Memory tier · 80 GB
Lowest-priced 80 GB GPU listings
2 GPU models carry 80 GB: A100 80GB and H100. The lowest-priced listing is $1.09 per GPU-hour for the A100 80GB at Thunder Compute. Read 12 November 2026.
- Lowest listed
- $1.09
- Per GB-hour
- $0.014
- Providers
- 19
- One day
- $26.16
A100 80GB, Thunder Compute
lowest listed over memory
with a listing in the tier
one GPU at the lowest listing
Tier
GPU models with 80 GB
| GPU | Memory | Lowest-priced listing | $/GPU-hr | $/GB-hr | Providers | Source |
|---|---|---|---|---|---|---|
| A100 80GBdatacenter | 80 GB | Thunder Compute | $1.09 | $0.014 | 12 | page |
| H100datacenter | 80 GB | Voltage Park | $1.99 | $0.025 | 19 | page |
The tier is the card's stated memory, exactly 80 GB. Listings are provider-set on-demand prices in effect today, per GPU-hour. The rule for the lowest-priced listing is simply the lowest price among them.
Listings
Lowest-priced 80 GB listings, provider by provider
| # | GPU | Provider | Configuration | $/GPU-hr | Source |
|---|---|---|---|---|---|
| 1 | A100 80GB | Thunder Compute | 1× · Instance A100 80GB | $1.09 | page |
| 2 | A100 80GB | Hyperstack | 1× · NVIDIA A100 | $1.35 | page |
| 3 | A100 80GB | Massed Compute | 1× · A100 (80GB) | $1.35 | page |
| 4 | A100 80GB | Runcrateours, live | 1× | $1.73 | page |
| 5 | A100 80GB | Civo | 1× · Small 1 x NVIDIA A100 - 80GB | $1.79 | page |
| 6 | A100 80GB | RunPod | 1× PCIe · Secure Cloud A100 PCIe | $1.79 | page |
| 7 | H100 | Voltage Park | 1× · On-demand H100 (FAQ starting price) | $1.99 | page |
| 8 | A100 80GB | Crusoe | 1× PCIe · NVIDIA A100 80GB PCIe | $2.00 | page |
| 9 | H100 | Vultr | 8× HGX · NVIDIA HGX H100 (8 GPU) | $2.30 | page |
| 10 | A100 80GB | Verda | 1× SXM · 1x A100 SXM4 80GB (22 vCPU, 120 GB RAM) | $2.34 | page |
| 11 | A100 80GB | Vultr | 1× PCIe · NVIDIA PCIe A100 (1 GPU) | $2.40 | page |
| 12 | H100 | Hyperstack | 1× PCIe · NVIDIA H100 | $2.50 | page |
| 13 | A100 80GB | CoreWeave | 8× · NVIDIA A100 | $2.70 | page |
| 14 | H100 | Massed Compute | 1× · H100 (80GB) | $2.73 | page |
| 15 | H100 | RunPod | 1× PCIe · Secure Cloud H100 PCIe | $2.89 | page |
One line per provider and GPU, its lowest price. The fifteen lowest are shown of 31.
Or more
The lowest-priced ways to get at least 80 GB
| GPU | Memory | Lowest-priced listing | $/GPU-hr | $/GB-hr | Providers | Source |
|---|---|---|---|---|---|---|
| RTX PRO 6000workstation | 96 GB | packet.ai | $0.76 | $0.008 | 14 | page |
| A100 80GBdatacenter | 80 GB | Thunder Compute | $1.09 | $0.014 | 12 | page |
| MI300Xdatacenter | 192 GB | Vultr8 GPU machine | $1.85 | $0.010 | 4 | page |
| GH200datacenter | 96 GB | Vultr | $1.99 | $0.021 | 2 | page |
| H100datacenter | 80 GB | Voltage Park | $1.99 | $0.025 | 19 | page |
| MI325Xdatacenter | 256 GB | Vultr8 GPU machine | $2.00 | $0.008 | 2 | page |
| MI355Xdatacenter | 288 GB | Vultr8 GPU machine | $2.59 | $0.009 | 2 | page |
| H200datacenter | 141 GB | Civo | $3.49 | $0.025 | 15 | page |
Every GPU model with 80 GB or more, ordered by lowest listed price, eight lowest. A bigger card can cost less than the card at exactly 80 GB; the memory is then free, but the card may be slower or older.
Arithmetic
What 80 GB holds
Weights alone, at 2 bytes per parameter (FP16), 80 GB holds about 40B parameters; at 1 byte (FP8), 80B; at half a byte (INT4), 160B. A 14 GB 7B model at FP16 fits in 80 GB, and a 26 GB 13B model at FP16 fits. These figures leave out the KV cache and activations, which take more memory as context and batch size grow, so treat them as a ceiling.
Questions
What is the lowest price for an 80 GB GPU?
$1.09 per GPU-hour for the A100 80GB at Thunder Compute, read 2026-11-12.
Which GPUs have 80 GB?
A100 80GB (12 providers) and H100 (19 providers).
What does an 80 GB GPU cost per GB of memory?
At the lowest listing, $0.014 per GB-hour. The median of provider lows across the tier is $1.54 per GPU-hour.
Other memory tiers
Disclosure
Low-Cost GPU Finder is published by the Runcrate team, who also run Runcrate, an AI cloud. Runcrate is listed in the table, marked "ours, live". It is sorted by price like every other row, with no pinned position. Other providers' prices are read from their own public pricing pages. We have no paid placement.