Comparisons

Two GPUs, one table, no opinions.

Every comparison is generated from the same catalogue that sets the prices: memory, bandwidth, node sizes, what fits, what it costs. Pick a ready-made pair or build your own.

Browse all 47 GPUs See prices

Build one

Any two models in the catalogue.

First GPU Second GPU

Both cards run the same templates and the same tiers here, so the comparison is about the hardware and the price — nothing else changes.

Result

A full side-by-side: specifications, the three prices, what fits in memory at each precision, and the cost of the same hundred GPU-hours.

Any pair works, including two cards from different families. The URL is stable, so you can share it.

Ready-made

The 18 comparisons people actually make.

All models

Comparison	Memory	Spot price	The short answer	Open
H100 vs A100Hopper vs Ampere	80 GB vs 80 GB	$0.55 vs $0.19	H100 holds more, A100 costs less	Open
H200 vs H100Hopper vs Hopper	141 GB vs 80 GB	$0.75 vs $0.55	H200 holds more, H100 costs less	Open
B200 vs H200Blackwell vs Hopper	180 GB vs 141 GB	$1.69 vs $0.75	B200 holds more, H200 costs less	Open
H100 vs H100 PCIeHopper vs Hopper	80 GB vs 80 GB	$0.55 vs $0.45	H100 holds more, H100 PCIe costs less	Open
MI300X vs H100CDNA 3 vs Hopper	192 GB vs 80 GB	$0.99 vs $0.55	MI300X holds more, H100 costs less	Open
MI325X vs H200CDNA 3 vs Hopper	256 GB vs 141 GB	$1.09 vs $0.75	MI325X holds more, H200 costs less	Open
L40S vs A100 PCIeAda Lovelace vs Ampere	48 GB vs 80 GB	$0.15 vs $0.09	A100 PCIe: more memory and cheaper	Open
RTX 5090 vs RTX 4090Blackwell vs Ada Lovelace	32 GB vs 24 GB	$0.09 vs $0.06	RTX 5090 holds more, RTX 4090 costs less	Open
RTX 4090 vs A100 40GBAda Lovelace vs Ampere	24 GB vs 40 GB	$0.06 vs $0.09	A100 40GB holds more, RTX 4090 costs less	Open
RTX 4090 vs L40SAda Lovelace vs Ada Lovelace	24 GB vs 48 GB	$0.06 vs $0.15	L40S holds more, RTX 4090 costs less	Open
RTX PRO 6000 vs H100Blackwell vs Hopper	96 GB vs 80 GB	$0.19 vs $0.55	RTX PRO 6000: more memory and cheaper	Open
RTX A6000 vs RTX 6000 AdaAmpere vs Ada Lovelace	48 GB vs 48 GB	$0.09 vs $0.09	RTX A6000: more memory and cheaper	Open
A10 vs L4Ampere vs Ada Lovelace	24 GB vs 24 GB	$0.09 vs $0.05	A10 holds more, L4 costs less	Open
T4 vs L4Turing vs Ada Lovelace	16 GB vs 24 GB	$0.04 vs $0.05	L4 holds more, T4 costs less	Open
A100 vs A100 PCIeAmpere vs Ampere	80 GB vs 80 GB	$0.19 vs $0.09	A100 holds more, A100 PCIe costs less	Open
GH200 vs H100Hopper vs Hopper	96 GB vs 80 GB	$0.49 vs $0.55	GH200: more memory and cheaper	Open
RTX 3090 vs RTX 4090Ampere vs Ada Lovelace	24 GB vs 24 GB	$0.05 vs $0.06	RTX 3090: more memory and cheaper	Open
V100 vs A100 40GBVolta vs Ampere	32 GB vs 40 GB	$0.06 vs $0.09	A100 40GB holds more, V100 costs less	Open

Method

What a comparison here does and does not claim.

Published specifications

Memory, bandwidth, throughput, power and interconnect come from each vendor's datasheet, linked on the model page. Where a vendor does not publish a figure, the row is absent instead of estimated.

Our real prices

The three tiers come from the live catalogue — the same table the console bills from. A price change moves every comparison at the same moment.

Not a benchmark

We do not publish tokens per second or images per minute, because those depend on your model, your batch size and your serving stack. Memory decides what runs; bandwidth decides how fast it usually runs; the rest is your workload.

Get started

The cheapest comparison is an hour of each.

Spot instances start at $0.03 an hour and stop when you say so.

Deploy a GPU See all prices

Billed per minute from the moment the instance is reachable. Minimum credit $40, no subscription.