Architectural Design Guidelines Based on Computer Core Evaluation and Processing Paradigms ^†

Abstract

The architectural design process may utilize decision-making algorithms; usually, this takes a long time when using analog methods. Today, sequential processing (CPU) or parallel processing (GPU) works with architectural design in different ways, such as Graphical User Interfaces using CPU or ray-tracing rendering using GPU. Architectural design paradigms have not yet fully adapted to the processing computer core paradigm. Thus, this research proposes design guidelines to take the computing core paradigm and transport it to architectural design with a practical case that tests the core’s performance with a genetic algorithm (form-finding process). The genetic algorithm was tested on a computer based on a central processing unit (CPU), another on a graphics processing unit (GPU), and the last one on quantum computers (Quantum Gates). The guidelines operate these cores under the cloud computing paradigm, supported by new components made in Rhinoceros–Grasshopper 3D. The discussion shows the integration of the computation core paradigm with the current digital architecture design process.

1. Introduction

The architectural design process shows a relationship between the design procedure and the calculation with or without computational tools. This research seeks a comparison criterion to adapt the decision-making process in architectural design based on the advantages of three different computing paradigms. The first is sequential computing with CPU, the second is parallel processing using GPU, and the third is quantum processing using quantum processors (QPU).

A genetic algorithm (GA) implemented in different computing paradigms proposes how the architectural design process can improve considering the computer core processing. The genetic algorithm requires an intensive calculation process. The CPU processor works in Amazon Web Services (AWS) based in Intel, a GPU processor based in Nvidia was used in AWS, and quantum computing works with AWS Bracket with a Rigetti server. A graphic user interface (GUI) created in Rhinoceros–Grasshopper 3D allows connecting an API connection located in AWS to access different computing processors.

Usually, studies on architectural design are associated with division and task integration into the abstraction process. Likewise, the abstraction process reduces the time process, which is possible with parametric design technologies. In this way, Rhinoceros-Grasshopper incorporation in this core computation evaluation is crucial.

The specific information technology architecture based on CPU, GPU, and QPU where the GA is implemented is shown in Figure 1.

(Figure 1) shows the information technology (IT) architecture. Grasshopper and Rhinoceros 3D in the right-side work as the frontend, which means a GUI; for this purpose, there are two main areas: Grasshopper with the new component that works with send, receive, and transform data, and Rhinoceros 3D with the visual geometry. Another area is on the top-left side, which consists of the EC2, a virtual machine that allows us to use a CPU processor (based on Intel) and a GPU processor (based on Nvidia). The last area, on the bottom-left side, is the Rigetti server, where the quantum algorithm runs, but to use this service we have to connect through AWS Bracket.

The design process shown in this research has two main branches: the digital morphogenesis process (A) and the analog morphogenesis process (B). In this way, a process guided by the simulation of natural evolution was used (A.1).

1.1. Digital Morphogenesis

1.1.1. Guided by Simulation of Natural Evolution through Genetic Algorithms

In these processes, the algorithm produces a population that has the parameters as genes. Once the selection and mutation mechanisms work, the best individuals can reach the goal; in this way, analyzing the performance of each individual is necessary through a mathematical procedure [1].

1.1.2. Guided by Emerging Patterns through Autonomous Agents

In this process, a population of agents is created, and each agent can communicate with others through basic state rules and constant analysis of their environment [2].

1.1.3. Guided by Parameters and Geometry Transformation Processes through Visual Programming

The process uses computational thinking to create shape relationships and data transformations, usually via visual programming.

1.1.4. Guided by Animation Sequences through Dynamic Patterns

In these processes, design decisions are usually made based on the achievement of frames in animated sequences where time is a variable to make decisions.

1.1.5. Guided by Materials’ Behavior and Computational Physics

In these processes, the behavior of materials is usually analyzed with variables and physical laws such as gravity or Hooke’s law to then make design decisions based on these observations.

1.1.6. Guided by Manual Control of Digital Curve Points

In these processes, the designer determines the position of control points that model a shape in space; Bezier or Nurbs curves are often used for these processes.

1.2. Morphogenesis with Analog Resources

1.2.1. Guided by Physical Model Transform to Digital Model

The process can transform the physical model made by hand or another method in bits through points digitalization or 3D scanning.

1.2.2. From Free-Hand Drawing to Digital Model

The process uses free-hand drawings to transform into pictures using a scanning process or a digital camera.

2. Information Technology Architecture

2.1. Genetic Algorithm Architecture

The genetic algorithm was made with the following parts:

• Generating a population method:

The population is generated randomly with individuals that have binary-encoded genes. These genes represent parameters in binary format.

2.

Iterating with a stopping method:

The selection and evolution mechanism iteration reaches a parameter combination to target approximate results and finally stop.

3.

Individual evolving method:

The evolution and selection mechanisms have a closer relation. The best parent genes (parameter value) are determined according to a selection rule for evolution.

4.

Selecting the parent method:

The performance rule defines which individuals are better to adapt to environmental conditions.

5.

Applying the crossover and mutation method to parents:

This operation uses a gene selection rule that determines, through operations in pairs, which genes will be taken for the next generation and, in some cases, gene mutation.

2.2. Computation Processors in the Cloud

Amazon Web Services allows the use of different computation technologies through servers such as AWS EC2 service (Intel i7 CPUs, the NVIDIA T4 Tensor Core GPUs) and AWS Bracket using Aspen-10-25Q Rigetti processors used in this research.

• AWS EC2 based on CPU

In this system architecture, connectivity was developed through a Rest API created with the Express system, and then the genetic algorithm was implemented with NodeJs that, through the Google V8 engine, can make use of the CPU (Figure 2).

2.

AWS EC2 based on GPU

In this architecture, the node-rapids technology connects the Nvidia GPU with NodeJs through CUDA; this technology is exposed by using Rest API (Figure 3).

3.

AWS BRACKET based on Aspen-10-25Q Rigetti processors

This system architecture works with two cloud computing servers. First, Amazon Web Services Bracket allows the use of Python-based Jupyter Notebooks to control load execution and find out the status of computing tasks; second, the Rigetti Aspen 10 can access this from AWS Bracket (Figure 4).

2.3. Frontend Anatomy in Grasshopper–Rhinoceros 3D

A new component was developed to connect the server with Grasshopper 3D. This component considers the sending and receiving values in JSON (JavaScript Object Notation) format, and their subsequent transformation into inputs to be used within parametric design processes.

Figure 5 shows the Grasshopper 3D functions. The API function allows inserting data from the user into the Grasshopper 3D component; after that, the transform class uses this information to obtain data from the server with the API function. The API function can obtain information from the server and transform the server response into data to use. The output function takes the information from the transform function and gives the information that the Grasshopper 3D user needs; after that, this information is shown in Rhinoceros 3D.

3. Results

Architectural design currently does not differentiate between which computing tasks to direct its processes towards. The different morphogenesis processes require different computing capacities related to the following tasks:

A.1 Searches for the best combination and the best route.

A.2 Determination of states based on neighboring states.

A.3 Re-calculation based on a time frame.

A.4 Re-calculation based on picture frames.

A.5 Subdivision to generate analysis mesh.

A.6 Analysis via raytracing.

A.7 Statistical analysis of point concurrence.

In this research, the comparison of the “Search for the best combination” task was carried out using a genetic algorithm implemented in the CPU, GPU, and QPU architectures.

For the first case (CPU), sequential tasks were developed based on the use of a server with the Intel Core i7 processor (Figure 6).

For the second case (GPU), the combination tasks were distributed in the different cores of the GPU to obtain information about the performance of the offspring (Figure 7).

For the third case (QPU), the tasks were executed at the same time thanks to the quantum entanglement and superposition mechanisms (Figure 8).

The results between the different processors show that the quantum computing processor performed better in finding the best combination of genetic algorithm genes to achieve a performance goal (Figure 9).

The objective of implementing a genetic algorithm is to show how the different processing cores behave to determine strategies in architectural design processes. In terms of processing speed, the QPU had better performance than the GPU, while the GPU was superior to the CPU. With this observation, the following steps are proposed.

• Define the type of problem to solve according to the morphogenesis process to take.
• Analyze the compatibility of the problem with the type of processing paradigm to be used, considering the following.
  • If it is a general-purpose task and requires design platforms and low-intensity computational capacity, it is recommended that the CPU is used; preferably, the tasks should be executed sequentially and interdependently (Figure 10).
  • If the task can be divided as in the determination of the state in autonomous agents, the GPU should be used. Tasks are not sequential and interdependent [3] (Figure 11).
  • If the task can be divided and consists of reaching an objective solution in multiple ways, as in the case of searching for the best route to reach an objective, it is recommended to use the QPU. It seeks to take advantage of superposition and quantum entanglement to execute tasks with Qubits (Figure 12).
• Define the design problem to be solved according to a computational design process; the use of a visual programming system such as Rhinoceros 3D–Grasshopper, or Revit-Dynamo is recommended, or if possible, the use of explicit programming.
• The indicators for this guideline are computing execution time and decision-making time in the architectural design.

4. Future Work

This research proposes guidelines based on the evaluation of processing paradigms of the CPU, GPU, and QPU. The implementation of a customizable genetic algorithm through a component created ad hoc in the Rhinoceros 3D–Grasshopper computer system was used as an example of implementation and demonstration.

The main selection criteria for the processing core were the computing tasks’ divisibility, the task’s flexibility, and the need for intensive computing. Future research will expand the range of implementation examples focused on the creation and performance analysis of machine learning engines; in this way, the focus will be on neural networks applied in computing core paradigms and architectural design processes.