Image-to-3D · 4 models · 8 GPUs measured first-party · Updated October 2026

Best GPU for Image-to-3D

Which graphics card to use for image-to-3D, from first-party measurements of TRELLIS.2 Image-to-3D, TRELLIS Image-to-3D, TripoSG Image-to-3D and more on 8 GPUs.

Fastest we measured

NVIDIA H200

90.1 assets/hour on TRELLIS.2 Image-to-3D, the ceiling. Measured on our bench. 141GB of VRAM, $31,000 at launch.

Pros

• 90.1 assets/hour on TRELLIS.2 Image-to-3D
• 141GB, clears the TRELLIS.2 Image-to-3D floor
• Rentable by the hour rather than bought

Cons

• 700W board rating
• Datacenter or workstation hardware, not a retail purchase

Best consumer card

NVIDIA GeForce RTX 4090

57.2 assets/hour on TRELLIS.2 Image-to-3D, fastest card you can buy at retail. Measured on our bench. 24GB of VRAM, $1,599 at launch.

Pros

• 57.2 assets/hour on TRELLIS.2 Image-to-3D
• 24GB, clears the TRELLIS.2 Image-to-3D floor

Cons

• 450W board rating

Cheapest card that runs it

NVIDIA GeForce RTX 3090

30.1 assets/hour on TRELLIS.2 Image-to-3D, lowest launch price that still fits. Measured on our bench. 24GB of VRAM, $1,499 at launch.

Pros

• 30.1 assets/hour on TRELLIS.2 Image-to-3D
• 24GB, clears the TRELLIS.2 Image-to-3D floor

Cons

• 350W board rating

Image-to-3D models turn a single picture into a 3D mesh you can drop into a game engine, a 3D printer slicer or Blender. They don't all hand you the same thing. TripoSR returns a rough mesh with vertex colours in under a second, and TripoSG returns detailed geometry with no texture. Both TRELLIS models return a finished, textured GLB, and our TRELLIS numbers include that export, so they're slower per asset because they do more of the job. That export, which simplifies the mesh and bakes the textures, takes most of each asset's time: 9-26 seconds of a TRELLIS.2 asset's ~40 on an H100, and 70-85% of TRELLIS's, whose 3D generation itself takes 3-7 seconds. TRELLIS numbers are measured on a warm worker: a fresh one spends its first few assets tuning GPU kernels (about 3 minutes for the first TRELLIS.2 asset on an H100), so in a real batch, keep one machine running rather than starting many.

We measured 4 models for image-to-3D on 8 GPUs. Speed is assets per hour (360 = ten seconds per mesh). Every number below is a first-party run on our own harness; cards absent from a model's chart have not been run on it yet.

What the numbers show.

TRELLIS.2 Image-to-3D: fastest on the NVIDIA H200 at 90.1 assets/hour, 3.82x the slowest card we measured (NVIDIA L4); best consumer result NVIDIA GeForce RTX 4090 at 57.2; it used about 14.4GB of VRAM.

TRELLIS Image-to-3D: fastest on the NVIDIA H100 80GB HBM3 at 167.2 assets/hour, 2.43x the slowest card we measured (NVIDIA L4); it used about 12.3GB of VRAM.

TripoSG Image-to-3D: fastest on the NVIDIA H100 80GB HBM3 at 559.4 assets/hour, 6.72x the slowest card we measured (NVIDIA L4); it used about 8.9GB of VRAM.

TripoSR Image-to-3D: fastest on the NVIDIA H100 80GB HBM3 at 2107.4 assets/hour, 2.18x the slowest card we measured (NVIDIA L4); it used about 5.4GB of VRAM.

Which model to pick. On the same card, the NVIDIA H100 80GB HBM3, TripoSR Image-to-3D runs at 2107.4 assets/hour in about 5.4GB; TripoSG Image-to-3D runs at 559.4 assets/hour in about 8.9GB; TRELLIS Image-to-3D runs at 167.2 assets/hour in about 12.3GB; TRELLIS.2 Image-to-3D runs at 87.9 assets/hour in about 14.4GB. TripoSR Image-to-3D gets through the work 24.0x as fast as TRELLIS.2 Image-to-3D, so the model you choose moves the speed as much as the card does.

About TRELLIS.2 Image-to-3D. TRELLIS.2 Image-to-3D: from microsoft, 4.0B parameters, on Hugging Face since December 2025, MIT licence. 1,903,026 downloads in the last 30 days and 39 community quantizations.

How it compares. RTX 4090: TRELLIS.2 Image-to-3D 57.2 assets/hour, Stable Fast 3D 9113.9 (1B), Hunyuan3D 2.0 12.27, Hunyuan3D 2mini 19.81, Shap-E (image to 3D) 1309.1. 2 of 4 beat TRELLIS.2 Image-to-3D here. These aren't like-for-like: TripoSR returns a rough vertex-coloured mesh and TripoSG bare geometry, while both TRELLIS numbers are timed to a finished, textured GLB.

Cost on a rented GPU. 1,000 assets of TRELLIS.2 Image-to-3D: $4.05 on a RTX 3090 ($0.12/hr, 33.2 hours), $39.84 on a H200 ($3.59/hr, 11.1 hours, 9.8x the cost).

Speed tiers for TRELLIS.2 Image-to-3D. 60-360 assets/hour: 3 (H200, H100 80GB HBM3, L40S); under 60 assets/hour: 5 (RTX 4090, RTX 3090). 360 assets/hour is ten seconds per mesh.

VRAM for TRELLIS.2 Image-to-3D. Measured peak 13.7GB, so 16GB is the smallest common card size; smallest card it ran on: RTX 4090 (24GB). That peak is PyTorch's own; the GLB exporter's mesh tools allocate on top of it and ran out of memory on a 24GB A10G until we freed PyTorch's cache first. Microsoft lists 24GB as the minimum, and that's the size to plan for.

Our verdict

For image-to-3D, the NVIDIA H100 80GB HBM3 is the fastest card we measured. Of the cards you can buy at retail, the NVIDIA GeForce RTX 4090 leads. Check the fit table before buying: VRAM, not speed, is what rules a card out.

FAQ

What is the fastest GPU for image-to-3D?

In our runs, the NVIDIA H100 80GB HBM3 at 2107.4 assets/hour on TripoSR Image-to-3D. We measured 4 models on 8 GPUs for this page.

How much VRAM do I need for image-to-3D?

The lightest model here, TripoSR Image-to-3D, used about 5.4GB. The table above shows which models fit 8, 12, 16, 24 and 32GB cards, from measured peaks.

Are these numbers measured or estimated?

Measured. Every number on this page is a first-party run on our own harness, with power and VRAM sampled during the run. Cards we have not run yet are simply absent, not filled in.

What GPU do I need to run TRELLIS.2 Image-to-3D?

About 14GB. Cheapest consumer card that ran it: NVIDIA GeForce RTX 3090 (24GB, 30.1 assets/hour).

How fast is TRELLIS.2 Image-to-3D on the NVIDIA GeForce RTX 4090?

57.2 assets/hour, 63% of the NVIDIA H200.

How much does it cost to run TRELLIS.2 Image-to-3D in the cloud?

$4.05 per 1,000 assets on a NVIDIA GeForce RTX 3090 at $0.12/hr, cheapest of 7 rentable cards we measured.

Can I run TRELLIS.2 Image-to-3D on a 12GB, 16GB or 24GB card?

It used 13.7GB at the precision we tested. 12GB: no; 16GB: yes; 24GB: yes.

Is the RTX 4090 or the RTX 3090 faster for TRELLIS.2 Image-to-3D?

The RTX 4090: 57.2 vs 30.1 assets/hour, 90% faster on our bench.

Should I buy or rent a RTX 4090 for TRELLIS.2 Image-to-3D?

Its $1,599 launch price buys 4,759 rented hours at $0.34/hr, enough for about 272x 1,000 assets of TRELLIS.2 Image-to-3D. Buy only if you'll run more than that.

How we test

Each model runs a fixed workload on every card: a warmup, then timed runs with power, temperature and VRAM sampled every half second through NVML. Models run at the precision and settings from their model card. Datacenter cards run on Modal; consumer cards on rented machines. Non-commercially licensed models are not part of this page. Both TRELLIS models are timed on five game-asset pictures (rifle, pistol, magazine, crate, tree) given as transparent cut-outs, after a full warm-up pass on the same five, from picture to finished GLB. TripoSR and TripoSG are timed on one fixed test picture (a plain disc on white) after a warm-up run, from picture to mesh, without a file export.