vCU Hardware Registry

Global normalisation table · GA-maintained. Sole source of valid hardware SKUs for all flexibility bids.

vCU (Virtual Compute Unit): 1 vCU = one normalised CPU core at 100% utilisation for one hour. Settlement = vCU rating × utilisation % × duration hours × spot rate. Only GA-approved SKUs are valid for flexibility bids. Tenants and DC Operators cannot add their own SKUs.

RISC_V_AI hardware class added Jun 2026: Introduced for SpacemiT K3 (X100) and equivalent RISC-V AI SoC hardware carrying dedicated NPU silicon. RISC_V_AI SKUs are assigned to COMPUTE_AI marketplace category ($0.35-$0.55/vCU-hr). They use physical node isolation via cgroups v2. Both DTO and DCT mechanisms supported. This class is structurally distinct from GPU_AI but economically equivalent at the COMPUTE_AI rate band.

Hardware Classes

GPU_AI

GPU with AI silicon

NVIDIA Tensor Cores H100 H200 A100 L40S

CategoryCOMPUTE_AI

RISC_V_AIAdded Jun 2026

RISC-V AI SoC

SpacemiT K3/X100 with dedicated NPU 60 TOPS

CategoryCOMPUTE_AI

CPU_SERVER

Server CPU

Intel Xeon AMD EPYC DTO only

CategoryCOMPUTE_GPU

CPU_WORKSTATION

Workstation CPU

High-core-count processors

CategoryCOMPUTE_GPU

FPGA

FPGA accelerator

Programmable logic nodes

Categorycase-by-case

CUSTOM

Custom/other

Non-standard requiring GA assessment

CategoryGA Review

Approved Hardware SKUs

Hardware SKU	Class	Category	vCU Rating	AI/TOPS	Mechanisms	Active Bids	Benchmark	Updated
SPACEMIT_K3_32GB_2TB Eko Atlantic K3 fleet	RISC_V_AI	COMPUTE_AI	40 vCU/node	60 TOPS NPU	DTO + DCT	1,752	SpacemiT AI SDK Jun 2026	Jun 2026
NVIDIA_H100_SXM5	GPU_AI	COMPUTE_AI	5,000 vCU/unit	3,958 TOPS	DTO + DCT	1,840	MLPerf v4.0	Mar 2026
NVIDIA_H200_SXM5	GPU_AI	COMPUTE_AI	8,200 vCU/unit	3,958 TOPS	DTO + DCT	240	MLPerf v4.1	Apr 2026
AMD_MI300X	GPU_AI	COMPUTE_AI	4,800 vCU/unit	1,307 TOPS	DTO + DCT	180	MLPerf v4.0	Mar 2026
NVIDIA_A100_SXM4	GPU_AI	COMPUTE_AI	3,200 vCU/unit	312 TOPS	DTO + DCT	480	MLPerf v3.1	Jan 2026
NVIDIA_L40S	GPU_AI	COMPUTE_AI	3,800 vCU/unit	362 TOPS	DTO + DCT	320	MLPerf v3.1	Feb 2026
INTEL_XEON_P8480	CPU_SERVER	COMPUTE_GPU	4 vCU/core	–	DTO only	840	SPEC CPU 2017	Jan 2026
AMD_EPYC_9654	CPU_SERVER	COMPUTE_GPU	4 vCU/core	–	DTO only	480	SPEC CPU 2017	Jan 2026

DTO vs DCT mechanisms: DTO = Dispatch Time Optimisation (batch workloads deferred, node execution paused and context fully freed to disk). DCT = Duty Cycle Throttling (active serving and inference rates safely throttled via cpufreq DVFS on K3 RISC_V_AI silicon). RISC_V_AI nodes support both DTO and DCT. Legacy CPU_SERVER nodes support DTO only.