Vast.aiUS-EastOn-demand
$0.29/ GPU / hr
Open provider
Cloud GPU · verified 2026-09-05
NVIDIA RTX 4090
GeForce24 GB
24 GB GDDR6X, 450 W, the marketplace default. Still the correct GPU for a shocking number of LoRAs. Still a gaming card in a rented box.
VRAM24 GB
From /GPU/hr$0.29
ArchitectureGeForce
VendorNVIDIA
Sponsored
Featured RTX 4090 offers
The four cheapest in-stock on-demand RTX 4090 hours in this sample. Each card is a sponsored placement.
RunPodUS-EastOn-demand
$0.39/ GPU / hr
Open provider
FluidstackEU-WestOn-demand
$0.44/ GPU / hr
Open provider
HyperstackEU-WestOn-demand
$0.49/ GPU / hr
Open providerSponsored does not change ranking. Commissions never move a row, a GPU, or a displayed price.
In brief
What to know before you rent
- 24 GB is tight for 70B, plenty for 7B–13B adapters and diffusion.
- No ECC, no NVLink, no MIG. Checkpoint or cry.
- Cheapest serious CUDA hour on this index, when the host is not a toaster.
Catalog listings
RTX 4090 Pricing and Availability
Published on-demand rates first. Spot and waitlists are labeled. Confirm the live rate before you provision.
ProvidersRegionBilling typeInterconnect/GPU/hr
Our reading
Who this GPU is for
I develop on 4090s. I move the run I care about to Secure Hopper or a PRO card. The 4090 is not a failed H100. It is the GPU you should have started on before you bought 80 GB of ego.
Rent it if LoRA, SD, small full fine-tunes, iteration, anyone who can lose a job.
Skip it if 70B full fine-tunes, compliance, multi-GPU NVLink, uncheckpointable trainers.
How we compare hoursDatasheet
Technical specifications
NVIDIA GeForce RTX 4090 · Per GPU · Consumer
Compute
- FP32
- 82.6 TFLOPS
- CUDA cores
- 16,384
- Tensor cores
- 5124th generation
- Precision support
- FP8, FP16, BF16, TF32, FP32, INT8
Memory
- Capacity
- 24 GB GDDR6X
- Bandwidth
- 1,008 GB/s
- Bus width
- 384-bit
- ECC
- No
Silicon
- Architecture
- Ada Lovelace
- Class
- GeForce / consumer
Fabric and host
- GPU interconnect
- None
- Host interface
- PCIe 4.0 x16
Power
- Board power
- 450 W
Laptop 4090s and 4090 D SKUs are not this card. Cloud rows should be 24 GB desktop-class. If VRAM is 16 GB, you bought a different GeForce.
Trade-offs
Pros and cons of RTX 4090
What this GPU does well, and when the hour is the wrong buy.
Strengths
- The floor price of useful CUDA
- Ada Tensor cores are enough for adapters
- Everywhere: Vast.ai, RunPod Community, random hosts
- Teaches you whether the model even works before Hopper tax
Limits
- 24 GB is a wall, not a suggestion
- Consumer reliability
- 450 W cooling lottery
- People train 70Bs here and then blog about “CUDA OOM meaning NVIDIA is bad”
Start here more often
The site’s whole personality is: do not rent Hopper because a thread told you to. A 4090 hour is how you find out the batch size, the tokenizer bug, and whether you even like the loss curve. Then you scale.
Two 4090s can beat a lonely H100 on small adapters. That is not a dunk on Hopper. That is VRAM and price doing math.
How I actually rent it
- 01
Fit 24 GB honestly
QLoRA is a strategy, not a moral failure.
- 02
Pick a host you can lose
Community/marketplace. Checkpoint.
- 03
Graduate on purpose
5090 for 32 GB, L40S for ECC, H100 for HBM.
FAQ
RTX 4090 questions
Who has the cheapest RTX 4090?+
As of September 5, 2026, the cheapest verified in-stock on-demand RTX 4090 tracked by GPUBeacon is $0.29 per GPU per hour from Vast.ai.
How much does the RTX 4090 cost per month?+
As of September 5, 2026, at 720 hours per month, one RTX 4090 costs an estimated $299 at the median on-demand price in this sample. Cheapest verified in stock: $209 per month on-demand.
Where can I rent an RTX 4090?+
GPUBeacon currently tracks RTX 4090 configs from 4 providers. The cheapest verified in-stock on-demand configs come from Vast.ai, RunPod, Fluidstack and Hyperstack. See the full price comparison above for every provider and config.
Are RTX 4090 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 4090 in this sample is $0.42 per GPU per hour. The cheapest verified in-stock hour is $0.29.
Why choose the RTX 4090?+
LoRA, SD, small full fine-tunes, iteration, anyone who can lose a job.
When is the RTX 4090 not a good fit?+
70B full fine-tunes, compliance, multi-GPU NVLink, uncheckpointable trainers.
Can I fine-tune Llama 70B on a 4090?+
Adapters, sometimes. Full fine-tune with optimizer states, no. Rent memory or more GPUs.
4090 or 5090?+
5090 if you need 32 GB or Blackwell. 4090 if 24 GB fits and the hour is cheaper. Do not pay a generation tax for a 7B LoRA.
Keep comparing
Alternatives to NVIDIA RTX 4090
Same job, different envelope. Compare VRAM, bandwidth and the hour before you lock a badge.
NVIDIA RTX 5090
32 GB GDDR7 · 78% more bandwidth than the RTX 4090
From$0.34/ GPU / hr
Read the verdictNVIDIA L40S
48 GB GDDR6 · 14% less bandwidth than the RTX 4090
From$0.79/ GPU / hr
CompareNVIDIA A10
24 GB GDDR6 · 40% less bandwidth than the RTX 4090
From$0.45/ GPU / hr
CompareNVIDIA H100
80 GB HBM3 · 3.3x the bandwidth of the RTX 4090
From$1.73/ GPU / hr
CompareSources
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.