Runyard is a free hardware-aware AI model browser. You enter your CPU, GPU, and VRAM and it instantly shows every local LLM that will run on your machine, ranked by speed and quality.

How much VRAM do I need to run local LLMs?

8GB of VRAM runs 7B models like Llama 3.1 8B and Mistral 7B at Q4 quantization. 16GB unlocks 13B models. 24GB lets you run Mixtral 8x7B and Llama 3 70B at lower quantization.

What is the best local LLM for my GPU?

Use Runyard at www.runyard.dev — enter your GPU and VRAM and the Model Radar will rank every compatible LLM for your exact hardware, showing estimated tokens per second for each model.

Can I run Llama 3 locally?

Yes. Llama 3.1 8B at Q4 runs on any 8GB VRAM GPU. Llama 3.1 70B needs around 40GB VRAM at Q4, or an Apple Silicon Mac with 64GB+ unified memory.

RRunyard Index

Runyard / GPU Index / NVIDIA H100 NVL

Rental price index

NVIDIA H100 NVL

PCIe H100 with 94 GB, sold in NVLink pairs.

Prices collected 2026-10-11 · 6 providers, 19 configurations

Jump to:At a glance Pricing Changes AI models Buy vs rent Specs FAQ Alternatives

Cheapest H100 NVL →vs H100 SXM →

On-demandSpot

Index median price per GPU per hour, by billing type. Nothing is interpolated.

At a glance

Median on-demand

$2.80/GPU/hr

Cheapest on-demand

$2.06

GPU.ai

Floor, any billing

$1.29incl. spot

Per month at 720h

$2,018

NVIDIA H100 NVL pricing by provider

One card per provider, cheapest first. A configuration is one priced unit — an instance type, a SKU, a marketplace offer — and the per-GPU price is that unit's price divided by its GPU count. Open a card to see every configuration with its vCPU, RAM and region where the provider publishes them.

In stock only6 providers · 17 configurations

GPU.aiIn stock

UAE2 configs1xcollected 2026-10-11

On-Demand from $2.06

From

$2.06 / GPU / hr

On-DemandVisit website →

Vast.aiIn stock

USA9 configs1x–4xcollected 2026-10-11

On-Demand from $2.33

From

$2.33 / GPU / hr

On-Demand · USVisit website →

RunPod Community CloudIn stock

USA1 config1xcollected 2026-10-11

On-Demand from $2.59

From

$2.59 / GPU / hr

On-DemandVisit website →

Massed ComputeIn stock

USA2 configs1x–2xcollected 2026-10-11

On-Demand from $2.92

From

$2.92 / GPU / hr

On-DemandVisit website →

RunPod Secure CloudIn stock

USA1 config1xcollected 2026-10-11

On-Demand from $3.19

From

$3.19 / GPU / hr

On-DemandVisit website →

Microsoft AzureIn stock

USA2 configs1x–2xcollected 2026-10-11

On-Demand from $6.98Spot from $1.29

From

$6.98 / GPU / hr

On-Demand · eastusVisit website →

Recent price changes

Every time a provider's median for the NVIDIA H100 NVL moved between two consecutive snapshots. A listing appearing or vanishing is not a price change and is not shown.

When	Provider	From	To	Change
2026-10-11 09:14	GPU.ai	$2.79	$2.23	-20.3%
2026-10-11 09:14	Vast.ai	$2.47	$2.40	-2.8%
2026-10-10 08:53	GPU.ai	$3.11	$2.79	-10.3%
2026-10-10 08:53	Vast.ai	$2.74	$2.47	-9.7%
2026-10-09 09:31	GPU.ai	$2.50	$3.11	+24.4%
2026-10-09 09:31	Vast.ai	$3.02	$2.74	-9.3%
2026-10-08 09:24	GPU.ai	$3.11	$2.50	-19.6%
2026-10-08 09:24	Vast.ai	$2.54	$3.02	+18.9%
2026-10-07 09:10	Vast.ai	$2.93	$2.54	-13.6%
2026-10-07 09:10	GPU.ai	$2.89	$3.11	+7.6%

Spot, reserved and where it is cheapest

Spot median

$1.2954% off · 1 providers

Reserved median

—mostly by quote

UAE providers

$2.23median of 1

USA providers

$3.02median of 5

What fits on a NVIDIA H100 NVL

Weights + KV cache at 8K context + runtime overhead, against 85 GB usable of the 94 GB on the card. Capacity follows total parameters, even for mixture-of-experts models — any expert may be needed next. Where one card is not enough, the table says how many are, and prices the set at today's median.

Model	Params	4-bit	8-bit	16-bit
Kimi K3	2.8T104B active	19× GPU$53.25/hr	36× GPU$100.89/hr	67× GPU$187.77/hr
DeepSeek V4.1-Flash	552B16B active	4× GPU$11.21/hr	7× GPU$19.62/hr	14× GPU$39.24/hr
Llama 4 Maverick	400B17B active	3× GPU$8.41/hr	6× GPU$16.82/hr	10× GPU$28.03/hr
gpt-oss-120b	117B5.1B active	✓ 67.9 GB~1088 tok/s · $0.72/M tok	2× GPU$5.61/hr	3× GPU$8.41/hr
Llama 3.1 70B	70B	✓ 44.5 GB~79 tok/s · $9.85/M tok	✓ 79.5 GB~42 tok/s · $18.54/M tok	2× GPU$5.61/hr
Gemma 4 31B	31B	✓ 21.2 GB~179 tok/s · $4.35/M tok	✓ 36.7 GB~95 tok/s · $8.19/M tok	✓ 65.7 GB~50 tok/s · $15.57/M tok
Qwen3 Coder 30B-A3B	30.5B3.3B active	✓ 19 GB~1681 tok/s · $0.46/M tok	✓ 34.3 GB~890 tok/s · $0.87/M tok	✓ 62.9 GB~473 tok/s · $1.65/M tok
Qwen3.8 27B	27B	✓ 18.7 GB~205 tok/s · $3.8/M tok	✓ 32.2 GB~109 tok/s · $7.14/M tok	✓ 57.6 GB~58 tok/s · $13.42/M tok
gpt-oss-20b	21B3.6B active	✓ 13.7 GB~1541 tok/s · $0.51/M tok	✓ 24.2 GB~816 tok/s · $0.95/M tok	✓ 43.9 GB~433 tok/s · $1.8/M tok
Llama 3.1 8B	8B	✓ 6.9 GB~693 tok/s · $1.12/M tok	✓ 10.9 GB~367 tok/s · $2.12/M tok	✓ 18.4 GB~195 tok/s · $3.99/M tok

tok/s is decode throughput from memory bandwidth, derated 20%, single stream — an estimate, not a benchmark. $/M tok is the hourly median divided by that rate: the one-user floor for a million generated tokens. Set cost uses today's median × GPUs required.

Buy or rent a H100 NVL?

Approx. list price

$28,000not tracked

Rent, per month

$2,018at median, 720 h

Break-even

9,991 h≈ 13.9 months

Rent for fewer than about 13.9 months of continuous use and renting wins on the card alone; past that, owning does — before the server, power, cooling and the person who keeps it running, which for a datacenter card are not small. The list price is a typical quoted figure and is not tracked by the index.

Specifications

Memory94 GB HBM3

Memory bandwidth3,900 GB/s

ArchitectureHopper

VendorNVIDIA

ClassDatacenter

Released2023-03

Bandwidth is listed because it governs generation speed: each token streams the active weights once, so tokens per second is bandwidth divided by bytes read. TFLOPs decide prompt processing and training, not how fast text appears.

Alternatives and comparisons

H100 NVL vs H100 SXMPrice, memory, what fits, cost per token

NVIDIA GH20096 GB · $3.45/hr median · 4 providers NVIDIA H100 SXM80 GB · $3.50/hr median · 43 providers NVIDIA A100 80GB80 GB · $1.79/hr median · 35 providers

Questions

How much does it cost to rent a NVIDIA H100 NVL?

As of 2026-10-11, the median on-demand price across 6 providers is $2.80 per GPU-hour, about $2,018 a month at 720 hours. The cheapest we saw was $2.06 at GPU.ai.

Why do prices for the same GPU vary so much?

Hyperscalers bundle CPU, RAM and network into an instance price and charge for the ecosystem around it. Neoclouds sell the GPU more directly. Marketplaces are independent hosts competing on price, with the trade-offs of shared, third-party hardware. The per-GPU figure is the instance price divided by GPU count, which is the only way to compare an 8-GPU node with a single-card listing.

Is the median the price I will pay?

No — it is the middle of the market. Each provider's own median is taken first so a provider with many instance sizes counts once, then the median across providers. Spot and reserved rates are excluded from it; they set the floor shown separately. Use it to judge whether a quote is high or low, then verify with the provider.

Which models can a NVIDIA H100 NVL run?

With 94 GB per GPU, the table above shows which reference models fit on one card at 4, 8 and 16-bit, using the same memory arithmetic as the rest of Runyard: weights plus KV cache plus runtime overhead, with capacity following total parameters even for mixture-of-experts models. Where a model needs more than one GPU it says how many and what that set costs at today's median.

Own the hardware instead? Work out what your card holds with the same arithmetic.

Check what fits →