I work with an MSP that services multiple architecture/engineering and medical customers. Providing systems with professional level 3D capabilites is crucial - customers look to balance cost with capability in order to provide the most usable system for a given staff member capability.
Unfortunately, there aren’t great resources for comparing hardware under professional workloads. Traditional graphics benchmarks favor consumer cards and games - the framerate achieved by a graphics card in Cyberpunk doesn’t tell me how quickly I can render a view in Autodesk Revit or Enscape.
Without some basis for comparison (besides just saying “trust me, I’m a professional”), I don’t have a good way of explaining the performance delta of a professional 3D application running on an Intel i7 vs. an i9, vs. a Threadripper. Other questions that I’d like to more definitively answer (in terms of professional software performance):
What kind of time/labor savings could different levels of video card produce in professional rendering applications?
What’s the performance difference between an engineering laptop and a tower workstation (from the customer’s point of view, since the same product names get used, doesn’t that mean a similarly spec’ed laptop and tower should perform the same)?
I’ve started compiling system benchmark scores for engineering systems that I have access to. So far, I’m running:
Cinebench r23 (for processor benchmarking)
EzBench (professional tools use the Unreal for rendering)
RFO Benchmark (for use with Autodesk Revit - scripts standard workflows and times how long it takes to complete - requires a valid license for Revit) RFO Benchmark v3
PugetBench for creators (for Adobe Creative Cloud applications)
Products that I’d like to use, but don’t have available licensing for:
SPECapc Benchmark for SOLIDWORKS (I have limited access to a Solidworks license, so this is unfortunately out)
Are there any other good tools that anyone could recommend?
You are absolutely right that almost all review sites tend to only use gamer or “creator” benchmarks for reviews which are completely different from CAE applications.
Techpowerup has started using some useful software in their hardware reviews as well, but only for their CPU reviews.
As much as people like to crap on it, passmark seems to be the most analogous to real word performance of professional applications when looking at the popular benchmarks. imo geekbench scores are completely divorced from meaningful reality.
Shoot - processor throttling has been a killer in laptops. A near ultrabook style system with an i9 and RTX video card just doesn’t have the capability to cool itself under heavy load. We’ve seen throttling on Dell Precisions when opening/browsing with Chrome.
I’ve done a lot of research in 3D cad computers for myself and other companies. (I have my own company doing some consulting)
There is a mix of needed requirements that are also based on the person working on them, a big difference is the cad software and usage. Inventor will have different requirements than Creo or solid works. Autocad will be limited by cpu more than Creo etc.
3D differences is a very varied one. Some applications don’t benefit at all while others do, but only depending on the workload.
I’m a big fan of using cinebench for comparing cpu’s together in single/multi threaded performance. but not more than that.
Video cards don’t really provide that much in professional rendering if you are working on a model. This is different than if you make visual renders, but people working in revit are usually just modellers.
Geforce GPU’s work just as well in autodesk suite of applications, this is a major saving in cost for computers. Even if companies like dell, hp and nvidia like to tell you otherwise. For some other applications like Creo it can make a difference.
For Inventor there is a benchmark called invmark that works well as a benchmark.
For autocad you have a cadalyst bench, which is ok.
Specviewperf is probably the best professional benchmark, but somewhat expensive.
Most Modelling software for people that make 3d models, depend mostly on single threaded cpu power and then a graphics card that is strong enough to not slow down showing their models. So for designing small parts a simple gpu already works fine.
As for laptops, i will generally spec game laptops for workstation use unless they are always at a single place or need a lot of multi threaded workloads. The workstations from dell or HP often don’t have cooling that is adequate to cool the laptop while not sounding like an airplane.
In the architectural companies that I service, at least, the push for designers to create 3D output has been significant. Enscape and Lumion are heavily used throughout all of the companies I work with. I’ve found that once Enscape gets introduced to an office, every designer will make some use of it.
re: consumer vs professional video cards - we only use pro level video cards now due to support issues we experienced years ago. We will still occasionally use consumer cards (I agree that performance can be as good or better than pro level cards); these are one-off situations though, and we won’t standardize on a consumer level card.
A lot of variables at play when it comes to 3D CAD performance, and not just limited to workstation hardware. I work as a systems engineer for a large Architectural Design and Construction company, specifically supporting around 70 designers (Lead Architects and Architectural Technicians) who mainly use Revit and associated Autodesk stuff such as Collaborate Cloud (BIM360) and Navisworks, but also Adobe CC, BlueBeam, BIMtrack etc.
Main fleet hardware spec is Dell Precision 5820’s (with recent addition of new 5860’s) many of which date back to 2018. This is a domain environment so “conservative” has been the common theme, but you can’t fault them spending top dollar on server-class hardware (Xeon’s, NVMe storage and RTX-grade GPU’s) under each desk.
With that out of the way, my point is that it’s not just about the hardware.
We have big choke-points with legacy document storage (DFS) and latency to BIM360 datacentre (270mS from here to Nth Virginia AWS) let alone inconsistencies around file management and optimisation (worksharing, linked files, families, templates etc). These are big projects, but I don’t see impacts of hardware performance (and I’ve measured them) on CAD workflows for the most part .
The Windows platform (OS) is another story though, as-is the Autodesk framework (see BIM360 above) let alone other dependencies such as impact of the chosen print-engine (required for exporting deliverables such as PDF’s, which is rendering after all).
Your best test is UAT - User Acceptance Testing. Ask an arch-tech who’s on Revit every day to try using a different platform and record their feedback. I guarantee you that they can tell you within 20 mins if it’s better or worse than what they’re usually using, and why. There’s no benchmark that’ll do that.
Finally, speaking of benchmarks (and having just done a whole pile on our physical machine fleet using RFO Benchmark 22 and 24) there are some more variables that you might not have considered. Firstly, all the Xeon machines (60+) performed worse than the high-performance laptops which is understandable (registered RAM for starters) but not that much slower. The laptops had 12th gen i9’s so they certainly have a speed advantage, but agreed there could be a cooling issue resulting in throttling after longer-term intense use, as someone has already said.
But I also noticed GPU has not as much impact as previously thought and in fact the system with RTX A3000 performed better on an identical system than the A4000 card … and also on the i9 systems, the IRIS GPU (on-die graphics) was heavily utilised, while the discrete GPU was left idling.
Finally, graphics configuration is also a factor: 2k vs 4k has obvious implications, but I hadn’t considered what the impact of RDP might be (most of our designers work some or all of the time remoting into their Revit machines) which then leverages Remote display adapter. Machines running headless (nothing connected but power and ethernet) ran almost as well as those with displays connected … and in fact the laptop workstations only enabled their discrete GPU when an external displayport was in use (ie the 17" 4k laptop screen was more than happy with the i9’s intel graphics, and performed just as as well).