aliteq.

RTX 5090: 32 GB, what it runs and what it costs

Specs checked · Rental prices from 55m ago · Aliteq Team · How we compute this

The short answer

The NVIDIA GeForce RTX 5090 has 32 GB of VRAM and 1792 GB/s of memory bandwidth (2025). At Q4_K_M quality or better with an 8K context it runs 32 of the open models we track; the largest that fits with room to spare is Qwen3.6 35B-A3B (MoE). The cheapest rental right now is $0.200/hr (Vast.ai, interruptible).

VRAM
32 GB
Bandwidth
1792 GB/s
Board power
575 W
Launched
2025

Source: NVIDIA specification · bandwidth not yet confirmed by a second source, so we don't estimate speed from it

Buy it or rent it

GeForce RTX 5090 32GB cards, maker's photo

Buy an RTX 5090

GeForce RTX 5090 32GB cards

New, from Amazon. Check the exact model says 32 GB before you buy.

Best for: Top-end gaming and workstation cards; 70B-class models at low-bit quants.

Check price on Amazon

Today's price shows on Amazon · Referral link

ProviderTypeFrom, per hourRent
Vast.aiSpot (can be interrupted)$0.200/hrVast.ai
Vast.aiOn demand$0.387/hrVast.ai
LiumSpot (can be interrupted)$0.670/hrLium
LiumOn demand$0.700/hrLium
RunpodOn demand$1.190/hrRunpod

Lowest price per provider in our hourly snapshot (55m ago). Rent buttons are referral links. Every card side by side: cloud GPU prices.

Which AI models run on the RTX 5090

Computed by our VRAM engine for an 8K context: the best quality each model fits at in 32 GB. Speeds are upper-bound estimates from memory bandwidth, not benchmarks.

ModelBest fitVRAM used~tok/s
Mixtral 8x7B (MoE)Q5_032 GB100% · tight—
Qwen3.6 35B-A3B (MoE)Q6_K28 GB89%—
Ornith-1.5 35B-A3B (MoE)Q6_K28 GB89%—
Qwen3.5 35B-A3B (MoE)Q6_K28 GB89%—
DeepSeek R1 Distill 32BQ6_K28 GB87%—
Qwen2.5 Coder 32BQ6_K28 GB87%—
Qwen2.5 32B InstructQ6_K28 GB87%—
Qwen3 32BQ6_K28 GB87%—
NVIDIA Nemotron 3 Nano 30B-A3B (MoE)Q6_K25 GB78%—
Gemma 4 31BQ6_K25 GB79%—
Qwen3 30B-A3B (MoE)Q8_032 GB99% · tight—
Qwen3-Coder 30B-A3B (MoE)Q8_032 GB99% · tight—

Fits only below Q4 quality: Qwen3-Next 80B-A3B (MoE) (IQ3_XS), Qwen3-Coder-Next 80B-A3B (MoE) (IQ3_XS), DeepSeek R1 Distill 70B (Q2_K).

Too big for 32 GB (rent a bigger card instead): what it costs to run GLM-4.5-Air (MoE), what it costs to run Llama 4 Scout 17B-16E (MoE), what it costs to run gpt-oss 120B (MoE), what it costs to run NVIDIA Nemotron 3 Super 120B-A12B (MoE).

Check a model and context of your own in the VRAM calculator.

AI video on the RTX 5090

Alibaba's Wan 2.2 is an open video model you can run yourself in ComfyUI. With 32 GB the RTX 5090 fits:

  • Wan 2.2 14B at 640x640 with fp8 weights (24 GB and up, per Runpod's Wan 2.2 guide)
  • Wan 2.2 5B, with ComfyUI's native offloading (8 GB and up, per ComfyUI's Wan 2.2 docs)

What each setup needs, the cheapest card to rent and what one clip costs: cloud GPUs for ComfyUI with Flux and Wan 2.2. Against an app or an API: AI video cost per clip, app vs API vs rented GPU. Rather not run a model yourself? Hosted video models such as Seedance, Veo and Hailuo are compared in Kling AI alternatives, priced per clip.

Our guides on the RTX 5090

Cards to compare it with

Every card by VRAM tier: all GPUs for gaming and local AI.

RTX 5090: common questions

How much VRAM does the RTX 5090 have?
32 GB, with 1792 GB/s of memory bandwidth, per the maker's specification.
What is the biggest AI model the RTX 5090 can run?
Of the open models we track, Qwen3.6 35B-A3B (MoE) is the largest that fits at Q4_K_M or better with an 8K context and room to spare (about 28 GB of 32 GB). Mixtral 8x7B (MoE) also squeezes in, but with almost no memory left for longer context. Bigger models only fit at lower-quality quants or need more memory.
How much does it cost to rent an RTX 5090?
From $0.200/hr on Vast.ai (spot, can be interrupted), the lowest price in our hourly snapshot of Nov 7, 2026. Prices move every hour.
Is 32 GB enough?
Top-end gaming and workstation cards; 70B-class models at low-bit quants.