ArchiCAD's use of Hardware

[Anonymous]

If my understanding of ArchiCAD operations is correct, during most 2D operation such as showing trace references, panning, zooming etc ArchiCAD can only use 1 cpu core (being specific not one physical CPU core but one logical processor. Why? I cannot understand it - for us (12+ staff) we waste huge amounts of time waiting for 2D redraw, just navigating around, and the very worst scenario is having something set as a trace reference which we do alot of to ensure the building is coordinated. We recently upgraded 2 machines and in my opinion spent a ridiculous amount of money on hardware (see signature) in an attempt to cure this issue and frankly it is almost no different at all. I can only think this is because the software was not designed to use all of the hardware available to it. What is the technical reason that in 2016 the operation of the software is restricted to use 3-ish Ghz of processing power when there is (on my machine) potentially 40Ghz+ sat idle? For such expensive software, sat on such an expensive computer - we are mighty disappointed. Particularly now that we have AC20 and have constant bizarre 'overwatchserver.exe' crashes and still no fix.

I would dearly like an explanation and please don't be afraid to give us a proper technical answer on why the software can't use more than one core. I hope to learn that this is something that is in development for AC21 because frankly the ££ we waste across an entire room sat waiting for ArchiCAD can only go on so long.

[Anonymous]

Agree with previous post 100% 2D navigation is very slow! The problem is in ArchiCAD, nowhere else. We need completely new software, not 21st version of old one. I see no increase in performnce since I worked 10 years ago in 10th version. I had 1-core average PC, ArchiCAD was slow. Now I have much more powerful machine, but Archicad is not faster!
Dan G spent a huge ammount of money to give workers opportunity to work faster, but no! ArchiCAD do not allow us to work faster.

Do you know how my day looks like? 20% working, 20% coffee pause, 60% internet surfing just because I wait ArchCAD to finish an operation.

[mikas]

Dan wrote:
If my understanding of ArchiCAD operations is correct, during most 2D operation such as showing trace references, panning, zooming etc ArchiCAD can only use 1 cpu core (being specific not one physical CPU core but one logical processor.

This is of no help, sorry, but ArchiCAD (20) seems to load all 4 of i7 cores when zooming in/out a 2D plan view. It seems that with this task AC does not use logical cores (i7 hyperthreading) though. Please see attachment (267% out of 800%), scrolling constantly in/out with mouse wheel. When panning in 2D plan view AC only loads one of the cores (100% out of 800%).

[Anonymous]

Thank you, you've just taught me how to properly monitor number of cores being used which is helpful. I see the same result here, panning and zooming seems to use more than one core. But look at my graph for cpu load attached. Most of the work is being done in my case by logical core 16 and it is constantly topping out. Maybe I am expecting too much here from technology, but why can't the load be split more evenly or across even more cores? What I'm trying to understand is why with such a seemingly high power machine the software seems so slow still?

[Anonymous]

Vahur wrote:
Agree with previous post 100% 2D navigation is very slow! The problem is in ArchiCAD, nowhere else. We need completely new software, not 21st version of old one. I see no increase in performnce since I worked 10 years ago in 10th version. I had 1-core average PC, ArchiCAD was slow. Now I have much more powerful machine, but Archicad is not faster!
Dan G spent a huge ammount of money to give workers opportunity to work faster, but no! ArchiCAD do not allow us to work faster.

Do you know how my day looks like? 20% working, 20% coffee pause, 60% internet surfing just because I wait ArchCAD to finish an operation.

At the moment without any other reasonable explanation I think I have to agree. We have had many years bolting on more and more 'functionality' but in the background the software seems to be very overdue a rework from the ground up so that it can keep up with the potential pace of the user. I can work much much faster than the software can (if that makes sense) and that is hugely frustrating.

[alemanda]

DanG,
as long as I'm going to buy a more powerful multi-core workstation, I'd like to hear from you what about the background processing? Don't you find any help with this amount of cores in 3D re/generation and sections/elevations regeneration? What about schedule generation? What about rendering times?
Can you share your configuration?

[mikas]

I tried to search Dan's machine too (with the specs in his the signature), but did not find the exact system with the resellers online tools (https://www.scan.co.uk/3xs😞 Scan 3XS System. Intel Xeon E5-2640 v4 @ 2.40Ghz. 128GB RAM. Nvidia Quadro M4000. Intel 750 PCI-E SSD. Windows 10 Pro.

It might be a custom configuration.

Still, that machine (Dan G´s) should be quite a beast with renders. At least compared to mine: i7 2.93GHz, 4 cores, 8 threads. I would bet it actually is a really, really fast renderer with cinerender. It also probably would use every available processor thread with cinerender (both actual and logical threads), and that means 20 pcs with that Xeon E5. Renderings always benefit of cores and gigahertz.

Rendering algorithm should be a naturally and easily threadable task. That means the job can be divided in multiple concurrent, parallel, jobs to ultimately achieve only one unified result - the rendered frame or picture. I have been told rendering jobs are one of the most easily threadable, and in that way, the tasks that benefit the most from multiple cores of a modern workstation processor. The more cores, the better, the more gigahertz per processor, the better.

Unfortunately that's not what we get with every task in ArchiCAD. I would not say it's impossible to use every thread of a processor with every given task of a given program. Still, I believe it is very difficult to achieve, because of the nature of the tasks. Most of the common tasks would not easily be dividable as multiple parallel jobs, and thus use all processors or cores, to quicker achieve what would be correct and an acceptable result.

I am not a programmer, but I've had/got friend(s) who do that as a day job. I remember them/him being very enthusiastic of massively parallel computing and algorithms associated to boost computing, both personal and scientific computing. This all included, besides scientific astral calculations, some realistic rendering algorithms. Those very fine algorithms were not suited for the most of the everyday tasks, though. I don't know, but I think the same still applies today.

I have also learnt that GPU's (graphics processor units) are by their design and by nature something close to massively parallel. Thats why they too are being used for rendering tasks nowadays. And they seem to be quite good doing that job.

ps. I've got a 2006 Mac Pro Hack with 8 real cores (Xeon X5355, 2,66 GHz). By my measurements it's a little bit faster with rendering than my i7 iMac. Thats how and why I've come to a conclusion; real cores usually mean more real power than logical cores. So, a higher gigahertz might sometimes mean more than the number of logical cores. I would say as a rough approximate, that a logical intel core is worth 0,25- to 0,33 of a real core (within the same production year).

[Anonymous]

With regards performance elsewhere in ArchiCAD of course the new machines are certainly quicker for other processes. I haven't thought of a good way to benchmark/compare other than rendering the same image on an old machine and new machine and comparing the time. From memory a render that was taking something like 24 hours on an old machine here was rendering in 8ish hrs on the new machine. If you have a method for making a fair comparison then I am all ears and happy to run a test for you.

I noticed having been back to the scan website that they have just changed the name of the base machine that we used. We customised the standard build slightly and opted to buy just one physical cpu instead of 2. Which saved us around £900GBP. I think we upped the amount of ram from the standard 64G to 128 too.

I hear what you are saying with regards some 'things' not being suitable for splitting across multiple cores. If that's the case then I partially accept that's just how it is - but it would be useful to have it clarified by Graphisoft. Also, it kind of suggests for day to day use you to buy the best performing single cpu you can afford right?

[mikas]

Dan wrote:
Also, it kind of suggests for day to day use you to buy the best performing single cpu you can afford right?

That's exactly what I have been thinking lately. We (our practice) should make an investment in workstations soon too. New 4GHz i7 seems to be a really good choice for price/performance and balance between 2D/3D.

On the other hand, I occasionally render a lot, and a machine with lot of cores, like yours, would make AC/me more effective.

We have a cinema 4D with unlimited client team render, so I can use it for distributing the renders across multiple machines. In my opinion it would still be more convenient to do these "in the middle" renderings straight in AC. Final renders can then be spread out to local network render nodes (workstations or dedicated machines).

With our firm I have to think about moving from Mac to PC also, because Apples high end choices (Mac Pro) are a bit old today, and the cost is pretty high.

But I agree, hopefully Graphisoft programmers can find new ways to improve AC speed.

[mikas]

Dan wrote:
I haven't thought of a good way to benchmark/compare other than rendering the same image on an old machine and new machine and comparing the time.

That will do the job.

To standardize the test, and to compare a lot of different machines (cross platform, WIN and OS X), I have used cinebench software. It's freely available from Maxon, the technology provider of AC cinerender. It would be really interesting to see your machines score with that test.