UK’s trusted IT infrastructure partner since 2003
Servnet
FinanceToolsConfiguratorGet in Touch
Composable infrastructure software

LIQID Matrix software
One control plane for GPUs, memory and storage

LIQID Matrix composable infrastructure software is the part of Liqid that turns pooled hardware into servers. Pick a host, choose how many GPUs, how much CXL memory and which NVMe devices it should have, and Matrix composes them as bare-metal devices. When the job is done, it gives them back to the pool. You can drive it from a GUI, a CLI, a REST API or the schedulers you already run.
POOL TYPES IN ONE UI: GPU, MEMORY, STORAGE
Seconds[src]
TO COMPOSE A BARE-METAL SERVER
UI · CLI · API[src]
WAYS TO DRIVE IT
NAMED INTEGRATIONS: K8S, SLURM, OPENSHIFT, VMWARE, ANSIBLE
Hot-swap[src]
REPLACE FAILED GPUS AND DIMMS IN SOFTWARE
LIQID DIRECTOR APPLIANCE PER BUNDLE
What it does

Compose, scale, reclaim

Liqid's own term for the approach is composable disaggregated infrastructure: GPUs, FPGAs, NVMe and storage-class memory are taken out of the server into enclosures, and Matrix “lives on the fabric”, configuring servers to explicit workload requirements in seconds[src].
Compose

Bare-metal servers to order

Spin up a precisely sized server from the GUI or API. Liqid's example of what conventional design cannot do: a 1U server with 30 GPUs and 100 TB of memory, composed on the fly[src].
Scale

Add resources without a rebuild

Add GPUs, CXL memory or storage to a running host as the workload grows. Liqid lists dynamic adds and removals without rebooting the OS[src].
Reclaim

Return idle devices to the pool

Reclaim and reassign in real time, so a GPU or terabyte that one host stops using is available to the next[src].
Recover

Swap failed parts in software

Power off slots and virtually replace failed GPUs and DIMMs non-disruptively[src].
Automate

OS deployment and orchestration

Automated OS deployment and cloud-automation integration are listed SmartStack features[src].
Observe

One view of every pool

GPU, memory and storage pooling in a single pane of glass, with RESTful APIs built in[src].

In short, LIQID Matrix is composable infrastructure software for bare-metal servers: it holds the map of which pooled device belongs to which host, and changes that map when you, a script or a scheduler asks.

Integrations

Matrix in the tools you already run

Matrix is meant to sit under your existing orchestration rather than replace it. Liqid lists native compatibility with Kubernetes, Slurm, OpenShift, VMware and Ansible[src]. Its Cisco brief adds RBAC, audit and API-based automation[src], and the Abaco page names Kubernetes and Slurm plug-ins and Nutanix NKP[src].
LIQID Matrix integrations and controlled resourcesIntegration names as Liqid publishes them; check version support for your stack on quote.LIQIDMatrixDRIVE IT WITHMatrix GUICLIREST APIORCHESTRATIONKubernetesSlurmOpenShiftAnsiblePLATFORMSVMwareNutanix NKPPOOLS IT CONTROLSGPUs and acceleratorsCXL memoryNVMe, FPGA, DPUIntegration names as Liqid publishes them; check version support for your stack on quote.
Where Matrix connects: how operators drive it, which schedulers and platforms consume it, and which device pools it controls. Sources: Liqid Matrix page, Cisco UCS brief and Abaco page.
Day and night

One pool, two shifts

Liqid names day/night workload shifts as a pooling use case[src]. A design studio running GPU virtual desktops by day and renders or AI batch jobs overnight shows why: the two peaks never overlap.
A 20-GPU Liqid pool recomposed between day and night workloadsRecomposition can be scripted through the Matrix API or a scheduler.VDI desktops (day)Render and AI batch (night)Idle in the pool00:0003:0006:0009:0012:0015:0018:0021:0001020GPUsFixed servers would need 32 GPUs to cover both shifts; one pool covers them with 18.Recomposition can be scripted through the Matrix API or a scheduler.
Illustrative schedule for a 20-GPU pool: 14 GPUs to VDI hosts from 08:00 to 18:00, then 18 GPUs to render and batch AI hosts overnight. Fixed servers sized for each shift would need 32 GPUs; Matrix recomposes one pool of 20 twice a day. Example figures, not a Liqid benchmark.
Refresh

Upgrade the server, keep the GPUs

Fixed GPU servers tie accelerators to the CPU generation they shipped with. Liqid's datasheet puts it plainly: instead of replacing an entire server to upgrade the CPU, buy a 1U server with new CPU and RAM, then compose the accelerators and storage from the old server to the new one[src].
Refreshing a GPU host with and without composabilityConceptual timeline based on Liqid's Matrix datasheet example.Year 0Year 1Year 2Year 3Year 4Year 5Year 6Fixed GPU serverCPU, RAM, GPUs replaced togetherServer + GPUsNew server + new GPUsComposable hostonly the server is refreshedHost gen 1Host gen 2Pooled GPUsre-composed to the new hostSame GPUs, same chassis, recomposed by softwarerefresh pointConceptual timeline based on Liqid's Matrix datasheet example.
Illustrative six-year view. With fixed GPU servers, the CPU refresh at year 3 replaces the GPUs too. With Matrix, only the host is refreshed and the same pooled GPUs are recomposed to it. Whether a GPU is still worth keeping at year 3 is a separate decision; composability makes it a choice rather than a consequence.

That changes how a refresh is planned. Use our IT hardware refresh planner to map host generations, and the new vs refurbished server calculator where a newer host is all you need.

Buying

What a Matrix deployment includes

ItemWhat Liqid publishes
Management appliance1 × Liqid Director 1U in every SmartStack and UltraStack bundle[src][src]
Host softwareThe HBA requires no additional software[src]; workload components (for example famfs on Abaco[src]) may still apply
LicensingNo general model published. A three-year slot licence on the MI350P system[src]; 30 licences, 36 months, for 30 GPUs in the Cisco design[src]
Support36- or 60-month software and hardware support on SmartStack bundles[src]
Host OSHBA hot-plug on Linux and Windows Server 2012 R2 or later, no desktop editions[src]; VMware and Nutanix hosts per the Cisco brief[src]
Supply & quote

Add Matrix to a new or existing Liqid estate

  • New estate: we size the bundle and the licence count to the devices you will pool.
  • Existing Liqid fabric: we quote additional chassis, devices, licences and support renewals.
  • We check the scheduler and hypervisor versions you run against Liqid’s integrations.
  • One quote, with the licence term and support path stated.

Get a Liqid quote

Tell us what you want to pool and which servers you run. We size it against Liqid's hardware compatibility list and quote the complete build.

Business enquiries only — please use your company email. Gmail, Hotmail and other personal addresses are not accepted.

We only ship hardware to the countries in the dropdown. Software and maintenance services are global.

We use these details to reply to your enquiry and prepare any quote. No marketing emails without your consent. Privacy Policy

No obligation · response within 4 business hours

FAQs

LIQID Matrix: common questions

What is LIQID Matrix?

LIQID Matrix is Liqid’s composable infrastructure software. It runs on the Liqid fabric and assigns pooled devices (GPUs, CXL memory, NVMe storage, FPGAs, DPUs and NICs) to bare-metal host servers, then reallocates them as demand changes. Liqid describes it as the only platform with a single UI for GPU, memory and storage pooling.

How do you manage Matrix?

Through its GUI, a CLI or the built-in RESTful API. Liqid lists native compatibility with Kubernetes, Slurm, OpenShift, VMware and Ansible, and Liqid’s Abaco page adds plug-ins for Kubernetes and Slurm and support alongside Nutanix NKP. Schedulers and automation tools can therefore request a GPU or memory allocation as part of a job.

Where does Matrix run?

On the Liqid fabric. Every SmartStack and UltraStack bundle Liqid publishes includes one Liqid Director, a 1U management appliance, and the fabric devices carry dedicated management ports. Liqid’s HBA datasheet says the adapter requires no additional software; workload-specific components, such as Micron’s famfs for shared memory on Abaco, may still be needed.

Can Matrix replace a failed GPU without downtime?

Liqid says Matrix can power off slots and virtually replace failed GPUs and DIMMs non-disruptively via software. In practice that means composing a spare device from the pool to the affected host: the failed card still has to be removed physically, but that no longer has to wait for the host to be taken down.

How is Matrix licensed?

Liqid does not publish a general licensing model. Its MI350P system includes a three-year slot licence, and its Cisco UCS reference design lists 30 Liqid software licences, 36 months, for 30 GPUs. SmartStack bundles carry 36- or 60-month software and hardware support; the MI350P system includes three years. We confirm the licence count and term on every quote.

Does Matrix support GPUs from more than one vendor?

Yes. Liqid lists multi-vendor GPU support as a SmartStack feature, and its compatibility list includes NVIDIA, AMD, Intel, Groq and NextSilicon accelerators alongside FPGAs and NVMe devices.

The Liqid range

More Liqid products we supply and quote

Keep exploring

Related Servnet guides, tools and hardware

Sources

Every specification on this page comes from Liqid's own product pages and datasheets, listed below with their dates. Where Liqid's web copy and a datasheet differ, we use the datasheet. Figures are confirmed against the current revision when we quote.

  1. LIQID Matrix Software · web page, read 29 Oct 2026
  2. Liqid Matrix CDI datasheet · dated 2 Nov 2023
  3. EX-5410P SmartStack datasheet (GPU) · dated 10 Dec 2025
  4. Cisco UCS C-Series with high-power GPUs (tech brief) · Nov 2025
  5. Micron Abaco, powered by Liqid · web page, read 29 Oct 2026
  6. HX-5160P-1C PCIe Gen5 HBA datasheet · dated 27 Oct 2025
  7. UltraStack with AMD MI350P datasheet · dated 13 Oct 2026
  8. LIQID UltraStack with AMD Instinct MI350P · web page, read 29 Oct 2026
  9. UltraStack L40S reference design (Dell PowerEdge) · dated 13 May 2024
  10. Liqid Hardware Compatibility List · dated 17 Aug 2026

LIQID, LIQID Matrix, SmartStack and UltraStack are trademarks of Liqid Inc. Servnet is an independent UK IT reseller; this page is our own guide to the range.

Talk to a UK specialist

Get expert advice or a no-obligation quote — servers, storage, networking, maintenance, finance and cloud. We reply the same working day.

or call 0800 987 4111