NVIDIA · Ada Lovelace · Consumer
NVIDIA GeForce RTX 4090
NVIDIA's Ada Lovelace consumer flagship, common on GPU marketplaces for image generation, small-model inference, and experiments.
01 / SPECIFICATIONS
What the hardware holds
- Memory
- 24 GB
- Memory bandwidth
- 1.01 TB/s
- FP8 dense
- 330.3 TFLOPS
- FP16 / BF16 dense
- 165.2 TFLOPS
- Variant
- GeForce RTX 4090
- Memory type
- GDDR6X
- Form factor
- PCIe consumer graphics card
- Interconnect
- PCIe Gen4
- Max power
- 450 W
Source notes for GeForce RTX 4090
From NVIDIA Ada GPU Architecture whitepaper v2.02, Appendix A Table 2. Tensor values given as dense/sparse: FP16 Tensor with FP32 accumulate 165.2/330.4 (recorded), FP16 Tensor with FP16 accumulate 330.3/660.6; BF16 with FP32 accumulate 165.2/330.4; FP8 Tensor with FP32 accumulate 330.3/660.6 (recorded), FP8 with FP16 accumulate 660.6/1321.2. Bandwidth 1008 GB/sec, 384-bit, 21 Gbps. TGP 450 W. PCI Express interface Gen 4. No NVLink.
Compute figures are dense, without sparsity. NVIDIA source for GeForce RTX 4090 Checked 8 Nov 2026.
02 / MEMORY FIT
How many do you need to hold a model?
| Model size | FP16 / BF16 | FP8 | INT4 |
|---|---|---|---|
| 8B parameters | 1 GPU16 GB of weights | 1 GPU8 GB of weights | 1 GPU4 GB of weights |
| 70B parameters | 7 GPUs140 GB of weights | 4 GPUs70 GB of weights | 2 GPUs35 GB of weights |
| 405B parameters | 41 GPUs810 GB of weights | 21 GPUs405 GB of weights | 11 GPUs203 GB of weights |
GPUs = parameters × bytes per parameter × 1.2 headroom ÷ 24 GB, rounded up. This covers weights only. Long contexts, large batches, and training need much more memory.
03 / OFFERS
Reviewed on-demand rates
No reviewed rate yet. The sellers below list the RTX 4090; check their pricing pages for current terms.
Per GPU-hour divides a machine's hourly price by its GPU count, so single GPUs and 8-GPU nodes compare fairly. The billed price is what one machine costs per hour. Storage, network transfer, and minimum terms can add to the bill.
04 / SELLERS
Who lists the RTX 4090
Runpod, Vast.ai list the RTX 4090 without a rate we have reviewed. Ask them for a quote, especially for clusters of 8 or more GPUs.
05 / COMPARE WITH