AMD

AMD · CDNA 2 · 2021

AMD MI250X.

Rent the AMD MI250X for AI training and inference. 128GB HBM2e, 95.7 TFLOPS FP16, deploy in 60 seconds.

128 GB
HBM2e
3.3 TB/s
Bandwidth
95.7
TFLOPS FP16
8× GPU
Infinity Fabric
Available now · 4 regions

Cloud rental

$1.35/hr

On-demand · per-second billing

$1.20$1.50/hr range

Or $0.94/hr reserved (30% off)

Deploy

One command. AMD MI250X in 60 seconds.

Skip the dashboard if you don't need it. SDK, Python, or cURL — copy the snippet, paste your API key, ship.

import Runcrate from "@runcrate/sdk";

const rc = new Runcrate({ apiKey: "rc_live_••••••••••••••••" });

const instance = await rc.instances.create({
  gpu: "mi250x",
  region: "auto",
  image: "runcrate/vllm:latest",
});

console.log(`SSH: ssh root@${instance.host}`);

Benchmarks

How AMD MI250X stacks up.

Side-by-side specs vs the closest alternatives. Bars are normalized to the highest value in each metric.

FP16

AMD MI250X95.7 TFLOPS
NVIDIA H100 PCIe102 TFLOPS
NVIDIA H100 SXM120 TFLOPS
NVIDIA A100 SXM78 TFLOPS

VRAM

AMD MI250X128 GB
NVIDIA H100 PCIe80 GB
NVIDIA H100 SXM80 GB
NVIDIA A100 SXM80 GB

Bandwidth

AMD MI250X3.3 TB/s
NVIDIA H100 PCIe2.0 TB/s
NVIDIA H100 SXM3.4 TB/s
NVIDIA A100 SXM2.0 TB/s

Full specs

AMD MI250X specifications.

Memory

VRAM128 GB
Memory typeHBM2e
Bandwidth3.3 TB/s

Compute

FP3247.9 TFLOPS
FP1695.7 TFLOPS
INT8383 TOPS
Stream processors14,080

Power & form factor

TDP560 W
Form factorOAM
ArchitectureCDNA 2
Released2021

Cluster

Max GPUs/node8
InterconnectInfinity Fabric
ManufacturerAMD
Available regions4

FAQ

Frequently asked.

Ready when you are

Deploy AMD MI250X
in 60 seconds.

Available across 4 regions · per-second billing · no commitments. Ship today.