PerfilGlobal designs, imports and deploys the supercomputing infrastructure that research depends on. From the first sketch to a running cluster.
We build with those who build the world's infrastructure
Who it is for
Research infrastructure with tax exemption and support from the proposal stage.
ViewGenomics, medical imaging and processing of sensitive clinical data.
ViewSimulation, fluid dynamics and computational engineering.
ViewDual-use projects, with restricted data and continental scale.
ViewCase · UFMG
A partnership between UFMG and the Brazilian Army will map the six biomes at five-metre detail. We sized and delivered the cluster that makes it possible.
Read the caseWho has worked with us
“We researchers do not have the know-how about market trends, price and quality, adapting technology to the structural dimensions of a project. It was, without a doubt, a two-way effort to buy the right machine at the right moment.”
“Without this supercomputer, we could not deliver what we set out to do.”
testimonial to collect
Lei 8.010/90
The law exempts from tax the import of equipment for scientific and technological research. We handle qualification, licensing and the entire process.
Value of the goods plus international freight and insurance.
State VAT is calculated on a gross-up basis.
Set by the equipment's tariff code.
Set by the excise tariff table, applied to customs value plus import duty.
Simulação ilustrativa, montada sobre a estrutura real de tributação na importação: II sobre o valor aduaneiro, IPI sobre valor aduaneiro mais II, PIS e COFINS a 9,25% sobre o valor aduaneiro, e ICMS calculado por dentro sobre o conjunto. As alíquotas de II e IPI variam conforme o NCM do equipamento, e a isenção de ICMS depende de o estado de destino ter ratificado o Convênio ICMS 93/98. O valor unitário de GPU é referência de mercado. A conta oficial do seu projeto é feita caso a caso.
From the outside it is a two-metre cabinet. What matters is inside, and that is where most projects get the maths wrong.
Processors, memory and storage per node. Alone, each solves little. Linked by a low-latency network, they become a single machine.
GPU, high-bandwidth memory and the path the data travels to reach it. When that path is narrow, nothing else helps.
Operating system, job scheduler, libraries, and the group's applications installed and tested. That is what separates delivered hardware from running science.
Solutions
Four fronts that combine according to the science you do. Click any one to see how it works inside.
HPC
Weeks of computation in hours of machine time. Compute and GPU nodes, parallel storage and a low-latency network, with a job scheduler and quotas per research group.
Artificial intelligence
GPU infrastructure to train, fine-tune and serve models inside the institution, without sending research data to a foreign cloud.
Data centre and network
We design the entire environment. In a large cluster, running costs usually exceed purchase costs, and that enters the plan from the first sketch.
Workstations
Before a project becomes a cluster, it starts on someone's desk. Workstations certified for the scientific software the group already uses.
How it works
Specification and budget to attach to your funding proposal.
Sizing based on the applications the team actually runs.
Qualification under Law 8.010/90 and import licensing.
Purchase from the manufacturer, international logistics and customs clearance.
Installation, HPC stack, scientific applications and acceptance benchmark.
Training, local support and expansion planning.
In numbers
three figures to confirm internally
We size, specify and import the right infrastructure for your research, at the right moment to buy.