Aixia’s Data Center, based in Gothenburg, is set to become one of the first in Europe to receive NVIDIA’s groundbreaking new supercomputer, the DGX H100!

NVIDIA’s DGX platform is the premier solution for initiating and developing AI projects. Since 2017, Aixia has been delivering all versions of NVIDIA’s DGX platform. We have designed, built, and supported some of the largest AI-focused data centers in the Nordics. NVIDIA’s DGX platform provides a fully supported hardware and software stack, enabling companies to concentrate on generating value rather than spending valuable time constructing, testing, and configuring complex infrastructure.

The innovative DGX H100 is built on NVIDIA’s H100 (Hopper) GPUs, the next-generation successor to the A100 (Ampere). In certain aspects, the DGX H100 is up to 9 times more powerful than the DGX A100. Each DGX H100 is equipped with 8 NVIDIA H100 GPUs, boasting a total of 640 Gigabytes of GPU memory. With 4 NVIDIA NVSwitches, the DGX H100 can transmit 7.2 Terabytes per second between GPUs, 1.5 times faster than the previous generation.

 

2

Those familiar with GPU-based computing understand the substantial power consumption of GPUs, as well as the recent surge in electricity costs. The DGX H100 can consume over 10kW at its peak, significantly more than the DGX A100’s maximum consumption of 6.5kW. However, when considering performance per kW, the DGX H100 offers a considerably better return on investment.

As one of the pioneering companies in Europe, Aixia has recently taken delivery of its first DGX H100 unit.

Aixia will make the DGX H100 accessible to anyone interested in testing and evaluating this cutting-edge technology through our proprietary AI Orchestration software, AiQu. Aixia has a history of being at the forefront of innovation, as we were one of the first to offer DGX 1 and DGX A100 as cloud services. Aixia is an Nvidia DGX Elite Partner.

Stay tuned to our website and social media channels for updates on when the DGX H100 will be available for testing at our data center.

Latest News

Från pilot till produktion: MLOps-fällorna när AI ska skalas

De flesta AI-piloter når aldrig produktion. Här är de fem vanligaste MLOps-fällorna – spårbarhet, övervakning, modell-drift, resursdelning och ägarskap -…
Läs mer

GPU-compute i Norden: var ska AI-infrastrukturen ligga?

Var ska beräkningskraften ligga? Publikt moln, eget datacenter, colocation eller en kombination? Så fattar ni beslutet, och vad försvar, medtech…
Läs mer

Arista tar AI-nätverket till 1,6T: första switcharna levereras i höst

Arista lanserar sin första 1,6T Ethernet-portfölj för AI-nätverk, byggd på Broadcom Tomahawk 6. Vad betyder det för svenska datacenter -…
Läs mer

Storage för AI: varför GPU-kluster svälter utan rätt dataplattform

En GPU kan bara arbeta så snabbt som datan når fram till den. Här är varför AI ställer andra krav…
Läs mer