weeteq CEO Dr Taner Dosluoglu will present at the 2026 Open Compute Project Foundation EMEA Summit in Barcelona later this month, introducing our work on AI data center power optimisation to the wider OCP community.
Taner’s talk, “How to Increase Tokens Per Watt by 30% — Without Compromise”, will introduce Ultra Edge® DIVA, weeteq’s synchronised dual-path power delivery architecture for AI workloads. The session covers the power delivery bottleneck constraining today’s HPC data centers, the architectural thinking behind DIVA, and the measured results that make a 30% improvement in tokens per watt achievable.
Wednesday 29 April 2026, 15:15 – 15:45. Barcelona, Spain.
Why this matters
As HPC data centers face unprecedented energy demands, power delivery networks are bottlenecked by fast load transients causing on-chip voltage droops. GPUs consume additional power due to the guard band compensating for PDN droop, and systems overcompensate with large capacitor banks and a 100mV regulation margin — wasting significant power before a single token is generated.
The industry’s response so far has been incomplete. Faster regulators (IVR / VRM) offer fast response and smaller components, but suffer higher losses and limited vin/vout – a net benefit of around 5%. Higher distribution voltage offers lower current and cost savings, but adds complexity and dynamic control challenges – a net benefit of around 1.5%. Existing power delivery innovations sacrifice efficiency for transient response.
The DIVA approach
DIVA eliminates the artificial boundary between higher voltages and on-chip power domains. The architecture pairs a high-efficiency bulk path at 91% efficiency with a high-bandwidth fast transient path at 81% efficiency to actively compensate for voltage droops in real time.
30% less processor chip power · 25% less overall consumption · 30% more tokens per megawatt
DIVA can work with existing IVR and digital LDO solutions. On-chip and in-package integrated voltage regulators can initially be used alongside the proposed parallel path delivery system, which matters for customers with existing investments in those technologies.
What Taner will unpack
The session goes into the technical detail behind the headline numbers: the High Voltage Matrix Transformer (HVMX) delivering estimated average current at greater than 90% efficiency; the FAST DCM path providing sub-10ns droop correction at the point of load with up to 100MHz droop compensation; and the Ultra Edge Controller synchronising the two paths and monitoring real-time deviation from ideal transient response.
At rack level, the numbers translate to 120kW per rack reduced to 90kW, and 13kW per compute node reduced to 10kW, for equivalent throughput.
Come and say hello
If you’re attending OCP EMEA, come and find Taner and Julie on the floor. We’d be glad to talk through how DIVA reduces voltage guardbands, removes the efficiency-versus-response trade-off, and unlocks 30% more tokens per watt.
