Hugging Face's logo

Datasets:
600me
/
gpu-cloud-h100-a100-pricing-benchmark

Tasks:
Tabular Classification
Modalities:
Tabular
Text
Formats:
json
Languages:
English
Size:
< 1K
Tags:
gpu
cloud-pricing
h100
a100
l40s
rtx-4090
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
Dataset Viewer
Auto-converted to Parquet Duplicate
provider
stringclasses
7 values
gpu_model
stringlengths
4
24
vram_gb
int64
24
80
tier
stringclasses
6 values
usd_per_gpu_hour
float64
0.34
12.3
usd_per_100_gpu_hours
float64
34
1.23k
bundle_min_gpus
int64
1
8
nvme_usd_per_gb_month
float64
0
0.1
cold_start_minutes
float64
0.4
2
egress_usd_per_gb
float64
0
0.12
storage_usd_per_hour_150gb
float64
0
0.02
cold_start_usd_per_hour
float64
0
0.01
effective_usd_per_useful_hour
float64
0.43
14.5
source_url
stringclasses
7 values
observed_at
timestamp[s]date
2026-10-19 00:00:00
2026-10-19 00:00:00
RunPod
H100 SXM
80
Community
2.69
269
1
0.1
0.6
0
0.0205
0.0034
3.193
https://api.runpod.io/graphql
2026-10-19T00:00:00
RunPod
H100 SXM
80
Secure
3.99
399
1
0.1
0.6
0
0.0205
0.005
4.724
https://api.runpod.io/graphql
2026-10-19T00:00:00
RunPod
H100 PCIe
80
Community
1.99
199
1
0.1
0.7
0
0.0205
0.0029
2.369
https://api.runpod.io/graphql
2026-10-19T00:00:00
RunPod
H100 PCIe
80
Secure
2.89
289
1
0.1
0.7
0
0.0205
0.0042
3.429
https://api.runpod.io/graphql
2026-10-19T00:00:00
RunPod
A100 80G SXM
80
Community
1.39
139
1
0.1
0.8
0
0.0205
0.0023
1.662
https://api.runpod.io/graphql
2026-10-19T00:00:00
RunPod
A100 80G SXM
80
Secure
1.79
179
1
0.1
0.8
0
0.0205
0.003
2.134
https://api.runpod.io/graphql
2026-10-19T00:00:00
RunPod
A100 80G PCIe
80
Community
1.19
119
1
0.1
0.9
0
0.0205
0.0022
1.427
https://api.runpod.io/graphql
2026-10-19T00:00:00
RunPod
A100 80G PCIe
80
Secure
1.79
179
1
0.1
0.9
0
0.0205
0.0034
2.134
https://api.runpod.io/graphql
2026-10-19T00:00:00
RunPod
L40S
48
Community
0.99
99
1
0.1
0.8
0
0.0205
0.0016
1.191
https://api.runpod.io/graphql
2026-10-19T00:00:00
RunPod
RTX 4090
24
Community
0.34
34
1
0.1
1.2
0
0.0205
0.0009
0.425
https://api.runpod.io/graphql
2026-10-19T00:00:00
RunPod
RTX 4090
24
Secure
0.89
89
1
0.1
1.2
0
0.0205
0.0022
1.074
https://api.runpod.io/graphql
2026-10-19T00:00:00
Vast.ai
H100 SXM
80
Verified on-demand
1.788
178.8
2
0.06
1.5
0
0.0123
0.0056
2.125
https://console.vast.ai/api/v0/bundles/
2026-10-19T00:00:00
Vast.ai
H100 PCIe
80
Verified on-demand
2.001
200.1
1
0.06
1.6
0
0.0123
0.0067
2.376
https://console.vast.ai/api/v0/bundles/
2026-10-19T00:00:00
Vast.ai
A100 80G SXM
80
Verified on-demand
0.989
98.9
1
0.06
1.8
0
0.0123
0.0037
1.182
https://console.vast.ai/api/v0/bundles/
2026-10-19T00:00:00
Vast.ai
A100 80G PCIe
80
Verified on-demand
0.802
80.2
1
0.06
1.9
0
0.0123
0.0032
0.962
https://console.vast.ai/api/v0/bundles/
2026-10-19T00:00:00
Vast.ai
RTX 4090
24
Verified on-demand
0.456
45.6
1
0.06
2
0
0.0123
0.0019
0.553
https://console.vast.ai/api/v0/bundles/
2026-10-19T00:00:00
Lambda
H100 SXM
80
On-demand
2.99
299
1
0
2
0
0
0.0125
3.532
https://lambda.ai/pricing
2026-10-19T00:00:00
Lambda
A100 40G SXM
40
On-demand
1.29
129
1
0
2
0
0
0.0054
1.524
https://lambda.ai/pricing
2026-10-19T00:00:00
Lambda
L40S
48
On-demand
0.86
86
1
0
2
0
0
0.0036
1.016
https://lambda.ai/pricing
2026-10-19T00:00:00
CoreWeave
H100 SXM
80
On-demand
4.76
476
1
0.05
1
0
0.0103
0.0099
5.624
https://www.coreweave.com/pricing
2026-10-19T00:00:00
CoreWeave
L40S
48
On-demand
1.99
199
1
0.05
1
0
0.0103
0.0041
2.358
https://www.coreweave.com/pricing
2026-10-19T00:00:00
AWS
H100 SXM (p5.48xlarge)
80
On-demand, 8-GPU node
12.29
1,229
8
0.08
0.4
0.09
0.0164
0.0102
14.49
https://aws.amazon.com/ec2/pricing/on-demand/
2026-10-19T00:00:00
AWS
A100 40G (p4d.24xlarge)
40
On-demand, 8-GPU node
4.1
410
8
0.08
0.4
0.09
0.0164
0.0034
4.847
https://aws.amazon.com/ec2/pricing/on-demand/
2026-10-19T00:00:00
AWS
L40S (g6e.xlarge)
48
On-demand, 1-GPU node
1.861
186.1
1
0.08
0.4
0.09
0.0164
0.0016
2.211
https://aws.amazon.com/ec2/pricing/on-demand/
2026-10-19T00:00:00
GCP
H100 SXM (a3-highgpu-8g)
80
On-demand, 8-GPU node
11.06
1,106
8
0.09
0.4
0.12
0.0185
0.0092
13.044
https://cloud.google.com/compute/gpus-pricing
2026-10-19T00:00:00
GCP
A100 40G (a2-highgpu-1g)
40
On-demand, 1-GPU node
3.67
367
1
0.09
0.4
0.12
0.0185
0.0031
4.343
https://cloud.google.com/compute/gpus-pricing
2026-10-19T00:00:00
Azure
H100 SXM (ND96isr v5)
80
On-demand, 8-GPU node
12.29
1,229
8
0.08
0.4
0.09
0.0164
0.0102
14.49
https://azure.microsoft.com/pricing/details/virtual-machines/linux/
2026-10-19T00:00:00
Azure
A100 80G (NC24ads v4)
80
On-demand, 1-GPU node
3.67
367
1
0.08
0.4
0.09
0.0164
0.0031
4.341
https://azure.microsoft.com/pricing/details/virtual-machines/linux/
2026-10-19T00:00:00

GPU Cloud H100/A100/L40S/RTX 4090 Pricing & Effective-Cost Benchmark (2026)

Catalogue $/GPU-hour is not what you pay. This dataset records the advertised on-demand rate and the effective cost of one useful GPU-hour after 150 GB of model storage, cold-start pre-pull, and a realistic 85% utilisation ceiling.

Files

File Rows Description
data/gpu_cloud_pricing_2026.csv 28 Wide table, one row per provider/GPU/tier
data/gpu_cloud_pricing_2026.jsonl 28 Same records as newline-delimited JSON

Schema

Column Type Meaning
provider string Vendor (RunPod, Vast.ai, Lambda, CoreWeave, AWS, GCP, Azure)
gpu_model string GPU SKU as sold
vram_gb int VRAM per GPU
tier string Community / Secure / on-demand / node size
usd_per_gpu_hour float Published on-demand list rate
usd_per_100_gpu_hours float Straight-line 100-hour cost
bundle_min_gpus int Minimum billable GPUs (bundle rounding)
nvme_usd_per_gb_month float Attached NVMe list rate
cold_start_minutes float Image pull + model load before first token
egress_usd_per_gb float Egress list rate where charged
storage_usd_per_hour_150gb float 150 GB NVMe amortised per hour
cold_start_usd_per_hour float Cold start amortised over an 8-hour session
effective_usd_per_useful_hour float (rate + storage + cold start) / 0.85 utilisation

Method (fully reproducible) 

storage_hour  = nvme_usd_per_gb_month * 150 / 730
cold_hour     = (cold_start_minutes / 60) * usd_per_gpu_hour / 8
effective     = (usd_per_gpu_hour + storage_hour + cold_hour) / 0.85

Utilisation 0.85 is an explicit planning assumption, not a measurement. Change it and the ranking between a cheap marketplace bundle and a fixed on-demand node can flip.

Headline result

Cheapest effective useful GPU-hour in this snapshot: RunPod RTX 4090 (Community) at $0.42/h, versus $0.34/h advertised — a 25% gap between the invoice line and the cost of useful compute.

Summary table

Provider GPU Tier $/GPU-h $/100 GPU-h Effective $/useful h
RunPod H100 SXM Community $2.69 $269.00 $3.19
RunPod H100 SXM Secure $3.99 $399.00 $4.72
RunPod H100 PCIe Community $1.99 $199.00 $2.37
RunPod H100 PCIe Secure $2.89 $289.00 $3.43
RunPod A100 80G SXM Community $1.39 $139.00 $1.66
RunPod A100 80G SXM Secure $1.79 $179.00 $2.13
RunPod A100 80G PCIe Community $1.19 $119.00 $1.43
RunPod A100 80G PCIe Secure $1.79 $179.00 $2.13
RunPod L40S Community $0.99 $99.00 $1.19
RunPod RTX 4090 Community $0.34 $34.00 $0.42
RunPod RTX 4090 Secure $0.89 $89.00 $1.07
Vast.ai H100 SXM Verified on-demand $1.79 $178.80 $2.12
Vast.ai H100 PCIe Verified on-demand $2.00 $200.10 $2.38
Vast.ai A100 80G SXM Verified on-demand $0.99 $98.90 $1.18
Vast.ai A100 80G PCIe Verified on-demand $0.80 $80.20 $0.96
Vast.ai RTX 4090 Verified on-demand $0.46 $45.60 $0.55
Lambda H100 SXM On-demand $2.99 $299.00 $3.53
Lambda A100 40G SXM On-demand $1.29 $129.00 $1.52
Lambda L40S On-demand $0.86 $86.00 $1.02
CoreWeave H100 SXM On-demand $4.76 $476.00 $5.62
CoreWeave L40S On-demand $1.99 $199.00 $2.36
AWS H100 SXM (p5.48xlarge) On-demand, 8-GPU node $12.29 $1229.00 $14.49
AWS A100 40G (p4d.24xlarge) On-demand, 8-GPU node $4.10 $410.00 $4.85
AWS L40S (g6e.xlarge) On-demand, 1-GPU node $1.86 $186.10 $2.21
GCP H100 SXM (a3-highgpu-8g) On-demand, 8-GPU node $11.06 $1106.00 $13.04
GCP A100 40G (a2-highgpu-1g) On-demand, 1-GPU node $3.67 $367.00 $4.34
Azure H100 SXM (ND96isr v5) On-demand, 8-GPU node $12.29 $1229.00 $14.49
Azure A100 80G (NC24ads v4) On-demand, 1-GPU node $3.67 $367.00 $4.34

Use the calculators and mirror

Deep dives

The 4-calculator / 4-Space / 4-Dataset matrix

Online calculators

HuggingFace Static Spaces

HuggingFace Datasets

Transparency & FTC affiliate disclosure

This dataset is published by the independent cost-research project at mybskm.top. Some outbound vendor links are affiliate links: if you open one and later buy, we may earn a commission at no extra cost to you. Commissions never change the measured numbers, the formula, or which vendor wins a scenario, and the free open-source baseline is included on purpose because it pays us nothing. Every price is a vendor-published list rate captured on the observed_at date; prices, trials, KYC rules and refund terms change without notice. Nothing here is a quote, a guarantee of eligibility, tax or legal advice, or a promise of results. Verify current pricing and terms with the vendor before purchasing.
Downloads last month
24
Number of rows:
28
Total file size:
26.3 kB