Compute Comparison
How-To Guides9 min read

How to Run Llama 3.3 70B on a Rented GPU (2026)

What GPU you need for Llama 3.3 70B, what it costs per hour right now across 120+ providers, and how to serve it with vLLM or Ollama.

1. What you need

Llama 3.3 70B has 70 billion parameters. The table below shows the minimum GPU memory required for the model weights alone, using the same formula as the VRAM calculator: multiply parameter count by bytes per parameter.

FormatBytes / paramWeights only (GB)
FP16 / BF162140 GB
Q8 / INT8170 GB
Q4 / INT40.535 GB

These are weights-only figures. Real deployments need additional memory for the KV cache and runtime overhead, both of which grow with context length and batch size.

Llama 3.3 weights are distributed by Meta under the Llama 3.3 Community License. Downloading the weights requires accepting Meta's terms on the Hugging Face model page before the download is permitted.

2. Pick a GPU

The table below shows live prices for GPU instances with 48 GB or more VRAM — the minimum for running Llama 3.3 70B at Q4 quantization with headroom for the KV cache. Prices update every 15 minutes from provider APIs.

Format guidance: Q4 quantization (35 GB weights) fits a single 48 GB card with headroom for the KV cache. Q8 (70 GB weights) requires an 80 GB card or two 48 GB cards. FP16 (140 GB weights) requires multiple GPUs or a card with 192 GB or more. Two RTX 3090 or RTX 4090 cards (24 GB each) can run the Q4 model split across both cards using tensor parallelism — see the serving section below.

Cheapest$0.35/hrThunder Compute
Cloud Providers90
GPU Listings282
Loading
Filter:
Dot = availability:HighMediumLow / unavailablen/a (static data)
← swipe to scroll →
Cloud Provider
GPU Model
vRAM
vCPUs
RAM
Storage
On-Demand $/hr
Spot $/hr
Availability
Region
RTX A600048 GB848 GB256 GB
$0.35
—n/aUS
L40S48 GB864 GB256 GB
$0.48
$0.19n/aVarious
RTX A600048 GB648 GB256 GB
$0.48Estimate
—n/aUS
A1664 GB832 GB200 GB
$0.56
—n/aEU-West
A100 80GB80 GB16128 GB512 GB
$0.58
—n/aCA/US
RTX A600048 GB1296 GB512 GB
$0.63
—n/aEU-West
A100 80GB80 GB16128 GB512 GB
$0.65Estimate
—n/aIN-West
A4048 GB848 GB256 GB
$0.65
—n/aIN
RTX PRO 6000 Blackwell96 GB696 GB150 GB
$0.66
—n/aUS-West
RTX A600048 GB1296 GB512 GB
$0.72Estimate
—n/aUS
L40S48 GB32128 GB960 GB
$0.77
—n/aEU-West
RTX A600048 GB848 GB256 GB
$0.79
—n/aEU-North
RTX A600048 GB848 GB256 GB
$0.79Estimate
—n/aEU-North
RTX A600048 GB1296 GB512 GB
$0.80
—n/aIS
RTX A600048 GB1296 GB512 GB
$0.82
—n/aSE
A100 80GB80 GB16128 GB512 GB
$0.85
—n/aVarious
RTX A600048 GB1296 GB512 GB
$0.85
—n/aEU
RTX A600048 GB1296 GB512 GB
$0.88
—n/aEU
A100 80GB80 GB16128 GB512 GB
$0.90Estimate
$0.33n/aUS/EU/APAC
L40S48 GB864 GB50 GB
$0.92
—n/aUK
A100 80GB80 GB16128 GB512 GB
$0.93Estimate
$0.34n/aVarious
A100 80GB80 GB16128 GB512 GB
$0.95Estimate
$0.35n/aEU/US
A100 80GB80 GB16128 GB512 GB
$0.95
$0.38n/aVarious
L40S48 GB864 GB256 GB
$0.95
—n/aGlobal
A100 80GB80 GB28119 GB400 GB
$0.98Estimate
—n/aEU/US
L40S48 GB864 GB256 GB
$0.98
—n/aUS/EU
L40S48 GB864 GB256 GB
$0.99
—n/aGlobal
A100 80GB80 GB16128 GB512 GB
$1.01Estimate
$0.36n/aUS/EU/APAC
A100 80GB80 GB16128 GB512 GB
$1.03Estimate
$0.38n/aUS
A100 80GB80 GB24192 GB480 GB
$1.05Estimate
—n/aUS-West
A100 80GB80 GB16128 GB512 GB
$1.05
—n/aEU-West
A100 80GB80 GB16128 GB512 GB
$1.05
—n/aUS
L40S48 GB864 GB256 GB
$1.05
—n/aEU
L40S48 GB864 GB256 GB
$1.05
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.06Estimate
$0.39n/aEU/US
A100 80GB80 GB16128 GB512 GB
$1.08Estimate
—n/aUS
A100 80GB80 GB24192 GB512 GB
$1.09Estimate
—n/aUS/EU
L40S48 GB864 GB256 GB
$1.09
—n/aUS
A100 80GB80 GB1280 GB512 GB
$1.10Estimate
—n/aUS
A100 80GB80 GB———
$1.10
—n/a
A100 80GB80 GB16128 GB512 GB
$1.10
—n/aUS
L40S48 GB864 GB256 GB
$1.10
—n/aEU/US
L40S48 GB864 GB256 GB
$1.10
—n/aUS
RTX 6000 Ada48 GB2064 GB2.0 TB
$1.13
—n/aCH
A100 80GB80 GB864 GB240 GB
$1.15Estimate
—n/aUS/EU/BR
A100 80GB80 GB16128 GB512 GB
$1.15
—n/aUS/APAC
L40S48 GB864 GB256 GB
$1.15
—n/aEU
L40S48 GB864 GB256 GB
$1.15
—n/aUS/EU
A100 80GB80 GB16125 GB200 GB
$1.19Estimate
$0.42n/aUS
A100 80GB80 GB16128 GB512 GB
$1.19
—n/aGlobal
L40S48 GB864 GB256 GB
$1.20
—n/aEU/US
L40S48 GB864 GB256 GB
$1.20
—n/aNL
A100 80GB80 GB16128 GB512 GB
$1.25Estimate
$0.45n/aVarious
L40S48 GB864 GB256 GB
$1.25
—n/aEU
A100 80GB80 GB16128 GB512 GB
$1.25
—n/aIN
A100 80GB80 GB16128 GB512 GB
$1.25
—n/aUS/EU
A100 80GB80 GB1296 GB240 GB
$1.26Estimate
—n/aEU-West
A100 80GB80 GB1290 GB500 GB
$1.28Estimate
—n/aUS-East
A100 80GB80 GB30200 GB512 GB
$1.29Estimate
—n/aus-east-1
A100 80GB80 GB16128 GB512 GB
$1.29
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.29
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.29
—n/aGlobal
L40S48 GB32128 GB960 GB
$1.30Estimate
—n/aEU-West
A100 80GB80 GB16128 GB512 GB
$1.30
—n/aEU/US
L40S48 GB864 GB256 GB
$1.30
—n/aEU
A100 80GB80 GB16128 GB512 GB
$1.35
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.35
—n/aUS
L40S48 GB864 GB256 GB
$1.35
—n/aSE
A100 80GB80 GB16128 GB512 GB
$1.38Estimate
—n/aEU-West
A100 80GB80 GB16256 GB3.4 TB
$1.39Estimate
$0.49n/aus-ashburn-1
L4048 GB864 GB256 GB
$1.39
—n/aUS
A100 80GB80 GB1290 GB400 GB
$1.39
—n/aUS/IN
A100 80GB80 GB16128 GB512 GB
$1.39
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.39
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.40
—n/aEU-South
A100 80GB80 GB16128 GB512 GB
$1.40
—n/aUS
RTX 6000 Ada48 GB848 GB256 GB
$1.41
—n/aEU-North
RTX 6000 Ada48 GB848 GB256 GB
$1.41Estimate
—n/aEU-North
A100 80GB80 GB1280 GB150 GB
$1.43
—n/aUK
A100 80GB80 GB16128 GB512 GB
$1.45Estimate
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.45
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.45
—n/aEU
A100 80GB80 GB16128 GB512 GB
$1.45
—n/aUA/PL/SE
A100 80GB80 GB16128 GB512 GB
$1.45Estimate
—n/aUS-East
L40S48 GB864 GB256 GB
$1.49
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.49
$0.60n/aVarious
A100 80GB80 GB16128 GB512 GB
$1.49
—n/aEU
A100 80GB80 GB16128 GB512 GB
$1.49
—n/aEU
H100 80GB80 GB20160 GB512 GB
$1.50
—n/aCA/US
A100 80GB80 GB16128 GB512 GB
$1.50
—n/aUS
A100 80GB80 GB961152 GB8.0 TB
$1.53Estimate
$0.51n/aus-east-1
RTX PRO 600096 GB———
$1.55
—n/a
A100 80GB80 GB16128 GB512 GB
$1.55
—n/aEU
A100 80GB80 GB16128 GB512 GB
$1.55
—n/aUS/EU
RTX PRO 600096 GB16128 GB512 GB
$1.56
—n/aCA-Central
L40S48 GB1664 GB400 GB
$1.57
—n/aUS
RTX 6000 Ada48 GB1664 GB400 GB
$1.57
—n/aUS
A100 80GB80 GB24240 GB512 GB
$1.59Estimate
$0.48n/aUS-East
A100 80GB80 GB16128 GB512 GB
$1.59
—n/aEU
A100 80GB80 GB———
$1.59Estimate
—n/aGlobal
A100 80GB80 GB16128 GB512 GB
$1.60
—n/aEU-West
L40S48 GB864 GB256 GB
$1.60
—n/aEU-South
A100 80GB80 GB16128 GB512 GB
$1.60
—n/aNL
A100 80GB80 GB96880 GB6.4 TB
$1.61Estimate
$0.56n/aeastus
A100 80GB80 GB16128 GB512 GB
$1.65
—n/aIS
A100 80GB80 GB16128 GB512 GB
$1.65Estimate
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.69
—n/aKR
A100 80GB80 GB16128 GB512 GB
$1.69
—n/aSE
A100 80GB80 GB16128 GB512 GB
$1.69Estimate
—n/aUS
RTX PRO 600096 GB2064 GB4.0 TB
$1.70
—n/aCH
A100 80GB80 GB16128 GB512 GB
$1.75
—n/aUS/EU/APAC
A100 80GB80 GB16128 GB512 GB
$1.79
—n/aUS
A100 80GB80 GB16128 GB512 GB
$1.79
—n/aEU
H100 80GB80 GB16128 GB512 GB
$1.80
—n/aVarious
A100 80GB80 GB16128 GB512 GB
$1.80
—n/aEU
A100 80GB80 GB1280 GB512 GB
$1.80Estimate
—n/aUS
L40S48 GB864 GB256 GB
$1.83
—n/aIN-North
A100 80GB80 GB16128 GB512 GB
$1.85
—n/aSE
A100 80GB80 GB16128 GB512 GB
$1.89Estimate
$1.40n/aUS
L40S48 GB864 GB256 GB
$1.90
—n/aEU-West
A100 80GB80 GB1280 GB512 GB
$1.90
—n/aUS
MI300X 192GB192 GB24192 GB512 GB
$1.99
—n/aUS
H100 80GB80 GB16128 GB512 GB
$1.99
$0.79n/aVarious
A100 80GB80 GB1290 GB400 GB
$1.99Estimate
—n/aUS
A100 80GB80 GB16128 GB512 GB
$2.05
—n/aEU
A100 80GB80 GB16128 GB512 GB
$2.10
—n/aUS/EU
A100 80GB80 GB16128 GB512 GB
$2.10
—n/aEU/US
A100 80GB80 GB16128 GB400 GB
$2.13Estimate
—n/aEU-North
A100 80GB80 GB16128 GB512 GB
$2.15
—n/aUS/EU
H100 80GB80 GB20160 GB512 GB
$2.19
—n/aUS
RTX A600048 GB1296 GB512 GB
$2.20
$0.84n/aVarious
A100 80GB80 GB20160 GB512 GB
$2.20
—n/aUS
H100 80GB80 GB20160 GB512 GB
$2.20
—n/aUS
A100 80GB80 GB16128 GB512 GB
$2.20
$0.66n/aAP-Southeast
H100 80GB80 GB20160 GB512 GB
$2.25
—n/aUS
H100 80GB80 GB20160 GB512 GB
$2.25
—n/aUS
H100 80GB80 GB20160 GB512 GB
$2.29
—n/aIN
H100 80GB80 GB20160 GB512 GB
$2.29
—n/aGlobal
H100 80GB80 GB20160 GB512 GB
$2.31Estimate
$0.90n/aVarious
H100 80GB80 GB20160 GB512 GB
$2.35
—n/aEU-West
H100 80GB80 GB20160 GB512 GB
$2.35
—n/aUS
H100 80GB80 GB20160 GB512 GB
$2.35
—n/aUS/EU
H100 80GB80 GB16128 GB512 GB
$2.36Estimate
$0.93n/aUS/EU/APAC
H100 80GB80 GB16125 GB200 GB
$2.39Estimate
$0.99n/aUS
H100 80GB80 GB20160 GB512 GB
$2.39
$0.95n/aVarious
H100 80GB80 GB20160 GB512 GB
$2.39
—n/aGlobal
RTX A600048 GB1296 GB512 GB
$2.40
$0.90n/aUS
A100 80GB80 GB16128 GB512 GB
$2.40
—n/aEU/US
H100 80GB80 GB20160 GB512 GB
$2.45
—n/aUS
H100 80GB80 GB20160 GB512 GB
$2.48Estimate
$0.95n/aEU/US
H100 80GB80 GB26200 GB512 GB
$2.49Estimate
—n/aus-east-1
H100 80GB80 GB20160 GB512 GB
$2.49
—n/aUS
H100 80GB80 GB20160 GB512 GB
$2.49
—n/aUS/APAC
H100 80GB80 GB20160 GB512 GB
$2.49
—n/aUS
H100 80GB80 GB20160 GB512 GB
$2.49Estimate
$1.80n/aUS
H100 80GB80 GB20160 GB400 GB
$2.55Estimate
—n/aEU/US
H100 80GB80 GB20160 GB512 GB
$2.59
—n/aUS
H100 80GB80 GB20160 GB512 GB
$2.59
—n/aEU
H100 80GB80 GB16128 GB512 GB
$2.60Estimate
$1.05n/aVarious
H100 SXM80 GB———
$2.60
—n/a
H100 80GB80 GB20160 GB512 GB
$2.61Estimate
$1.02n/aUS/EU/APAC
H100 80GB80 GB16128 GB512 GB
$2.63Estimate
$1.05n/aUS
H100 80GB80 GB20160 GB512 GB
$2.65
—n/aUS
H100 80GB80 GB20160 GB512 GB
$2.65Estimate
—n/aUS
H100 80GB80 GB16128 GB512 GB
$2.68Estimate
$1.07n/aEU/US
H100 80GB80 GB20160 GB512 GB
$2.69
—n/aEU
H100 80GB80 GB20160 GB512 GB
$2.69
—n/aEU
H100 80GB80 GB12100 GB512 GB
$2.70Estimate
—n/aUS
H100 80GB80 GB24192 GB480 GB
$2.74Estimate
—n/aUS-West
H100 80GB80 GB20160 GB512 GB
$2.75
—n/aUS
H100 80GB80 GB24240 GB512 GB
$2.79Estimate
$0.84n/aUS-East
H100 80GB80 GB24192 GB512 GB
$2.79Estimate
—n/aUS/EU
H100 80GB80 GB20160 GB512 GB
$2.79
—n/aEU
H100 80GB80 GB20160 GB512 GB
$2.79Estimate
—n/aUS
RTX A600048 GB1296 GB512 GB
$2.80
—n/aUS-East
H100 80GB80 GB20160 GB512 GB
$2.80Estimate
—n/aIN-West
H100 80GB80 GB16128 GB512 GB
$2.83Estimate
—n/aUS
H100 80GB80 GB20160 GB512 GB
$2.85
—n/aUS/EU
H100 80GB80 GB20160 GB512 GB
$2.85Estimate
—n/aUS
H100 SXM80 GB20160 GB512 GB
$2.85Estimate
—n/aUS
H100 80GB80 GB16120 GB400 GB
$2.89
—n/aUS/IN
H100 80GB80 GB20160 GB512 GB
$2.89
—n/aEU
H100 80GB80 GB20160 GB512 GB
$2.89
—n/aNL
A1664 GB832 GB200 GB
$2.90
$1.10n/aUS
H100 80GB80 GB20160 GB512 GB
$2.90
—n/aEU/US
H100 80GB80 GB16128 GB480 GB
$2.95Estimate
—n/aUS/EU/BR
H100 80GB80 GB20160 GB512 GB
$2.95
—n/aIS
H100 80GB80 GB26188 GB480 GB
$2.98Estimate
—n/aEU-West
H100 80GB80 GB20160 GB512 GB
$2.99
—n/aKR
H100 80GB80 GB20160 GB512 GB
$2.99
—n/aSE
H100 80GB80 GB16120 GB400 GB
$2.99Estimate
—n/aUS
H100 80GB80 GB12100 GB512 GB
$3.01Estimate
—n/aUS
H100 80GB80 GB20160 GB512 GB
$3.05
—n/aEU
H100 80GB80 GB20160 GB512 GB
$3.05
—n/aEU
H100 80GB80 GB12115 GB500 GB
$3.09Estimate
—n/aUS-East
A4048 GB1296 GB512 GB
$3.10
$1.20n/aVarious
H100 80GB80 GB20160 GB512 GB
$3.10Estimate
—n/aEU-West
H100 80GB80 GB20160 GB512 GB
$3.15
—n/aSE
A1664 GB832 GB200 GB
$3.20
$1.25n/aVarious
H100 80GB80 GB20160 GB512 GB
$3.20
—n/aEU-West
H100 80GB80 GB20160 GB512 GB
$3.20
—n/aEU-South
H200 141GB141 GB24192 GB512 GB
$3.20
—n/aCA/US
H100 80GB80 GB12100 GB512 GB
$3.20
—n/aUS
H100 80GB80 GB20160 GB512 GB
$3.20Estimate
—n/aUS-East
A4048 GB32256 GB960 GB
$3.40Estimate
—n/aUS/EU/BR
H100 80GB80 GB20160 GB512 GB
$3.44
—n/aUS
H200 141GB141 GB24192 GB512 GB
$3.49
—n/aGlobal
RTX 6000 Ada48 GB1296 GB512 GB
$3.50
$1.30n/aVarious
H200 141GB141 GB24192 GB512 GB
$3.50
—n/aUS
H100 80GB80 GB20160 GB512 GB
$3.50
$1.05n/aAP-Southeast
H100 80GB80 GB20256 GB3.4 TB
$3.55Estimate
$1.28n/aus-ashburn-1
A4048 GB1296 GB512 GB
$3.60
$1.40n/aUS
L4048 GB1296 GB512 GB
$3.60
$1.35n/aVarious
RTX 6000 Ada48 GB1296 GB512 GB
$3.60
$1.35n/aEU/US
H200 141GB141 GB24192 GB512 GB
$3.60
—n/aEU-West
H200 141GB141 GB24192 GB512 GB
$3.60
—n/aUS
B200180 GB16180 GB250 GB
$3.75
—n/aUS-West
H200 141GB141 GB24192 GB512 GB
$3.75
—n/aEU
A4048 GB1296 GB512 GB
$3.80
—n/aus-west-1
A1664 GB832 GB200 GB
$3.80
—n/aUS-East
RTX 6000 Ada48 GB1296 GB512 GB
$3.80
$1.45n/aUS
H200 141GB141 GB24192 GB512 GB
$3.80
—n/aUS/APAC
H100 80GB80 GB20160 GB512 GB
$3.87
—n/aUS
H200 141GB141 GB24192 GB512 GB
$3.89
—n/aKR
H100 80GB80 GB20160 GB512 GB
$3.89
—n/aUS/EU
L4048 GB1296 GB512 GB
$3.90
$1.50n/aUS
L40S48 GB848 GB256 GB
$3.90
—n/aUS/EU
H200 141GB141 GB24192 GB512 GB
$3.99Estimate
—n/aUS
H100 80GB80 GB1922048 GB3.8 TB
$4.10Estimate
$1.55n/aus-east-1
H100 80GB80 GB96900 GB6.4 TB
$4.15Estimate
$1.64n/aeastus
A4048 GB1296 GB512 GB
$4.20
—n/aUS-East
H200 141GB141 GB24192 GB512 GB
$4.20Estimate
—n/aIN-West
H200 141GB141 GB24192 GB512 GB
$4.29Estimate
—n/aUS
H200 141GB141 GB20188 GB400 GB
$4.39Estimate
—n/aEU/US
RTX 6000 Ada48 GB1296 GB512 GB
$4.40
—n/aUS-East
H200 141GB141 GB24192 GB512 GB
$4.40
$1.76n/aVarious
L40S48 GB1296 GB512 GB
$4.50
—n/aEU/US
L4048 GB1296 GB512 GB
$4.50
—n/aUS-East
H200 141GB141 GB24192 GB512 GB
$4.50Estimate
—n/aVarious
H100 80GB80 GB24192 GB512 GB
$4.50
—n/aUS/EU
H200 141GB141 GB24192 GB512 GB
$4.50Estimate
—n/aUS
H100 80GB80 GB20160 GB512 GB
$4.51Estimate
—n/aEU-North
H200 141GB141 GB20188 GB400 GB
$4.59Estimate
$1.89n/aUS
L40S48 GB1296 GB512 GB
$4.60
$1.75n/aVarious
L40S48 GB864 GB256 GB
$4.70
$1.80n/aUS/EU/APAC
H200 141GB141 GB24192 GB512 GB
$4.79Estimate
—n/aUS/EU
L40S48 GB1296 GB512 GB
$4.90
$1.90n/aUS
H200 141GB141 GB26240 GB512 GB
$4.99Estimate
—n/aus-east-1
A100 80GB80 GB2401920 GB3.0 TB
$5.07Estimate
$0.48n/aus-central1
L40S48 GB1296 GB512 GB
$5.10
—n/aUS/EU
B200 180GB180 GB32384 GB960 GB
$5.10
—n/aEU-Central
H200 141GB141 GB24192 GB512 GB
$5.20
—n/aEU-South
H200 141GB141 GB26240 GB512 GB
$5.25Estimate
—n/aUS-East
L40S48 GB1296 GB512 GB
$5.40
—n/aus-east-1
L40S48 GB1296 GB512 GB
$5.80
—n/aUS-East
H200 141GB141 GB24192 GB512 GB
$5.89Estimate
—n/aEU-North
B200 180GB180 GB———
$6.30
—n/a
MI355X 288GB288 GB———
$6.30
—n/a
A100 80GB80 GB864 GB256 GB
$6.80
—n/aUS/EU
B200180 GB24180 GB350 GB
$6.99
—n/aUS-West
B300 288GB288 GB———
$8.49
—n/a
H200 141GB141 GB24192 GB512 GB
$8.70
—n/aEU-West
H100 80GB80 GB128512 GB3.8 TB
$8.76
—n/aEU-West
H200 141GB141 GB2081872 GB3.0 TB
$10.60Estimate
—n/aus-central1
A100 80GB80 GB28200 GB1.0 TB
$10.80Estimate
—n/aUS/EU/APAC
H100 80GB80 GB2081872 GB3.0 TB
$11.06Estimate
$1.35n/aus-central1
MI300X 192GB192 GB24192 GB512 GB
$28.50
—n/aUS-West
MI300X 192GB192 GB24192 GB512 GB
$29.00
—n/aVarious
MI300X 192GB192 GB24192 GB512 GB
$31.50
—n/aEU/US
MI300X 192GB192 GB24192 GB512 GB
$34.00
—n/aus-east-1
MI325X 256GB256 GB24256 GB512 GB
$36.00
—n/aVarious
MI325X 256GB256 GB24256 GB512 GB
$36.00
—n/aUS-West
MI325X 256GB256 GB24256 GB512 GB
$38.50
—n/aEU/US
MI325X 256GB256 GB24256 GB512 GB
$42.00
—n/aus-east-1
B200 192GB192 GB32384 GB960 GB
$50.00
—n/aUS-West
B200 192GB192 GB32384 GB960 GB
$52.00
—n/aUS/EU
B200 192GB192 GB32384 GB960 GB
$54.50
—n/aus-east-1
B200 192GB192 GB32384 GB960 GB
$58.00
—n/aUS-East
GB200 NVL72192 GB64576 GB1.9 TB
$68.50
—n/aus-east-1
GB200 NVL72192 GB64576 GB1.9 TB
$72.00
—n/aUS-East
GB300 NVL72288 GB64768 GB1.9 TB
$90.00
—n/aus-east-1
GB300 NVL72288 GB64768 GB1.9 TB
$95.00
—n/aUS-East

Showing 282 of 467 listings · Prices updated every 15 min via provider APIs · Data is indicative only

■ Cheapest  ■ Mid  ■ Expensive

Affiliate link — we may earn a commission.

Announced / Quote-Only — not included in pricing table

i

Vultr — AMD MI455X: announced, capacity available soon

Pricing: Contact SalesAvailability: Capacity available soon

Full-rack and data center scale reservation. No public per-GPU hourly rate published. Pricing requires direct sales engagement.

Affiliate disclosure: some provider rows include a "Visit" link. Compute Comparison may earn a commission if you sign up through that link, at no extra cost to you.

3. Rent and connect

The steps below apply to any GPU cloud provider. Provider-specific dashboards differ in layout but follow the same sequence.

  1. Create an account and add a payment method.
  2. Launch an instance with your chosen GPU. Select a PyTorch or CUDA base image if available — it saves installing the GPU driver stack manually.
  3. Copy the SSH command from the provider dashboard and connect:
ssh -i ~/.ssh/id_rsa user@<instance-ip>

Verify the GPU is visible before installing anything:

nvidia-smi

4. Serve the model

Two paths: vLLM for an OpenAI-compatible API server, or Ollama for a simpler single-command setup.

Path A — vLLM (OpenAI-compatible API)

Install vLLM and serve the model. You will need a Hugging Face token with access to the gated Llama 3.3 repository.

pip install vllm

vllm serve meta-llama/Llama-3.3-70B-Instruct \
  --dtype bfloat16 \
  --max-model-len 8192

Test the server with a curl request:

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "meta-llama/Llama-3.3-70B-Instruct",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

For a dual 24 GB card setup (e.g. two RTX 3090 or RTX 4090), add the tensor-parallelism flag to split the model across both GPUs:

vllm serve meta-llama/Llama-3.3-70B-Instruct \
  --dtype bfloat16 \
  --tensor-parallel-size 2 \
  --max-model-len 4096

For production deployment detail — autoscaling, quantization options, and multi-node setups — see the vLLM deployment guide.

Path B — Ollama (simple single-command)

Ollama handles model download and quantization automatically. It defaults to Q4 quantization, which fits a single 48 GB card.

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull and run the model
ollama run llama3.3:70b

5. What it costs

Your hourly cost is the on-demand or spot rate for the GPU you choose — see the live table in section 2 above. Multiply the hourly rate by the number of hours your instance runs to get your total compute cost.

To compare self-hosted cost against per-token API pricing for Llama 3.3 70B, see the Llama 3.3 70B model page, which shows current managed inference prices across providers. For the break-even framework — at what token volume self-hosting beats managed APIs — see the LLM API vs self-hosted cost comparison.

FAQ

Which GPU is cheapest for running Llama 3.3 70B?
Check the live table in section 2 — prices change daily. For price stability over time, see GPU rankings, which tracks observed price history.
Is spot pricing safe for inference?
Spot instances can be interrupted with little warning, which makes them unsuitable for latency-sensitive production inference without a checkpoint or failover strategy. For a full analysis of when spot is appropriate, see the GPU spot vs on-demand pricing guide.
Can I run Llama 3.3 70B on a single 24 GB card?
No. Q4 quantization requires approximately 35 GB for the model weights alone, which exceeds a single 24 GB card. You need a 48 GB card or larger, or two 24 GB cards running tensor parallelism as described in section 4.
Is a managed API cheaper than renting a GPU?
It depends on your token volume. At low volumes, managed APIs are almost always cheaper because you pay only for what you use. At high volumes, self-hosting can be significantly cheaper. The LLM API vs self-hosted cost comparison provides the break-even framework.
Was this guide useful?