OTOY Forums

I'm fine with spending $2,000 on a card if it's works 4 times better than the $500 gaming card...

I'm a person who likes to have options!

Poly

Truth be known those cards are probably exactly the same except one has some severed areas to cripple it.

So according to that post on BA, the 680 will be bad for OpenGL and for gpgpu.

In this case users of Blender and Octane should stay clear of it.

EDIT: the 680 is a perfect marketing trick. Call it 680 and give it 1500 CUDA cores (which are not the same as fermi cores). People will look at the numbers and automatically think it will perform 3x faster than the previous family. Such a ripoff

well I am not quite sure about that. Octane uses single precision shaders and that's what is strong about the 680. Double precision is what is the weak point of the 680.
So maybe the 680 is just right for Octane, we will see.

Leiurus wrote:The thing is that the product performs very poorly on all platform, it hasn't been a cold shower for the Octane community only but for all users of GPU accelerated render engine.

The thread I'm refering to is this one, from a Blender users forum (don't know if it's allowed to post links rederecting to other similar websites, however the discussion thread treats of the gtx680 specifically and of GPU rendering in general, so I guess it should be OK. If not just let me know and I'll edit my post):

http://blenderartists.org/forum/showthread.php?249681-Nvidia-GeForce-GTX680-released...

A particular post retained my attention, written by Stargeizer:

Ok, let's talk about the last line of this posting: "It's all about business".
That's correct, but what kind of business are we talking about? If Nvidia thinks, that people will start to buy their 2000-4000€ Quadro and Tesla cards instead of 300-500€ gamer cards, just because of a misleading marketing strategy, they will fail miserably. Why? Because Quadro cards are mostly bought by big studios who can afford them. But here in Germany we have many small companies, struggling to survive in this sick industry. In this area it's not so much about quality, but about being cheaper than the competition, therefore prices for CGI jobs were ruined over the last few years. Some companies have to work on 2-3 projects at the same time, working their asses off through the night and over the weekends just to pay their bills.
There is only small financial headroom to spend money on software licenses and hardware. Quadros are simply not affordable. What we need are efficient solutions at fair price levels. This is why Octane is so attractive. It is available at a fair price level, runs on affordable hardware and delivers stunning results in a short amount of time. But if I have to spend 2000+ € just for a CUDA card in the future, it might be better to buy 2-3 CPU-based renderclients instead and continue to use traditonal biased renderers.
As mentioned by Stargeizer, AMD could turn things around. While CUDA seems to become slower with every newer 4.x release, OpenCL makes nice progress. The LuxGPU benchmark is an indicator for the performance that can be achieved with OpenCL on AMD's current line of gamer cards.
So, if Nvidia continues to focus on "big business" with their CUDA products, the situation could change quickly, making OpenCL the favourable solution for both, software developers and customers.
Brecht van Lommel, the developer of Cycles, the new render engine in Blender, which can use either OpenCL or CUDA, mentioned in an Interview, that OpenCL is very similar to CUDA and as long as you restrict your code to the feature set of OpenCL it's easy to convert it to CUDA and to maintain both code branches. It would be interesting so see a speed comparison between Cycles OpenCL on a Radeon 7970 and Cycles CUDA on a GTX680. Wouldn't make me wonder, if the Radeon would be faster.

But time will tell. Maybe all the drama will settle down in a few weeks/months, when drivers and software are optimized for the new architecture. It's too early to abandon all hope.

Maybe it is all a big marketing trick where they're deliberately restricting performance of certain cards, so then maybe they will come out with cheaper Quadros and Teslas that everyone will be sure to jump on.

One can dream, right?

My decision to purchase a GTX 680 4GB depends on compatibility with Octane Render.
There are anticipated update? Or can I buy a GTX 580 with 3gb?

Nice read,

http://www.eetimes.com/electronics-news ... s-analyst-

Regards,
Poly

Good find. I haven't seen that info mentioned anywhere before.
To my way of thinking there is a fair amount of BS in there though. ECC memory and double precision are not necessary for Geforce and dont justify snail like performance with single precision.
Its fairly obvious Nvidia have been embarrassed by the very poor compute performance of Geforce Kepler and are quickly publicising a coming revision without actually calling it one and even though it will actually take 6 months or more for the cards to appear.
Probably what they will show in May will be an early engineering sample of the Quadro/Tesla build but they will call it Geforce Plus or something. Something to save face anyway and hold customers interest from straying to ATI.
Given the time and expense to make a chip I can't see how this one can be basically any different other than to either not be deliberately crippled or have extra duplicate modules hastily tacked on to extra real estate.
If it is a bigger chip then that gives an excuse to charge a premium price for how it should have been to begin with.
However you have to ask if the performance is substantially better from this second tier how that ties in with Quadro and Tesla Kepler. It seems likely the performance will be somewhere in between for an in between price.
The good news I suppose is that Nvidia are doing something about it.
I still think Refractive should defer revising their code for Kepler in the meantime and concentrate on finishing v1.
Looks too, given this info, like Octane users should hold off upgrading hardware for at least 6 months in anticipation of something 3/4 decent instead of hoping on hope to coax performance out of something fairly pathetic.

pixelrush wrote:I still think Refractive should defer revising their code for Kepler in the meantime and concentrate on finishing v1.

I think you meant to say they should revise the code for opencl