Price spread · the aligned conclusion nobody else does
Same GPU card, different marketplace, different price
The same GPU card can differ by up to 20.2% across marketplace platforms: RTX 5090 32GB is cheapest on Clore.ai ($0.5554/GPU·h) and dearest on vast.ai ($0.6676/GPU·h) — an 8-GPU box rented for a month differs by $655.25. Below is the aligned spread table for all 37 qualifying models (updated 2026-11-13).
Anyone can scrape vast.ai and Clore.ai listings. What nobody else does is align them: same model, same time window, whole-machine prices normalized to per-GPU, dimension-inconsistent quotes dropped — then work out the real spread and what it costs per month.
| Model | Spread | Cheapest | Dearest | 8-GPU box, monthly diff | Machines listed | Series |
|---|---|---|---|---|---|---|
| RTX 5090 32GB
nv.rtx5090.pcie.32g |
20.2% | Clore.ai $0.5554/GPU·h |
vast.ai $0.6676/GPU·h |
$655.25 | 300 / 2,014 | 9d |
| RTX 4090 24GB
nv.rtx4090.pcie.24g |
31.8% | Clore.ai $0.375/GPU·h |
vast.ai $0.4941/GPU·h |
$695.54 | 207 / 1,146 | 9d |
| RTX 3090 24GB
nv.rtx3090.pcie.24g |
36.5% | Clore.ai $0.1601/GPU·h |
vast.ai $0.2185/GPU·h |
$341.06 | 222 / 881 | 9d |
| RTX 3060 12GB
nv.rtx3060.pcie.12g |
43.6% | Clore.ai $0.0521/GPU·h |
vast.ai $0.0748/GPU·h |
$132.57 | 112 / 565 | 9d |
| NVIDIA RTX5060TI PCIE 16G
nv.rtx5060ti.pcie.16g |
41.5% | Clore.ai $0.0958/GPU·h |
vast.ai $0.1356/GPU·h |
$232.43 | 88 / 506 | 9d |
| NVIDIA RTX5080 PCIE 16G
nv.rtx5080.pcie.16g |
17.7% | Clore.ai $0.2285/GPU·h |
vast.ai $0.2689/GPU·h |
$235.94 | 189 / 394 | 9d |
| NVIDIA RTX5070TI PCIE 16G
nv.rtx5070ti.pcie.16g |
9.3% | vast.ai $0.2279/GPU·h |
Clore.ai $0.249/GPU·h |
$123.22 | 300 / 206 | 9d |
| NVIDIA V100 SXM 32G
nv.v100.sxm.32g |
36.4% | Clore.ai $0.0599/GPU·h |
vast.ai $0.0817/GPU·h |
$127.31 | 281 / 162 | 9d |
| RTX 3080 10GB
nv.rtx3080.pcie.10g |
12.6% | Clore.ai $0.1204/GPU·h |
vast.ai $0.1356/GPU·h |
$88.77 | 220 / 152 | 9d |
| RTX 3070 8GB
nv.rtx3070.pcie.8g |
72.4% | vast.ai $0.0967/GPU·h |
Clore.ai $0.1667/GPU·h |
$408.8 | 117 / 236 | 9d |
| NVIDIA RTXPRO6000MAXQ PCIE 96G
nv.rtxpro6000maxq.pcie.96g |
33% | Clore.ai $1.10/GPU·h |
vast.ai $1.47/GPU·h |
$2,129.85 | 6 / 222 | 9d |
| NVIDIA RTX5070 PCIE 12G
nv.rtx5070.pcie.12g |
14.6% | Clore.ai $0.1521/GPU·h |
vast.ai $0.1743/GPU·h |
$129.65 | 84 / 110 | 9d |
| NVIDIA RTX3080TI PCIE 12G
nv.rtx3080ti.pcie.12g |
29.8% | Clore.ai $0.125/GPU·h |
vast.ai $0.1622/GPU·h |
$217.25 | 75 / 113 | 9d |
| NVIDIA RTXA4000 PCIE 16G
nv.rtxa4000.pcie.16g |
50.3% | Clore.ai $0.0729/GPU·h |
vast.ai $0.1096/GPU·h |
$214.33 | 9 / 169 | 9d |
| NVIDIA RTX3060TI PCIE 8G
nv.rtx3060ti.pcie.8g |
65.6% | Clore.ai $0.0561/GPU·h |
vast.ai $0.0929/GPU·h |
$214.91 | 76 / 96 | 9d |
| NVIDIA RTXPRO4000 PCIE 24G
nv.rtxpro4000.pcie.24g |
40.3% | Clore.ai $0.2/GPU·h |
vast.ai $0.2807/GPU·h |
$471.29 | 7 / 112 | 9d |
| A100 SXM 40GB
nv.a100.sxm.40g |
2.5% | vast.ai $0.5359/GPU·h |
Clore.ai $0.5492/GPU·h |
$77.67 | 103 / 11 | 9d |
| NVIDIA CMP170HX PCIE 64G
nv.cmp170hx.pcie.64g |
32.9% | Clore.ai $0.3125/GPU·h |
vast.ai $0.4152/GPU·h |
$599.77 | 15 / 94 | 9d |
| NVIDIA RTX4070 PCIE 12G
nv.rtx4070.pcie.12g |
30.8% | Clore.ai $0.1042/GPU·h |
vast.ai $0.1363/GPU·h |
$187.46 | 48 / 53 | 9d |
| NVIDIA RTX4070TI PCIE 12G
nv.rtx4070ti.pcie.12g |
5.4% | Clore.ai $0.1334/GPU·h |
vast.ai $0.1406/GPU·h |
$42.05 | 32 / 56 | 9d |
| NVIDIA RTX5060TI PCIE 8G
nv.rtx5060ti.pcie.8g |
38.2% | Clore.ai $0.0938/GPU·h |
vast.ai $0.1296/GPU·h |
$209.07 | 53 / 32 | 8d |
| RTX 6000 Ada 48GB
nv.rtx6000ada.pcie.48g |
67.2% | Clore.ai $0.3867/GPU·h |
vast.ai $0.6465/GPU·h |
$1,517.23 | 4 / 77 | 9d |
| RTX 4080 16GB
nv.rtx4080.pcie.16g |
24.6% | Clore.ai $0.2052/GPU·h |
vast.ai $0.2556/GPU·h |
$294.34 | 16 / 58 | 8d |
| NVIDIA RTX3070TI PCIE 8G
nv.rtx3070ti.pcie.8g |
3.7% | vast.ai $0.1356/GPU·h |
Clore.ai $0.1406/GPU·h |
$29.2 | 11 / 54 | 4d |
| NVIDIA RTXA5000 PCIE 24G
nv.rtxa5000.pcie.24g |
27.6% | Clore.ai $0.1806/GPU·h |
vast.ai $0.2304/GPU·h |
$290.83 | 3 / 59 | 9d |
| NVIDIA RTX5060 PCIE 8G
nv.rtx5060.pcie.8g |
48.3% | Clore.ai $0.0729/GPU·h |
vast.ai $0.1081/GPU·h |
$205.57 | 23 / 35 | 9d |
| NVIDIA GTX1070 PCIE 8G
nv.gtx1070.pcie.8g |
192.8% | Clore.ai $0.0167/GPU·h |
vast.ai $0.0489/GPU·h |
$188.05 | 30 / 19 | 8d |
| NVIDIA GTX1080TI PCIE 11G
nv.gtx1080ti.pcie.11g |
65.2% | Clore.ai $0.0417/GPU·h |
vast.ai $0.0689/GPU·h |
$158.85 | 20 / 26 | 5d |
| NVIDIA RTX3080 PCIE 20G
nv.rtx3080.pcie.20g |
18.6% | Clore.ai $0.1042/GPU·h |
vast.ai $0.1236/GPU·h |
$113.3 | 13 / 31 | 9d |
| NVIDIA RTX2080TI PCIE 11G
nv.rtx2080ti.pcie.11g |
15.7% | vast.ai $0.1009/GPU·h |
Clore.ai $0.1167/GPU·h |
$92.27 | 35 / 7 | 9d |
| NVIDIA RTX3090TI PCIE 24G
nv.rtx3090ti.pcie.24g |
21.1% | Clore.ai $0.2031/GPU·h |
vast.ai $0.2459/GPU·h |
$249.95 | 8 / 30 | 7d |
| NVIDIA GTX1070TI PCIE 8G
nv.gtx1070ti.pcie.8g |
32.9% | vast.ai $0.0529/GPU·h |
Clore.ai $0.0703/GPU·h |
$101.62 | 21 / 10 | 8d |
| NVIDIA V100 SXM 16G
nv.v100.sxm.16g |
77.8% | Clore.ai $0.05/GPU·h |
vast.ai $0.0889/GPU·h |
$227.18 | 17 / 10 | 7d |
| NVIDIA RTX3070LAPTOP PCIE 8G
nv.rtx3070laptop.pcie.8g |
16.3% | vast.ai $0.0896/GPU·h |
Clore.ai $0.1042/GPU·h |
$85.26 | 3 / 19 | 5d |
| NVIDIA GTX1660TI PCIE 6G
nv.gtx1660ti.pcie.6g |
87.5% | Clore.ai $0.0407/GPU·h |
vast.ai $0.0763/GPU·h |
$207.9 | 8 / 12 | 7d |
| NVIDIA RTX2070 PCIE 8G
nv.rtx2070.pcie.8g |
13.8% | vast.ai $0.0622/GPU·h |
Clore.ai $0.0708/GPU·h |
$50.22 | 15 / 4 | 7d |
| NVIDIA P40 PCIE 24G
nv.p40.pcie.24g |
12.3% | Clore.ai $0.0958/GPU·h |
vast.ai $0.1076/GPU·h |
$68.91 | 7 / 7 | 6d |
Why only 37 models when the ledger holds 252? The other 215 fail the publication bar: fewer than 3 machines per platform (median not trustworthy), fewer than 2 valid platforms (that's not cross-platform), or a platform median deviating 4x from the overall median (usually an un-converted unit — per-day or per-month — which would produce absurd 1000%-plus fake spreads). We don't pad the table — a fake spread hurts credibility more than no spread.
Method: latest real listing per machine, per-platform median; every price normalized to $/GPU/hour; these are asking prices, not completed sales. A spread is not an arbitrage: card variant, uptime, SLA, bandwidth and region differ — the cheap side may not be rentable. Machine-readable: GET /v1/spread (free, per-platform samples and exclusion reasons). Asking which cloud is cheapest on demand? That's a different layer: /en/cloud.