ZXO / FLAGSHIP AI DESIGN STUDY

GB300
ROSSO.

A stronger presence.
A considered architecture.

Racing red. Graphite. An unmistakable ZXO signature. A full-size liquid-cooled workstation concept for the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip.

01 / 12THE FLAGSHIP EXTERIOR
Tall red and graphite ZXO GB300 ROSSO tower concept with black honeycomb intake, three red diagonal accents and ZXO emblem
ZXO GB300 ROSSO / EXTERIOR CONCEPT

Design concept · Engineering validation pending. NVIDIA figures describe the reference platform. The proposed ZXO enclosure, cooling loop and final configuration have not been validated as a complete system.

ONE PLATFORM.
MORE ROOM TO EXPLORE.

Reference sources ↗
Coherent memory
748 GB
252 GB GPU + 496 GB CPU
AI compute peak
20 PFLOPS
Up to · FP4 with sparsity
GPU memory bandwidth
7.1 TB/s
HBM3e reference
NVMe storage target
32 TB
4 × 8 TB · raw capacity

01 / A PURPOSEFUL BASELINE

The configuration,
clearly defined.

Start with one GB300 Desktop Superchip and a substantial local storage target. The table separates NVIDIA reference capabilities from the ZXO design choices that still require OEM and complete-system validation.

RECOMMENDED DEVELOPMENT TARGET

GB300 ROSSO / Core AI

1 Grace CPU · 1 Blackwell Ultra GPU · 748 GB coherent memory · 32 TB raw NVMe

Network / headless baseline. A local accelerated desktop requires an OEM-supported add-in graphics configuration.

Swipe to view all columns

Recommended ZXO GB300 ROSSO development configuration compared with the NVIDIA reference platform.
ComponentReference platformZXO ROSSO development target
Core platformNVIDIA GB300 Grace Blackwell Ultra Desktop SuperchipOne Desktop Superchip; OEM platform selection and integration pending.
CPU1 × Grace CPU; 72 Arm Neoverse V2 coresUse the reference CPU configuration. Validate Arm64 application and container support.
Primary GPU1 × NVIDIA Blackwell Ultra GPUDedicated core AI accelerator. No additional graphics card is shown in the interior concept.
GPU memory252 GB HBM3e ECC; 7.1 TB/s bandwidthKeep bandwidth-sensitive tensors in GPU-local memory where the software permits.
CPU memory496 GB LPDDR5X ECC; 396 GB/s bandwidthHost memory and coherently accessible capacity; distinct from GPU-local HBM.
Coherent linkNVLink-C2C; 900 GB/s reference bandwidth748 GB total across CPU and GPU memory. Actual placement and access depend on the software.
Local storageFour M.2 NVMe positions; generation and lane mapping depend on the selected OEM system4 × 8 TB NVMe = 32 TB raw. Confirm supported drives, physical fit, slot mapping, cooling and firmware before specifying a BOM.
Storage protectionStorage layout depends on the OEM system and softwareDefine OS, working data, checkpoints and backup separately. Usable capacity depends on formatting and any redundancy.
High-speed networkingConnectX-8; 2 × QSFP112, up to 400 GbE per port; up to 800 Gb/s aggregateSpecify compatible Ethernet switches, cables/transceivers and network settings for the deployment.
LAN & management1 × 10 GbE RJ45 + 1 × 1 GbE management RJ45Plan separate workload and management networks. Final rear I/O follows the selected OEM board.
Local displayPrimary host display uses a PCIe add-in GPU; BMC Mini DisplayPort is for management-console accessCore AI is a network/headless target. For a local accelerated desktop, specify an OEM-supported add-in graphics option.
Power planning1600 W reference system; 20 A circuit requirementConfirm the complete system's PSU, input voltage, cable, connector and site circuit with the OEM. This is not a measured ZXO power rating.
CoolingOEM-specific implementationProposed closed liquid loop with parallel CPU/GPU branches, front/top radiator zones and rear case exhaust. Thermal qualification is pending.
SoftwareUbuntu with NVIDIA AI Developer Tools; Arm64 platformValidate OS image, driver/CUDA versions, container images, frameworks and application licensing as one supported stack.
Product statusNVIDIA reference specifications describe the platformExterior and internal-layout concept. A final shipping configuration, price and availability have not been announced.

Reference basis: NVIDIA platform specifications and DGX Station documentation [1–3]. The enclosure renderings illustrate a design direction and do not establish motherboard dimensions, exact port placement or production component fit.

OPTIONAL / VISUALIZATION EXPANSION

Professional graphics,
on a separate memory pool.

A proposed visualization option is 1 × NVIDIA RTX PRO 6000 Blackwell Workstation Edition, with 96 GB GDDR7 ECC and a 600 W maximum GPU power specification. It is an optional target, rather than part of the Core AI baseline or the rendered internal layout. A local accelerated desktop requires an OEM-supported graphics configuration, which need not be this particular card.

Its 96 GB remains separate graphics memory. It does not turn the 748 GB coherent platform into an 844 GB memory pool. Applications, Arm64 drivers, PCIe support, physical clearance and OEM whole-system power and cooling support must be confirmed.

The reference platform shares power between the GB300 module and an optional RTX card. Supported power caps and OEM policy determine the operating combination. The 600 W card maximum does not establish simultaneous 1600 W plus 600 W operation.

Graphics [4] · Memory separation [6] · Power sharing [7] ↗

02 / UNDERSTAND THE ARCHITECTURE

One coherent system.
Two memory tiers.

Coherence lets compatible software work across CPU and GPU memory. The total capacity is large; the two memory types and their access paths still have different bandwidth and latency characteristics.

GPU-LOCAL / HBM3e ECC

252 GB

7.1 TB/s reference bandwidth

Blackwell Ultra GPU memory for high-throughput local tensor access. Resident weights, activations and KV cache all compete for capacity.

CPU-LOCAL / LPDDR5X ECC

496 GB

396 GB/s reference bandwidth

Grace CPU memory provides host capacity and coherent access for supported workflows. Remote access does not have GPU-local HBM bandwidth.

252 GB + 496 GB748 GB coherent total

Capacity is the sum. Performance depends on where the data lives and how the workload accesses it.

Model fit needs headroom.

Raw model weights are only one allocation. Leave capacity for the operating system, runtime, KV cache, activations, batching and working data. Quantization and offload behavior are model- and framework-specific.

Placement affects response time.

A model can fit within the coherent capacity while its speed changes with CPU-memory traffic, quantization kernels and context length. Validate time to first token and sustained decode rate using the intended configuration.

Memory capacity, bandwidth, NVLink-C2C and coherency: NVIDIA references [1, 2, 6]. No ZXO throughput or model-size guarantee is implied.

03 / MATCH THE WORKLOAD

A peak is a starting point.
The workload is the test.

NVIDIA specifies up to 20 PFLOPS of FP4 Tensor Core performance with sparsity, or 15 PFLOPS without sparsity. These are precision-specific compute peaks, not measured application results for ZXO ROSSO.

01 / LOCAL INFERENCE

Large models.
Defined conditions.

Evaluate the exact model revision, quantization, context, batch size and concurrency. Measure prompt processing and token generation separately; a peak FP4 figure does not establish tokens per second or latency.

02 / TUNING & EVALUATION

Budget beyond
the weights.

Parameter-efficient tuning, evaluation and data preparation have different memory requirements. Gradients, optimizer state and activations can dominate training memory. Inference fit does not establish an equivalent training limit.

03 / DEVELOPMENT STACK

Build for
Arm64 Linux.

Check framework builds, container architecture, CUDA extensions, custom kernels and application licensing. An x86 Windows application does not automatically run on the Grace Arm64 platform.

NETWORKING

Plan the path,
not just the port.

The reference ConnectX-8 design provides two 400 GbE ports and up to 800 Gb/s aggregate Ethernet bandwidth. A single transfer is not automatically 800 Gb/s; link configuration, switches, cables, storage and application protocols determine achieved throughput.

Multiple connected workstations do not automatically become one larger coherent-memory system. Distributed applications need a supported software and network design.

SOFTWARE

A versioned stack.
A reproducible result.

Start with the OEM-supported Ubuntu image and NVIDIA AI Developer Tools. Record firmware, driver, CUDA, container and framework versions alongside the workload.

Confirm optional graphics support with the platform OEM and application vendor. A supported combination must include physical integration, PCIe behavior, drivers and power management.

Compute and connectivity references [1–3]; optional graphics [4]. Real application performance remains to be measured on a validated complete system.

04 / THE PROPOSED THERMAL PATH

Two cold plates.
Parallel liquid paths.

An internal layout study built around separate CPU and GPU cold plates. Supply and return headers create independent branches, with radiator and case airflow zones expressed in the enclosure.

Open ZXO GB300 ROSSO concept showing smaller Grace CPU cold plate, larger Blackwell Ultra GPU cold plate and separate supply and return headers with parallel branches
07 / OPEN LIQUID-COOLING LAYOUT

One Grace CPU, one Blackwell Ultra GPU. The interior is illustrative; board layout, tubing, fittings and mounting remain engineering targets.

01 / LIQUID

Supply. Branch. Return.

The proposed supply header feeds the CPU and GPU cold plates on parallel branches. Both return branches join the return header and pump/reservoir loop. Branch balance, flow rate, coolant and pressure drop require validation.

02 / HEAT REJECTION

Two radiator zones.

The concept shows front and roof radiator/fan assemblies. Core thickness, fan curves, radiator capacity and chassis pressure must be selected around the actual OEM thermal load; the rendering does not prove cooling capacity.

03 / SERVICE

Design for access.

A removable service panel and lower pump/reservoir zone express the service intent. Final drain/fill access, leak testing, electrical isolation, filters and maintenance clearances need a complete mechanical design.

Exploded ZXO GB300 ROSSO concept separating red enclosure panels and the proposed liquid-cooled internal core
08 / ENCLOSURE & INTERNAL CORE

Illustrative assembly and service relationships. No production tolerances or exact component fit are established.

Close-up concept of CPU and GPU cold plates with supply and return manifolds and independent parallel coolant branches
09 / COLD-PLATE & MANIFOLD STUDY

Separate supply and return headers. Connections, seals, flow balance and coolant compatibility require engineering qualification.

Illustrative airflow concept with cyan front intake and orange roof and rear exhaust around the liquid-cooled core
10 / PROPOSED CASE AIRFLOW

Cyan intake and orange exhaust show intended direction only. This is not CFD, a temperature map or measured test data.

THE NEXT ENGINEERING STEP

Validate the
complete system.

No measured temperatures, acoustic levels, sustained clocks or application benchmarks are claimed for this concept.

  1. Mechanical & electricalOEM board fit, PCIe clearances, supported PSU, site circuit, safety and service access.
  2. Liquid & airflowPressure/leak tests, branch balance, pump behavior, radiator curves and all component temperatures.
  3. Sustained operationDefined ambient, workload duration, clock behavior, measured input power and acoustics.
  4. ApplicationsNamed model/software versions, memory allocation, time to first token, decode rate and stability.

A commercial OEM GB300 Desktop implementation demonstrates closed-loop cooling as a platform approach [5]. It does not validate the ZXO loop, fan layout or performance shown here.

SOURCE NOTES / REVIEWED NOVEMBER 5, 2026

Follow the references.

  1. [1]
    NVIDIA DGX Station platform specifications ↗

    GB300 Desktop CPU/GPU, memory tiers, bandwidth, compute peaks, networking and reference power.

  2. [2]
    NVIDIA DGX Station development guide ↗

    Desktop Superchip architecture, coherent memory and platform development context.

  3. [3]
    NVIDIA DGX Station software stack ↗

    Ubuntu/Arm64, AI software, add-in graphics for the host display and reference storage-position details. Confirm final slot generation and lane mapping with the OEM.

  4. [4]
    NVIDIA RTX PRO 6000 Blackwell Workstation Edition ↗

    Optional graphics reference: 96 GB GDDR7 ECC and 600 W maximum GPU power; OEM integration remains conditional.

  5. [5]
    Supermicro Super AI Station OEM platform example ↗

    Official OEM GB300 Desktop implementation and closed-loop liquid-cooling context. This is a separate product from the ZXO design study.

  6. [6]
    NVIDIA GB300 mixed-coherency environment ↗

    Grace–Blackwell coherent-memory behavior and the separate memory of an optional PCIe graphics card.

  7. [7]
    NVIDIA DGX Station dynamic power sharing ↗

    Power allocation between the main GB300 module and optional RTX graphics; supported limits depend on platform policy.

START WITH YOUR WORKLOAD

Define your
next flagship.

Share the models, context lengths, concurrency, applications, storage and installation power available. Together, these requirements define the next configuration and validation step.