[JTB]

I think GPU is here to stay.... There is Octane, now Blender has Cycles, hundreds of apps start using GPU power.... so I was wondering... what about Maxwell?
I mean, Fire is great but it is still too slow for some scenes, interiors for example... where the small things, the ones you will need to correct, come after SL12...

Is there anything new about Maxwell and GPU rendering?

[jespi]

Yeah, I also think GPU is the way to go. I've my eyes on Cycles, it's really impressive and well implemented within Blender. I wish Maxwell 3 will support GPU, if not I'll jump to another render engine, GPU's render time are a dream.

Have a nice day,

josé

[polynurb]

but has any existing system solved the memory issue?

[JTB]

I can only talk for myself... When I visit Octane's forum and see really nice arch renderings done in minutes, I am jealous... I have a EVGA GTX550ti with 2Gb and I think it is more than enough, considering what other in the forum have...
So, I guess the memory issue exists only for some large scenes.
I also found very interesting the fact that there is a biased rendering option in Octane with really nice results, it is very fast and if I want unbiased rendering, I don't have to change materials or anything else...
So, I guess I can live with a rare memory issue...

[numerobis]

or maybe wait for intels answer...

http://www.intel.com/content/www/us/en/ ... cture.html

http://www.xbitlabs.com/news/other/disp ... Shape.html
http://blogs.nvidia.com./2012/04/no-fre ... c-or-gpus/

the first card was announced to be released this year, but it looks more like 2013 now. I'm a bit afraid to see the price tag...

[JTB]

Impressive but in Greece, especially now, in the middle of the economical crisis, we don't get paid very well... So I will choose the cheap solution for 1-2 years...

[Michael Betke]

I use Octane Render on a regular base and mostly do not look back. Only instances are missing in Octane for my rendering jobs which need to be realistic mostly but not too realistic like if I would do as a product designer.
Arion Render seems to be solved memory issues like its written on their website. But never tried it.

One drawback hit the market with the GTX680. Nvidia crippled calculation performance and shifted this professional stuff only towards Quadro/Tesla. This makes is hard for small business like I run to build a GPU renderfarm. I will go and check whats more effective. Stay with my current gaming 580GTX set, go to a Hybrid renderer or buy CPU rendernodes.

One good thing with the Kepler GPU generation are 4GB Ram cards now on main-stream. With Tesla/Quadro there will be even more.

[bograt]

One drawback hit the market with the GTX680. Nvidia crippled calculation performance and shifted this professional stuff only towards Quadro/Tesla.

I bought a nvidia gtx 580, then saw the next day the 680 came out. I sent back and got the 680 and saw a big performance increase, like 2 580s almost (this was when testing arion render) this was (I presumed) because of the 1536 CUDA cores over the 580's 512 Cores.
why do you think the 580 was a 'drawback'?

[Michael Betke]

The 680 is "optimized" for gaming performance.
For computing it performs worse as a 560TI or so.

This is what I found in a hurry: http://www.anandtech.com/show/5805/nvid ... ra-fast/15
A 580SLI is significantly faster as a 690.

We wil lsee what Nvidia announces later this month.

[bograt]

I know what you are saying, but the ati cards are at the top and they dont even support cuda,
The 680 has 3 times the cuda cores of the 580

[dmeyer]

I know what you are saying, but the ati cards are at the top and they dont even support cuda

Exactly. No CUDA. Still a no-go unless you're totally OpenCL.

The 680 has 3 times the cuda cores of the 580

680 also has roughly 1/3 the double precision performance and a lower memory bandwidth. Kind of a fail as far as GPU compute is concerned. Real Kepler (GK110) is due this fall. The K20 Tesla was announced yesterday which will feature this, so Quadro versions likely to follow.

I have a feeling that nVidia is making efforts to increase differentiation between consumer and pro lines, like it was back in the 1998-2003 time frame. Lots of people using 'gaming' cards for 'pro' purposes cutting into profits.

[numerobis]

I know what you are saying, but the ati cards are at the top and they dont even support cuda,
The 680 has 3 times the cuda cores of the 580

yes, but it looks like the new cuda cores can't be compared directly to the older ones. The chip architecture is different for Kepler.

And the Radeon cards are faster in non CUDA benches. This is no CUDA benchmark.

I think nvidia wants to take the cheap consumer cards out of the GPGPU market to sell more Quadros and Teslas.

Some news from the ongoing GPU Technology Conference:

http://www.tomshardware.com/news/nvidia ... 15636.html
http://www.tomshardware.com/news/nvidia ... 15654.html

http://www.gputechconf.com/

[sandykoufax]

When NL adopt GPU rendering technology for Maxwell Fire, I'll buy a new graphic card.

[bograt]

And the Radeon cards are faster in non CUDA benches. This is no CUDA benchmark.

I think nvidia wants to take the cheap consumer cards out of the GPGPU market to sell more Quadros and Teslas.

I'm sure you are right, and I know this is no CUDA benchmark, its just that I use arion which supports CUDA cards only.
I am not an expert on graphics cards but I can confirm: the 680 is lightning fast with arion, substantially faster than the 580 is returned for it.
I checked the toms hardware post, It is very interesting so thanks for posting but I don't see it favouring the fermi based cards, look at where it mentions raytracing...

Quote: '11:01 - Now on to raytracing, and how the movie industry spends hundreds of computation hours on it. Now Huang is demoing realtime raytracing. Pretty impressive! Realtime raytraced lighting and reflections. How about realtime raytracing with fluid simulations? Yup.'

And just below the title: Nvidia shows the world that GPUs are fun, but not always about games.
It seems to me that the whole article is on how powerful the Kepler based cards such as the 680 are... have I missed something?

[polynurb]

... have I missed something?

i haven't read through all the links posted, but from earlier readings the issue with the new 6xx series is that double-precision float calculations are crippled on the gforce but not on the tesla, and the quadro i expect.

but that does not necessarily mean that render engines are affected by this, afaik it is more relevant for industrial and scientific calculations that require greater precision.