SOLIDWORKS 2019: Quadro RTX Series in Enhanced Graphics Performance Mode
Written on March 18, 2019 by William GeorgeIntroduction
Our past testing of graphics cards, in SOLIDWORKS 2018, wasn't all that impressive - because for a long time the CPU has been a limiting factor when it comes to graphics performance in SW. The 2019 version came out recently, though, and looks to change that story. It has a new feature, currently in beta status, called "Enhanced graphics performance". It can be enabled in the System Options, under the Performance section, and its description is pretty straight-forward: "Improves graphical performance which will scale with your graphics card." We looked at how this mode improves frame-rates when modeling on the previous generation of Quadro video cards, and now that we have the full line of new Quadro RTX cards available we are going to repeat the same testing in SW 2019 SP2.
Please note that the Enhanced Graphics Performance feature is in beta. We are testing it only as a preview of future performance. We recommend heeding this warning shown when you click to enable this option: "Pre-production version for quality evaluation. May be unstable and have critical bugs; should never be used in a production environment or to modify production data."
Test Hardware & Methodology
For my testbed system, I used the following hardware:
| Test Platform | |
| CPU | Intel Core i9 9900K |
| CPU Cooler | Noctua NH-U12S |
| Motherboard | Gigabyte Z370 AORUS 5 |
| RAM | 4x DDR4-2666 16GB (64GB total) |
| Video Card | NVIDIA Quadro RTX 8000 48GB NVIDIA Quadro RTX 6000 24GB NVIDIA Quadro RTX 5000 16GB NVIDIA Quadro RTX 4000 8GB |
| Hard Drive | Samsung 960 Pro 1TB |
| Software | Windows 10 Pro 64-bit (version 1809) SOLIDWORKS 2019 SP2 |
This platform is built around an Intel Core i9 9900K, with very high clock speeds, to avoid the CPU being a bottleneck in this testing. That processor also gives the best performance right now in SOLIDWORKS for general usage and modeling. More than enough RAM was included, to avoid that being a bottleneck of any kind, and a super-fast M.2 SSD was used for the same reason. For the video cards, we included the full Quadro RTX line this time around - for earlier cards, including the Quadro P-series, see our previous article.
The only options changed from default settings in SW 2019 SPs were the aforementioned "Enhanced graphics performance" mode being enabled and disabling use of "Large assemblies". No LOD settings or other graphics quality sliders were changed.
To perform the actual benchmarking, I used the same basic testing we've used here at Puget for analyzing graphics performance in SOLIDWORKS in the past, just updated slightly for the 2019 release: a mix of AutoIt scripts and SOLIDWORKS macros to set the different quality settings, load the relevant model, and record the average frames per second while rotating the model. To recorded the FPS, a macro is used with a timer to rotate the model 45 degrees to the left and right for a set number of frames. From the number of frames and the total time it took to render those frames, our software is able to determine the average FPS.
For test samples, we have utilized models available from GrabCad.com that provide a range of complexities based on the total number of parts and number of triangles - along with one extremely large assembly provided by the gentleman who organizes the CAD Monkey Dinner that we have sponsored at SWW for the last several years. Here are details about the models we use in our testing:
Steam Engine
by Ridwan Septyawan
80 parts - .26 million triangles
Spalker
by Andy Downs
364 parts - .5 million triangles
Lego Tower Bridge
by Daniel Herzberg
4372 parts - 40.9 million triangles
One note that I would like to make is that if you do not know how many triangles the models you work with have, the easiest method I know of to find out is to simply save the model as an .STL file. During the save process, a window pops up with information about the model including the number of files, the file size, and the number of triangles.
Results
There are a lot of data points that come from testing four models with four different graphics settings across a four video cards at two different resolutions. To present all of that information, we have put together charts showing the frames per second (FPS) we measured in each model + settings + GPU combination at both 1080P (1920x1080) and 4K (3840x2160):
Analysis & Conclusion
This is where I would usually try to sum up all the data above in a simpler, easy-to-understand graph... but there isn't really much point. All four of these cards maintained over 30fps in all of our testing, even with our most complex model and demanding settings. There is a difference in performance across the cards, but it is above the level I would consider to be my expectation for a good usage experience. As such, any of these Quadro RTX cards will serve SOLDIWORKS users very well! And that is good news, since the RTX 4000 is far less expensive than its siblings. For folks working with smaller assemblies, an older Quadro P-series card might be fine too.
Please keep in mind that these recommendations are for SOLIDWORKS 2019 with "Enhanced graphics performance" enabled, and that mode is still in beta. At some point it will be released as a full feature, presumably, and may eventually even be the default or only option - but right now, it should be used carefully for evaluation or testing and avoided for critical work.
Looking for an
Engineering Workstation?
Puget Systems offers a range workstations tailored for CAD, BIM, photogrammetry, and other engineering applications. Even in the most demanding situations, our workstations and designed to minimize downtime and allow you to work as efficiently as possible.
Great, thanks for this, any plans to test with SOLIDWORKS Visualize?
Not at this time. While it isn't exactly the same, we already cover GPU-based rendering via OctaneRender and Redshift - so for those who want to see how different cards stack up in that type of workload the info from those articles & recommended systems should help.
Ok it's just that when develop3d tested they saw strange results with the new AI denosing performance. This feature was added to 2018 SP3 onwards(check it's on under options) They tested the Quadro rtx 4000 with 2019 Sp1 but Visualize 2019 SP2 includes a newer version of the denoiser. In theory Turing gen cards should be quicker with tensor cores but develop3d found it slower. I'm wondering if it is drivers, software as I'm pretty sure the hardware will be way quicker than the previous generation long term
https://www.develop3d.com/w...
There is a lot more in these cards when the registry is done right.
https://www.tomshw.de/2019/03/22/workaround-von-igorslab-beschleunigt-solidworks-2017-mit-allen-rtx-quadros-und-der-rtx-titan-igorslab/
It's in German, so people who prefer it in an other language might want to run it to an online translator.
Igor know's what he is talking about, he's kind of very long in the business.
Maybe the translation didn't come through properly, but it looks like the article you linked to is about altering the Windows registry in order to get full support for newer Quadro video cards (the RTX series, specifically) on older versions of SOLIDWORKS (2017, in this case). That is a handy trick, certainly, but for 2019 it is not necessary since the Quadro RTX cards are already on the supported list for that version :)
hi
for GPU rendering is base core clock important, as I know.
So is this Redshift3D config good? (I need the best viewport-realtime performance)
GPU: ASUS Turbo GeForce RTX™ 2080 Ti 11GB GDDR6
CPU: 9900X i9 19.25MB Cache (3.5GHz Base Clock 4.5GHz TBM)
MoBo: ASUS Prime 299X Deluxe ll
VROC SSD: ASUS HyperX x16 + 4x Intel 256 GB PCIe SSD (in VROC RAID0 ) (I use for music composing too, read/s is important for me)
New ACER Predator XB273K, 4K,DCI-P3 color,DisplayHDR 400 certified $1300
EKWB Water cooling for CPU+GPU
The GeForce RTX 2080 Ti works well in Redshift, as long as you are on a recent version (we tested with 2.6.22 and it worked - but older versions did not). However, Redshift also scales very well with multiple GPUs... so if you want to optimize for fast rendering, having two or even more video cards might be worth considering.
However, your comment about needing the best viewport / realtime performance and the fact that you asked this on a SOLIDWORKS article (even though you mention Redshift) has me confused. Are you using SOLIDWORKS for modeling? If so, that program really needs a Quadro series video card (instead of a GeForce) for the best performance and full feature support.
Hi, I have been trying to enable the enhanced graphics mode on my rtx 4000 but it is greyed out. I added Solidworks as a program in the Nvidia control panel and installed the solidworks certified drivers, but no luck. I tried with my old p2000, and that still works with enhanced graphics. Any ideas as to what might be wrong?
Hmm, what driver version are you using with the RTX 4000?
412.16 but also tried 419.67
Hmm, that seems strange. I've got an RTX 4000 installed right now, with 412.16 (the driver version specified on the SW hardware compatibility website) running SW 2019 SP2 and the Enhanced Graphics Performance option is available and checked. May try doing a clean re-install of the 412.16 drivers?
Hey,
Thanks for all these helpful reviews. I'm in the market to upgrade my old Quadro card right now because I do need to be able to handle large assembly more often.
The result is pretty interesting, with most complex model(Tower Bridge), Quadro RTX5000 is about twice as fast as the old P6000/P5000. Any reason why? Is this because of the solidworks issue or because of the new tensor core in RTX? I mean the result of other model is reasonable, but I can't believe the lower end RTX4000 will outperform the old flagship in the most complex model test.
The thing is right now, I have a chance to buy the P6000 with the same price as RTX4000, but after reading the reviews, I got confused. Please help me make the decision, I love the large memory in P6000, so I really hope this is just a mistake or software/drive issue. Any chance you may update the driver or test it in other version of Solidworks maybe?
Thank you
I have not revisited this testing since the article, and I don't have any certainty on why the RTX cards performed so well... but remember that this is looking at performance in the new "enhanced graphics" mode. Last I checked, that is still a "beta" feature in Solidworks - so keep that in mind when deciding if performance in that sort of feature is most important to you or not.
Beyond that, though, the RTX series cards do bring some new features to the table that older models simply don't have. RT cores for hardware ray tracing are really nice in some applications (GPU rendering, primarily, as render engines begin to support it) and tensor cores are handy for machine learning inference. If you think that your workload could utilize the massive 24GB of memory on the P6000, though, that is definitely one area that it outshines the RTX4000 (and even 5000).
Thanks for getting me back. I looked at your review again, it looks like even RTX4000 with 8GB memory can obtain 30+fps in Tower Bridge 4K test. So I guess massive 24GB will be overkill on P6000.....
90% do not depend on tensor cores or RT cores...it depends on the basic cuda cores that had a huge upgrade in volta and Turing, compared the maxwell/pascal ones...
they separate the data path of integer and floating point units, so that they can be used concurrently...if you search they can do N Tflops + N Tops (integer AND float concurrently), while older cuda core can execute only one of the two in a moment...
The result is that those card had a great capacity in address manipulation (integer units) and do not stop the floating point execution...if you do some programming in many vectorial operation you find address manipulation...
For example extract a matrices W from a vector V and multiply each element for a constant K:
w[i, j] = v[ n_col * i + j ] * K
These new card can prepare the index n_col * i+j in integer units while multiplying other V extracted before v [(n_col-1) * i + j ] * K in floating point units (this is a basic example only to give a sense, in reality is more complex since is distributed between more execution units)...
This is why in some benchmarks per core performance seems doubled (and without modifying the source code)...
Another upgrade is the a stack Pointer and a program counter for every thread (while before it was for a group of thread)... This allows to execute better some vectorial work coded with an IF statement inside a loop (see their blog/news for a better explanation of this.. Type volta SIMT execution)...(this need some modification in source code to be exploited completely, probably a big part of why all the programs took about 1 year to be optimized)...when I see the presentation of volta and know of these improvement I immediately think that they will lead to some non standard gains