SSD-1B · 7 GPUs measured first-party · text-to-image · Updated October 2026
SSD-1B on 7 GPUs, measured first-party: NVIDIA GeForce RTX 5090 leads at 20.49 images/min, RTX 3060 trails at 2.72 images/min, and it peaked at 8GB of VRAM.
Benchmarked weights: segmind/SSD-1B

20.49 images/min on SSD-1B, the ceiling. Measured on our bench. 32GB of VRAM, $1,999 at launch.

15.77 images/min on SSD-1B, fastest card you can buy at retail. Measured on our bench. 24GB of VRAM, $1,599 at launch.

2.72 images/min on SSD-1B, lowest launch price that still fits. Measured on our bench. 12GB of VRAM, $329 at launch.

4.56 images/min on SSD-1B, most speed per dollar. Measured on our bench. 16GB of VRAM, $429 at launch. That is 10.63 images/min per $1,000 of launch price.
What GPU Do You Need for SSD-1B?, images/min by GPU
Efficiency: images/min per 100W drawn
Power is the average pulled during the run, sampled at 1Hz. The fastest card is often not the one here, and for anything left running this is the number that shows up on the bill.
Value: images/min per $1,000 of MSRP
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
SSD-1B. Measured image generation speed by GPU
| GPU | Images/min | s per image | img/W·min | Avg power |
|---|---|---|---|---|
| NVIDIA GeForce RTX 5090 | 20.49 | 2.929 | 0.035 | 588.8 W |
| NVIDIA GeForce RTX 4090 | 15.77 | 3.805 | 0.038 | 413.5 W |
| NVIDIA GeForce RTX 3090 | 7.48 | 8.021 | 0.022 | 344.6 W |
| GeForce RTX 5060 Ti | 4.56 | 13.16 | 0.049 | 93.0 W |
| NVIDIA GeForce RTX 4060 Ti 16GB | 4.49 | 13.361 | 0.032 | 139.1 W |
| NVIDIA GeForce RTX 3070 Founders Edition | 4.31 | 13.92 | 0.02 | 212.0 W |
| NVIDIA GeForce RTX 3060 | 2.72 | 22.036 | 0.017 | 158.9 W |
What the numbers show. Across 7 GPUs measured on our own bench, RTX 5090 is fastest at 20.5 images/min. The slowest, RTX 3060, manages 2.72, so the spread is 7.5x from top to bottom. RTX 5060 Ti is the most efficient, 4.56 images/min at 93W. Per dollar of launch price, RTX 5060 Ti gives the most (13.9 images/min per $1,000). The fastest card with 16GB or less is RTX 5060 Ti at 4.56 images/min.
How it compares. RTX 4090: SSD-1B 15.77 images/min, Stable Diffusion XL 16.28, Stable Diffusion 3 Medium 13.95 (2B), Playground v2.5 9.96 (3B), Kolors 9.88. 1 of 4 beat SSD-1B here.
Cost on a rented GPU. 1,000 images of SSD-1B: $0.22 on a RTX 3060 ($0.036/hr, 6.1 hours), $0.32 on a RTX 5090 ($0.39/hr, 49 min, 1.4x the cost).
SSD-1B: cost per 1,000 images on rented GPUs
| GPU | Cheapest rate | Speed (images/min) | Cost per 1,000 images |
|---|---|---|---|
| NVIDIA GeForce RTX 3060 | $0.036/hr | 2.72 | $0.22 |
| NVIDIA GeForce RTX 3090 | $0.12/hr | 7.48 | $0.27 |
| NVIDIA GeForce RTX 3070 Founders Edition | $0.075/hr | 4.31 | $0.29 |
| NVIDIA GeForce RTX 5090 | $0.39/hr | 20.49 | $0.32 |
| NVIDIA GeForce RTX 4090 | $0.34/hr | 15.77 | $0.36 |
| GeForce RTX 5060 Ti | $0.14/hr | 4.56 | $0.50 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for SSD-1B. 6-30 images/min: 3 (RTX 5090, RTX 4090, RTX 3090); under 6 images/min: 4 (RTX 5060 Ti, RTX 4060 Ti 16GB, RTX 3070 Founders Edition). 30 images/min means two seconds or less per picture.
Time per image. SSD-1B at 1024px and 50 steps: 2.9s per image on the RTX 5090, 22.1s on the RTX 3060. A batch of 100 takes 5 min on the fastest card and 37 min on the slowest.
VRAM for SSD-1B. Measured peak 7.1GB, so 8GB is the smallest common card size; smallest card it ran on: RTX 3070 Founders Edition (8GB).
Power on SSD-1B. Most efficient: RTX 5060 Ti, 93W, 0.34 kWh per 1,000 images. Hungriest: RTX 5090, 589W, 0.48 kWh. At $0.15/kWh: $0.051 per 1,000 images.
Fastest on SSD-1B: NVIDIA GeForce RTX 5090, 20.49 images/min. Cheapest consumer card that ran it: NVIDIA GeForce RTX 3060 ($329, 2.72 images/min). Cheapest to rent per job: NVIDIA GeForce RTX 3060, $0.22 per 1,000 images.
SSD-1B at 1024px in diffusers, bf16, native precision with no offload, timed over three generations after a warmup, with power and VRAM sampled throughout. Image speed tracks tensor compute and architecture generation more than memory bandwidth, so the order here differs from our LLM boards.