Puget Systems print logo


Read this article at https://www.pugetsystems.com/guides/895
Article Thumbnail

SOLIDWORKS Visualize 2017 Quadro GPU Performance Comparison

Written on February 7, 2017 by Matt Bach


SOLIDWORKS Visualize (formerly known as bunkspeed) is a relatively new addition to the wide range of product offered by Dassault Systemes. Based on the NVIDIA Iray rendering engine, Visualize is able to utilize the power of both the CPU and the GPU to complete renders extremely quickly. However, the choice of your GPU (and the use of multiple GPUs) is much more important than your choice of processor when it comes to raw render times.

Dassault Systemes does not yet have a formal hardware requirements list or an official stance of whether they recommend you use a workstation or consumer card, but in this article we will be focusing on various NVIDIA Quadro cards to see how they perform relative to one another. A GeForce card should give you more performance for your dollar, but not only is there a high likelihood that Dassault Systemes will eventually decide to only certify workstation cards but Quadro cards are also more reliable. Overall, here at Puget Systems we see about half the failure rate with Quadro compared to GeForce. Considering the fact that GeForce cards are actually very reliable already, this makes Quadro cards ideal if uptime and reliability is more important to you than getting the best performance for your dollar.

If you are interested in how GeForce cards perform or how Visualize scales with multiple GPUs, we recommend reading our other Visualize articles:

Update 2/7/2017: Added results for the NVIDIA Quadro GP100 16GB

Test Setup

To see how the different Quadro cards perform, we used the following hardware and software:

Our test platform is based on the Dual GPU Workstation from our SOLIDWORKS Visualize Recommended Systems. We will be testing in GPU only mode, but to ensure that we do not have any CPU or RAM bottlenecks we are using the fastest Core i7 CPU currently available and have many times more RAM in the system than we need. The Quadro cards we are testing are a mix of Maxwell and Pascal based cards, but with the exception of the M6000 they are all the latest cards available at each price point. The M6000 was replaced by the P6000, but we decided to include it as a way to compare the older Maxwell architecture to the newer Pascal architecture.

We will be using two of the samples files provided by Dassault Systemes to see how quickly each card is able to complete a render:


1969 Camaro



The render settings used were:

  • 1920x1080
  • Accurate
  • 1000 passes
  • GPU only mode

Quadro Rendering Performance

SOLIDWORKS Visualize Quadro Render Benchmark GP100
The two projects used in our testing certainly rendered at very different speeds, but the relative performance between the different video cards was actually very consistent across both of them. Putting these results into relative numbers, the Quadro M4000 was about 50% faster than the M2000 and the M6000 was just a bit more than twice as fast as the M4000. We unfortunately did not have a M5000 available to test, but we would expect it to fall almost exactly in between the M4000 and the M6000.

Moving up to the newer Pascal cards, the P5000 was just a hair faster than the M6000 but in a practical sense it effectively performs the same. The P6000 didn't give as significant increase in performance as the other cards, but it is still 30% faster than the P5000/M6000.

The latest addition to the Quadro lineup - the GP100 - is a very interesting card. According to the raw CUDA core count and core frequency, it should technically be a bit slower than the P6000 but oddly it is actually 20-30% faster. There are a few things unique to this card like increased double precision performance, but the most likely thing affecting performance here is the addition of the High Bandwidth Memory. This is much faster than even the GDDR5X VRAM used on the P6000 and appears to make a pretty significant difference.

We have already shown that Visualize scales almost perfectly with more GPUs in our GPU scaling article, but we did want to go ahead and include the results with dual GP100 cards for those that are interested. In short, the scaling even at this high level is still extremely good and results in render times that are just under half of what it would take with a single card.


Overall, we did not really find any surprises in our testing. There is a very nice increase in performance going from the M2000 to the M4000, and an even better increase going from the M4000 to the P5000. The performance benefit to the P6000 is definitely much less (only about 30%) but even that is not terrible if you need the absolute fastest render times. Going all the way up to the new GP100, we saw a further 20-30% increase in performance compared to the P6000. Considering the price difference between the P6000 and the GP100 should be much smaller than the pricing difference between the P5000 and the P6000, if you are already considering the P6000 then the GP100 should definitely be on your radar as a way to get a decent bump in performance for a (relatively) moderate increase in price.

This excellent increase in performance across the various models has a very interesting side effect in that it vastly simplifies your choice of GPU. With other GPU-based rendering engines (such as Octane Render) you run into situations where using a larger number of cheaper cards is actually faster than using fewer, more expensive cards. Just like Octane, Visualize scales almost perfectly with multiple cards but due to the great performance improvements across the different models you really don't have to make this choice. Until you get all the way up to the P6000 you get pretty much the same increase in performance as the increase in cost so you might as well spend the same dollar amount on a single GPU instead of spreading it out over multiple cards.

For example, the P5000 is about twice as expensive as the M4000 but is actually a bit more than twice as fast. So while you could spend the same money on two M4000 cards, you might as well just get a single P5000 (or two P5000 vs four M4000). Not only are fewer cards a much simpler configuration (which typically means less problems with heat, drivers, and other things), it will also be much easier to expand in the future if you need even more performance.

As long as you stick to the newest generation cards, really the only cards with any caveats to them are the P6000 and GP100. All the others have great relative performance for their price points, but these cards at the very top of the performance stack are definitely a case of diminishing returns. If you need the absolute best performance regardless of budget, however, the GP100 in particular provides incredible performance in Visualize that can't be matched by any other card currently on the market.

SOLIDWORKS Workstations

General Modeling


Simulation & Rendering


Tags: Visualize, Rendering, GPU


Please add the benchmarks for Quadro P4000, P2000, P1000 in the future.

Posted on 2017-02-12 13:25:19

We definitely will once they are available. We were lucky enough to get a pair of GP100s to use in our testing but we are likely going to have to wait until the other P-series cards are available through distribution to test them.

Posted on 2017-02-14 18:05:53
Niko Nikolov

+1 for a quadro p4000 when it comes out.Ive been holding for 3 months on just to get 2 of them.I saw that there is some info about the p4000 since a week on diferent pages but is soooo confusing as everyone is stating different specs.(example: some say 243 gb/s and other say 192gbs,same with the clock speeds).I suspect they are mixing the laptop version of the p4000 with the desktop p4000 dohhh :) Please guys,when you get one,share some 100% gpuz info on the gpu. I suspect that in raw power is just a gtx980 but mannn,for 105w its worth buying a few(maya>some stable temp gpu rendering etc)

Posted on 2017-02-15 19:00:10

I would go off what PNY has on their site as far as specs right now: https://ns40.pny.com/promo/... . They have full datasheets and I trust their numbers more than anyone else when it comes to Quadro products. One thing to note is that NVIDIA really doesn't like to talk about clock speed on Quadro so any numbers you find there are more than likely made up.

As far as performance, I would expect the P4000 to be around 20% faster than the current M4000. The P5000 and P6000 was a bit more than that, but they got GDDR5X VRAM so their performance gain is probably a bit higher than what we will see on the other models. 100% accurate numbers will of course have to wait until we get our hands on them, but once we do I plan on testing a range of different software packages.

Posted on 2017-02-15 19:10:55
Niko Nikolov

I think it shoud be more than 20% as it looks like these guys got one


And the 3d mark orange room shows some serious improvement even over the m5000.I hope they dont lie about the nice scores.

By the way i did worked with the m4000.Liked it a lot but thanks to the low mhz in brute force i find it to be just 5-10% better than a old gtx 680.

By the way,nvidias page say that the p4000 is 1792 cuda and the one you gave me states 1729?

Posted on 2017-02-15 20:07:56
Aaron Schmidt

Be interesting to see FirePro vs Quadro card performance results.

Posted on 2017-02-21 01:42:42

Hey Aaron, Visualize is based on the NVIDIA Iray rendering engine which currently does not support AMD Firepro (or Radeon Pro as they are now branded) cards. You can still use it in CPU-only mode if you have an AMD card, but you won't anywhere near the performance of using a CUDA-based NVIDIA GPU.

Posted on 2017-02-21 01:51:59

Awesome & useful review as always, thanks! I have not yet seen any other GP100 benchmarks anywhere btw! Nice to see the vram bandwidth does so much. I was wondering, have you done any other compute benchmarks on the GP100? I'm even interested in some gaming benchmarks just to see what the vram does :)

Posted on 2017-03-14 19:02:43

We will be doing more compute benchmarks in the future (Don has a short blog post with raw GFLOPS numbers if you are interested: https://www.pugetsystems.co.... The difficulty with these cards is that they are very new and somewhat different than standard GPUs since they have NVLink and the high bandwidth memory. A few of the applications we've tried to test don't work quite right with them (or not at all) so we're having to spend some time simply getting them to work. For example, easy things like rendering shouldn't be a problem, but some of the simulation packages we are looking at right now seem to have some issues

Gaming, however, we probably won't have the time to cover. I wish we could (I'm curious too!), but our to-do list of testing seems to always be getting longer and longer so I doubt we'll be able to do that kind of testing.

Posted on 2017-03-14 19:11:54
Mirel Aretu

What about a comparison between the new GTX 1080TI FE and these pro cards?
I think that would be interesting since the new GTX 1080 is a lot cheaper.

Posted on 2017-03-21 08:24:57

Based on the results we've seen in other 1080 Ti tests, both here at Puget and elsewhere, it is almost not worth testing the 1080 Ti any more. It will perform within 1-2% of the Titan X (Pascal version), usually just a hair ahead of that older and more expensive card. Any time we have test data with that model included, just imagine the 1080 Ti being right there beside it. And we do happen to have Solidworks performance data for the Titan X:


Posted on 2017-03-21 16:13:25

I just ran a test with my 1080Ti and completed the 1969Camero render in 146 seconds, although it is factory overclocked.

Posted on 2017-06-12 19:56:06