You can scroll on ahead. On this page, you will find:

For more details head Down the Rabbit Hole.

Which Computer Should You Buy For Architecture in the United States?

Step 1 - Workflow

Do you use Revit + AutoCAD + Rhino + SketchUp + 3DS Max + VRay + Adobe + Microsoft Project + Microsoft Office? This constellation of tools describes the majority of architectural workflows in the US. Within this constellation, a few tools will push the limits of whatever computer you get - Photoshop, V-Ray, Rhino, Revit. Some programs won’t, especially Microsoft Office. Since Revit is most likely to dominate both time and push your computer the most of these programs, it makes sense to focus on Revit model performance what making a purchasing choice. That usage doesn’t describe all users, so we will return to user Types in a moment.

If we are purchasing based on Revit, then we have to land on the level of complexity in the typical model your firm produces. Larger models, like detailed hospital and large airport terminal buildings, will require more power than a six unit apartment building. Single family homes are the easiest of all, unless you model with lots of details and iterations.

So, a typical designer working on most projects is a Type 1 user. They may use other tools in their workflow, but moderately complex Revit models will drive the purchase of computer hardware.

Type 2 users add on the ability to produce quick renderings in Lumion or Enscape, and maybe even a simple virtual reality (VR) walkthrough - but without too many trees and entourage details.

Type 3 users can produce more detailed real-time, interactive, and VR renderings. They may produce more iterations of sketch renderings before sending the render file to a service, or they may produce final images locally.

Type 4 users are primarily visualization artists producing final quality images in V-Ray or another CPU-based rendering package.

Type 5 users are visualization artists like Type 4, but they use a GPU-based rendering program like Octane or Redshift or V-Ray GPU. Their systems are also adept with BIM.

Step 2 - Type of Architecture Computer

Once the workflow is chosen, or whatever software dominates your particular workflow in terms of computing power and time is understood fully, then you match that users workflow to the type of computer. One way to think of these systems is on a continuum - where all computers start with the base Designer / Project Manager workstation and then have various things added on to them. One danger in thinking this way is that rather quickly the entire computer chassis / case needs to be swapped, so it may just be better to think of them as discrete types of workstations.

The need to spend more on individual workstations decreases as the firm gets large enough to justify purchasing servers to handle some rendering work. In some cases, to host entire teams of workstations in a VDI server farm that designers can access from any relatively modest computer and even over the internet. Those sorts of systems only start to make sense when firm headcount passes 20 or so designers. There are systems like DesignAirSpace that work with smaller teams, but in my opinion these technologies are not ready to supplant traditional hardware purchases. Fully virtual users are described as Type 0, and if you want to go that direction for your 5+ person firm, you will need someone able to run your VDI system on a permanent support basis.

Starting with the Type 1 workstation, we find a reasonably priced computer that can run pretty complex Revit models. There isn’t a point in spending much less than $2,500 on the system, since the value and longevity will diminish so rapidly and early replacement will cost more than a slightly larger capital outlay. But, for the fairly significant capital, one gets a fairly significantly capable machine.

Upgrade the Type 1 GPU to make a machine capable of running very detailed or complex models and even perform VR walkthroughs.

Upgrade the Type 1 CPU and create a machine that can do more CPU rendering on the desktop while maintaining modeling abilities. Continue by upgrading the GPU and you have a do-it-all or freelancer machine. Adding one or up to three additional GPUs with a processor upgrade creates a GPU rendering focused machine.

Upgrade the Type 3 CPU further to create a CPU rendering machine four times faster (eight times faster than the base Type 1), but you do give up a significant amount of modeling capability. If you downgrade the GPU in such a machine you have a render node.

Upgrade the Type 1 CPU a bit and the GPU quite a lot and you have a GPU visualization workstation that can also handle modeling tasks well. These machines are the most expensive but they can produce final quality images without relying on render nodes, cloud rendering, or any other machines. The simplicity of iterating with final quality directly on the artist’s computer is a great advantage if you can afford it.

Although these five roles are distinct, the machines that support the roles can be fluid. Based on the workflow of your specific firm, you can fluidly move between the types in order to support people who fluidly move between these roles.

 Step 3 - Specification Soup

In order to effectively match the Type of user to a specific computer, you unfortunately have to know a bit about computer components. There’s plenty more in the Computer Components section Down the Rabbit Hole. Here’s the short version.

CPU / Processor / Central Processing Unit:

Type 1 / Type 2 / Type 3

Tier 1: Look for AMD Ryzen 5600xfor Type 1 and Type 2. The Ryzen 5900x is a nice upgrade. Choose the 5950x for Type 3 and top performing systems.

Tier 2: The 10900K / 10850K is Intel’s top tier, even or just behind the options from AMD. They are in the second tier due to their aging platform and high power requirements. The Intel 10600k or 10700k are great choices for modeling workstations. Last generation’s AMD Ryzen 3700X, 3800X, are 3900X are good options especially for more rendering-oriented workflows, but these lack in modeling performance compared to cheaper Intel options like the 10600K.

Tier 3: Several years old now but still great for modeling workstations are the Intel 9900K / 9900KS / 9900KF, and the 8700K / 8086K. AMD’s 3950X performs about as well in modeling applications, while its rendering performance is quite a bit faster, but at double the price.

Tier 4: Much older systems like AMD’s Ryzen 2xxx and 1xxx CPUs, Intels 7700K and below and older 6xxx models are in this range. You would need to be in dire financial straits to consider these for a new system purchase for your main tool.

For rendering-focused systems - Type 4 / Type 5

Tier 1: AMD Ryzen Threadripper 3990x for Type 5 and AMD Ryzen Threadripper 3960x for Type 4

Tier 2: AMD Ryzen Threadripper 3970x

Tier 3: Intel Xeon and Core Extreme CPUs are good rendering options, but they are very expensive compared to their competitors from AMD.

GPU / Video Card / Display Adapter:

Type 1 / Type 5

Tier 1: The brand-new NVIDIA RTX 3060 Ti and RTX 3070 are great choices, performing about on par with 2018’s flagship RTX 2080 Ti. Most 3D modeling viewports perform better with NVIDIA GPUs, but if you know that your viewport does just fine with AMD, the new 6900 XT, 6800 XT, and 6800 are extremely compelling choices. The RTX 3080 is very fast, but you likely won’t notice the increased speed nearly as much as the increased cost.

Tier 2: Last generation’s RTX 2070 Super, RTX 2080 Super, and RTX 2080 Ti are also good choices. AMD doesn’t offer many choices that compete with these cards. These cards are plenty for Revit models of moderately high complexity. The Quadro RTX 4000 is a more expensive option at this performance level but is only useful if you require first-rate support from Autodesk and your computer vendor.

Tier 3: AMD’s Radeon 5700XT is a fine choice if you can run AMD with your workflow. If you don’t know, avoid it.

Type 2 / Type 3

Super Tier: The NVIDIA Quadro A6000 48GB is the top of the heap. Able to scale up to four GPUs and can run absolutely massive scenes at top speed. If you might transition to rendering final images on GPU, consider purchasing an A6000 to start with. Also excellent for VR as it can hold every detail and tree model you can throw at it.

Tier 1: The NVIDIA RTX 3090 24GB is an incredible card. Consider a blower style card (like the Gigabyte Turbo) if you’re planning on using two GPUs. If you want to use 3 or 4 GPUs, step up to a AMD Ryzen Threadripper 3960x CPU and blower cards become a necessity. You’ll also need to consider power requirements, cooling, and case, as well as room cooling and noise. AMD’s new 6900XT, 6800XT, and 6800 are great cards with a healthy 16GB of VRAM - worth considering if your software works with AMD.

Tier 2: NVIDIA RTX 3080 10GB is constrained in terms of scene complexity but for users rendering single buildings it should be enough. Look for new models with higher VRAM in the next few months.

Type 4

Tier 1: NVIDIA A6000 48GB

Tier 2: NVIDIA RTX 3090 24GB, NVIDIA RTX 6000 24GB, NVIDIA RTX 8000 48GB. Four of any of these cards will provide very quick final scene renders. The older RTX 6000 and RTX 8000 cards are much easier to find than the brand new RTX 3090. You’ll need a blower style card to use multiple GPUs - the RTX 6000 and RTX 8000 are made by PNY and come only in that style. Most RTX 3090 cards are unsuitable, excepting the Gigabyte Turbo.

Tier 2: NVIDIA Quadro RTX 5000, NVIDIA RTX 3080, NVIDIA RTX Titan. AMD’s new 6900XT, 6800XT, and 6800 are great cards with a healthy 16GB of VRAM - worth considering if your rendering software works with AMD, although most prefer NVIDIA.

RAM / Memory:

32GB of RAM is a good baseline. Look for 64GB if you’re loading large models with lots of context or working across many models and applications at the same time. As far as speed goes, faster is better and 3200 MHZ is a sweet spot for price to performance. For the Ryzen 5950X, the absolute best performance is achieved at 4000 MHZ, but at significant cost for quantities sufficient for 3D modeling.

For Type 4 and Type 5 systems, look for 64GB as a minimum with 128GB preferred. Generally, having 2GB of RAM per CPU core is a good idea in a render system.

SSD / Storage / Hard Drive:

Look for a PCIe 3.0 or PCIe 4.0 (bandwidth) NVME (connector) SSD. These are the fastest available.

Since you’re likely storing models on a server, 256GB is fine for most users but 500GB is not much more expensive to have a great deal of breathing room.

If you are working locally without a NAS or a server, look for a 1TB or 2TB model. The Samsung 970 EVO is a great drive at a good price. The ultra-fast WD Black SN850, Samsung 980 Pro, and Sabrent Rocket 4 don’t provide a substantial enough benefit in architectural workloads. Most file opening and saving operations are still limited by CPU.

Add an internal hard drive and run Windows Backup for an added layer of protection against downtime.

You Should Probably Use This Computer

Return to Top

Buy This Display

Return to Top

Two displays is the ideal workspace for efficiency. I like 4K (aka UHD) screens at 125% or 150% scaling for the clear lines and details they show. With recent price drops, you can now have 4K for about the price that used to max at WQHD (2560x1440 pixels).

Top Upgrade Picks

Return to Top

If you want the best that can be bought for an architectural workflow, choose this set up. With cloud rendering as inexpensive as it is currently, a dedicated render node only makes sense if you’re doing more than 10 projects per year that include in-house final rendered images. One accounting bonus of cloud rendering is you can add it to your contract as a reimbursable and transparently pass the cost to the client - harder to do if you already own the rendering computer.

Boxx

Other Great Vendors

Other excellent builders include gaming-focused Maingear, Origin PC, and value-focused gaming builders CyberPower and iBuyPower. Puget Systems offers the best knowledge of the industry, and if you can forgo on-site support, offer far better value than Boxx, high quality components, and better support agents than Dell.

The Office Set Up

Return to Top

If you are more than one or two people, it makes sense to start optimizing employee computers for modeling and offloading high resolution rendering tasks onto dedicated machines in the office. A good render node starts under $2,000 and a very high performance render node can be had for about $7,000-$10,000. It would be wise to wait until your cloud rendering budget climbs to about $2,500 per year before jumping into buying Nodes. Even then, you may want to stick with cloud-based rendering or move to GPU-rendering, or just keep your options open as GPU-based renderers mature into production-ready speed demons.

Mix the Modeling workstations with the Render Only workstation above. Add more Render workstations if you need more speed. Laptops at desks are usually not worth the tradeoff in price unless you very much need portability. If you just need something to bring to client meetings, get a recent Lenovo Thinkpad P1 or the XPS 17 and share it across the whole office. If you need to have the mobility, expect to pay about $1,000 more than the equivalent desktop, limit your future upgrade options, and live with a definite ceiling on performance (avoid realtime rendering and there won’t be a way to get great VR performance). If you’re interested in laptops as desktops, check out this Technology Connections YouTube video on Desktop-Based Laptop Setups.

Store Your Files on This NAS

Return to Top

If you have another person in your office, you need to share your files. Store them on this. Backup your NAS with Dropbox or Backblaze B2. Why a NAS rather than relying on Dropbox directly? Because you need a copy of all your files where you can see and control them. These files are your work - you should know where they are at all times. The flexibility of having everything on Dropbox is a real boon to workflows, and it keeps a backup copy of everything online as well. Dropbox is the most reliable online service useful for architecture since it can upload just changes in large files (very useful for frequent saves to large files).