- Factical loads burn up to 45% of energy just to maintain stable power levels
- Skeleton graphenegpu provides up to 40% more calculation with the same equipment
- The workloads have killed energy efficiency, graphenegpu Smooths Power Draw in real time
While artificial intelligence systems become more demanding, many data centers have found themselves consuming almost double the energy they technically need.
This overuse is not due to the faults of the system or obsolete equipment, it is rooted in the way GPUs behave, because their request for power can swing radically in a few seconds, from the complete accelerator to idle.
To cope, operators often deploy dummy charges, deliberate energy waste, to maintain a stable power draw – but these data centers deliberately slow down the performance of tens of thousands of GPUs to prevent power outages
Factical loads mean a massive waste of energy
Although this avoids damage and current failures, this means that up to 45% energy is lost as heat, making no useful calculation.
Skeleton Technologies now claims that it can have a more effective alternative, which allows GPUs to operate at full capacity without overwhelming the grid.
The Estonian company has developed Graphenegpu, a advanced shaving system using owners’ curved graphene supercakers.
Unlike lithium -based systems, these capacitors can respond in just 10 microseconds, absorbing energy during periods of inactivity and unloading instantly when the GPU loads.
The result, according to Skeleton, is the ability to maintain coherent GPU performance without emphasizing the grid or resorting to limitation.
Their tests suggest that the system can deliver up to 40% more flops – operations of floating points per second – using the same GPUs, simply by removing the performance penalties associated with thermal dedication and the instability of the energy.
“Graphènegpu offers up to 40% more calculation with the same energy imprint, while reducing capital and operating costs by reducing network upgrade needs, energy waste and cooling,” said Taavi Madiberk, CEO of Skeleton Technologies.
“Propelled by our patented curved graphene, this is a fundamental change in the way the AI infrastructure can evolve – permanently and economically”.
The company also reports a 44% reduction in the electrical capacity that data centers must book from the network.
The central unit, the PCS 50 graphenegpu, offers up to 80 kW of cutting -edge power in a standard 1 or factor of 1 or, compatible with the existing infrastructure and cooled by air or liquid.
Above all, it completely avoids lithium, using the patented material of Skeleton graphene.
According to Skeleton, this technology has been tested under rigorous hyperscular quality GPU workloads with positive results. However, it was not tested independently for the performance and sustainability of the real world.
The first expedition of this technology will start in Germany by June 2025. The company also has an American production site scheduled for the beginning of 2026.
