deploybase

Sesterce GPU Cloud Pricing: Complete Guide ($/hr for Every GPU)

Deploybase · November 18, 2025 · GPU Pricing

Contents

Sesterce Overview

Sesterce is cheap GPU cloud. 50k+ active GPUs across NA and Europe, Asia coming 2026. Fixed hourly pricing, no auction nonsense. Startups and researchers use it.

Sesterce Pricing Tiers

Tier 1: Entry-Level Inference

  • L4 GPU: $0.35/hour
  • T4 GPU: $0.25/hour
  • A10 GPU: $0.65/hour
  • Use cases: Model experimentation, small-batch inference
  • Monthly cost (24/7): $255-475

Tier 2: Production Inference

  • RTX 4000 GPU: $0.48/hour
  • L40 GPU: $0.62/hour
  • L40S GPU: $0.71/hour
  • A100 PCIe GPU: $1.05/hour
  • Use cases: Production endpoints, sustained inference
  • Monthly cost (24/7): $450-765

Tier 3: Training and Complex Tasks

  • H100 PCIe GPU: $2.09/hour
  • H100 SXM GPU: $2.35/hour
  • H200 GPU: $3.10/hour
  • Use cases: Model training, fine-tuning, complex computations
  • Monthly cost (24/7): $1,275-2,263

Tier 4: Production Clusters

  • 8xH100 bundle: $13.60/hour (per bundle)
  • 8xH200 bundle: $24.80/hour (per bundle)
  • 8xL40S bundle: $5.68/hour (per bundle)
  • Use cases: Large-scale training, production clusters
  • Monthly cost (24/7): $9,920-18,084

Competitive with Lambda and Vast AI individually. Bundles beat manual rentals, lose to CoreWeave.

Pricing by GPU Model

RTX 3090 Series

  • RTX 3090: $0.18/hour
  • RTX 3090 Ti: $0.22/hour
  • Suitable for: Development, gaming workloads
  • Availability: High (common inventory)

RTX 40 Series

  • RTX 4090: $0.30/hour
  • A10G (AWS equivalent): $0.58/hour
  • RTX 4000 series: $0.48/hour
  • Suitable for: Inference, lightweight training
  • Availability: Good

Data Center Cards (A-Series)

  • A10 GPU: $0.65/hour
  • A40 GPU: $0.95/hour
  • A100 40GB: $0.98/hour
  • A100 80GB: $1.05/hour
  • Suitable for: Mixed training and inference
  • Availability: Moderate (lower demand than H100)

Data Center Cards (H-Series)

  • H100 PCIe (80GB): $2.09/hour
  • H100 SXM (80GB): $2.35/hour
  • H200 (141GB): $3.10/hour
  • Suitable for: Large model training, production deployments
  • Availability: Lower (high demand period)

Data Center Cards (L-Series)

  • L4 GPU: $0.35/hour
  • L40 GPU: $0.62/hour
  • L40S GPU: $0.71/hour
  • GH200 GPU: $5.20/hour
  • Suitable for: Inference specialization, CUDA-compute balance
  • Availability: Moderate

Spot Pricing (Preemptible Capacity) Sesterce offers preemptible instances at 30-50% discounts:

  • H100 PCIe spot: $1.05-1.25/hour (40-50% discount)
  • A100 spot: $0.53-0.63/hour (40% discount)
  • L40S spot: $0.36-0.43/hour (40% discount)
  • Caveat: Interruptions possible with 24-hour notice

Comparison with Competitors

Single A100 Pricing:

  • Sesterce: $1.05/hour
  • Lambda: $1.48/hour
  • Vast AI: $0.85-1.50/hour (marketplace dependent)
  • JarvisLabs: $0.95/hour
  • Winner: Vast AI at low end, Sesterce comparable to JarvisLabs

Single H100 PCIe Pricing:

  • Sesterce: $2.09/hour
  • Lambda: $2.86/hour
  • Vast AI: $1.80-2.50/hour (marketplace dependent)
  • JarvisLabs: $1.85/hour
  • Winner: Sesterce offers competitive fixed pricing

8xH100 Cluster Pricing:

  • Sesterce: $13.60/hour = $1.70 per GPU
  • Lambda: No bundles available
  • Vast AI: Varies by provider (typically $1.90-2.20 per GPU)
  • CoreWeave: $49.24/hour = $6.16 per GPU
  • Winner: Sesterce significantly cheaper for bundled deployments

L40S Pricing:

  • Sesterce: $0.71/hour
  • RunPod: $0.79/hour
  • Lambda: $0.92/hour
  • Vast AI: $0.70-0.95/hour (marketplace dependent)
  • Winner: Sesterce and Vast AI comparable

Strength: H-series fixed pricing beats marketplace volatility. Weakness: A-series, Vast AI sometimes cheaper.

Cost Optimization Strategies

Use L40S ($0.71/hr). It handles 40-60% of inference workloads. Skip H100 unless developers need it. Cuts costs 60-75%.

Spot instances: 40-50% off. Batch jobs with checkpoints survive interruptions. Save $5k+/month on clusters.

Batch inference at night. Use spot for stored data, daytime requests on regular instances. Utilization jumps 40-60%.

Quantize models aggressively 8-bit quantization fits 80GB A100 models on 40GB A100 hardware. Costs drop 50% while accuracy loss stays below 2%. 4-bit quantization enables H100 models on L40S GPUs, providing 60% cost reduction.

Use distributed inference Split large models across multiple smaller GPUs. A 70B model on single H100 costs $2.09/hour; distributed across 8x L40S GPUs costs $5.68/hour but adds fault tolerance and scales throughput 8x.

Negotiate reserved capacity Sesterce offers 15-30% discounts for 3-6 month commitments. Only commit for proven workloads with stable forecasts. Savings reach $20,000-50,000 annually on steady-state infrastructure.

Monitor utilization metrics Idle GPU time represents wasted budget. Implement monitoring to catch idle instances. Target 70%+ GPU utilization; below 50% indicates engineering problems (queuing, network bottlenecks) worth investigating.

FAQ

Is Sesterce cheaper than Lambda?

Sesterce undercuts Lambda by 15-40% depending on GPU type. Lambda pricing ($1.48 for A100) exceeds Sesterce ($1.05 for A100 80GB). This advantage narrows for H-series (Sesterce $2.09 vs Lambda $2.86 for H100), where Sesterce's discount reaches ~27%. Choose Sesterce for pure cost; Lambda if integration and support matter more.

How does Sesterce compare to Vast AI?

Vast AI's marketplace model creates pricing competition and volatility. Sesterce's fixed pricing eliminates uncertainty but sacrifices potential savings from marketplace deals. I'd use Vast AI for experimental workloads tolerating price volatility, Sesterce for production systems valuing cost predictability.

What about JarvisLabs pricing?

JarvisLabs matches Sesterce on most GPUs ($0.95 for A100, $1.85 for H100). JarvisLabs adds integrated Jupyter notebooks and easier setup for researchers. Sesterce serves teams comfortable with command-line infrastructure. For ease of use, JarvisLabs; for pure cost, Sesterce.

Are there hidden fees?

Sesterce's pricing includes GPU compute only. Storage, data transfer, and support cost extra. Disk storage typically runs $0.10-0.30/GB/month. Data transfer to internet adds $0.05-0.10/GB. Budget 10-20% overhead beyond GPU costs for real deployments.

What's Sesterce's uptime guarantee?

Sesterce guarantees 99.0% uptime for standard instances. This falls below hyperscaler standards (99.9%+) but matches budget-focused providers. For production critical systems, I'd implement multi-provider failover rather than relying on single-provider SLA.

Should I use Sesterce for long-term projects?

Sesterce works well for sustained training projects (2-3 months duration). Lock in reserved capacity pricing and commit with confidence. For projects with uncertain duration or changing requirements, month-to-month on-demand pricing provides flexibility at 15-30% premium.

Sources