Jump to content

herbieherb

Member
  • Posts

    291
  • Joined

  • Last visited

Posts posted by herbieherb

  1. You need the GPU for 2D and 3D geometric calculations and OpenGL-view. Basically the GPU only needs enough VRAM to fit your model. For larger projects I wouldn't go below 8GB, but only because even midrange graphics cards have that much memory today.

     

    The same for RAM, you basically only need enough to fit your model. You don't need ECC, registered etc. But the Ryzen CPU's like fast clocked RAM e.g. 3200-3600Mhz and of course low timings around CL16 is the sweet spot.

     

    Here is an example of AMD's current dominance in the multicore CPU values: The maxed out Mac Pro reaches 9.705 points in Cinebench R20 Multicore. That's about the level of a 16 core AMD Ryzen 9 3950X consumer cpu. But the Ryzen PC would only cost about 2.500 $.

     

    Apple is of course still unbeaten in the Macbook Pro's when it comes to CPU-Rendering. But the Ryzen Mobile CPU's coming onto the market right now seem to catch up.

    • Like 1
  2. The more multicore GPU power the better.
    RAM only needs to be large enough to fit the model.
    No CUDA, no METAL, no fancy GPU etc.

    Look at the Cinebench R15/R20 Multicore Benchmarks. Twice as many points means twice as fast renderworks renderings.

     

    https://www.cpu-monkey.com/de/cpu_benchmark-cinebench_r15_multi_core-8

     

    AMD Ryzen 3000 has best price-performance ratio, AMD Threadripper best performance atm.

    Intel/Apple are currently far behind in performance and price for GPU rendering.

  3. This is not worth a wish list entry. The developers are aware of the potential and write new code in a way that makes the best use of the hardware. Nevertheless, it will never be possible to parallelize all processes, because for many it is mathematically impossible.

    • Like 2
  4. Your CPU is already used at 100%, but only on one core. The calculations for the terrain model were not or maybe can't be parallelized, they cannot therefore be distributed over several cores. For example, with a 4-core CPU, Windows shows a load of 25% (100%/4 cores), although the full single-core performance is used. Not all calculations can be parallelized. For example, everything that has to do with rendering can be parallelized well while Geometry calculations are difficult to parallelize because the result of the calculations depends on the previous calculations.

    • Like 1
  5. As a student you will probably rather need a laptop. The most important thing about a computer for Vectorworks is that it has a dedicated graphics card. Without one, Vectorworks will only work very limited, if at all.

     

    With a device that meets these criteria, you will be able to work with Vectorworks very well:

    • dedicated graphics card with about 4GB of VRAM
    • 16 GB RAM
    • Processor with as many cores as you can afford 😄

    On the Apple side, only a Macbook Pro meets the requirements. It's expensive, but it's worth the money.

     

    On the Windows side, I would look for gaming laptops. You don't need workstation hardware like Xeon processors or Quadro graphics cards. Very interesting at the moment are the laptops with Ryzen processors that are just appearing.

     

    If you have a selection of devices, please check back here so that you can be sure to purchase the most suitable one.

    • Like 1
  6. Whether you will notice a difference depends on whether the current card is running at its limit or not. This depends entirely on the level of detail of your models. Vectorworks 2020 introduced a LOD system for OpenGL. It automatically controls the display detail level so that your graphics card is not overwhelmed. A new graphics card would cause more parts of your model to switch to the detailed view more quickly. The Quadro P2000 is surely good enough, you wouldn't need to change it. With a graphics card, your old computer will sell better, and a new, much faster graphics card wouldn't be very expensive. How much VRAM you need depends on your usage. As long as the VRAM is not filled, a GPU with 2 GB works as fast as one with 12 GB (assuming all other stats are the same).

    The usage of RAM and VRAM can be viewed in the task manager.

  7. Then you will need a processor with many cores. For example, a Ryzen 3000 with 12-16 cores. In exchange, you'll save on the graphics card. You don't need a Quadro for Vectorworks. You need a decent gaming graphics card. At the moment I find the Radeon 5000 series very interesting in terms of price and performance. Especially because it offers a lot of VRAM for a small price.

     

    With RAM and VRAM, you can see the load on one of your largest projects while you have all layers turned on in OpenGL view, and once on a high-resolution Renderworks image. If they are sometimes more than 80% full, you can add a little more RAM respectively VRAM to your new computer.

     

    I would sell your old computer in one piece.You won't be able to use the mainboard and RAM with a new processor anyway.

     

    If you have a specific machine in mind, please contact the forum again, so you are sure to make a good purchase.

    If you do so, post your results from the RAM/VRAM tests and the exact specs of the machine you have in mind.

  8. 5 hours ago, Claes Lundstrom said:

    As for output quality of realtime renders, there is nothing that says that they deliver inferior final quality renderings.

    Of course there are differences in the render quality. Realtime renderers work with the tricks of game engines. They come close to reality, but show weaknesses in some points. These are mainly indirect lighting, reflections and above all refractions and shadowing. Especially when rendering glass, all realtime renderers show weaknesses. The most accurate results are still provided by path tracing engines. So they have a right to exist beside the realtime renderers. Which software you use depends on the result you want to achieve, the media type, your own skills, your budget, the time you want to invest, the software and OS you use etc.

     

    Here are some examples of pathtracing (left) and realtime (right) in blender:

     

    large-a9f04513d4f6329548346619b91c616e.j

     

    1551978660642-1551978660642.png

     

    grafik.thumb.png.3ef9325b59d8af579a225376c20d4dfc.png

     

    4cd04a8d4d69ed892eff355a0e07de315ed0f85e

    grafik.pnggrafik.png

    • Like 2
  9. Realtime renderers are great when it comes to fast light setups, render time and learning curve. In exchange, you have to deal with exports, change textures and objects, etc. Of course, the render quality is never as realistic as it is possible with Renderworks. But well done renderings come very close. Although you can often see on the reflections of glass that there are not really light beams calculated. I would say that a beginner in Renderworks and a real-time renderer will get better results faster with the real-time renderer. For an experienced user it is the other way around.

    I tried Twinmotion when it was free for a while. It was great for rendering videos and for live presentations to clients. That's also the purpose for which I would still use live renderers today. For rendering single images I am faster with Renderworks and have the better quality.

    • Like 2
  10. If you are satisfied with the speed, why do you want to change?
    What kind of plans do you draw? 2D/3D? What is the current maximum RAM/VRAM usage while using Vectorworks?
    The i7-4790K is still pretty good for Vectorworks 2D/3D. Only rendering with Renderworks could be faster on a modern processor. If that's your concern, I'd rather not use an i9, but a Ryzen, since you get more cores for less money. If you want to swap the processor to a newer one, your DDR3 RAM won't work anymore. It is not compatible with today's DDR4 motherboards.

  11. You can test the performance and behavior of your software by simulating a higher setting via the graphics card driver. The function is called DSR for Nvidia and VSR for AMD graphics cards. Simply set a DSR/VSR factor of 4.00 of the native resolution (or any other factor). Now you can increase the screen resolution to 4k. The graphics card will then calculate the picture in 4k resolution and scale it down to HD for the monitor output. The objects on the screen will be as small as 4k while the resolution is still the maximum that can be displayed on the screen. After that you can also try out the screen scaling of the operating system. So you can see how the performance behaves, but also how the operating system or software reacts to a screen scaling.
    This function is a qualitatively better method for antialiasing. So it is intended for 3D applications like games. In 2D, the function causes a blurred image, especially when displaying text. Therefore you should disable this function after testing.

  12. Yeah, that would make sense. Especially since the 1070 is still one of the faster models and is well equipped with 8GB VRAM. I would certainly look at used ones there as well. You can save a lot of money there.

     

    Just make sure that you are not limited by VRAM. The model must be loaded as a whole into the VRAM of both cards. The fact that both cards together now have twice as much VRAM does not mean that you have twice as much available.

     

    Nevertheless, the additional shader units can bring better performance to Vision. How much that will be, I can't say, though, because I'm not using SLI or Vision. Maybe another user here has some experience?

  13. GeoTiffs with elevation information are basically tifs that come with a text file stating how they are georeferenced. The only difference is that instead of RGB values for the pixel heights they contain numbers. It is quasi a matrix of height numbers, for example

     

    705.1, 705.7, 705.9, 706.2
    704.8, 705.4, 706.0, 706.1
    704.6, 705.2, 705.8, 706.2
    704.2, 705.1, 705.5, 705.9

     

    The separate .tfw file contains information about how big a pixel is, where it is located and how it is rotated.

    In order to be able to use something like this in Vectorworks, you need a file in .xyz format or similar. They look like this:

     

    x-coordinate y-coordinate z-coordinate


    for example:

     

    2677000.25000000 1246999.75000000 569.249939
    2677010.75000000 1246999.75000000 569.999939
    2677021.25000000 1246999.75000000 570.769958
    2677031.75000000 1246999.75000000 571.219971
    2677042.25000000 1246999.75000000 571.859985
    2677052.75000000 1246999.75000000 572.940063
    2677063.25000000 1246999.75000000 574.370056
    2677073.75000000 1246999.75000000 575.350037

     

    You can use an online converter to convert geotiffs into xyz format.


    Then import this xyz point cloud into the open source program CloudCompare. There you reduce the point density (Edit-Subsample), so that in the end you only have about 10'000 points. These can then be imported into Vectorworks via 'Survey Input'. Select 'Create 3D Locus Objects'. You can then use these 3D locus objects to generate a terrain model. This way you get a fully editable, georeferenced terrain model.

    • Like 3
×
×
  • Create New...