Software consultancy · Stockholm, Sweden

Engineering Performance at AI Scale.

Qdisc AB builds high-performance, scalable software — from AI HPC datacenters and orchestration to enterprise distributed systems and embedded edge devices.

Trusted on the systems behind

CoreWeaveFluidstackMastercard EricssonSwedbankIntel

High performance is the whole point.

We are a consultancy for teams who can't trade away speed, scale, or reliability. Three domains, one engineering philosophy — measure, optimize, and make it hold under load.

AI-Driven Software

GPU clusters, training infrastructure and the orchestration layer beneath modern AI — engineered to scale horizontally without losing a millisecond.

Embedded Systems

Real-time Linux, board bring-up and kernel-to-userspace drivers with sub-100 µs response — where the deadline is physics, not a service-level target.

Enterprise Distributed Systems

Mission-critical, high-throughput backends — real-time payments at thousands of TPS, redesigned for resilience and 3× the performance.

Competence areas

Deep expertise, proven on real clusters.

Drawn from 15+ years across telecom, payments, EV charging and AI datacenters.

Flagship focus

AI HPC Datacenters — Infra & Orchestration

The layer that turns thousands of GPUs into one reliable supercomputer.

Design and deployment of Kubernetes + Slurm clusters up to 4,000 GPUs on bare-metal and VMs — H100/H200 nodes over 400 Gbps InfiniBand, NVLink/NVSwitch fabric automation, and validation with NCCL, HPL and ClusterKit. Track-specific, fully automated provisioning of orchestrators, networking and hardening.

Kubernetes Slurm / Slurm-over-K8s Kubespray NVLink / NVSwitch InfiniBand / UFM NCCL · HPL · ClusterKit DGX SuperPod ArgoCD Ansible LustreFS · RoCE-v2 NVIDIA GPU & Network Operators

Observability & Reliability at Scale

A custom, horizontally-scaling Prometheus exporter tested on hundreds of nodes; GPU, InfiniBand and Slurm alerting; Mimir, Grafana and ECK — load-tested with K6 and Rally.

Prometheus / Mimir Grafana ECK Custom exporters

Performance Engineering

Reengineering for throughput and latency — including a rewrite that delivered a 50× boost — backed by profiling, tracing and large-scale log analysis.

perf · ftrace · LTTng valgrind · gdb Python · Pandas

Enterprise Distributed Systems

Real-time payment platforms — ISO20022, ISO8583, SWIFT, SEPA — redesigned for 3× throughput at 3000 TPS, with full-system containerization for portability.

Erlang / Elixir Java · JMS · MQ Oracle Docker

Embedded & Real-Time Systems

Preempt-RT Linux with CPU/IRQ isolation hitting <100 µs, Yocto/Buildroot images, NXP i.MX8 bring-up, and GPIO/ADC drivers bridged to Elixir/Nerves.

Preempt-RT Yocto · Buildroot Qt / QML FreeRTOS · VxWorks

Linux Systems Engineering

Kernel driver development, eBPF/XDP, DPDK, kprobes and kdump, plus custom distributions — Bootlin-certified across kernel, embedded Linux, Yocto and debugging.

eBPF / XDP DPDK kprobes · kdump Kernel drivers

Linux, everywhere.

Linux runs through every line of our story — from kernel modules and real-time patches to datacenter fleets. The depth is certified, not assumed.

Cloud-native & Kubernetes.

Kubernetes is the backbone of our AI HPC work — orchestrating GPU fleets, Slurm workloads and the operators that keep them observable and self-healing at thousands of nodes.

Kubernetes Slurm-over-K8s Kubespray ArgoCD GPU / Network Operators Prometheus Operator ECK Operator Docker · containerd · LXC Namespaces Ansible

Where we've delivered

A decade-plus on the systems that matter.

Industry leaders in AI infrastructure, payments, telecom and embedded — across 8 countries.

CoreWeaveHPC NVLink networking
FluidstackAI datacenter HPC
MastercardReal-time payments
SwedbankPayments DevOps
EricssonTelecom & Linux core
IntelWireless via SysDSoft
CTEKEV charging / EVSE
Herrmann UltraschallReal-time embedded
SK TelecomWiMAX integration
WavesatWiMAX chipsets
WiproWireless USB / Intel PDK
8countries on-site

AI labs served via Fluidstack clusters

Character.AI Mistral Pika Poolside MBZ / G42

By the numbers

Measured impact, not adjectives.

0+

Years engineering

0

Countries on-site

0

GPUs orchestrated

0

Payments TPS

0×

Peak perf gain

<100µs

Real-time response

Let's make it fast.

Building an AI cluster, scaling a distributed backend, or squeezing microseconds out of an embedded system? Qdisc AB is ready to engineer it.

Evenemangsgatan 24 · Stockholm 169 56 · Sweden