gpu': 0,
'ram': 98} gc={'gpu': 0, 'py': 17} fn=reload_model_weights:load_diffuser why=load time=0.25
16:32:02-640037 INFO Load model: family=sdxl time={'total': 48.21, 'options': 24.14, 'sdnq': 22.16, 'load': 1.78,
'compile': 0.12} native=1024 memory={'ram': {'total': 15.71, 'rss': 11.83, 'used': 15.34,
'free': 0.36, 'avail': 0.36, 'buffers': 0, 'cached': 0}, 'gpu': {'total': 0, 'used': 0,
'error': 'Torch not compiled with CUDA enabled', 'swap': 0}, 'job': 'Load model'}
16:32:02-665112 DEBUG Save: file="F:\sdnext\config.json" json=15 bytes=663 time=0.021
16:32:02-668612 DEBUG Save: file="F:\sdnext\secrets.json" json=2 bytes=139 time=0.002
16:32:25-273771 TRACE Server: alive=True requests=81 memory=15.43/15.71 status='idle' task='' timestamp=None
current='' id='' job=0 jobs=0 total=3 step=0 steps=0 queued=0 uptime=125 elapsed=22.63 eta=None
progress=0
16:32:48-693394 DEBUG Sampler: "Default" cls=EulerAncestralDiscreteScheduler config={'num_train_timesteps': 1000,
'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'trained_betas':
None, 'prediction_type': 'epsilon', 'timestep_spacing': 'trailing', 'steps_offset': 1,
'rescale_betas_zero_snr': False, 'clip_sample': False, 'interpolation_type': 'linear',
'sample_max_value': 1.0, 'set_alpha_to_one': False, 'skip_prk_steps': True}
16:32:48-776425 INFO Processing modifiers: apply
16:32:48-850036 DEBUG Tokenizer: words=49408
file="models\Diffusers\models--stabilityai--sdxl-turbo\snapshots\71153311d3dbb46851df1931d3ca6e
939de83304\tokenizer\vocab.json"
16:32:52-447743 INFO Base: pipeline=StableDiffusionXLPipeline task=TEXT_2_IMAGE batch=1/1x1 set={'prompt_embeds':
torch.Size([1, 77, 2048]), 'pooled_prompt_embeds': torch.Size([1, 1280]),
'negative_prompt_embeds': torch.Size([1, 77, 2048]), 'negative_pooled_prompt_embeds':
torch.Size([1, 1280]), 'guidance_scale': 1.5, 'generator': 'cpu:[4196989155]',
'num_inference_steps': 6, 'eta': 1.0, 'guidance_rescale': 0, 'denoising_end': None,
'output_type': 'latent', 'width': 512, 'height': 768, 'parser': 'native'}
Progress ?it/s 0% 0/6 00:00 ? Base16:34:01-850139 DEBUG OpenVINO compile: device=AUTO backend=openvino_fx hints={} file="None"
16:34:25-998842 TRACE Server: alive=True requests=192 memory=15.47/15.71 status='running' task='Inference'
timestamp='20260414163252' current='task(9522i3lnoolqv2x)' id='f98902cda0df40f' job=0 jobs=1
total=10 step=0 steps=0 queued=0 uptime=246 elapsed=93.36 eta=None progress=0
Progress 22.73s/it ███████████████████████████████████ 100% 6/6 02:16 00:00 Base
16:35:08-928431 INFO Processing modifiers: unapply

16:35:08-968248 DEBUG Decode: type="taesd" variant="TAESD 1.3 Mocha Croissant"
fn="models\TAESD\taesd_13_decoder_sdxl.pth" layers=3 load
16:35:10-147806 DEBUG Setting decode: component=vae {'slicing': True, 'tiling': False}
16:35:23-331303 DEBUG OpenVINO compile: device=AUTO backend=openvino_fx hints={} file="None"
16:35:40-224874 DEBUG Decode: vae="default" scale=8 upcast=False slicing=True tiling=False latents=[1, 4, 96, 64]:cpu
dtype=torch.float32 time=30.073
16:35:40-434060 INFO Save: image="outputs\text\00167-2026-04-14-sdxl-turbo.jpg" type=JPEG width=512 height=768
size=146381
16:35:40-448560 INFO Processed: images=1 its=0.03 ops=['txt2img']
16:35:40-453576 DEBUG Processed: timers={'total': 176.61, 'pipeline': 140.11, 'decode': 31.32, 'prompt': 3.59,
'preview': 1.16, 'post': 0.2, 'pre': 0.12}
16:35:40-457059 DEBUG Processed: memory={'ram': {'total': 15.71, 'rss': 11.84, 'used': 15.21, 'free': 0.5, 'avail':
0.5, 'buffers': 0, 'cached': 0}, 'gpu': {'total': 0, 'used': 0, 'error': 'Torch not compiled
with CUDA enabled', 'swap': 0}, 'job': ''}
16:36:25-966100 TRACE Server: alive=True requests=201 memory=15.45/15.71 status='idle' task='' timestamp=None
current='' id='' job=0 jobs=0 total=12 step=0 steps=0 queued=0 uptime=366 elapsed=45.49
eta=None progress=0
16:36:30-295340 DEBUG Processing: batch=1 total=4 progress=0.25
16:36:30-382340 DEBUG Sampler: "Default" cls=EulerAncestralDiscreteScheduler config={'num_train_timesteps': 1000,
'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'trained_betas':
None, 'prediction_type': 'epsilon', 'timestep_spacing': 'trailing', 'steps_offset': 1,
'rescale_betas_zero_snr': False, 'clip_sample': False, 'interpolation_type': 'linear',
'sample_max_value': 1.0, 'set_alpha_to_one': False, 'skip_prk_steps': True}
16:36:30-391837 INFO Processing modifiers: apply
16:36:30-610781 INFO Base: pipeline=StableDiffusionXLPipeline task=TEXT_2_IMAGE batch=1/4x1 set={'prompt_embeds':
torch.Size([1, 77, 2048]), 'pooled_prompt_embeds': torch.Size([1, 1280]),
'negative_prompt_embeds': torch.Size([1, 77, 2048]), 'negative_pooled_prompt_embeds':
torch.Size([1, 1280]), 'guidance_scale': 1.5, 'generator': 'cpu:[2719592433]',
'num_inference_steps': 6, 'eta': 1.0, 'guidance_rescale': 0, 'denoising_end': None,
'output_type': 'latent', 'width': 512, 'height': 768, 'parser': 'native'}
Progress 3.14s/it ███████████████████████▎ 67% 4/6 00:13 00:06 Base16:36:47-131164 WARNING Decode: type="taesd" variant="TAESD 1.3 Mocha Croissant" long decode time=3.26
Progress 4.05s/it ███████████████████████████████████ 100% 6/6 00:24 00:00 Base
16:36:58-305388 INFO Processing modifiers: unapply
16:37:01-507772 DEBUG Setting decode: component=vae {'slicing': True, 'tiling': False}
16:37:03-286295 DEBUG Decode: vae="default" scale=8 upcast=False slicing=True tiling=False latents=[1, 4, 96, 64]:cpu
dtype=torch.float32 time=1.785
16:37:03-340974 ERROR Decode: sample=(768, 512, 3) invalid=1179648 dtype=float32 vae=torch.float32 upcast=True failed
to validate
16:37:03-366477 INFO Save: image="outputs\text\00168-2026-04-14-sdxl-turbo.jpg" type=JPEG width=512 height=768
size=3164
16:37:03-368975 DEBUG Processing: batch=2 total=4 progress=0.50
16:37:03-371973 DEBUG Sampler: "Default" cls=EulerAncestralDiscreteScheduler config={'num_train_timesteps': 1000,
'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'trained_betas':
None, 'prediction_type': 'epsilon', 'timestep_spacing': 'trailing', 'steps_offset': 1,
'rescale_betas_zero_snr': False, 'clip_sample': False, 'interpolation_type': 'linear',
'sample_max_value': 1.0, 'set_alpha_to_one': False, 'skip_prk_steps': True}
16:37:03-374476 INFO Processing modifiers: apply
16:37:03-389975 INFO Base: pipeline=StableDiffusionXLPipeline task=TEXT_2_IMAGE batch=2/4x1 set={'prompt_embeds':
torch.Size([1, 77, 2048]), 'pooled_prompt_embeds': torch.Size([1, 1280]),
'negative_prompt_embeds': torch.Size([1, 77, 2048]), 'negative_pooled_prompt_embeds':
torch.Size([1, 1280]), 'guidance_scale': 1.5, 'generator': 'cpu:[2719592434]',
'num_inference_steps': 6, 'eta': 1.0, 'guidance_rescale': 0, 'denoising_end': None,
'output_type': 'latent', 'width': 512, 'height': 768, 'parser': 'native'}
Progress 2.97s/it ███████████████████████████████████ 100% 6/6 00:17 00:00 Base
16:37:21-235109 INFO Processing modifiers: unapply
16:37:22-249232 DEBUG Setting decode: component=vae {'slicing': True, 'tiling': False}
16:37:23-964158 DEBUG Decode: vae="default" scale=8 upcast=False slicing=True tiling=False latents=[1, 4, 96, 64]:cpu
dtype=torch.float32 time=1.715
16:37:24-003181 ERROR Decode: sample=(768, 512, 3) invalid=1179648 dtype=float32 vae=torch.float32 upcast=True failed
to validate
16:37:24-026181 INFO Save: image="outputs\text\00169-2026-04-14-sdxl-turbo.jpg" type=JPEG width=512 height=768
size=3164
16:37:24-028698 DEBUG Processing: batch=3 total=4 progress=0.75
16:37:24-033197 DEBUG Sampler: "Default" cls=EulerAncestralDiscreteScheduler config={'num_train_timesteps': 1000,
'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'trained_betas':
None, 'prediction_type': 'epsilon', 'timestep_spacing': 'trailing', 'steps_offset': 1,
'rescale_betas_zero_snr': False, 'clip_sample': False, 'interpolation_type': 'linear',
'sample_max_value': 1.0, 'set_alpha_to_one': False, 'skip_prk_steps': True}
16:37:24-036692 INFO Processing modifiers: apply
16:37:24-050681 INFO Base: pipeline=StableDiffusionXLPipeline task=TEXT_2_IMAGE batch=3/4x1 set={'prompt_embeds':
torch.Size([1, 77, 2048]), 'pooled_prompt_embeds': torch.Size([1, 1280]),
'negative_prompt_embeds': torch.Size([1, 77, 2048]), 'negative_pooled_prompt_embeds':
torch.Size([1, 1280]), 'guidance_scale': 1.5, 'generator': 'cpu:[2719592435]',
'num_inference_steps': 6, 'eta': 1.0, 'guidance_rescale': 0, 'denoising_end': None,
'output_type': 'latent', 'width': 512, 'height': 768, 'parser': 'native'}
Progress 2.92s/it ███████████████████████████████████ 100% 6/6 00:17 00:00 Base
16:37:41-608285 INFO Processing modifiers: unapply
16:37:42-607135 DEBUG Setting decode: component=vae {'slicing': True, 'tiling': False}
16:37:44-339547 DEBUG Decode: vae="default" scale=8 upcast=False slicing=True tiling=False latents=[1, 4, 96, 64]:cpu
dtype=torch.float32 time=1.733
16:37:44-366685 ERROR Decode: sample=(768, 512, 3) invalid=1179648 dtype=float32 vae=torch.float32 upcast=True failed
to validate
16:37:44-389671 INFO Save: image="outputs\text\00170-2026-04-14-sdxl-turbo.jpg" type=JPEG width=512 height=768
size=3164
16:37:44-391690 DEBUG Processing: batch=4 total=4 progress=1.00
16:37:44-396175 DEBUG Sampler: "Default" cls=EulerAncestralDiscreteScheduler config={'num_train_timesteps': 1000,
'beta_start': 0.00085, 'beta_end': 0.012, 'beta_schedule': 'scaled_linear', 'trained_betas':
None, 'prediction_type': 'epsilon', 'timestep_spacing': 'trailing', 'steps_offset': 1,
'rescale_betas_zero_snr': False, 'clip_sample': False, 'interpolation_type': 'linear',
'sample_max_value': 1.0, 'set_alpha_to_one': False, 'skip_prk_steps': True}
16:37:44-399673 INFO Processing modifiers: apply
16:37:44-415172 INFO Base: pipeline=StableDiffusionXLPipeline task=TEXT_2_IMAGE batch=4/4x1 set={'prompt_embeds':
torch.Size([1, 77, 2048]), 'pooled_prompt_embeds': torch.Size([1, 1280]),
'negative_prompt_embeds': torch.Size([1, 77, 2048]), 'negative_pooled_prompt_embeds':
torch.Size([1, 1280]), 'guidance_scale': 1.5, 'generator': 'cpu:[2719592436]',
'num_inference_steps': 6, 'eta': 1.0, 'guidance_rescale': 0, 'denoising_end': None,
'output_type': 'latent', 'width': 512, 'height': 768, 'parser': 'native'}
Progress 2.93s/it ███████████████████████████████████ 100% 6/6 00:17 00:00 Base
16:38:01-993474 INFO Processing modifiers: unapply
16:38:02-834485 DEBUG Setting decode: component=vae {'slicing': True, 'tiling': False}
16:38:04-536091 DEBUG Decode: vae="default" scale=8 upcast=False slicing=True tiling=False latents=[1, 4, 96, 64]:cpu
dtype=torch.float32 time=1.702
16:38:04-560088 ERROR Decode: sample=(768, 512, 3) invalid=1179648 dtype=float32 vae=torch.float32 upcast=True failed
to validate
16:38:04-579088 INFO Save: image="outputs\text\00171-2026-04-14-sdxl-turbo.jpg" type=JPEG width=512 height=768
size=3164
16:38:04-605087 INFO Save: grid="outputs\grids\00172-2026-04-14-sdxl-turbo.jpg" type=JPEG width=1024 height=1536
size=10074
16:38:04-608587 INFO Processed: images=5 its=0.32 ops=['txt2img']
16:38:04-611088 DEBUG Processed: timers={'total': 104.32, 'pipeline': 80.66, 'decode': 12.86, 'preview': 9.87,
'post': 0.5, 'pre': 0.18, 'prepare': 0.1, 'validate': 0.1}
16:38:04-614089 DEBUG Processed: memory={'ram': {'total': 15.71, 'rss': 10.04, 'used': 14.48, 'free': 1.23, 'avail':
1.23, 'buffers': 0, 'cached': 0}, 'gpu': {'total': 0, 'used': 0, 'error': 'Torch not compiled
with CUDA enabled', 'swap': 0}, 'job': ''}
16:38:26-535336 TRACE Server: alive=True requests=258 memory=14.68/15.71 status='idle' task='' timestamp=None
current='' id='' job=0 jobs=0 total=40 step=0 steps=0 queued=0 uptime=486 elapsed=21.92
eta=None progress=0

[Disty0]

Z-Image is not compatible with FP16. Try setting OpenVINO execution mode to accuracy.

[ak-sangal]

If F16 not work then what setting should I use for z image turbo SDNQ quantization

[vladmandic]

btw, for general questions pls use discussions, transferring this thread

[ak-sangal]

Step by step guide me which setting i use to run z image turbo SDNQ quantization model in sdnext as I have Intel irisX igpu with 16gb ram,what image size , sampler,steps i should use

[Ethan101Potter]

SDNQ doesn’t work with OpenVINO or Intel iGPU — it’s CUDA/NVIDIA only.
For Iris Xe, stick to OpenVINO FP16 models.
Z-Image Turbo runs well, SDXL is usable with optimizations, but FLUX is too heavy.
Convert models using Optimum Intel to OpenVINO IR (FP16 or INT8) and run via SD.Next OpenVINO backend.

[ak-sangal]

I am using openvino GenAi pipeline and converted sdxl and flux schell model using optimum Intel command into int4 and int8 and run in openvino GenAi pipeline successfully but when I move same model into sdnext and run it it does not run can u guide me how use openvino model in sdnext with openvino