Intel Accelerator
INTEL ACCELERATOR
COMPUTE PLATFORMS

Start with what you are deploying

Facility infrastructure is derived from the rack platform. Where a current official facility power figure exists we apply it; where it does not, we say SOURCE REQUIRED rather than assume a number.

Demo data
NVIDIA · Blackwell
NVIDIA GB200 NVL72

Rack-scale liquid-cooled Blackwell system with 72 GPUs connected as one NVLink domain. Widely deployed reference for 2025–2026 AI factories.

GPUs
72
CPUs
36
Cooling
Direct-to-chip liquid cooled (rack-scale)
Rack architecture
NVL72 — 72-GPU NVLink domain, single rack
Networking
NVIDIA Quantum-X800 InfiniBand or Spectrum-X Ethernet + BlueField-3
FACILITY POWER REQUIREMENT
Approx. 120 kW per rack (design reference)
OFFICIAL NVIDIA SOURCE · SOURCE QUALITY HIGH
NVIDIA ecosystem referenceNVIDIA MGX reference
LAST UPDATED 2026-08
NVIDIA · Blackwell Ultra
NVIDIA GB300 NVL72

Blackwell Ultra rack-scale system: 72 GPUs and 36 Grace CPUs in one liquid-cooled NVLink rack, with a published enterprise reference architecture.

GPUs
72
CPUs
36
Cooling
Direct-to-chip liquid cooled (rack-scale)
Rack architecture
NVL72 — 72-GPU NVLink domain, single rack
Networking
NVIDIA Quantum-X800 InfiniBand or Spectrum-X Ethernet + BlueField-3 / ConnectX-8
FACILITY POWER REQUIREMENT
Up to 142 kW per rack (full rack electrical requirement)
OFFICIAL NVIDIA SOURCE · SOURCE QUALITY HIGH
NVIDIA ecosystem referenceNVIDIA MGX referenceNVIDIA DSX reference
LAST UPDATED 2026-08
NVIDIA · Rubin
NVIDIA Vera Rubin NVL72

Next-generation Vera Rubin rack-scale platform. Facility power figures must come from the current official NVIDIA specification — none is applied by default.

GPUs
72
CPUs
36
Cooling
Liquid cooled (rack-scale)
Rack architecture
NVL72 — 72-GPU NVLink domain, single rack
Networking
NVIDIA scale-up NVLink + scale-out InfiniBand / Spectrum-X (confirm generation)
FACILITY POWER REQUIREMENT
FACILITY POWER REQUIREMENT — SOURCE REQUIRED
NOT AVAILABLE · SOURCE REQUIRED
NVIDIA ecosystem reference
LAST UPDATED 2026-08
NVIDIA · Vera
NVIDIA Vera CPU Rack

CPU-centric Vera rack for data-processing and head-node capacity alongside GPU racks.

GPUs
0
CPUs
NA
Cooling
Liquid cooled (confirm variant)
Rack architecture
Rack-scale CPU compute
Networking
Scale-out Ethernet / InfiniBand (confirm generation)
FACILITY POWER REQUIREMENT
FACILITY POWER REQUIREMENT — SOURCE REQUIRED
NOT AVAILABLE · SOURCE REQUIRED
NVIDIA ecosystem reference
LAST UPDATED 2026-08
Customer defined · Custom
Custom GPU / AI platform

Define your own rack platform. Enter the design rack power you have been given by your OEM.

GPUs
NA
CPUs
NA
Cooling
Customer defined
Rack architecture
Customer defined
Networking
Customer defined
FACILITY POWER REQUIREMENT
Enter your design rack power
PROJECT RFP · SOURCE QUALITY MEDIUM
LAST UPDATED 2026-08
·
Other / not decided yet

Not decided yet. Configure by capacity and density; compute platform can be applied later.

GPUs
NA
CPUs
NA
Cooling
To be determined
Rack architecture
To be determined
Networking
To be determined
FACILITY POWER REQUIREMENT
Not selected — configure by density
NOT AVAILABLE · SOURCE REQUIRED
LAST UPDATED 2026-08
INTEL ACCELERATOR. All figures shown are demo data for structural demonstration. Nothing is presented as verified market intelligence.