// Tools
GPU Benchmark Lookup
Real inference and training benchmarks across consumer GPUs. Filter by model, GPU, or workload type. Your rigs are highlighted.
Your Rig
RTX 5080
16GB VRAM · Primary rig — fastest local inference
Your Rig
RTX 3080 16GB
16GB VRAM · Secondary — solid for training runs
Your Rig
GTX 1660 Ti
6GB VRAM · Aux — light inference only
32 results
GPU
Model
Resolution
VRAM
Steps
Time
Speed
Precision
Flags / Notes
RTX 5090
SDXL FP16
1024×1024
32GB
20
4.8s
50.0 img/m
FP16
--gpu-only --highvram
RTX 5090
LTX Video 2.3
768×512
32GB
25
2.8s
34.6 fps
FP16
--gpu-only --highvram· 97 frames
RTX 5080
SDXL FP16
1024×1024
16GB
20
12.8s
18.8 img/m
FP16
--gpu-only --highvram
RTX 5080
LTX Video 2.3
768×512
16GB
25
5.4s
18.0 fps
FP16
--gpu-only --highvram· 97 frames
RTX 4090
SDXL FP16
1024×1024
24GB
20
14.2s
16.9 img/m
FP16
--gpu-only --highvram
RTX 5090
FLUX Dev FP8
1024×1024
32GB
20
4.1s
14.6 img/m
FP8
--gpu-only --highvram· Top consumer card
RTX 4090
LTX Video 2.3
768×512
24GB
25
6.8s
14.3 fps
FP16
--gpu-only --highvram· 97 frames
RTX 4080 Super
LTX Video 2.3
768×512
16GB
25
9.2s
10.5 fps
FP16
--gpu-only· 97 frames
RTX 4080 Super
SDXL FP16
1024×1024
16GB
20
14.0s
8.6 img/m
FP16
--gpu-only
RTX 3090
LTX Video 2.3
768×512
24GB
25
12.4s
7.8 fps
FP16
--gpu-only· 97 frames
RTX 5080
FLUX Dev FP8
1024×1024
16GB
20
8.1s
7.4 img/m
FP8
--gpu-only --highvram
RTX 4090
FLUX Dev FP8
1024×1024
24GB
20
9.2s
6.5 img/m
FP8
--gpu-only --highvram
RTX 3080 16GB
LTX Video 2.3
768×512
16GB
25
16.8s
5.8 fps
FP16
--gpu-only· 97 frames
RTX 3080 16GB
SDXL FP16
1024×1024
16GB
20
22.0s
5.5 img/m
FP16
--gpu-only
RTX 4080 Super
FLUX Dev FP8
1024×1024
16GB
20
11.4s
5.3 img/m
FP8
--gpu-only
RTX 3080 10GB
LTX Video 2.3
512×512
10GB
25
24.0s
4.0 fps
FP16
--lowvram· Reduced res, 49 frames
RTX 4070 Ti
FLUX Dev FP8
1024×1024
12GB
20
18.2s
3.3 img/m
FP8
--gpu-only
RTX 3090
FLUX Dev FP8
1024×1024
24GB
20
22.1s
2.7 img/m
FP8
--gpu-only
RTX 3060 12GB
SDXL FP16
1024×1024
12GB
20
24.0s
2.5 img/m
FP16
--gpu-only
RTX 3080 16GB
FLUX Dev FP8
1024×1024
16GB
20
28.4s
2.1 img/m
FP8
--gpu-only
RTX 3080 10GB
FLUX Dev FP8
1024×1024
10GB
20
38.0s
1.6 img/m
FP8
--lowvram· Requires --lowvram
RTX 3060 12GB
FLUX Dev FP8
1024×1024
12GB
20
52.0s
1.2 img/m
FP8
--lowvram
GTX 1660 Ti
SDXL FP16
768×768
6GB
20
1.0m
1.0 img/m
FP16
--lowvram· Reduced resolution only
RTX 5090
FLUX LoRA Training
1024×1024
32GB
1000
1.2h
1.2h
BF16
--gradient_checkpointing· ~70 min / 1000 steps
RTX 4090
FLUX LoRA Training
1024×1024
24GB
1000
1.5h
1.5h
BF16
--gradient_checkpointing· ~90 min / 1000 steps
RTX 5080
FLUX LoRA Training
1024×1024
16GB
1000
2.0h
2.0h
BF16
--gradient_checkpointing· ~2h / 1000 steps
RTX 4080 Super
FLUX LoRA Training
1024×1024
16GB
1000
2.5h
2.5h
BF16
--gradient_checkpointing· ~2.5h / 1000 steps
RTX 3090
FLUX LoRA Training
1024×1024
24GB
1000
3.0h
3.0h
BF16
--gradient_checkpointing· ~3h / 1000 steps
RTX 3080 16GB
FLUX LoRA Training
1024×1024
16GB
1000
4.0h
4.0h
BF16
--gradient_checkpointing· ~4h / 1000 steps
RTX 3080 10GB
FLUX LoRA Training
1024×1024
10GB
1000
6.0h
6.0h
BF16
--gradient_checkpointing --cpu_offload· ~6h / 1000 steps, CPU offload required
GTX 1660 Ti
FLUX Dev FP8
1024×1024
6GB
20
OOM
—
FP8
N/A· OOM — not supported
GTX 1660 Ti
LTX Video 2.3
512×512
6GB
25
OOM
—
FP16
N/A· OOM — not supported
Fast (70%+ of max)MediumSlow / OOMYour rig
// Hardware Partner
Benchmarks too slow
for your workload?
ComputeAtlas is an AI workstation planning platform. Use it to spec the exact hardware you need — GPU, CPU, RAM, storage — before you buy.