Dummy -Laadings kill your AI performance, the Grafenegpu from Skeleton can simply change everything
- Advertisement -
- Advertisement -
- Dummy loads burn up to 45% of the energy to keep the power levels stable
- Skeleton’s graphenegpu supplies up to 40% more computing with the same hardware
- Ai -workloads kill energy -efficiency, graphenegpu smooth out the power collection in real time smooth
As artificial intelligence systems become more demanding, many data centers Have noticed that they consume the energy they need technically almost twice.
This excessive use is not due to system errors or outdated hardware, it is rooted in how GPUs behave because their power question can drastically swing within a few seconds, from full throttle to inactive.
In order to make the head, operators often implement dummy loads, deliberate energy viss, to maintain stable electricity – but these data centers deliberately delay the performance of tens of thousands of GPUs to prevent power outages
Dummy -Laadings mean mass waste of energy
Although this avoids damage and blackouts, this means that up to 45% of the energy is lost as heat, not performing a useful calculation.
Skeletal technologies now claims that it can have a more efficient alternative, an alternative with which GPUs can work at full capacity without overwhelming the grid.
The Estonian company developed Grapheneegpu, a peak shear system using its own curved graphene-supercondensators.
Unlike lithium -based systems, these capacitors can react in just 10 microseconds, absorb energy during stationary periods and immediately discharge when GPU loads peak.
According to Skelete, the result is the ability to maintain consistent GPU performance without emphasizing the schedule or resorting to a throttle.
Their tests suggest that the system can deliver up to 40% more flops – floating point operations per second – with the same GPUs, simply by removing the performance buds that are associated with thermal rating and power instability.
“Graphenegpu supplies up to 40% more computing with the same energy-contest printing, while both capital and operating costs are discussed by reducing the needs of the net upgrade, energy waste and cooling,” said Taavi Madiberk, CEO of Skeleton Technologies.
“Driven by us patented curved graphene, this is a fundamental shift in how AI infrastructure can be scaled – sustainable and economic”.
The company also reports a reduction of 44% in the power capacity that data centers must reserve from the grid.
The core unit, the graphenegpu pcs 50, supplies up to 80 kW peak power in a standard 1OU shape factor, compatible with existing infrastructure and cooled by air or liquid.
It is important that the lithium completely avoids, instead with Skelet’s patented on graphene -based material.
According to Skeleton, this technology has been tested under rigorous Hyperscaler-grade GPU-Deskloads with positive results. However, it has not been tested independently on Real-World performance and sustainability.
The first shipment of this technology will start in Germany in June 2025. The company has also planned an American production site for the beginning of 2026.
Maybe you like it too
- Advertisement -