Benchmarks

SDXL benchmarks. Images per dollar.

Name: SDXL images per dollar on four GPUs: Eager vs torch.compile, batch 1 and 4
Creator: Runcrate
Published: 2026-10-27

SDXL base at 1024 x 1024 on four rented GPUs, eager and compiled, batch 1 and 4: seconds per image and what 1,000 images cost.

Launch an RTX A6000 All benchmarks

Measured

$0.91per 1,000 images

RTX A6000, compiled, batch 1, at the run-day price.

Run 2026-10-27, single runSee the table

Run 2026-10-27

SDXL images per dollar on four GPUs. Eager vs torch.compile, batch 1 and 4.

At run-day prices an RTX A6000 made 1,000 images for $0.91, an RTX 5090 for $0.93, an L40S for $1.01 and an H100 for $1.80. The H100 took 1.67 s per image against 4.94 s on the RTX A6000.

Single run: one run per GPU, 8 to 16 timed images per configuration. First published 2026-10-27.

The full write-up, with every command

Hardware

GPUs: RTX A6000 48 GB, L40S 48 GB, RTX 5090 32 GB and H100 PCIe 80 GB, one per machine

Software

Model: stabilityai/stable-diffusion-xl-base-1.0, fp16 weights, with the madebyollin/sdxl-vae-fp16-fix VAE; base model only, no refiner
Libraries: diffusers 0.42.1, Python 3.12, PyTorch's built-in SDPA attention

Method

Image: 1024 x 1024, 30 steps, the default scheduler and guidance, one fixed prompt
Configurations: Batch 1 and batch 4, eager and torch.compile(pipe.unet, fullgraph=True) with channels-last; one warm-up generation, then 8 images timed at batch 1 and 16 at batch 4

Price per hour

RTX A6000$0.66 on 2026-10-27, from $0.57/hr per GPU today
L40S$1.14 on 2026-10-27, from $0.54/hr per GPU on a 2-GPU machine today
RTX 5090$1.33 on 2026-10-27, from $0.84/hr per GPU today
H100$3.88 on 2026-10-27, from $3.87/hr per GPU on a 4-GPU machine today

Seconds per image at batch 1, eager and compiledEagerCompiled

Measured on 2026-10-27, from the table below.

Seconds per image, images per hour and cost per 1,000 images, GPU kept busy

GPU	Config	Seconds per image	Images per hour	$ per 1,000, run day	Peak VRAM	$ per 1,000, today
RTX A6000	eager, batch 1	5.79	622	1.06	9.0 GB	0.92
RTX A6000	eager, batch 4	5.65	637	1.04	14.6 GB	0.90
RTX A6000	compiled, batch 1	4.94	728	0.91	9.0 GB	0.79
RTX A6000	compiled, batch 4	5.13	701	0.94	14.6 GB	0.81
L40S	eager, batch 1	3.48	1,034	1.10	9.0 GB	0.52
L40S	eager, batch 4	3.39	1,064	1.07	14.6 GB	0.51
L40S	compiled, batch 1	3.21	1,121	1.02	9.0 GB	0.48
L40S	compiled, batch 4	3.19	1,130	1.01	14.6 GB	0.48
RTX 5090	eager, batch 1	2.77	1,301	1.02	9.0 GB	0.64
RTX 5090	eager, batch 4	2.66	1,355	0.98	14.6 GB	0.62
RTX 5090	compiled, batch 1	2.53	1,424	0.93	9.0 GB	0.59
RTX 5090	compiled, batch 4	2.53	1,421	0.94	14.6 GB	0.59
H100	eager, batch 1	2.25	1,603	2.42	9.0 GB	2.41
H100	eager, batch 4	1.94	1,853	2.09	14.6 GB	2.08
H100	compiled, batch 1	1.80	2,003	1.94	9.0 GB	1.94
H100	compiled, batch 4	1.67	2,156	1.80	14.6 GB	1.80

Run day: the price on 2026-10-27, as the run paid it.

Today: the run-day figure times today's price per GPU-hour over the run-day price per GPU-hour. Today's price is the "from" price per GPU-hour shown on /pricing, which can be on a machine with more GPUs than the run used (the list of GPUs below names it). Prices read 2026-11-04 04:00 UTC.

What torch.compile costs up front (the first compiled call) and when it pays back at batch 1

GPU	Compile, batch 1	Compile, batch 4	Saved per image (batch 1)	Break-even
RTX A6000	188 s	255 s	0.85 s	about 220 images
L40S	82 s	121 s	0.27 s	about 300 images
RTX 5090	96 s	133 s	0.24 s	about 400 images
H100	94 s	130 s	0.45 s	about 210 images

Batch 4 recompiles (new input shape), so a server that mixes batch sizes pays both.

Setup on each machine, before the first image

GPU	Launch to SSH	diffusers install	Model download (about 7 GB)	Pipeline load
RTX A6000	288 s	29 s	73 s	3.9 s
L40S	190 s	under 1 s	24 s	2.0 s
RTX 5090	122 s	15 s	25 s	2.8 s
H100	307 s	7 s	27 s	2.5 s

What stood out

Batch 4 changed throughput by -4% to +4% on the RTX A6000, L40S and RTX 5090. Only the H100 gained (16% eager, 8% compiled).
Peak memory was 9.0 GB at batch 1 and 14.6 GB at batch 4, so SDXL base fits on any of these cards with room to spare.
The model download varied 3x between machines (24 s to 73 s for the same 7 GB). For short jobs, keep the weights on a volume.

Keep in mind

One prompt, one resolution, one scheduler, 30 steps. Fewer steps, a distilled model or another scheduler change the absolute numbers.
No refiner, no ControlNet, no LoRA, no attention kernels beyond PyTorch's built-in SDPA, no TensorRT.
Treat differences under about 5% as noise.

Costs

How the costs are worked out. Run day and today, side by side.

Run day

The cost the write-up printed: the hourly price the machine was launched at that day, against what it measured. Prices move daily, so each run-day figure carries its date. Every run was a single run.

Today

The run-day figure moved by today's price: the "from" price per GPU-hour shown on /pricing, over the run-day rate per GPU-hour. A GPU that is not in stock right now reads "Not in stock" rather than a stale price. Prices read 2026-11-04 04:00 UTC.

GPUs

The GPUs on this page. From price per GPU-hour, today.

RTX A6000from $0.57/hr per GPU
L40Sfrom $0.54/hr per GPU on a 2-GPU machine
RTX 5090from $0.84/hr per GPU
H100from $3.87/hr per GPU on a 4-GPU machine

Benchmarks

More benchmarks.

LLM inferenceCost per GPU.$0.12 per 1M output tokens. RTX A6000, vLLM, 64 concurrent requests, at the run-day price, run 2026-10-26, single run.A100 vs H100 vs H200One script, six GPUs.718.46 bf16 TFLOPS. H200, dense bf16 matmul at 8192, run 2026-10-25, single run.Speech to textAudio hours per dollar.315x real time. RTX 5090, faster-whisper large-v3-turbo, batch 16, run 2026-10-28, single run.Text embeddingsCost per passage.$0.032 per 1M passages. L40S, bge-small, 128-token passages, at the run-day price, run 2026-10-29, single run.Training and fine-tuningMemory, speed, cost.14.5 GiB peak memory. LoRA of Qwen2.5-1.5B on an L40S, micro-batch 4, run 2026-10-28, single run.GPU startup timeTimed from the API call.186 s to the first CUDA op. L40S, from the create call to a 4096 x 4096 matmul, run 2026-10-26, single run.