Puget Systems print logo
Read this article at https://www.pugetsystems.com/guides/1105
Article Thumbnail

SOLIDWORKS 2018 CPU Comparison: Coffee Lake vs Skylake-X vs Threadripper

Written on February 4, 2018 by William George


Dassault Systèmes launched the initial version of SOLIDWORKS 2018 (SP0.1) late last year, but with the recent release of SP1 we expect that customers will soon be using it in production environments. In preparation for that, we have tested the field of current Intel Core i7 and i9 processors to see how they stack up in SW 2018. We hadn't yet had a chance to test AMD's Threadripper processors in SOLIDWORKS either, so they are also included in this round of benchmarks. Ryzen was not included, though, as we already looked at it in SW 2017 and didn't find its performance compelling enough to test again.

AMD vs Intel

Test Hardware and Methodology

To see how these different CPUs perform in SOLIDWORKS 2018, we used the following configurations:

The tests conducted on these systems were originally developed by my colleague here at Puget Systems: Matt Bach. He put together a series of AutoIt scripts that run through testing a variety of the capabilities in SOLIDWORKS, so rather than reinvent the wheel I used his. Only minor updates were needed to bring the scripts up to speed for SW 2018. In the future we plan to overhaul our SOLIDWORKS testing, hopefully adding even more complex assemblies to really push CPUs and GPUs, but we aren't quite ready for that yet.

So, for now, what does our testing include?

Test Files
Assembly Rotation Audi R8 by ma73us
Rebuild/Rendering Vertical Twin Steam Engine with Reverse Gear (by Ridwan Septyawan)
Motion Study Gear Train Mechanism with Fixed and Swaying Axes (by trinityscsp)
FEA Simulation (Stress) FEA Benchmark V3

Flow Simulation (Airflow & Thermal)

Billboard - Lesson14 Case Study (SOLIDWORKS 2015 Flow Sim. training files)

FEA Simulation (Stress)

Flow Simulation
(Airflow & Thermal)

The results are broken up into their own sections below, followed by the conclusion. Each test was run three times on every CPU: the worst result was tossed out and the other two were averaged. This was done to help eliminate any influence from background applications running on the systems, but without cherry-picking only the very best result either.

Feel free to skip ahead if you are just interested in a specific set of results, or if you want to get right to the conclusion.

Results - Assembly Rotation (FPS)

Here are the results for rotating an assembly of an Audi R8 with 434 parts and about 1.4 million triangles:

In this first set of tests, Intel's Coffee Lake Core i7 processors - both the 8700 and the 8700K - perform very well. The enthusiast-grade Skylake X chips aren't bad, but only the top-end of that line comes close to matching the much less expensive Coffee Lake models.

Unfortunately, AMD's Threadripper processors do very poorly here. Mind you, frame rates around 30fps like this should still be usable, though not quite as smooth as speeds up closer to 60fps like many of the Intel chips displayed. What seems most odd is that there is less variance between the different graphics modes on the Threadripper processors, so maybe there was another factor at play (compatibility with the Quadro P6000, for example).

Results - Rebuilding Assembly

This graph shows how long it took to rebuild an assembly on each CPU:

Rebuilding an assembly appears to be pretty much single-threaded, which means the high core counts of the Skylake X and Threadripper processors aren't going to help them here. With clock speed being the focus, the 8700K does the best - with the vanilla 8700 coming in only about 4% slower. Once again, AMD's Threadripper processors end up in last place.

Results - Motion Study

Here is a chart showing how long it took to perform a motion study on each of the test systems:

SW 2018 Motion Study

In this test, all the Intel processors landed within 1.5 seconds of each other - and close enough that a clear winner is impossible to call. That also makes it hard to say how much impact either core count or clock speed have, but suffice it to say that any modern Intel processor should do just fine here. AMD, on the other hand, came in last place again.

Results - Simulation

These charts shows the time taken to perform three different simulations on each processor:

SW 2018 Stress Simulation

SW 2018 Airflow Simulation

SW 2018 Thermal Simulation

Simulations in SOLIDWORKS appear to benefit from both core count and clock speed, with different types of simulations behaving slightly differently. The airflow and thermal simulations, both examples of flow simulations, gained enough from high core counts that some of the Skylake X processors outpaced the mainstream Coffee Lake models, which had otherwise been leading in the tests. In the case of the stress simulation, though, core count only went so far: the 10-core i9 7900X was fastest, but as the number of cores rose from there the performance dropped, presumably due to the decline in clock speed associated with having more cores.

In all three cases, AMD's Threadripper chips continued to be the worst performers - rivaled only once by the Core i7 7820X (in the thermal simulation). It is looking like the calculations involved in many tasks in SOLIDWORKS may simply not be optimized for AMD's CPU architecture.

Results - Rendering

Here are the times taken for both the render pre-pass and the main render itself, at 1920x1080 resolution, on the various CPUs:

SW 2018 Rendering

This final test is where CPU core count really shines, and the one place that AMD's Threadipper processors do not end up in last place. They aren't the fastest either, but the TR 1950X does just barely outperform the Core i9 7900X - which is notable, as these processors have roughly the same MSRP ($999). The fastest chip overall, though, turned out to be the 16-core i9 7960X, meaning there is probably no reason to consider the 7980XE for SOLIDWORKS at all.

While the more mainstream Core i7 8700 and 8700K are certainly capable of rendering in SW, if a lot of your time is spent on rendering then you will definitely be better served by a higher core count processor.


Combining the results of the previous charts, here is a summary of how these CPUs perform in SOLIDWORKS 2018. Please note that performance here is relative to the Core i7 8700K, our current recommendation for general SOLIDWORKS usage based on these tests:

SW 2018 CPU Comparison Summary

This is a rather complicated chart, but it is the best way I could find to show all of the necessary data without resorting to multiple graphs. What we can see here is that the Intel Core i7 8700K is clearly the best current-gen CPU for general usage in SOLIDWORKS 2018. It tops the charts when it comes to manipulating models, rebuilding assemblies, and motion studies - while also punching above its price point in simulation workloads.

The story changes when you move to rendering. There, while the 8700K is no slouch, higher core count CPUs can and do surpass it by wide margins. The top choice for rendering ends up being the Core i9 7960X, which also happens to be one of the best choices for heavy simulation work, though other Skylake X and even Threadripper CPUs do well too.

If you are looking for a new workstation for SOLIDWORKS, we have recommended systems for both general usage and rendering.

SOLIDWORKS Workstations

Puget Systems offers a range of powerful and reliable systems that are tailor-made for your unique workflow.

Configure a System!

Labs Consultation Service

Our Labs team is available to provide in-depth hardware recommendations based on your workflow.

Find Out More!
Tags: Dassault, Systemes, Solidworks, 2018, CPU, Processor, Intel, Coffee, Lake, Core, i7, i9, Skylake X, AMD, Threadripper
Jim Kiefer

very interesting, and easy to digest.

Posted on 2018-02-04 20:26:21

There has got to be something wrong with your setup when flow simulation testing a 16 core/16 Thread 3.4 GHz quad memory processor is being beaten (by a long way) by a 6 core/ 12 thread 3.2GHz dual channel memory processor. Most the Threadripper testing online has been done on ASRock X399 Taichi not a gaming motherboard, probably for a good reason. My experience with simulation is that Thread count, CPU frequency, quad channel memory, memory frequency all improve the simulation time

Posted on 2018-02-10 03:04:34

I am open to ideas about what might be causing the Threadripper CPUs to perform poorly here, or how we might be able to improve our testing, if you have any suggestions. Looking at the rendering section, though, the performance is about where I'd expect based on other CPU-based rendering tests we've done with these same CPUs in different applications. For example: https://www.pugetsystems.co...

So if the multi-threaded rendering performance is accurate, I can't think why the simulation tests wouldn't be. I suspect it is just the case that SOLIDWORKS' built-in simulation engine isn't terribly well threaded, which is evident on the Skylake-X processors as well, and may be optimized more for Intel CPUs than AMD. We've run into that in other programs too.

Posted on 2018-02-12 17:37:40


First observation when you tested the i7 8700K (Solidworks 2017:Coffee Lake Comparison) the test score was 125 now its 102. Changes I can see are Solidworks 17 to 18, PCE-I SSD vs SATA and mother board. Same processor now getting 25% better results. Are you sure the results didn't get mixed up, the 1920X got 126 in the flow test?
Based on your own testing previously, moving from dual to quad channel memory should give about a 30% improvement in Flow simulation time (see attached chart) The benefit from going from i7 8700K to i7 7820X (dual to quad memory, 6 vs 8 cores and similar CPU speed) giving you a 9% decrease!! in performance. The best i9 7980XE only getting 10% improvement despite having a massive 18 cores and 32 threads. Second chart shows how good threads are https://uploads.disquscdn.c... ). This is my i7 5960X (8 cores) not using your test file so results not comparable, except the effect of Threads in reducing simulation time. The chart suggest that using all threads to do two simulations lose about 50% of the computing time compared to doing only one of the same simulations but still quicker than running the same simulation twice.

This test by PCWorld (https://www.pcworld.com/art... suggests the i9 7960X being better that AMD's 1950X but not by much. Your rendering tests are showing a much bigger gap. That may just be Solidworks

Might be worth checking some of the simple things, memory timing in BIOS defaulting to lower number?, page swap size not the same on all machines in Windows, Solidworks not told to use all CPUs (that means threads to Solidworks), where you are booting Solidworks from SSD or SATA drive?.

Posted on 2018-02-14 11:25:59

I double-checked the results files, and the simulation scores in this article for the CPUs are correct (I didn't mix up any of them). There is one additional possibility in terms of difference between this article's scores and the previous Coffee Lake results, though: I can't remember if the previous article included the Mesh Creation time when reporting simulation speeds. I did *not* include those this time around, as something in our benchmark is acting up and not always reporting that time. Since I have that for some CPUs but not for others, I thought it best to just exclude it across the board and report the pure simulation time only in this article.

There are often cases with applications that are only somewhat threaded (where there isn't amazing scaling as you add cores) in which running two or more concurrent jobs will provide better overall results than running jobs in sequence (one at a time). It looks from your data like that may be the case with SOLIDWORKS Simulation, but that is not something we've specifically tested here.

As for other differences, all systems that I tested in this roundup had only a single drive for Windows and applications: the same 1TB Samsung 960 Pro M.2 SSDs. All had default Windows settings for things like the pagefile, and default SOLIDWORKS settings - except where noted, like the custom 1920x1080 render size. I can't go back and double-check things like the memory timings in the BIOS now, but I am pretty confident that they were all set to 2666MHz and the memory modules' default timings at that speed.

Posted on 2018-02-14 16:01:18

Probably just poor threading.
It's also highly dependent on clockspeed.

Posted on 2018-03-01 06:37:17

The Assembly rotation result is not surprising. The Intel is 20-30% faster in math per core. Up to 6 cores, it will certainly be faster than a Ryzen/TR. Frequencies drop, but not dramatically. After that, it can't compete, even is balancing is not a requirement, the threads from 7 to 12 are half performance while the TR ones are full performance up to 16.

So, an 8700K could theoretically be 50-55% slower than a 1950x and that can happen in some specific scenarios, but when you don't have enough threads, it's 20-30% faster.

Memory frequency and quad channel shouldn't make a difference in this.

It's all about the problem. When I write software, even if development cost and lack of competition is not affecting decisions, I can't optimize everything. Imagine you are making a cake and have 20 cooks at your disposal. Perhaps you can make it a little faster if you have 2 cooks working on it, but you can't make good use of all of them. There are things that can be very parallel, such a some types of rendering, or some types of encoding, or using 100 audio plugins, but those are not as common as we would like. There are all sorts of bottlenecks. Each Threadripper cook is a little slower than the intel ones. When you only need a handful of cooks, the Threadripper cooks will be slower.

Posted on 2018-02-24 18:02:09

To me, based on the benchmarks here, the i7-8700 (non-K) seems like a clear choice for highest performance to value period. I know Puget Systems stays strict to 2666 memory speeds, but with 8700K version, I personally would bump the memory to 3200 CL14 flat timings. This will bring extra bandwidth in other uses as well. With the regular 8700, 2666 is perfect.

As usual thanks for these benchmarks.

P.S. I've ran Solidworks 17 on a Core-M laptop with 8GB RAM, if you are just learning (small assemblies, low dependencies) you can get by with bare minimum hardware, though an SSD is a must. The minute you start loading larger assemblies it falls through its butt.

I am still hopeful AMD can make a snappy chip, maybe next year.

Posted on 2018-02-13 19:58:26

If you look at the CPU alone, you are correct: dollar for dollar the i7 8700 (vanilla / non-K) is the best. This is a way of looking at components and prices that I have long used myself... but honestly, I now think it is the wrong approach. I wrote a blog post about this a few months ago, which you might find interesting:


Posted on 2018-02-13 20:03:18

I've read that article and I completely agree. A bit on the off topic do you guys have any plans to start including Optane 900p drives in the benchmarks?

Posted on 2018-02-18 04:23:47

I haven't done any work with Optane drives myself, but from my colleagues who have it didn't sound like there was much (if any) benefit in real-world applications. That is assuming a comparison versus a nice M.2 drive like the Samsung 960 Pro that is our go-to for benchmarks and reviews. The place Optane seems to shine is as a cache drive for traditional hard drives, but since we never test with a HDD as the primary - and rarely, if ever, sell systems with that setup anymore - Optane doesn't seem to have much of a purpose / role at Puget.

Posted on 2018-02-19 16:52:03

As always thanks for taking the time for a solid reply. Yeah I don't know anyone who told me "yeah I'll buy an HDD and "turbocharge" it with an Optane stick". I just want to believe Windows file system will be rewritten one day to take full advantage of Optane and Optane-like low latency storage. A seriously wishful thinking.

Posted on 2018-02-19 17:09:04

committing draw is a central point for us, someone has tried draw comparison? cuda acceleration ec.
or only cpu brute force (i'm planning to switch from solid 2016-2018), and seems no improvement about drafting capability.

Posted on 2018-03-08 10:56:37

I'm sorry, but I am not familiar with the term "committing draw" - can you elaborate on what you mean? If you mean just general 3D modeling, that is pretty much entirely based on single threaded CPU speed - so clock frequency and instructions per clock.

Posted on 2018-03-09 17:16:10

i wonder if the results would be the same if you set the affinity to have all the higher cpus have the same 8 or 10 number of allocated cores during the test?

you're previous cpu core count article showed a significant drop after 10 cores, so it would be interesting to see if that helps yield better results for the higher core cpus.

while it's not ideal to most people to do this, if it does yield better performance, it would allow for more multi-tasking during the use of simulation.

Posted on 2018-10-28 14:23:48

Hi William,

Being one of the specialists for airflow/CFD at SOLIDWORKS, I am very familiar with the effects of hardware on the performance of SOLIDWORKS Flow Simulation and the model that you used in your benchmark of this product. The comment that I would make in regards to the above is that the number of mesh cells will influence how efficiently the Flow Simulation software will utilize the hardware. You should include cell count (CV, for computational volumes) as a parameter in your Airflow chart and customer decision process; it doesn't make sense why the Core i9 7960X or 7980XE wouldn't perform better than the 7940X. Based on your business model, you should ask the customer how many cells they would typically solve in an Airflow calculation. For example, a problem with 200k cells may top out its speed-up at 6 cores, so if CPUs have at least this number of cores, a higher clock speed would dominate this problem. But a problem with a magnitude larger number of cells at 5-10 million could efficiently use more cores and this may be the determining factor in selecting a CPU. It's also interesting to note that Flow Simulation will utilize hyperthreaded virtual cores about as efficiently as physical cores, so now we are talking about 32 cores for the 7960X and 7980XE.

Solving a steady-state airflow problem in less than 3 minutes (180 secs) is not a real judge of a common problem solved with Flow Simulation. Typically steady-state solutions to convergence can take nearly an hour (20x above) or more. Transient airflow problems can take days to solve for the full physical time of the simulation.

The attached image shows testing that was done on solver performance improvements from the 2016 to 2017 software versions. In the Scalibilty chart, for the smallest problem, it tails down with more cores, but with larger problems it continues to go up. I believe this is what you are seeing with the sample problem you presented here and why the 7960X and 7980XE does not perform better than the 7940X.

If you'd like, I can work with you to provide a more suitable test model to use with variations of the number of CV.


Posted on 2018-10-30 15:28:53

Hi Joe,

I am always open to improving our benchmarking tools, and I very much appreciate your insight into how the number of cores being utilized is limited by the complexity of a simulation! I saw that you emailed me directly, so lets continue this conversation that way :)

Posted on 2018-10-30 16:37:19
Mihkel Gysson

SW 2017 had i3 and i5 processors in the review. Why are those excluded here?

Posted on 2018-11-28 11:44:17

It has been a while since this was written, and I cannot remember what my reasoning was for selecting the specific CPUs in this test. In our next round of SW testing, though, I think we will at least include a Core i5 (the 9600K). We hardly sell any Core i3 processors, though, so the odds of a model in that line showing up in any of our testing is pretty low.

Posted on 2018-11-28 18:02:49

Could you please benchmark the new Intel 9th series vs. Ryzen 7 2700X in Solidworks?

Posted on 2018-12-09 23:55:13

We are planning to do updated Solidworks testing with the latest CPUs, but I am not sure if we are going to end up doing it in SW 2018 or 2019. It will require a bit of updating if we want to be on the latest version, and we are also looking at improving some of the testing (especially simulations) at the same time.

Posted on 2018-12-10 17:02:01
Alex M

Can't yet find the updated testing with AMD Zen2 Ryzens. Soonish?

Posted on 2019-08-29 17:31:31


Posted on 2019-08-29 17:37:40