AMD MI250 vs Nvidia H100

The MI250 has 60% more memory, with about the same bandwidth. 1 provider lists it, against 57 for the H100.

At a glance

Memory

60% more on MI250

MI250: 128 GB

H100: 80 GB

Memory bandwidth

Within 5%

MI250: 3.20 TB/s

H100: 3.35 TB/s

Cheapest on-demand

MI250 is not sold on demand

MI250: --

H100: $1.80 /GPU/hr

Price comparison

Median price per GPU per hour by billing type, with the cheapest listing under it.

Billing	MI250	H100
On-demand	No on-demand listings	$3.69 /GPU/hr from $1.80
Reserved	-- /GPU/hr from $1.30 (12 mo)	$3.32 /GPU/hr from $1.53 (1 mo)
Spot	No spot listings	$2.31 /GPU/hr from $1.29

Heads up: A provider's own page may quote a different figure, for example on tax, region or a promotion. A monthly-only plan shows a derived hourly rate. Medians are across providers, one vote each, so a large catalog does not outweigh a small one. No median for reserved on the MI250: we only show one when at least 3 providers list the GPU on that billing type. Verify before provisioning. More on how we price.

Price history

MI250 vs H100 price history

Weekly median on-demand price per GPU per hour

Last 12 months

Aggregating historical prices...

Specs comparison

Spec	MI250	H100	Difference
Memory
Capacity	128 GB HBM2e	80 GB HBM3 94 GB on NVL	60% more on MI250
Bandwidth	3,200 GB/s	3,350 GB/s 2,000 GB/s on PCIe 3,938 GB/s on NVL	Within 5%
Bus width	8,192-bit	5,120-bit	60% wider on MI250
Compute · dense
FP8	Not supported	1,979 TFLOPS	H100 only
FP16 / BF16	362.1 TFLOPS	989.5 TFLOPS	2.7x higher on H100
INT8	362.1 TOPS	1,979 TOPS	5.5x higher on H100
FP32	90.5 TFLOPS	67 TFLOPS	35% higher on MI250
FP64	45.3 TFLOPS	34 TFLOPS	33% higher on MI250
Platform
Interconnect	Infinity Fabric 400 GB/s	NVLink 900 GB/s NVLink 600 GB/s on PCIe, NVL	2.2x higher on H100
Board power	500 W	700 W	40% higher on H100
Released	Q4 2021	Q3 2022	10 months newer on H100
Precision
Supported formats	FP16BF16FP32FP64INT4INT8	FP8FP16BF16TF32FP32FP64INT8

Figures are for each model's headline configuration. Where a form factor carries a different number, it is named under the figure.

Compute: Vendor peak figures, dense and per GPU. Not measured throughput.

Source: MI250 datasheet and H100 datasheet.

Where to rent them

Providers that rent these GPUs, with each one's cheapest rate. No provider we track offers both right now.

Provider	MI250	H100
Cirrascale USA	$1.30 /GPU/hr Reserved	--
HyperAI Hong Kong	--	$1.80 /GPU/hr On-demand
GPU.ai UAE	--	$1.98 /GPU/hr On-demand PCIe
Lium Distributed	--	$2.10 /GPU/hr On-demand SXM
Spheron Singapore	--	$2.65 /GPU/hr On-demand PCIe
Massed Compute USA	--	$2.73 /GPU/hr On-demand
Sesterce France	--	$2.75 /GPU/hr On-demand
Runpod USA	--	$2.89 /GPU/hr On-demand PCIe
Scaleway France	--	$3.21 /GPU/hr On-demand PCIe