GPUBeacon
Language
NVIDIA

Cloud GPU · verified 2026-09-05

NVIDIA RTX 5090

GeForce32 GB

32 GB of GDDR7 on Blackwell, 575 W, no ECC, no NVLink. The marketplace favorite when 24 GB died and you still refuse Hopper rent. Treat it as a fast workstation GPU in a rented PC, because that is what it is.

Author Valentin FourierLast modified

VRAM32 GB
From /GPU/hr$0.34
ArchitectureGeForce
VendorNVIDIA

In brief

What to know before you rent

Catalog listings

RTX 5090 Pricing and Availability

Published on-demand rates first. Spot and waitlists are labeled. Confirm the live rate before you provision.

ProvidersRegionBilling typeInterconnect/GPU/hr
Vast.aiIn stockUS-EastOn-demandPCIe$0.34Open provider
RunPodIn stockUS-EastOn-demandPCIe$0.49Open provider
FluidstackLimitedEU-WestOn-demandPCIe$0.55Open provider
HyperstackIn stockEU-WestOn-demandPCIe$0.59Open provider

Our reading

Who this GPU is for

I rent 5090s on Vast.ai and RunPod Community for LoRA, image, and 7B–32B work that fits 32 GB. I do not put a production trainer I cannot checkpoint on a consumer card in someone else’s tower. The FLOPs are real. The machine is a hobby PC with a price tag.

Rent it if Adapters, diffusion, small full fine-tunes, people who will lose a job and shrug.

Skip it if ECC requirements, multi-week pretrains, anything that needs NVLink or 80 GB.

How we compare hours

Datasheet

Technical specifications

NVIDIA GeForce RTX 5090 · Per GPU · Consumer

Compute

FP32
~105 TFLOPS
CUDA cores
21,760
Tensor cores
6805th generation
RT cores
170
Precision support
FP4, FP8, FP16, BF16, TF32, FP32, INT8

Memory

Capacity
32 GB GDDR7
Bandwidth
1,792 GB/s
Bus width
512-bit
ECC
No

Silicon

Architecture
Blackwell
Class
GeForce / consumer

Fabric and host

GPU interconnect
None
Host interface
PCIe 5.0 x16

Power

Board power
575 W

Platform

Partitioning
No MIG

Cloud 5090s are gaming boards in rented boxes. Clock, cooler, and PSU quality are the real spec sheet. The NVIDIA consumer datasheet assumes a proper chassis. Marketplace hosts do not.

Trade-offs

Pros and cons of RTX 5090

What this GPU does well, and when the hour is the wrong buy.

Strengths

  • 32 GB GDDR7 at 1,792 GB/s is a lot of consumer card
  • Blackwell features without datacenter rent
  • Deep marketplace stock vs B200 theatre
  • Often the correct GPU when the model is 7B–13B and the ego wants 80 GB

Limits

  • No ECC. Bit flips are a lifestyle
  • 575 W in a random chassis. Thermal throttling is the hidden SKU
  • 32 GB still loses to 70B full states
  • Consumer CUDA quirks and driver roulette on unverified hosts

Versus 4090

The upgrade is 8 GB and Blackwell. If you already fit in 24 GB, a 4090 hour can still win. If you were paging, 32 GB is the product. I do not pay 5090 rent for a 7B LoRA that was happy on Ada.

Versus H100

Different goods. H100 is ECC, HBM, MIG, NVLink, a node. 5090 is VRAM/$. Mixing them in one spreadsheet is how people convince themselves a $0.34 hour is “the same GPU class.” It is not.

How I actually rent it

  1. 01

    Checkpoint

    If you cannot resume, do not buy a consumer host.

  2. 02

    Read the host

    Verified vs unverified. Reliability scores are not SLAs, but they are not nothing.

  3. 03

    Stay in 32 GB

    If you need 48–96 GB, L40S or RTX PRO 6000. If you need 80 GB HBM, H100.

FAQ

RTX 5090 questions

Who has the cheapest RTX 5090?+

As of September 5, 2026, the cheapest verified in-stock on-demand RTX 5090 tracked by GPUBeacon is $0.34 per GPU per hour from Vast.ai.

How much does the RTX 5090 cost per month?+

As of September 5, 2026, at 720 hours per month, one RTX 5090 costs an estimated $374 at the median on-demand price in this sample. Cheapest verified in stock: $245 per month on-demand.

Where can I rent an RTX 5090?+

GPUBeacon currently tracks RTX 5090 configs from 4 providers. The cheapest verified in-stock on-demand configs come from Vast.ai, RunPod and Hyperstack. See the full price comparison above for every provider and config.

Are RTX 5090 prices going up or down?+

GPUBeacon publishes a catalog sample, not a 90-day price index. As of September 5, 2026, the median on-demand RTX 5090 in this sample is $0.52 per GPU per hour. The cheapest verified in-stock hour is $0.34.

Why choose the RTX 5090?+

Adapters, diffusion, small full fine-tunes, people who will lose a job and shrug.

When is the RTX 5090 not a good fit?+

ECC requirements, multi-week pretrains, anything that needs NVLink or 80 GB.

Is a cloud 5090 good for LLM fine-tunes?+

Yes for adapters and smaller full fine-tunes that fit 32 GB, if you checkpoint. No for 70B full Adam on one card.

Why is it so much cheaper than H100?+

Because you are not buying HBM, ECC, NVLink, or a datacenter node. You are buying a GeForce in someone else’s PC.

Keep comparing

Alternatives to NVIDIA RTX 5090

Same job, different envelope. Compare VRAM, bandwidth and the hour before you lock a badge.

NVIDIA RTX 4090

24 GB GDDR6X · 44% less bandwidth than the RTX 5090

From$0.29/ GPU / hr

Read the verdict

NVIDIA RTX PRO 6000

96 GB GDDR7 · similar bandwidth to the RTX 5090

From$0.66/ GPU / hr

Compare

NVIDIA L40S

48 GB GDDR6 · 52% less bandwidth than the RTX 5090

From$0.79/ GPU / hr

Compare

NVIDIA H100

80 GB HBM3 · 87% more bandwidth than the RTX 5090

From$1.73/ GPU / hr

Compare

Browse all GPUs

Sources

Where these numbers come from

TFLOPS and TDP are from the vendor datasheet. Hourly prices are from the GPUBeacon catalog sample. Open the originals, then confirm the live SKU with the provider before you rent.