digital.auto
The First Open Benchmark for
Software-Defined Vehicle Platforms.
THE OPEN SDV BENCHMARK GIVES OEMS AND SUPPLIERS COMPARABLE, LOAD-TESTED EVIDENCE.
OPENLY PUBLISHED, METHODOLOGY FIRST, NO VENDOR AGENDA
WORKSTREAM
Platform Freeze is Closer than it Looks
Every SDV program reaches a point of no return. Once the SoC, OS, middleware, hypervisor, and network choices are locked, they travel with every vehicle for the next decade. A wrong combination means delayed SOP, rework, and supplier relationships that are very hard to exit.
The demand is real, but the standard does not yet exist. This initiative is hosted by digital.auto, the open and vendor-neutral SDV initiative, because no single company should own a benchmark the whole industry relies on.
The problem is scale: >10 SoC options × >5 OS × >5 middleware × >5 hypervisors × >5 network configurations. Well over 6,250 combinations to consider before integration enters the picture.
The question is not whether we should benchmark.
The question is how we benchmark in a way that every step of the decision process can rely on.
Automotive Industry already knows how to
YOU HAVE EURO NCAP. YOU HAVE VLTP. YOU DO NOT HAVE THIS.
No manufacturer tests every car in every crash scenario. A standardized synthetic test runs under identical conditions, producing a star rating every OEM, buyer, and regulator can compare.
Fuel efficiency works the same way. The WLTC drive cycle is synthetic, precise, repeatable, and accepted industry-wide as meaningful enough to make decisions on.
For SDV platform performance, none of this exists. Every OEM is on its own.
Start with Use Case. Derive The Tests
(Not the Other Way Around
MAKING AN AUTOMOTIVE RELEVANT BENCHMARK
Every benchmark starts with a real automotive workload.
We derive measurable requirements from actual vehicle functions and break them down into repeatable tests.
Example: Control Engineering
We show control engineering here, but this is valid in principle for steering, braking, active suspension, wiper control, etc. Typically we require sensor-to-actuator latency budget of 0.5–3 ms end to end, with nothing to spare.
Inside central compute: four cooperating real-time processes, 9 IPC boundaries. Every single one adds latency.
From that architecture: 12 synthetic tests isolating individual resource axes, and 5 composite tests covering full chains. Every test runs under 5 load profiles × 3 isolation modes.
Performance is only Part of the Story.
Usability, Flexibility, Maintainability are the Rest.
ASSESSING NON-TECHNICAL PROPERTIES
Integrate with a real automotive middleware stack and you find out quickly. Getting it running takes longer than it should. When something goes wrong, the system rarely tells you why. Could you actually observe what was happening, or were you reading tea leaves?
We track these dimensions alongside the performance numbers: integration effort, observability, reproducibility, vendor neutrality, and usability.
Performance tells you whether a stack can do the job. Everything else tells you whether you want to build with it.
Build the Standard Before Someone Else Does
GET INVOLVED
Several OEMs have already asked for this. If the industry does not define it together, someone will define it alone. Everyone else will spend the next decade living with those choices.
This initiative is pre-competitive and IP-free. Tests come from publicly documented automotive architectures. No participant exposes proprietary implementations. Results can be published or kept internal. Every organization shapes what gets tested and how.
What exists:
-
Test derivation methodology (v1.0)
-
Control engineering domain (12+5 test specifications)
-
IPC test framework running on two stacks.
What is left to be built:
-
load profile tooling
-
test implementations for scheduler jitter
-
compute, network and memory
-
a raw data pipeline
-
implementations on additional stacks
-
domain derivations for ADAS, Body / Comfort and Infotainment.
"Early participants shape the methodology before it hardens. That window closes, once the suite is established."
Dieter Schwarzmann, lead of work stream.
Meet the BYOD Workstream Chair
DRIVING THE INITIATIVE
Chair
Dieter Schwarzmann
Dieter has spent the past seven years working on software-defined vehicles, from control engineering through platform architecture. He is an adjunct professor at the University of Stuttgart (Institute for System Technologies), where he teaches control engineering at the master’s level. As project lead for the “Rolling Chassis”, a built-from-scratch electric vehicle prototype including ECUs and software, he saw firsthand how hard it is to evaluate platform performance before committing to a stack.
He initiated the Open SDV Benchmark because the gap between vendor claims and production reality needed a neutral, repeatable way to close it.
digital.auto Benchmark Workstream participants
