Help understanding AC20 use of hardware

[Anonymous]

Hello,

We are in the process of trying to specify a number of new workstations. We are looking at a decent jump in performance and as such are trying to really get to the bottom of what hardware ArchiCAD uses and when.

We are trying to decipher if ArchiCAD 20 can make use of 2 physical cpu's, each with potentially 10 cores (or 20 cores including hyperthreading).
I understand AC will only use more than one thread(?) for "background processes" although I believe this includes opening/updating sections, elevations, 3D window, details and also panning/zooming/navigating 2D drawings and handling trace references.
This is where in the past we find ourselves waiting around hence the desire to address the CPU issue in our new generation of machines.

We plan to run 128Gb of RAM, with the view that we see a noticeable difference in render time on existing machines with the most RAM in the office. My basic understanding of this is that ArchiCAD/Cinerender does some pre-processing which relies on RAM.

Lastly, we are trying to decide on the benefits of either a Quadro M2000 or M4000. Looking at ArchiCAD bench mark testing there only seems to be a significant jump in performance when running 'large' (10 million polygon) projects. Our current largest projects are at about 5 million but work on the horizon could well hit the 10 million polygon mark.

Just a slight grumble, I have only managed to find information from Graphisoft online that addresses how AC uses cores, multi-processing, hyper-threading etc written in 2012. Is there newer data/explanation for how AC utilises hardware since then that we can use to reliably select and configure new workstations?

[Jacques Toerien]

Dan,

I cannot answer most of your queries, however, I know on my Mac Pro (2010) AC uses all 24 cores provided by my two six core Xeon processors (12 cores + 12 HT). When I update sections, elevation etc all the cores (including the hyperthreaded ones) are used. They aren't used fully though, they peak at about 30%...but they are all used. The same goes for renders. I've not kept an eye on 2D performance though but I can do a quick test tonight. I'm on AC 19 but I cannot see AC20 being worse. So yes, dual CPU machines do have some benefit.

but...

You may be better off with a few dedicated dual 12 core Xeons render machines and your actual 'drawing' machines having faster 8/10/6 core i7 Extreme CPU's like the 6950x or 6900k cpu's. These run at 3Ghz and 3.2Ghz. Feed these with a good motherboard, lots of ram and a M4000 and you should be doing very well. You may find the sweet spot is a 6850k @ 3.6Ghz with 6 cores (12 total)

The Xeon E7 CPU's top out at 2.8Ghz for a 10 core, these are going to run into ££££ just for the CPU. The E5 series 12 core sits at a comfortable 3Ghz per core, that will be 48 cores at 3Ghz!

Can you test drive a few machines to see what you would benefit from most?

There is still al lot of ambiguous information about whether a Quadro card actually provides real benefit. The new Pascal GTX are fast, really fast! but that is for shifting poly's around in a game engine using directX.

[Anonymous]

Hi Jacques,

Thanks for your response. That is useful feedback - although I hope someone can confirm if Mac's and PC's operate in the same way? I feel like I read somewhere that Mac's can utilise more CPU/cores that Windows machine but that may well have been a much older version of ArchiCAD.

Hear what you are saying with regards dedicated render machines, we debated that but decided in the end that for the frequency we would need to run 'final' size/quality renders we would end up having some very expensive render machines sat idle most of the time.

Presently we are looking at Xeon E5-2640 v4 - and a motherboard that could potentially accommodate 2 physical CPU's. However at £900 per CPU we really want to be sure whether 2 physical CPU's is going to be of real world benefit.

[Anonymous]

Regardless of the final choice, it will be a major disappointment.
5 times cheper PC will perfom ~ 5% slower. I have faced this problems before, AC uses all CPUs, but for something what noone knows. The same situation with GPU. With W8100 I see only ~ 50% performance growth comparing to GTX670 but only in 3D perspective.

[Anonymous]

Vahur,
The graphics card will only improve speed and quality of 2D and 3D navigation after the elements have been generated from the database (CPU intensive), the graphics card does not participate in cinerender rendering, section generation, elevation generation, schedules etc
Scott

[Anonymous]

I understand that. GPU is used in 3D window while turning model, correct? I do not talk about rendering, this is not an important part of my work. 2D navigation is the most annoying part and there is almost no acceleration with professional cards. Simple 2D drawing with 10K 2D elements slows down ArchiCAD dramatically, while AutoCAD works smoothly. It takes forever just to move equipment overlay provided by other company.
Even without overlay 2D elements AC becomes too slow when building becomes more detailed. Just a simple 1500 m2 office building has lag when working in plan. Almost no difference on my old i5-750 12GB 275GTX, my current i7 16 GB W8100, our architects new i7 6700k 32GB, W8100. Navigation in plan is slow. Our architect renders images, but it also takes forever! He saves time using Artlantis, 1 hour in ArchiCAD VS 3 min in Artlantis.
There is no point to invest into top professional hardware is you are going to work only in ArchiCAD. The cause of its slowness is in software, not hardware.

[Anonymous]

Vahur,
you talk about slowness of ARCHICAD compared to AutoCAD, I find no slowness in using ARCHICAD compared to AutoCAD in 2D display of same drawings (hospital floor plan of over 3500m2 / floor with complete fitout - over 220000 lines / arcs fills if exploded to 2D elements)
BIM is not CAD and each element needs to be generated from the database to display on screen / drawing with the attributes that are setup for that element for each view which is slower than simple drawing, but the same element can be represented differently depending on the view context which is what makes BIM faster than CAD as you add/edit an element once and it is added / edited in all relevant views. I can add a single door in plan in a few seconds and it will be added to plan, wall hole to slab setout plan, exterior elevations, internal elevations, and door schedule in AutoCAD adding the same door would take at least 5 times as much work and much more coordination/risk of error.
Scott

[Anonymous]

I am talking mostly about ArchiCAD slowness. Yes I compare to AutoCAD as well, but the main target is still ArchiCAD.
My point is that just an average workstation at price ~1500€ is almost equal to insane level workstation costing 5000+ €. You will see almost no diference. It will be just a disappointment. There is no difference if you must wait 10 seconds to complete the operation on ordinary workstation or you must wait 8 seconds on insane workstation. And when 2D geometry coomes into the game, seconds turn into minutes! Is it worth 3500€? You are the one who decides. As for me, I've made my choice.

[Anonymous]

Vahur,
I agree there is little point in a $10k workstation for BIM as it generally does not have a linear payback, most of the time the workstation is waiting for our input but a fraction of the time we are waiting for it to regenerate (this may feel like it is constantly happening but would probably only amount to 1 to 2% of the time we are actually using the program), this is why the new workstations I am rolling out in the office are single processor i7s with fast SSD and 4Gb quadro graphics.
I am also looking at one workstation to be setup with multi CPU lots of RAM and multiple graphics cards for GPU rendering (Octane) for our graphics person to use.
Scott

[shtarkel]

Yep, buy a fast INTEL CPU with 6 cores. RAM is used in Cinerender a lot, so more is better. Don't buy expensive GPUs k620 will do the job for most projects or try using gaming GPU like GTX 970
SSD for sure!