Skip to content
Updated minutes ago
nvidia

RTX 5090

nvidia · blackwell · 32 GB GDDR6X · 575W TDP

VRAM

32 GB

BF16 TFLOPS

210

Bandwidth

1792 GB/s

From

$0.89/hr

Calculate ROI with this GPU →

Spec Sheet

VRAM32 GB GDDR6X
Memory Bandwidth1792 GB/s
BF16 TFLOPS210
FP16 TFLOPS210
FP8 TFLOPS420
INT8 TOPS420
TDP575W
InterconnectPCIE
Max per Node4
PCIe Gen5
CUDA Compute Capability8.9
Tensor CoresYes

Pricing by Provider

ProviderOn-DemandReservedSpotBadge
vast_ai$1.29/hr-$0.89/hrCheapest
tensordock$1.19/hr-$0.89/hr
runpod$1.69/hr-$1.29/hr

Compatible Models (246)

Training Capabilities

Estimated GPU count for full fine-tuning (AdamW, BF16) and QLoRA

Model SizeFull Fine-TuneQLoRA
7B model5 GPUs1 GPU
13B model8 GPUs1 GPU
70B model42 GPUs2 GPUs

Energy Efficiency

Estimated tokens/second per Watt for popular models

Mistral 7B
0.43 t/s/WFP8
Qwen 2.5 7B
0.41 t/s/WFP8
Llama 3.1 8B
0.39 t/s/WFP8
Llama 3.1 70B
0.04 t/s/WFP8
Qwen 2.5 72B
0.04 t/s/WFP8

Similar GPUs

GPUVRAMBF16 TFLOPSBW (GB/s)From
B200 SXM180 GB22508000$4.49/hr
B100 SXM192 GB17508000$4.50/hr
GB200 NVL72 (per GPU)192 GB22508000$6.50/hr
GB300 NVL72 (per GPU)192 GB25008000$7.50/hr
B300288 GB280012000$0.00/hr

Methodology Note

Performance estimates for the RTX 5090are based on InferenceBench's roofline performance model with CUDA kernel-level optimization including FlashAttention v2 and PagedAttention. Memory calculations account for model weights (32 GB GDDR6X available), KV-cache allocation, and activation memory. Throughput predictions use the RTX 5090's rated 1792 GB/s memory bandwidth and 210 BF16 TFLOPS compute capacity as roofline ceilings, with empirical correction factors per GPU architecture (blackwell). See our full methodology.

Frequently Asked Questions

How many AI models can run on RTX 5090?

The RTX 5090 can run 246 AI models from our database within a single node. Compatible models range across various parameter sizes depending on the quantization precision (BF16, FP8, INT4). Smaller models fit on a single GPU while larger models may require multi-GPU setups up to 4x RTX 5090.

What is the RTX 5090 inference throughput?

The RTX 5090 delivers 210 BF16 TFLOPS and 420 FP8 TFLOPS with 1792 GB/s memory bandwidth. Actual inference throughput (tokens/sec) depends on the model size, precision, and batch size. Use our calculator for model-specific throughput estimates.

How much does RTX 5090 cost per hour?

The RTX 5090 is available starting from $0.89/hour via vast_ai. Prices vary by provider and pricing tier (on-demand, reserved, spot). Compare pricing across all providers in the table above.