RTX 5090 vs H100

BlackwellvsHopperUpdated yesterday

The H100 wins for the most common use case of LLM inference. Its 80 to 94 GB VRAM capacity combined with 989 TFLOPS FP16 dense performance and 3350 GB per second memory bandwidth enables larger models and higher throughput than the RTX 5090 can achieve with 32 GB VRAM and 209.5 TFLOPS FP16 dense performance.

RTX 5090 from $0.53/GPU/hrH100 from $2.27/GPU/hr

Right now, from live stock

  • Cheapest right now: RTX 5090 at $0.53/hr on Vast.ai

    Deploy
  • Most providers in stock: H100 (11)

    See all offers
  • Best $ per TFLOPS FP16 dense right now: H100 ($0.0023 per TFLOPS-hour at 989 TFLOPS)

    Deploy

Specifications Compared

SpecRTX 5090H100
TDP575W700W
VRAM32 GB80-94 GB
CUDA Cores21,76016,896
FP4 (dense)1,676 TFLOPSNot published
FP8 (dense)419 TFLOPS1,979 TFLOPS
Memory TypeGDDR7HBM3
ArchitectureBlackwellHopper
FP16 (dense)209.5 TFLOPS989 TFLOPS
Form FactorsPCIeSXM5, PCIe, NVL
INT8 (dense)838 TOPS1,979 TOPS
InterconnectPCIe 5.0NVLink, PCIe 5.0, InfiniBand
Tensor Cores680528
FP32 Performance104.8 TFLOPS67 TFLOPS
FP64 Performance1.6 TFLOPS34 TFLOPS
Memory Bandwidth1,792 GB/s3,350 GB/s
FP4 (with sparsity)3,352 TFLOPSNot published
FP8 (with sparsity)838 TFLOPS3,958 TFLOPS
FP16 (with sparsity)419 TFLOPS1,979 TFLOPS
INT8 (with sparsity)1,676 TOPS3,958 TOPS

Performance Analysis

The H100 delivers 989 TFLOPS FP16 dense performance compared with 209.5 TFLOPS FP16 dense performance on the RTX 5090. This ratio means the H100 processes FP16 dense workloads in training and inference at roughly 4.7 times the rate of the RTX 5090. The H100 also delivers 1979 TFLOPS FP16 with sparsity compared with 419 TFLOPS FP16 with sparsity on the RTX 5090. Memory bandwidth of 3350 GB per second on the H100 versus 1792 GB per second on the RTX 5090 permits larger batch sizes during both training and inference. The RTX 5090 delivers 104.8 TFLOPS FP32 performance against 67 TFLOPS FP32 performance on the H100 so FP32 workloads favor the RTX 5090.

Current On-Demand Offers

Cheapest secure on-demand offer per provider that is reported in stock right now, per GPU per hour; multi-GPU instances show the whole-instance price alongside. Deploy opens the DeployGPU console for providers on the platform, otherwise the provider's own site. Stock is rechecked every minute.

RTX 5090

ProviderRegionGPUsPer GPU / hrInstance / hrDeploy
Vast.aiSouth Korea, KR1$0.53Deploy
RunPodglobal1$0.99Deploy
LeaderGPUThe Netherlands1$2.29Deploy

3 providers in stock, 7 offers (cheapest per provider shown). All RTX 5090 offers, price history and alerts

H100

ProviderRegionGPUsPer GPU / hrInstance / hrDeploy
Vast.aiCzechia, CZ1$2.27Deploy
QuantaCloudus-midwest-21$2.59Deploy
Massed Computeus-central-31$2.73Deploy
LyceumEurope2$2.79$5.58Deploy
Oridallas-24$2.90$11.60Deploy

11 providers in stock, 32 offers (cheapest per provider shown). All H100 offers, price history and alerts

Which GPU to watchWatch the price of

Notify me when RTX 5090 drops below a price

One email when the cheapest in-stock on-demand price per GPU-hour falls below your threshold. Cheapest right now: $0.53/GPU-hr.

QuantaCloud

Comparing H-series providers? We broker across all of them.

Hopper stock changes by the hour and prices differ by provider. If you need 16+ GPUs reserved or a cluster in the next 90 days, we quote H-series or B300 inventory at partner rates: one quote, 24h turnaround.

No waitlist24hr quote turnaroundInfiniBand fabric

Compare real-time pricing across 25+ providers

When to Choose the RTX 5090

The RTX 5090 suits Stable Diffusion workloads because its 104.8 TFLOPS FP32 performance exceeds the 67 TFLOPS FP32 performance of the H100. The RTX 5090 also suits environments where TDP must remain at 575 watts rather than 700 watts. Its PCIe form factor further supports single GPU consumer deployments.

When to Choose the H100

The H100 suits LLM training because its 80 to 94 GB VRAM capacity exceeds the 32 GB VRAM capacity of the RTX 5090. The H100 also suits large scale inference because its 3350 GB per second memory bandwidth exceeds the 1792 GB per second memory bandwidth of the RTX 5090. Its 989 TFLOPS FP16 dense performance supports higher throughput on dense low precision workloads.

Use Cases

LLM Training
H100

The H100 provides 80 to 94 GB VRAM and 989 TFLOPS FP16 dense performance while the RTX 5090 provides only 32 GB VRAM.

LLM Inference
H100

The H100 provides 3350 GB per second memory bandwidth and 1979 TFLOPS FP16 with sparsity while the RTX 5090 provides 1792 GB per second memory bandwidth and 419 TFLOPS FP16 with sparsity.

Fine-tuning
H100

The H100 provides 80 to 94 GB VRAM and 1979 TOPS INT8 dense performance while the RTX 5090 provides 32 GB VRAM and 838 TOPS INT8 dense performance.

Stable Diffusion
RTX 5090

The RTX 5090 provides 104.8 TFLOPS FP32 performance while the H100 provides 67 TFLOPS FP32 performance.

Scientific Computing
Either

The RTX 5090 provides higher FP32 performance at lower TDP while the H100 provides higher memory bandwidth and larger VRAM capacity.

Frequently Asked Questions

What is the FP16 dense performance difference?

The RTX 5090 delivers 209.5 TFLOPS FP16 dense. The H100 delivers 989 TFLOPS FP16 dense.

Which GPU has higher memory bandwidth?

The H100 has 3350 GB per second memory bandwidth. The RTX 5090 has 1792 GB per second memory bandwidth.

What TDP does each GPU list?

The RTX 5090 lists 575 watts TDP. The H100 lists 700 watts TDP.

Does the RTX 5090 support NVLink?

The RTX 5090 uses PCIe 5.0 interconnect only. The H100 supports NVLink in addition to PCIe 5.0 and InfiniBand.

Which is cheaper to rent, the RTX 5090 or the H100?

Cloud rental prices for both the RTX 5090 and H100 vary by provider, configuration, and availability. This page shows live pricing from 25+ providers updated every 60 seconds. Scroll to the Live Cloud Pricing section to compare current rates.

How much VRAM does the RTX 5090 have compared to the H100?

The RTX 5090 has 32 GB of GDDR7 memory. The H100 has 80 to 94 GB of HBM3 memory.

Can I find RTX 5090 and H100 GPUs available to rent right now?

Yes. This page shows real-time availability across 25+ cloud GPU providers. The Live Cloud Pricing section displays only in-stock offers with current pricing.

What is the main difference between the RTX 5090 and the H100?

The RTX 5090 uses the Blackwell architecture (2025) while the H100 uses Hopper (2022). The H100 delivers 4.7x the dense FP16 throughput (989 vs 209.5 TFLOPS, both without sparsity) and 1.9x the memory bandwidth of the RTX 5090.

Rent these GPUs

Each GPU page lists every current on-demand offer by provider, per GPU-hour, updated every minute.

Related comparisons

How this page is made

  • Specifications come from the NVIDIA, AMD and Intel datasheets for the RTX 5090 and the H100. Dense and sparse throughput are listed separately.
  • Prices and availability are live: every offer shown is an in-stock, on-demand listing from a provider we track, rechecked every minute.
  • The written comparison (overview, performance notes, when to choose each card, verdict, use cases and FAQ) was drafted with an AI model from the specification table above and passed an automated check that rejects any figure not in that table. It was last generated on . It contains no prices; those are always read live.
  • Read how we collect the data, or report an error on this page.

Next steps