grab▲gpu

GPU model · price + stock · hourly snapshot

Nvidia RTX 5070 cloud rental prices

3 providers list the Nvidia RTX 5070, 3 have it in stock. Prices run from $0.13 to $0.17 per GPU hour — the same card, 1.3x apart.

Blackwell 12 GB VRAM Q1 2025 5 configurations

LIVE Prices and stock checked 1 minute ago

Cheapest in stock
$0.13
Clore.ai
Median
$0.14
across 3 providers
Most expensive
$0.17
same card, 1.3x the price
You overpay by
9%
if you take the median instead of the floor

Providers

Every provider renting the Nvidia RTX 5070

All providers →

Billing:

On-demandfrom $0.13/hr3 providers in stockPay as you go, stop any time.Spotfrom $0.096/hr2 providers in stockCheaper, but the provider can take the machine back.Reservednone listedPrepay a term for the lowest hourly rate.
Providers renting the Nvidia RTX 5070, in stock first, then cheapest first
Provider Country Configs Scale From Stock Days in stock Grab
Clore.ai — 2 1x-1x $0.13 In stock 14/14 Grab
OctaSpace — 1 1x-1x $0.14 In stock 14/14 Grab
Vast.ai United States of America 2 2x-4x $0.17 per sec In stock 60/60 Grab

In stock first, then cheapest. Stock is what the provider reported at our last check. One GPU for a month (730h) at $0.13: $93.

Price history

Price trend

Cheapest Nvidia RTX 5070 rate has fallen 8.0% since Sep 6 — $0.14 → $0.12/GPU/hr.

$0.125 /GPU/hr · Nov 4

One point per day: the lowest listed $/GPU/hr across every provider we track, captured daily. 54 offers in the latest snapshot.

Configurations

All 5 configurations

5 of 5 configurations
Nvidia RTX 5070 configurations from every provider
ConfigurationProviderRegionVRAMvCPURAMPer GPU/hrTotal/hrStock
1x NVIDIA GeForce RTX 5070SpotClore.aiClore.aiBY12 GB—16 GB$0.10$0.10In stock
2x RTX 5070SpotVast.aiVast.aiUS24 GB1663 GB$0.12per sec$0.24In stock
1x NVIDIA GeForce RTX 5070Clore.aiClore.aiES12 GB—32 GB$0.13$0.13In stock
1x NVIDIA GeForce RTX 5070OctaSpaceOctaSpaceUS12 GB2430 GB$0.14$0.14In stock
4x RTX 5070Vast.aiVast.aiCA48 GB32126 GB$0.17per sec$0.67In stock

Model fit

What fits in 12 GB

The question behind most rentals: will the model load. Sized against one Nvidia RTX 5070.

VRAM required to serve open-weight LLMs on a Nvidia RTX 5070, by quantization precision
Model 16-bit8-bit4-bit
DeepSeek-V3 DeepSeek · 671B params · 37B active 1,610 GB 135× · 10 GB free 805 GB 68× · 11 GB free 403 GB 34× · 5 GB free
Llama 3.1 405B Meta · 405B params 972 GB 81× · 0 GB free 486 GB 41× · 6 GB free 243 GB 21× · 9 GB free
Qwen3 235B-A22B Alibaba · 235B params · 22B active 564 GB 47× · 0 GB free 282 GB 24× · 6 GB free 141 GB 12× · 3 GB free
Mixtral 8x22B Mistral · 141B params · 39B active 338 GB 29× · 10 GB free 169 GB 15× · 11 GB free 85 GB 8× · 11 GB free
Mistral Large 2 Mistral · 123B params 295 GB 25× · 5 GB free 148 GB 13× · 8 GB free 74 GB 7× · 10 GB free
gpt-oss-120b OpenAI · 117B params · 5.1B active 281 GB 24× · 7 GB free 140 GB 12× · 4 GB free 70 GB 6× · 2 GB free
Qwen2.5 72B Alibaba · 72.7B params 174 GB 15× · 6 GB free 87 GB 8× · 9 GB free 44 GB 4× · 4 GB free
Llama 3.3 70B Meta · 70.6B params 169 GB 15× · 11 GB free 85 GB 8× · 11 GB free 42 GB 4× · 6 GB free
Qwen3 32B Alibaba · 32.8B params 79 GB 7× · 5 GB free 39 GB 4× · 9 GB free 20 GB 2× · 4 GB free
Gemma 2 27B Google · 27.2B params 65 GB 6× · 7 GB free 33 GB 3× · 3 GB free 16 GB 2× · 8 GB free
Mistral Small 3 Mistral · 24B params 58 GB 5× · 2 GB free 29 GB 3× · 7 GB free 14 GB 2× · 10 GB free
Llama 3.1 8B Meta · 8B params 19 GB 2× · 5 GB free 10 GB 1× · 2 GB free 5 GB 1× · 7 GB free

Required memory is weights plus 20% for KV cache and runtime at short context; long contexts and large batches need more. Cells tinted green fit on a single Nvidia RTX 5070.

Specifications

Nvidia RTX 5070 specifications

Nvidia RTX 5070 hardware specifications
ArchitectureBlackwell
Memory per GPU12 GB GDDR7
Memory bandwidth672 GB/s
Release dateQ1 2025
FP4 compute rate493.9 TFLOPS (dense), 987.8 with sparsity
FP8 compute rate123.5 TFLOPS (dense), 246.9 with sparsity
FP16 / BF16 compute rate61.7 TFLOPS (dense), 123.5 with sparsity
INT8 compute rate246.9 TOPS (dense), 493.9 with sparsity
Process nodeTSMC 4N
Board power250 W

Alternatives

Alternatives to the Nvidia RTX 5070