EFFICIENT SIMULATION SOFTWARE


POWERFUL HARDWARE ALONE DOES
NOT MAKE EFFICIENT SOFTWARE
PERFORMANCE STARTS IN THE CODE – NOT IN
THE DATA CENTER
MAKING ENERGY EFFICIENCY VISIBLE – WITH
EMA
Modern data centers oﬀer enormous computing power – but
is it being used eﬀectively? Our project experience shows:
Even the most expensive high-end hardware cannot reach its
full potential if the software isn't designed to leverage it. Idle
cores, ineﬃcient parallelization, or poor conﬁgurations often
lead to long runtimes and high energy consumption.
Eﬃciency doesn’t come from more hardware, but from its
deliberate and software-driven utilization.
We repeatedly encounter the same pattern: Teams go
straight for the most powerful – and usually most expensive
– hardware available on the market, expecting that
multi-core CPUs, high clock speeds, or GPU acceleration will
automatically result in faster runtimes. But without proper
software adaptation, those expectations often remain
unmet.
Today, runtime is no longer the only critical metric – energy
eﬃciency is equally essential, for both ecological and
economic reasons. With our open-source tool EMA (Energy
Measurement for Applications), we measure the energy
consumption of entire applications or speciﬁc code sections –
even without access to the source code. A breakdown by
components such as CPU or GPU enables well-founded
optimization decisions.
Complemented by our Slurm plugin EPS, energy
consumption can also be analyzed per job across clusters –
ideal for monitoring and billing.
MEASURE – UNDERSTAND – OPTIMIZE
We are a team of specialized Software Performance
Engineers. Our work typically begins when systems fail to
deliver the expected performance – despite powerful
hardware. The bottleneck is rarely in the computing power
itself, but rather in the software. Factors such as
parallelization, memory access patterns, or the choice of
appropriate tools determine whether resources are used
eﬃciently.
VALIDATE, DON’T GUESS – WITH RTV
Whether it's runtime or energy savings: optimizations must
be veriﬁable. Our RTV (Reference Target Validation) tool
enables automated comparison of results – accounting for
numerical tolerances. This builds trust and helps prevent
regression errors during further development.
OUR APPROACH
• precise measurements
• in-depth analysis
• targeted optimization
Tobias Jaeuthe
OPTIMIZE
PERFACCT GmbH
CEO & Co-Founder
RTV
ANALYZE
VALIDATE
UNOPTIMIZED
SOFTWARE VERSION
OPTIMIZED
SOFTWARE VERSION
MEASURE
EMA, EPS
24
CONTINOUS OPTIMIZATION
Paperturn flip book viewer