anyone disappointed in upgrading video card for gpu rendering?

[SgWRX]

hello. so i have an nvidia k4000 card, two actually one at work and one at home. i was disappointed when bringing in my home card and installing it in my work computer and trying iray. it wasn't that much faster rendering.

 

my question is, has anyone upgraded from a more basic card like the k4000 to a new one and seen very noticable increase in performace (for either active shade or gpu rendering - either in VRAY or IRAY), or have you been disappointed.

 

i'm thinking of going the route of 2 quadro M4000's.

 

i realize there are many factors to be taken into account in terms of how detailed your scene is, what type of lighting etc...

 

thanks,

steve

[Nikolaos M]

What kind of model did you try to render with iray? Was it a complicated one (heavy on the Vram) or a simple one? The K4000 has 3gb of Vram and a model that demands an amount bigger than that in rendering would surpass the gpu and would be rendered with the cpu. Also, the K4000 has only 768 cuda cores of Kepler architecture. Don't expect to be lightning fast in gpu rendering (still, it should be faster than the cpu I think).

 

If you are really interested in gpu rendering, then avoid proffesional cards like quadros. They offer too small performance for their price. Gpu rendering uses single precision floating points, and any consumer card would perform nicely. Buy the newer architecture with the biggest Vram you can afford. The M4000 are not in the market yet, but I think they are expected to have a price close to 1000$ (more or less). 1664 cuda cores (same as the gtx 970's which costs 300+$= 1/3 of M4000's price) and 8gb of memory. On the other hand, with 1000$ one could get the Titan X which is a beast in performance, with 3072 cuda cores (almost twice the cores of M4000's) and 12gb of Vram.

 

So, if you really decide to go to gpu rendering for sure, then invest your money in Titans. That's my opinion.

[SgWRX]

just wish i could test a card. yeah the titan x has stuck out in my mind. worried about burning it out though.

 

so i have a bathroom i just tested and i got 2900 passes (which is at a point where it subjectively looks good - perhaps 2000 would have done the trick? i didn't have the time to watch) that took 2hr17m on the k4000 and 3.2Ghz Xeon 6core/12thread using 10 threads. render size = 1440px by 1800px.

 

no caustics, 12 photometric lights and self-illuminated light surfaces, 100k faces.

 

so where would i like it to be? 30mins

 

i did a tiny bump on the wall paint a little glossier than egg-shell, reflection map on the tile floor which was striated black/brown tile, so those two things being noticeable to me equates to "looks good".

 

at work, it'd be great to hammer out something in 1m that would give a pretty good idea to the people looking over my shoulders. which for me is about 70 passes at 720px by 900px - but you can't see the bump on the wall paint, and you can barely make out the bump map on a sofit over the sink which represents planked wood.

 

well heck, i'll attach a couple of images. 1 of the full size 2900 pass and 1 of the half size 70 pass. (the 70 pass one has a change of material orientation and reflection/bump on the doors and geometry)

 

[ATTACH=CONFIG]53584[/ATTACH][ATTACH=CONFIG]53585[/ATTACH]

[Nikolaos M]

The Titan X wont burn out with rendering. It's a monstrous gpu alright, but its default clocks (core and memory) are pretty low and its thermal behaviour during extreme tests is fine (as far as I know, I don't own one ).

 

As for your rendering times, are you sure the K4000 is enabled in iray's settings? Look at the "hardware resources" section of the iray settings window to enable it. I think your rendering times are huge. I've tryed iray with a car model I've made (almost 2 million polygons) and rendering times were fast. And all the examples I've watched in youtube were fast in rendering. I can't be totally sure of course.

[SgWRX]

oh yeah. the k4000 is enabled. i don't think poly count is a slow-down as long as it fits in memory on the card. i think it's # of lights and the amount of light. exterior renders with hdri or mrsun/sky are much faster!

 

i'll enable both CPU and GPU resources when i render. it's still the fastest.