Windows 10 - GTX 980 no performance gain, demand explanation

[Anonymous]

Dear graphisoft,

my client is using the new Archicad 19 and just went through an upgrade from
Win8.1 to Win10
GTX 770 2GB to GTX 980 4GB

Let me list the specifications for each card first:
GTX 770
2GB VRAM,
1536 shading units,
3.2GFLOPS Floatingpoint performance

GTX 980
4GB VRAM,
2048 shading units,
4.6GFLOPS Floatingpoint performance

We did document the 770 performance before the upgrade on a normal neigbourhood model, with a couple of trees and houses with enabled sun shadows ofc.
before:
We had aprox. 55-60FPS orbiting around the model on the 770
after:
We have in the same model orbiting at same distance 50-55 on 980
** also the redraw cicle for shadows seems to be longer/slower

I demand an explanation for my client, why was his upgrade in vain (wasted money)
he is using a brand new 4790K cpu with 4.4GHz and 16GB of ram so there is nothing such as CPU botleneck.

Let me give you a note, that its not a problem on our side, because we run a test in Maya and we got a performance increse from 69fps to 116fps which was expected.

I took a look at this: http://helpcenter.graphisoft.com/technotes/video-cards/recommended-video-cards-for-archicad-19/#Test...

Let me tell you in a proper maner that this is not helpful AT ALL, and the informations is not correct I think.
Just look at it yourselfes. What does this FPS ration thing even mean?

Let us take a look at 10M polygon test in Win8.1:


So you claim that a gtx 560 (kepler) is faster as a brand new K2200 (maxwell) quadro and at the same time the gtx 750 (maxwell) is 5x slower ???!!
Now just a fast check:
-gtx -560 - 1GB vram - 336 shading units - 1.1 GFLOPS - kepler
-gtx -750 - 1GB vram - 512 shading units - 1.1 GFLOPS - maxwell
gtx 750Ti - 2GB vram - 640 shading units - 1.3 GFLOPS - maxwell
K2200 ---- 4GB vram - 640 shading units - 1.3 GFLOPS - maxwell
for contrast the two above
gtx 770 ---2GB vram - 1536 shadin units - 3.2 GFLOPS - kepler
gtx 980 ---4GB vram - 2048 shadin units - 4.6 GFLOPS - maxwell

Does this make any sense to you ?

My client who upgraded because your suposedly new faster Optimised Opengl ,wants an explanation from me, and I want from you...

With this logic it seems that a gtx 560 from 2011 which is selling on ebay for 50€ is faster than a brand new K2200, gtx 980 and even gtx 770 ??

So should my client now throw away his expensive 980 and buy a K2200 for even more, so maybe it will be faster than a 50€ 560, altough your images are showing that its actually slower ?

I hope this is an official Archicad support forum, because I think that me myself and a lot of people need some real explanations here

[Anonymous]

matyasm wrote:
Dear graphisoft,
So you claim that a gtx 560 (kepler) is faster as a brand new K2200 (maxwell) quadro and at the same time the gtx 750 (maxwell) is 5x slower ???!!

what i've found - AC excerpt - "To avoid the potential accuracy problems, ArchiCAD is using Double Precision to calculate coordinate transformations. In case you need to work with huge coordinates, a new, invisible origin is created in the centre of weight of the placed elements every time you open the file. It results a new reference coordinate system.
Elements across all storeys and views (Section, Elevation, 3D, etc.) placed to the model will be included in the same calculation".

From what ive seen
the GTX 560 - SPP of 1088.6 with a ratio of 1/12 making DPP 90 (dpp unconfirmed)
vs
K2200 - SPP of 1280 with a ratio of 1/32 making DPP 40

I could not confirm the DPP number for the GTX 560 - but i did find the ratio.
correct me if im wrong, but i think if that number is correct, that's why the 560 does better.

https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#Comparison_tables:_Desktop_GP...

http://helpcenter.graphisoft.com/troubleshooting/3d-modeling/far-from-origin/

Ive been doing so much research lately, as i need a new machine, (despite the fact that my mbp easily moves around in 3d model at around 50fps) i'm very annoyed that it's taken me days to source the information above, also that it hasn't been provided by GS. My view is that Quardo cards for Archicad are a waste of money. The new Quadro M6000 with a freaking huge 24gb mem has worse double precision point numbers than a 2 year old Geoforce Titan BLACK or TITAN Z! (which is what i'll be buying if buy a desktop).

http://www.pcworld.com/article/2896411/the-brutal-graphics-war-continues-as-nvidia-reveals-the-gefor...

gtx 560 ratio source - https://devtalk.nvidia.com/default/topic/497325/double-precision-and-geforce-card-capable-of-double-...