Pages

Friday, June 8, 2018

Product Review: Dell Precision 5820 Tower - Superior Performance for the AEC Market

I got my hands on a Precision 5820 Tower from Dell recently and have been running a bunch of tests - from benchmarks to trying massive models in VR. It ranked near the top for various benchmarks, including the RFO Revit Benchmark.
Disclaimer: This computer was provided by Dell for testing through their "Try and Buy" program with no strings attached. I was not obligated to write this post nor did I get paid anything:)
First things first, the computer specs:

Precision 5820 Tower
Here are the specs for the computer as tested:
  • Processor: Intel Xeon W-2125 4.0GHz, 4.5GHz Turbo, 4C, 8.25M
    Cache, HT, (120W) DDR4-2666
  • Memory: 128GB (8x16GB) 2666MHz DDR4 RDIMM ECC
  • Storage: M.2 512GB PCIe NVMe Class 40 Solid State Drive
  • Graphics: NVIDIA, various units tested (see list below)

Photos
The exterior of the unit looks like the typical Precision tower. I love the built-in handles...
as I take a tower with me when VR is needed outside the office.


A couple USB-C connections on the front, which I like; fasts port that work with my new phone!


Interesting photo comparing the Precision tower with an Alienware Area 51 tower.


Now we head inside... normally covered by an air shroud, seen here is lots of RAM and an impressive CPU heatsink.


Several PCIe slots...


This image shows the air shroud, which directs airflow over the CPU and RAM...



The unit was ordered with two Quadro GP100's with NVLINK to make sure the power supply was sized properly. This GPU configuration is not really practical for the typical Revit or VR use case. However, that said, some computer benchmarks (not Revit) came out on top with just one of these cards. Plus, one LHB user switched from a GeForce GTX 1080 to a GP100 and experienced significantly better animation/video export performance in Lumion.



A "Normal" Computer
When I say this is a normal computer, I mean it is from a 'big manufacturer' and not optimized for a benchmark. For example:
  • Running antivirus
  • Connected to the office network (not authenticated)
  • Running a network-licensed version of Revit (except the one 2019 run)
  • Steam and Oculus installed for VR
  • As shipped Window 10 Pro and other Dell apps running
  • Running a 24" monitor at 1920 x 1200
  • No overclocking (I don't think you can overclock a Xeon anyway?)
  • No liquid cooling

Intel Xeon Processor
While Revit is known to not be a fully multi-threaded application, the fact is there is some support and growing. In the results below I do some testing with CPU cores turned off and the results are noticeable. Plus, this Xeon processor has one of the best RFO benchmarks. Check out this Autodesk AKN post:  Which function in Revit will take use of multiple processors
  • The RFO Revit Benchmark is 20% faster with all Xeon cores enabled, compared to just one.
The image below shows the CPU activity with only two instances of Revit open.



NVIDIA Quadro vs. GeForce
I am a fan of the Quadro cards as they are professional-grade and designed to run 24/7 with high quality thermal properties. The Quadro cards came out on top in the RFO benchmarks compared to a couple different GeForce gaming cards.
  • Check out this white paper HP published (PDF):
    The professional advantage: Quadro versus GeForce
    This paper is a few years old but highlights the quality of construction between Quadro and GeForce cards. As I understand it, NVIDIA makes the Quadro cards themselves.
These are the NVIDIA graphics cards I tested (for fun):
  • GeForce GTX 1070
  • GeForce GTX 1080
  • Quadro GP100 options
    • Quadro GP100x1
    • Quadro GP100x2 with one dedicated to PhysX
    • Quadro GP100x2 SLI
    • Quadro GP100x2 with NVLINK
    • Quadro GP100x2 in Tesla mode (TCC) with a Quadro P2000 for graphics
  • Quadro P2000
  • Quadro P5000
  • Quadro P6000
I know there are other GeForce options (e.g. 1080ti,) as well as the Titan, but I don't have them to test. Some may argue that the difference in performance is not worth the extra cost - that is for you to decide, and I hope this information helps you with that decision (either way you go).
One final point; the image below shows the GPU activity for a massive VR model running in Fuzor. Fuzor and SteamVR are the only applications running. As you can see, when GPU memory is 13GB you will need a Quadro card.


RFO Bechmark

Using Revit 2018.3, I ran several tests using the RFO Revit Benchmark test; Full Standard and Graphics Hammer. As Gordon Price says in the following quote, the provided graphics tests are:
"Good for evaluating graphics card purchase decisions, as long as you run all tests on the same machine, with just the graphics card and graphics driver changing."
So that is what I did.


RFO Full Standard
This is the go-to test that most people run within the RFO options. The P6000 configuration came out on top, with the GTX 1080 coming in last. Keep in mind, this is not just about the GPU... if you compare these numbers to the published results on RFO, the computer itself is awesome!

Since starting to write this post, the RFO Benchmark for Revit 2019 came out. The benchmark runs slightly different than the 2018 version so you cannot necessarily compare the number between them. Here are the results for the 2019 version using the P6000 card.



RFO Full Standard - Different # of CPU Cores
For this test I went into the BIOS and turned off CPU cores. The Dell Precision Optimizer will try to do this using built-in AutoCAD and Revit profiles - don't let it! Doing so clearly has a big impact on performance. Plus, those cores can be running other apps as well (Outlook, Excel, Enscape, etc.).


RFO Graphics Hammer
As Gordon also indicates in the RFO notes, this test is borderline silly. But I had the computer for 90 days and just let it run these multi-day tests to see what would happen. The relative results where similar to the Full Standard tests.

Computer Benchmarks

There are a number of general computer benchmarks that have nothing to do with Revit. Below are the results from five separate 'free' tests I did using the various graphics cards in the same computer. The P2000 came in last in all of them. The GP100 has the most wins, with 4 out of the 5 tests; the P6000 won the other one. The GeForce cards landed in the middle.

Pass Mark ResultsTest Download


3D Mark 11 Results - Test Download


3D Mark - Test Download


3D Mark, Vantage Results a sub-test of 3D Mark


3D Mark, Firestrike - a sub-test of 3D Mark


VR Experience

I am not aware of any formal VR benchmarks, but I tried a couple heavy models in Enscape and Fuzor. Both performed very well. But compared to the Alienware Area 51 with a GP100 in it, I did not notice much difference in Enscape, which I assume is due to it being more GPU dependent. I did notice a slight improvement in Fuzor. Glad no one was watching me test this, as I was turning my head super fast and doing a face-plant in a bush!

The image below shows a massive model in Fuzor using 13GB of GPU memory. So if you work on big projects, you may need the P6000 with 24GB GDDR5X or the GP100 with 16GB HBM2.


On a related note, Dell created this promotional video of our firm using their hardware to deliverer cutting edge VR experiences: LHB visualizes architectural designs through virtual reality powered by Dell Precision.

Conclusion

The conclusion is that Dell and NVIDIA can create a powerful computer for the AEC market. While the smaller bespoke computer companies using overclocking and liquid cooling are super cool, they are not the only ones who can pull off solid numbers. And the prices are pretty close. Especially if you cut the RAM in half, which I tried and the results did not really change for the Revit benchmark.

The main goal is to share my results based on curiosity related to Revit performance and optimal VR experience. I hope you find this interesting and/or useful!


For BIM Chapters updates, follow @DanStine_MN on Twitter