AI in Design: In Need of New Hardware
New hardware options are emerging to support artificial intelligence use cases.
Altair says its new DesignAI product “provides a cost-effective acceleration of engineering processes that integrates with existing engineering workflows and toolchains.” Image courtesy of Altair.
Engineering Computing News
Engineering Computing Resources
AMD
Dassault Systemes
Latest News
May 8, 2024
Just when you thought the buzz surrounding artificial intelligence (AI) couldn’t get any more pervasive, it seems engineering vendors are working overtime to bring AI to design. From initial design to simulation and detail drawing, companies across the gamut are adding AI features now.
The following three software companies are in three different aspects of product development. They are representative of what the entire industry is doing to bring AI to engineering:
Altair offers DesignAI, which it says combines “physics-based simulation-driven design and machine learning-based AI-driven design to create high-potential designs earlier in development cycles.”
Dassault Systèmes’ SOLIDWORKS is researching how to use AI for the complicated task of creating parametric models: “There is no company right now that can generate parametric models from picture or text input,” says Shrikant Savant, who works on AI and machine learning for Dassault Systèmes. “What we want to do is give users a generative AI model that works out of the box, but we also understand that customers would like to further customize the model by training it on their own data.”
Graebert recently released AI Assist for ARES Commander. The new feature is powered by OpenAI technology and trained on the specifics of both drafting and engineering. Sample requests could include “How do I calculate the wire gauge for a given amount of power?” or “I need technical advice on sheet metal blending.”
These new features—and many more like them—represent a significant evolution for product design software. These new powers come with their own set of hardware requirements, distinct from those of traditional CAD/CAE software. From power consumption to CPU performance and motherboard capabilities, AI-enhanced design will make new demands from workstations.
More Is Not Always Better
The transition to AI-enhanced CAD/CAE applications presents new challenges for the person shopping for engineering workstations. For example, high clock speeds rule for existing CAD/CAE work, where most of the processing is sequential. These applications benefit more from the speed of the cores than how many cores are in a CPU.
In today’s CPU market, existing CAD/CAE applications generally run well on Intel Core i7, i9 or the equivalent AMD Ryzen 7 or 9. For AI-enhanced CAD/CAE, more cores are better. Because AI is heavily dependent on parallel processing, AI-enhanced CAD/CAE workstations should have AMD Ryzen Threadripper or Intel Xeon series with high core counts (16, 32 or more), because they can better process parallel tasks more efficiently.
With graphics processing units (GPU), the key metric is newness. The latest generations from both AMD and NVIDIA include new capabilities designed with AI in mind. Traditional CAD/CAE does just fine with earlier midrange professional GPUs like NVIDIA Quadro or AMD Radeon Pro. AI-Enhanced CAD/CAE, on the other hand, will benefit from newer high-end GPUs with substantial CUDA cores and tensor cores (for NVIDIA) or Stream Processors (for AMD) and significant VRAM (8GB or higher). NVIDIA’s RTX A4000-A6000 or the Tesla series are specifically designed for AI tasks, offering tensor cores optimized for machine learning algorithms.
In a bit of AI product name irony, Tesla, the electric vehicle manufacturer, is working on a $1 billion supercomputer for AI development. The new supercomputer will be powered by NVIDIA Tesla-generation GPUs (and, to be fair, truckloads of AMD GPUs as well).
The differences extend to other workstation components. For traditional CAD/CAE, 16-32 GB of RAM is typically sufficient for most CAD tasks. Speeds of DDR4-3200 or higher can offer performance benefits. For AI-enhanced CAD/CAE, consider 64GB of RAM as the minimum, and get the fastest RAM your vendor offers. This ensures smooth multitasking and data handling for AI processes, which can be memory-intensive.
In traditional CAD/CAE, solid-state drives (SSD) are preferred over the spinning hard disk drives (HDD) for their speed. It is not uncommon to specify both. A 512 GB or larger SSD can serve as the boot drive, with additional HDDs used for bulk storage. AI-enhanced CAD/CAE will do better with the newest SSD technology NVMe, at 1 TB or more capacity. This will be the primary drive to accommodate large datasets and ensure rapid data access times. Other storage technology can be included for long-term data storage. Consider RAID configurations to ensure data redundancy and greater between-drive speed.
High-end GPUs use a lot of electricity and generate a lot of heat. Traditional CAD/CAE workstations do just fine with standard cooling solutions and a power supply of around 650-750 watts. AI-enhanced CAD/CAE workstations should come with advanced liquid cooling systems and power supplies of 1,000 watts or more.
It might be easy to overlook networking as a category for upgrade. AI can generate very large data sets. If you are using cloud storage or processing, you will want the faster transmission rate. Gigabit Ethernet is usually sufficient for traditional CAD/CAE. Consider 10 Gigabit Ethernet or faster for AI-enhanced CAD/CAE solutions. Also, be sure the workstations have the latest Wi-Fi standard (e.g., Wi-Fi 6) if wireless connectivity is important.
Finally, make sure the motherboard supports the latest I/O standards (e.g., PCIe 4.0/5.0) and offers plenty of expansion slots for additional GPUs, memory, and storage options.
Representative Models
Following is a sampling of workstations that meet or exceed the specifications suggested above.
Puget Systems Single GPU Tower Workstation for Machine Learning / AI
The base model ($4,782) is a midtower workstation that ships with either Windows or Linux. It is the model Puget Systems uses for development with Stable Diffusion, a generative AI model that produces unique photorealistic images from text and image prompts.
- CPU: AMD Ryzen 9 7900X 4.76 GHz 12 Core
- RAM: 64GB DDR5-5600 (2x32GB)
- GPU: NVIDIA GeForce RTX 4080 SUPER 16GB
- Open Air
- Storage: Two 1TB NVM3 PCIe Gen4 M.2 SSD (Peak 7000 MB/s)
- Power: Super Flower Leadex VII Gold 1000W (80 PLUS Gold)
- OS: Windows 11 Pro 64-Bit or (subtract $98.39) Ubuntu Linux 22.04 LTS 64-Bit Installation with Gnome Desktop
Dell Precision 7960 Tower
The base model of this top seller ($4,129) offers Intel Xeon and NVIDIA, and more than 100 options for storage, RAM and other components.
- CPU: Intel Xeon w5-3425 30MB cache, 12 cores, 24 threads, 3.2GHz
- RAM: 32GB (2x16GB) DDR5, ECC
- GPU: NVIDIA T1000 8GB, GDDR6
- Storage: 512GB, M.2, PCIe NVMe, SSD, Class 40
- Power: Not specified, not Energy Star qualified
Xi Computers MTower 2P64X
A dizzying array of options await the online buyer. The unit spec’d here ($6,195) is a midrange option with two CPUs and two GPUs. Xi recommends Supermicro motherboards.
- CPU: Two Intel 2nd Gen Xeon Scalable Refresh Silver, 10 cores, 20 threads, 2.4GHz
- RAM: 128GB DDR5-4800 (8x16GB)
- GPU: Two NVIDIA RTX A4000 16GB GDDR6 ECC, 6144 compute unified device architecture (CUDA) cores, 192 Tensor cores, 140W
- Storage: Primary is 4TB Samsung SSD 870 QVO 2.5 SATA 6Gb/Sec. Secondary is 4TB 7200 RPM SATA 6.0Gb/s WD Gold Datacenter
- Power: 1000W Gold Certified 80 Plus Corsair RM1000e
- OS: Windows 11 Pro 64-Bit or a variety of Red Hat Linux options
More Altair Coverage
More AMD Coverage
More Dassault Systemes Coverage
More Graebert Coverage
More NVIDIA Coverage
Subscribe to our FREE magazine, FREE email newsletters or both!
Latest News
About the Author
Randall S. Newton is principal analyst at Consilia Vektor, covering engineering technology. He has been part of the computer graphics industry in a variety of roles since 1985.
Follow DE
