NVIDIA RTX 3090 Cloud Pricing & Specs (2026)
The RTX 3090 pairs 24 GB of GDDR6X with 0.94 TB/s bandwidth at a 350W envelope. It remains a common second-hand and colo card for running 7B-13B models at INT4 and for QLoRA fine-tuning where datacenter GPUs are unnecessary.
Target workload: Local 7B-13B INT4 inference & QLoRA fine-tuning
24 GB VRAM
Spec fields mirrored from the verified telemetry row; board attributes per NVIDIA RTX 3090 specifications. No rate rows tracked.
0.94 TB/s
Peak memory bandwidth from the manufacturer specification.
NVIDIA RTX 3090: Key Numbers at a Glance
Rental cost: no verified rate rows are tracked yet — this page asserts specifications only, never a price.
70B fit: Does not fit Llama 3.3 70B single-GPU — INT4 alone needs 40 GB vs 24 GB available. 7B-13B class is the single-GPU target.
Bottleneck: Memory-bandwidth bound — 0.94 TB/s GDDR6X limits real-time generation throughput
OpenGPU Radar does not currently track this GPU rows in data/providers.json. No price is asserted on this page — the specifications below are manufacturer-sourced.
Compare GPUs with tracked rates →Compatible Models for NVIDIA RTX 3090
Models from the VRAM registry whose minimum INT4 footprint (weights + KV-cache + runtime overhead) fits 24 GB. FP16 shows where full precision also fits single-GPU.
VRAM breakdown →
VRAM breakdown →
VRAM breakdown →
VRAM breakdown →
VRAM breakdown →
Specifications
The RTX 3090 is an Ampere consumer card without FP8 support and without NVLink P2P: multi-GPU workloads fall back to PCIe round-trips (2-way training yields well under 2x speedup). 24 GB fits 13B INT4 (≈7 GB) and 30B INT4 (≈16 GB) with KV-cache headroom; Llama 70B (≈38 GB INT4) does not fit at any practical quantization. At 0.94 TB/s, token generation is bandwidth-bound — larger batches increase throughput linearly until VRAM pressure, not compute, becomes the limit. No rate rows are tracked for the 3090 in OpenGPU Radar's provider table.
Next steps