Speak With An Expert

Available now

LIQID UltraStack with AMD Instinct MI350P
In Stock.

Complete systems, 4 to 30 GPUs. Delivered in 4 weeks. Accelerate your AI projects.
liqid 10-slot gpu chassis 5410p with 10 amd instinct mi350p's inside
0
Weeks
lead time
Up to
0
%
less power vs traditional servers
0
→
0
GPUs on one server
0
GB
HBM3E per GPU
WHY LIQID + MI350P

Waiting On a GPU Server? You Don't Have To.

Available Now

  • In stock and shipping in 4 weeks.

No Waiting On Lead Times

  • A complete, tested system: server, LIQID chassis, MI350P GPUs and software. Start running while others wait.

Grow Without a Rebuild

  • Start from 4 GPUs. Scale up to 30 GPU's on the same server.
Waiting on a GPU server order
long lead time
LIQID MI350P System
4 Weeks
Reserve → connected with a LIQID partner → optional PoC demo → order → delivered in 4–6 weeks.
Size your system

How Many GPUs Do You Need?

4GPUs
AMD Instinct MI350P 1 Gen5 Chassis
Configuration
4 × MI350P · 1 LIQID Chassis · Server included
MSRP $280,950

3yr slot license, 3yr HW warranty and install all included.

576 GB
HBM3E
9.2 PF
FP8 dense
~4.95 kW
est. system power
6 RU
rack space
Air
cooling
4 weeks
lead time
Workloads that Fit
Contract review
Clinical summarisation
Triage

Your selection is added to the form.

What it runs

Private Inference For the Jobs You Already Have

See the system in action.Arrange a demo.
Arrange a Demo
Available while stocks last

Reserve Your MI350P GPUs

1

Reserve here

Pick your size, tell us what you'll run.

2

We connect you with a LIQID partner

Or with your existing one, if you have one.

3

See it run: Demo Optional

See a system in action

4

Order through your partner

Quote, order and support from one partner.

5

Delivered in 4 weeks

A complete, tested system.

A reservation is not an order and carries no commitment. All orders are placed through LIQID partners. Available while stocks last.

FAQ

Questions, answered

What's included?

You get a complete system, not a kit:

  • Server: a standard Supermicro dual-socket AMD EPYC 9005 server, not a custom build.
  • LIQID hardware: 1 to 3 EX-Series GPU chassis and LIQID HBA cards in the server.
  • GPUs: AMD Instinct MI350P GPUs, each with 144 GB of HBM3E and air-cooled.
  • AI stack: open source. vLLM runs on AMD's ROCm software.

I'm waiting on a GPU server order. Can this get me running inference workloads sooner?

Yes. The MI350P shipped in May. Server makers are still qualifying their own systems, so customers ordering through them see new deployments land around Q1 2027. LIQID + AMD orders ship in as soon as 4 weeks, to unlock your Q4 AI initiatives.

How do I buy?

Work with your LIQID account executive and/or an authorized LIQID partner. Each system is a fixed configuration, quoted after a short call about your workload, your models and your site.

Is the 4-week lead time from reservation or from order?

The 4-week lead time is from when the purchase order is received and confirmed by LIQID.

Can I start with 4 GPUs and add more later?

Yes. Start at 4 GPUs, then grow in pairs of 2, up to 30 GPUs on the same server. You add GPUs, not servers: no new network, no rebuild and nothing to migrate.

What power and cooling is required?

Air cooling only. There's no liquid cooling requirement and no facility redesign. Each GPU draws 600 W and can be set to 450 W.

  • 4 GPUs: about 2.4 kW for the GPUs
  • 8 GPUs: about 5 kW for the GPUs (an estimate)
  • 30 GPUs: about 22 kW for the whole system

We confirm your rack power and cooling on the first call.

What does a PoC demo involve, and is there a cost?

It is a LIQID + AMD system demo. Tests can run remotely on a LIQID-hosted system.