FLUX image benchmarks on AMD GPUs

72 reported configurations · 15 sources · 100 measurements

Documented results

24 selected configurations

Selected across sources, not a performance ranking. Match model weights, resolution, steps, software and timing scope before comparing.

Source	Model	GPUs	Output	Measurements
View ReportComfyUI	FLUX.1-devFull precision (16-bit)Model reference ↗Text → image	Ryzen AI Max+ 395 (Radeon 8060S)1 GPU Configuration Engine ComfyUI ab0d8a92 Backend ROCm / HIP OS Windows11 Pro24H2 build26100 Parallelization Not reported Memory Not reported Reported 2026-07-14 Full record →	1024 × 1024Batch 120 stepsBatch 1	103.89 s/imageTime / image · reported 3.65 s/stepStep time · reported Whole workflow · Cold
View ReportDiffusion workflow	FLUX.1-devBF16 + INT8 weight-onlyModel reference ↗Text → image	Radeon RX 7800 XT1 GPU Configuration Engine Diffusers + TorchAO Diffusers0.37.1 / TorchAO0.14.1 / torch2.9.1+rocm7.1.1 Backend ROCm / HIP OS Linux Parallelization Not reported Memory 16 GB per GPU Reported Not reported Full record →	1024 × 1024Batch 128 stepsBatch 1	128.3 s/imageTime / image · reported 4.58 s/stepCalculated · Pipeline time / step · reported · pipeline per stepDerived by source: total pipeline wall time / 28 steps, including VAE decode; not isolated denoising-step timing. 6.39 GBPeak VRAM · reported Whole workflow · Warm
View ReportComfyUI	FLUX.2 klein 4BSDNQ 4-bit storage / INT8-BF16 runtimeTested weights ↗Text → image	Radeon AI PRO R97001 GPU Configuration Engine ComfyUI / Paiton 0.34.0 / Paiton1.0.1 Backend ROCm / HIP OS Linux Parallelization Single GPU Memory 32 GB per GPU Reported 2026-09-07 Full record →	1024 × 1024Batch 14 stepsBatch 1	1.406 sTime / image · mean Whole workflow · Warm · mean
View ReportPaiton	FLUX.2 klein 4BSDNQ 4-bit storage / INT8-BF16 runtimeTested weights ↗Text → image	Radeon AI PRO R97001 GPU Configuration Engine Paiton Backend ROCm / HIP OS Linux Parallelization Single GPU Memory 32 GB per GPU Reported 2026-09-07 Full record →	1024 × 1024Batch 14 stepsBatch 1	104.074 sTime / image · reported Whole workflow · Cold
View ReportComfyUI	FLUX.2 Klein Base 9BFP8Model reference ↗Text → image	Radeon RX 9060 XT1 GPU Configuration Engine ComfyUI Backend ROCm / HIP OS CachyOS (Arch Linux), kernel 6.19 Parallelization Single GPU Memory 16 GB per GPU Reported Not reported Full record →	1024 × 102420 stepsBatch not reported	82.18 sReported time · reported Timing details missing
View ReportComfyUI	FLUX.1-devFull precision (source label)Model reference ↗Text → image	Ryzen AI Max+ 395 (Radeon 8060S)1 GPU Configuration Engine ComfyUI 0.42.0 Backend ROCm / HIP OS Windows 11 26H2 Parallelization Single GPU Memory Not reported Reported 2026-10-30 Full record →	1024 × 102420 stepsBatch not reported	51.65 s/imageTime / image · meanMean of images2-8 for the controller-product prompt; first image with model loading excluded. Timer boundary otherwise unreported. 38.25 GBReported max GPU memory · peak · unified gpu memory usageSource-reported maximum GPU-memory use for the model-generation sequence; unified system memory, not dedicated VRAM or a paired seven-sample timing statistic. Warm · Timing details missing · mean
View ReportComfyUI	FLUX.2 devFP8Model reference ↗Text → image	Radeon AI PRO R97001 GPU Configuration Engine ComfyUI Backend ROCm / HIP OS Arch Linux Parallelization Single GPU Memory 32 GB per GPU Reported Not reported Full record →	1024 × 102430 stepsBatch not reported	126.2 sGeneration time · reported 2.58 s/stepStep time · reportedDenoising stage only Whole workflow · Cold
View ReportComfyUI	FLUX.1 devQuant not reportedModel reference ↗Text → image	Radeon RX 9070 XT1 GPU Configuration Engine ComfyUI 0.38.0 Backend ROCm / HIP OS Windows 11 Home 26200 Parallelization Single GPU Memory 16 GB per GPU Reported 2026-10-12 Full record →	1024 × 1024Batch 120 stepsBatch 1	39.82 sReported time · mean Timing details missing · mean
View ReportComfyUI	FLUX.2 Klein Base 9BFP8 fastModel reference ↗Text → image	Radeon AI PRO R97001 GPU Configuration Engine ComfyUI Backend ROCm / HIP OS Not reported Parallelization Single GPU Memory 32 GB per GPU Reported Not reported Full record →	1024 × 102420 stepsBatch not reported	25.38 sReported time · reported Timing details missing
View Reportstable-diffusion.cpp / sd2R	FLUX.1 devQ4_K_SModel reference ↗Text → image	Radeon RX 90701 GPU Configuration Engine sd2R / stable-diffusion.cpp Backend Vulkan OS Not reported Parallelization Single GPU Memory 16 GB per GPU Reported Not reported Full record →	768 × 76810 stepsBatch not reported	13.72 sReported time · reported Timing details missing
View Reportstable-diffusion.cpp / sd2R	FLUX.2-klein-4BINT4 ConvRot (source label)Model reference ↗Text → image	Radeon RX 7900 XTXGPU count not reported Configuration Engine stable-diffusion.cpp Backend ROCm / HIP OS Windows (version unreported) Parallelization Not reported Memory Not reported Reported Not reported Full record →	1024 × 10248 stepsBatch not reported	9.56 sReported time · reportedSource-reported duration; the arm does not explicitly restate the timer boundary. Timing details missing
View ReportComfyUI	FLUX dev (generation unreported)Quant not reportedFamily reference ↗Text → image	Radeon AI PRO R97001 GPU Configuration Engine ComfyUI Backend Not reported OS Linux (ArchivistaBox) Parallelization Single GPU Memory 32 GB per GPU Reported 2026-04-15 Full record →	1920 × 1088Steps not reportedBatch not reported	120 sReported time · range lowerApproximate two-to-three-minute range converted to seconds. 180 sReported time · range upperApproximate two-to-three-minute range converted to seconds. Timing details missing · range lower
View ReportSD WebUI	black-forest-labs/flux.1-krea-devQuant not reportedFamily reference ↗Text → image	Radeon RX 7900 XT1 GPU Configuration Engine SD.Next b56d508a Backend ZLUDA OS Windows Parallelization Not reported Memory 20 GB per GPU Reported Not reported Full record →	512 × 512Batch 1Steps not reportedBatch 1	0.5 image-steps/sPipeline throughput · reportedHarness computes steps times batch divided by process_images wall time; not sampler iterations per second or images per second.
View ReportComfyUI	FLUX.1 (variant unreported)Quant not reportedFamily reference ↗Text → image	Radeon AI PRO R97001 GPU Configuration Engine ComfyUI Backend ROCm / HIP OS Linux (Ubuntu; version unresolved) Parallelization Single GPU Memory 32 GB per GPU Reported 2026-05-29 Full record →	1920 × 1080Steps not reportedBatch not reported	77.24 sReported time · reported Timing details missing
View Reportstable-diffusion.cpp / sd2R	FLUX.1-schnellQuant not reportedModel reference ↗Text → image	Radeon RX 7600 XT1 GPU Configuration Engine stable-diffusion.cpp Backend ROCm / HIP OS Linux Parallelization Not reported Memory 16 GB per GPU Reported 2026-01-19 Full record →	Resolution not reported4 stepsBatch not reported	19.45 s/stepStep time · reported Sampling only · Timing details missing
View ReportComfyUI	FLUX.1-devFull precision (16-bit)Model reference ↗Text → image	Ryzen AI Max+ 395 (Radeon 8060S)1 GPU Configuration Engine ComfyUI ab0d8a92 Backend ROCm / HIP OS Windows11 Pro24H2 build26100 Parallelization Not reported Memory Not reported Reported 2026-07-14 Full record →	1024 × 1024Batch 120 stepsBatch 1	77.56 s/imageTime / image · reported 3.64 s/stepStep time · reported Whole workflow · Warm
View ReportDiffusion workflow	FLUX.1-devBF16 + INT8 weight-onlyModel reference ↗Text → image	Radeon RX 7800 XT1 GPU Configuration Engine Diffusers + TorchAO Diffusers0.37.1 / TorchAO0.14.1 / torch2.9.1+rocm7.1.1 Backend ROCm / HIP OS Linux Parallelization Not reported Memory 16 GB per GPU Reported Not reported Full record →	1024 × 1024Batch 128 stepsBatch 1	88.5 s/imageTime / image · reported 3.16 s/stepCalculated · Pipeline time / step · reported · pipeline per stepDerived by source: total pipeline wall time / 28 steps, including VAE decode; not isolated denoising-step timing. 6.39 GBPeak VRAM · reported Whole workflow · Warm
View ReportPaiton	FLUX.2 klein 4BSDNQ 4-bit storage / INT8-BF16 runtimeTested weights ↗Text → image	Radeon AI PRO R97001 GPU Configuration Engine Paiton 1.0.1 Backend ROCm / HIP OS Linux Parallelization Single GPU Memory 32 GB per GPU Reported 2026-09-07 Full record →	1024 × 1024Batch 14 stepsBatch 1	19.365 sTime / image · reported Whole workflow · Cold
View ReportComfyUI	FLUX.2 Klein Base 9BFP8Model reference ↗Text → image	Radeon RX 9060 XT1 GPU Configuration Engine ComfyUI Backend ROCm / HIP OS CachyOS (Arch Linux), kernel 6.19 Parallelization Single GPU Memory 16 GB per GPU Reported Not reported Full record →	1024 × 102420 stepsBatch not reported	62.61 sReported time · reported Timing details missing
View ReportComfyUI	FLUX.2-devBF16 mixedModel reference ↗Text → image	Ryzen AI Max+ 395 (Radeon 8060S)1 GPU Configuration Engine ComfyUI 0.42.0 Backend ROCm / HIP OS Windows 11 26H2 Parallelization Single GPU Memory Not reported Reported 2026-10-30 Full record →	1024 × 102420 stepsBatch not reported	249.9 s/imageTime / image · reportedFirst image explicitly includes model loading; not part of the seven-image warm mean. Overall timer boundary otherwise unreported. Cold · Timing details missing
View ReportComfyUI	FLUX.2 devFP8Model reference ↗Image editing	Radeon AI PRO R97001 GPU Configuration Engine ComfyUI Backend ROCm / HIP OS Arch Linux Parallelization Single GPU Memory 32 GB per GPU Reported Not reported Full record →	1024 × 102430 stepsBatch not reported	220 sGeneration time · reported 5.73 s/stepStep time · reportedDenoising stage only Whole workflow · Cold
View ReportComfyUI	FLUX.1 devQuant not reportedModel reference ↗Text → image	Radeon RX 9070 XT1 GPU Configuration Engine ComfyUI 0.38.0 Backend ROCm / HIP OS Windows 11 Home 26200 Parallelization Single GPU Memory 16 GB per GPU Reported 2026-10-12 Full record →	1024 × 1024Batch 120 stepsBatch 1	39.26 sReported time · mean Timing details missing · mean
View ReportComfyUI	FLUX.2 Klein Base 9BFP8 fastModel reference ↗Text → image	Radeon AI PRO R97001 GPU Configuration Engine ComfyUI Backend ROCm / HIP OS Not reported Parallelization Single GPU Memory 32 GB per GPU Reported Not reported Full record →	1024 × 102420 stepsBatch not reported	16.85 sReported time · reported Timing details missing
View Reportstable-diffusion.cpp / sd2R	FLUX.1 devQ4_K_SModel reference ↗Text → image	Radeon RX 90701 GPU Configuration Engine sd2R / stable-diffusion.cpp Backend Vulkan OS Not reported Parallelization Single GPU Memory 16 GB per GPU Reported Not reported Full record →	1024 × 102410 stepsBatch not reported	24.84 sReported time · reported Timing details missing

Explore all 72 configurations →

All image benchmarks · Methodology and source review