AMD MI300X servers, bought and hosted in India.

The MI300X carries 192 GB of memory per GPU, 2.4 times an H100. Large models fit on fewer GPUs, and PyTorch and vLLM run on AMD’s ROCm stack.

Illustration of a 8-gpu oam platform, 1.5 tb of gpu memory with its lid removed.
GPU memory
192 GB HBM3 per GPU
GPU power
Up to 750 W per GPU
Server power
6 kW of GPU power for 8 GPUs, plus CPUs and fans
Rack space · cooling
8U typical · Air

Shipping Lead time, October 2026: Quoted per order · All GPU lead times

Good fit for

Who buys AMD MI300X.

GPU server hardware in detail.

The hardware

What an AMD MI300X server looks like.

8-GPU OAM platform, 1.5 TB of GPU memory. 192 GB HBM3 per GPU, up to 750 W per GPU, 8u typical of rack space.

  • Server power: 6 kW of GPU power for 8 GPUs, plus CPUs and fans
  • Cooling: Air

Hosting

Where an AMD MI300X server runs.

An 8-GPU MI300X server is in the same power class as an HGX H100 system. We place it against its full rated draw.

How managed hosting works

Yotta NM1 Navi Mumbai

  • Uptime Institute Tier IV certified
  • Dual utility feeds from separate substations
About Yotta NM1

Yotta D1 Greater Noida

  • Dual 220 kV feeds with an on-site substation
  • 48 hours of on-site backup power
About Yotta D1

AMD MI300X questions

Is MI300X a good alternative to H100?

For memory-bound inference, often yes: 192 GB per GPU versus 80 GB. Check that your stack runs on ROCm before you buy.

Does PyTorch run on AMD GPUs?

Yes, through ROCm. vLLM and most major inference frameworks support MI300X.

How much memory does an 8-GPU MI300X server have?

1.5 TB of HBM3 across eight GPUs.

Other GPU servers

Owning one

Talk to OwnGPU

Give your GPUs a managed home.

One line is enough. We’ll call or email you back.

I want to

Or email [email protected] · Privacy