Cloud GPU prices, normalized by VRAM

Compare hourly rental rates across cloud providers per gigabyte of memory and per FP8 TFLOP. Calculate open-weight model memory footprints and evaluate self-hosting break-even economics.

Snapshot · 43 configurations across 16 providers · updated 2026-10-31

Provider / GPU	Architecture	VRAM	$/GB-hr	$/hr	Spot	FP8 TF/$	730h/mo
HyperstackSponsored NVIDIA GeForce RTX 5090	Blackwell · PCIe Gen5	32 GB	$0.0453	$1.45	$0.95	889.7	$1,058.5	Rent
TensorWave AMD Instinct MI300X	CDNA 3 · Infinity Fabric 896GB/s	192 GB	$0.0089	$1.71	$1.25	1,526.3	$1,248.3	Rent
Genesis Cloud NVIDIA RTX A6000	Ampere · PCIe Gen4	48 GB	$0.0115	$0.55	$0.35	—	$401.5	Rent
Vast.ai NVIDIA GeForce RTX 3090	Ampere · PCIe Gen4	24 GB	$0.0121	$0.29	$0.14	—	$211.7	Rent
RunPod AMD Instinct MI300X	CDNA 3 · Infinity Fabric 896GB/s	192 GB	$0.0124	$2.39	$1.75	1,092.1	$1,744.7	Rent
Vast.ai NVIDIA GeForce RTX 4090	Ada Lovelace · PCIe Gen4	24 GB	$0.0154	$0.37	$0.22	1,783.8	$270.1	Rent
RunPod NVIDIA A40	Ampere · PCIe Gen4	48 GB	$0.0165	$0.79	$0.45	—	$576.7	Rent
Crusoe Cloud AMD Instinct MI300X	CDNA 3 · Infinity Fabric 896GB/s	192 GB	$0.0180	$3.45	—	756.5	$2,518.5	Rent
RunPod NVIDIA GeForce RTX 3090	Ampere · PCIe Gen4	24 GB	$0.0183	$0.44	$0.29	—	$321.2	Rent
RunPod NVIDIA RTX 6000 Ada Generation	Ada Lovelace · PCIe Gen4	48 GB	$0.0206	$0.99	$0.70	740.4	$722.7	Rent
Vast.ai NVIDIA GeForce RTX 5090	Blackwell · PCIe Gen5	32 GB	$0.0209	$0.67	$0.48	1,925.4	$489.1	Rent
Vast.ai NVIDIA GeForce RTX 4080	Ada Lovelace · PCIe Gen4	16 GB	$0.0213	$0.34	$0.22	1,147.1	$248.2	Rent
Vast.ai 8x NVIDIA GeForce RTX 5090	Blackwell · PCIe Gen5	256 GB	$0.0228	$5.84	$3.84	1,767.1	$4,263.2	Rent
Lambda Labs NVIDIA RTX 6000 Ada Generation	Ada Lovelace · PCIe Gen4	48 GB	$0.0229	$1.10	—	666.4	$803	Rent
Lambda Labs NVIDIA GH200 Grace Hopper (96GB)	Hopper + Grace · NVLink-C2C (900 GB/s)	96 GB	$0.0239	$2.29	—	864.2	$1,671.7	Rent
Vast.ai 8x NVIDIA GeForce RTX 4090	Ada Lovelace · PCIe Gen4	192 GB	$0.0246	$4.72	$3.04	1,118.6	$3,445.6	Rent
RunPod NVIDIA L40S	Ada Lovelace · PCIe Gen4	48 GB	$0.0248	$1.19	$0.79	616	$868.7	Rent
Specialized B300 Neocloud NVIDIA B300 (Ultra-VRAM)	Blackwell · NVLink 1.8TB/s	288 GB	$0.0260	$7.50	$0.50	600	$5,475	Rent
RunPod NVIDIA L4	Ada Lovelace · PCIe Gen4	24 GB	$0.0271	$0.65	$0.40	746.2	$474.5	Rent
RunPod NVIDIA A100 SXM4 (80GB)	Ampere · NVLink (600 GB/s)	80 GB	$0.0274	$2.19	$1.49	—	$1,598.7	Rent
Vast.ai NVIDIA A100 SXM4 (80GB)	Ampere · NVLink 600GB/s	80 GB	$0.0281	$2.25	$1.45	—	$1,642.5	Rent
Lambda Labs NVIDIA H200 SXM	Hopper · NVLink 900GB/s	141 GB	$0.0291	$4.10	—	482.7	$2,993	Rent
Vultr NVIDIA H100 NVL (94GB)	Hopper · NVLink 600GB/s	94 GB	$0.0307	$2.89	$2.10	577.9	$2,109.7	Rent
RunPod NVIDIA GeForce RTX 4090	Ada Lovelace · PCIe Gen4	24 GB	$0.0308	$0.74	$0.44	891.9	$540.2	Rent
RunPod 4x NVIDIA GeForce RTX 4090	Ada Lovelace · PCIe Gen4	96 GB	$0.0308	$2.96	$1.76	891.9	$2,160.8	Rent
Lambda Labs NVIDIA A100 SXM4 (80GB)	Ampere · NVLink 600GB/s	80 GB	$0.0311	$2.49	—	—	$1,817.7	Rent
Lambda Labs 4x NVIDIA A100 SXM4 (80GB)	Ampere · NVLink 600GB/s	320 GB	$0.0311	$9.96	—	—	$7,270.8	Rent
Vultr NVIDIA L40S	Ada Lovelace · PCIe Gen4	48 GB	$0.0313	$1.50	$0.98	488.7	$1,095	Rent
Microsoft Azure (ND MI300X v5) 8x AMD Instinct MI300X	CDNA 3 · Infinity Fabric + 400Gbps IB	1536 GB	$0.0313	$48.00	$28.80	435	$35,040	Rent
RunPod NVIDIA H100 PCIe (80GB)	Hopper · PCIe Gen5	80 GB	$0.0336	$2.69	$1.85	562.5	$1,963.7	Rent
Hyperstack NVIDIA B200	Blackwell · NVLink 1.8TB/s	192 GB	$0.0348	$6.69	$4.90	672.6	$4,883.7	Rent
CoreWeave NVIDIA H200 SXM	Hopper · NVLink 900GB/s	141 GB	$0.0372	$5.25	$3.85	377	$3,832.5	Rent
CoreWeave 8x NVIDIA H200 SXM	Hopper · NVLink + 3.2Tbps IB	1128 GB	$0.0372	$42.00	$30.80	377	$30,660	Rent
RunPod NVIDIA GeForce RTX 5090	Blackwell · PCIe Gen5	32 GB	$0.0372	$1.19	$0.69	1,084	$868.7	Rent
Vast.ai NVIDIA H100 SXM5	Hopper · NVLink 900GB/s	80 GB	$0.0374	$2.99	$1.95	661.9	$2,182.7	Rent
Lambda Labs NVIDIA H100 SXM5	Hopper · NVLink 900GB/s	80 GB	$0.0411	$3.29	—	601.5	$2,401.7	Rent
AWS EC2 (g5.xlarge) NVIDIA A10G	Ampere · PCIe Gen4	24 GB	$0.0419	$1.01	$0.42	—	$734.38	Rent
RunPod NVIDIA A100 SXM4 (40GB)	Ampere · NVLink 600GB/s	40 GB	$0.0473	$1.89	$1.19	—	$1,379.7	Rent
Lambda Labs NVIDIA GeForce RTX 4090	Ada Lovelace · PCIe Gen4	24 GB	$0.0521	$1.25	—	528	$912.5	Rent
Modal NVIDIA H100 SXM5 (Serverless)	Hopper · NVLink 900GB/s	80 GB	$0.0531	$4.25	—	465.6	$3,102.5	Rent
AWS EC2 (p5.48xlarge) 8x NVIDIA H100 SXM5	Hopper · NVLink + EFA 3200Gbps	640 GB	$0.1536	$98.32	$39.30	161	$71,773.6	Rent
Google Cloud (A3 High) 8x NVIDIA H100 SXM5	Hopper · NVLink + GPUDirect-TCPX	640 GB	$0.1538	$98.40	$38.80	160.9	$71,832	Rent
Microsoft Azure (NDv5) 8x NVIDIA H100 SXM5	Hopper · NVLink + Quantum-2 IB	640 GB	$0.1653	$105.80	$42.30	149.6	$77,234	Rent

Showing 43 configurationsNormalized metric: Hourly cost ÷ Total VRAM (GB)

How these metrics are calculated

1. Normalized VRAM cost ($/GB-VRAM-hr):

$/GB-VRAM-hr = Hourly on-demand rate ($/hr) / Total VRAM (GB)

Normalizing by memory reveals pricing differences across identical silicon. For instance, an A100 40 GB at $1.89/hr costs $0.0473/GB-hr, while an A100 80 GB at $2.25/hr costs $0.0281/GB-hr.

2. Model memory sizing:

Total VRAM (GB) ≈ ((Parameters in Billions × Quantization Bits) / 8) × Dynamic KV Factor

Raw weights use integer precision math. A baseline 1.20× factor is applied for activations and context, scaling dynamically for extended context windows (up to 128k tokens) and concurrent batch requests.

3. Data source & disclaimer:

Pricing data is based on public catalog listings across neoclouds (RunPod, Vast.ai, Lambda Labs, CoreWeave, Vultr, Hyperstack) and hyperscalers (AWS, Azure, Google Cloud). Prices are indicative and exclude taxes or committed-use discounts. Outbound deployment links may contain referral identifiers.