Performance per watt matters
AI infrastructure is constrained by energy, cooling and economics as much as chip benchmarks. A system that delivers more useful work for less power can be more important than one that wins a narrow speed test.
The Long View
Advanced Computing
The next computer may look less like a faster version of the last one.
RESEARCHVISION
The idea
The next computer may look less like a faster version of the last one.
AI accelerators, specialized chips, distributed systems and quantum computing are all changing the boundaries of what is practical.
Our research interest is broad but not mystical: which workloads benefit from specialized hardware, what should run locally, what belongs in large shared compute, and where emerging architectures can provide a real advantage?
Quantum computing in particular should be approached as a tool for specific classes of problems, not a magic adjective attached to every future product.
Imagine broadly. Promise carefully.
Advanced Computing
eIntelligence
FindUnderstandCreateAct
What would have to be true
Faster is interesting. Useful per watt is more interesting.
Advanced computing is not one exotic machine waiting to replace everything in the data center. It is a continuing search for the right architecture for each workload—general purpose, accelerated, edge, specialized and, where the evidence supports it, genuinely new forms of computing.
USEFUL
Workload first
Start with the problem: training, inference, retrieval, simulation, networking or another task. The hardware decision should follow the workload rather than the excitement surrounding a particular chip or architecture.
TRUST
Energy per useful result
Compute cost includes power, cooling, embodied materials and infrastructure. Performance that ignores energy simply moves the problem into another column of the spreadsheet.
ECONOMICS
Heterogeneous by design
CPUs, GPUs, accelerators and specialized systems may work together. The software and scheduling layer becomes important because the best machine for one task may be an expensive heater for another.
SYSTEM
Research is not procurement
Quantum, photonic and other emerging approaches deserve careful experimentation without assuming they belong in production. A lab result and a dependable service are separated by a great deal of engineering.
Status
The future of compute will probably be plural.
Advanced Computing is a research and vision area. eFind can explore new architectures while continuing to use conventional systems wherever they are the best tool for the job.
New hardware is allowed to be exciting. The electricity bill is allowed to remain unimpressed.
Compute as a portfolio
The fastest computer is the wrong computer if the workload does not fit.
Advanced computing covers accelerators, specialized hardware, distributed systems and carefully bounded quantum research. The objective is not to collect exotic technologies; it is to match workloads to the most useful architecture.
Specialization is normal
Training, inference, search indexing, video processing, cryptography and simulation can favor different hardware. The infrastructure layer should be able to mix architectures without forcing every workload into the same box.
Quantum belongs where it earns the complexity
Certain optimization, simulation or cryptographic research may eventually benefit from quantum systems. That does not turn every server room into a cryogenic science project.