[seghier]

new subdivision tool is great .

[juan]

Hi all,

First a bit of history to help we all are in the same context.

Between 1.0 and 2.0 we released four version, with tons of things for free including improvements in speed, a new physical sky model, new sss, on the fly displacement, instances, shift lens, etc

Between 2.0 and 3.0, apart from all the improvements of 2.0 which were many, we released for free the following features (I am only pointing out the main ones, full lists are easy to find in our knowledge base or googling):
-2.1: shadow pass with GI, multishadow, new skydome model.
-2.5: FIRE
-2.6: Hair and Fur, Particle Rendering, MXS references, faster motion blur, scripting, vector displacement, integration with RealFlow
-2.7: Grass system, MXS/MXI compression

I am sure most of you think that we could had released more non free upgrades with some of the features above. I am myself a customer of several software tools and I am not remotely used to get that amount of things for free; I truly believe Maxwell has been unique on that too. Since the first alpha released in 2004 we have released two non free updates, in nine years, from time to time we need to feed our kids :).

So anyone who upgrades to v3 will not only get 3.0 but 3.x upgrades, as we have always done. The v3.x path is still not totally defined, because things evolve quickly in rendering and we need to keep flexible, so nothing can be said about that, but of course we have part of the team already working on that.

Now let's talk about speed. We would be insane if we did not work on improving speed, it is just that speed is a very general thing and in this release we have focused mainly on massive scenes, with lots of geometry, motion blur, particles, grass, hair, etc. These optimizations are huge, sometimes you can get a speed up of more than x10. It is true that other areas maybe are not getting such a strong direct benefit from v3 optimizations but maybe they are where we will put the focus in the next update. It is true that interiors and caustics through specular surfaces are two of the most important open problems in unbiased raytracers, solving them is possible, but solving them without tricks and without sacrificing performance in other scenes is not easy at all. One aspect of speed that is not mentioned often is how multithreading friendly Maxwell is. A big CPU manufacturer just sent us a prototype with 120 cores and it renders benchwell in 68 seconds (we cannot publish these results because the machine has not been released yet), Maxwell scales in such a linear way that if you want it to be 2x faster, at least you have the option to get a machine 2x more powerful, something that is usually not so easy because most of the applications stop scaling linearly when the number of cores is too fast.

I really expected a real improvement in terms of speed and finally a fix of the well known old engine problems and limitations ...

There are many limitations fixed that of course are not pointed out in the highlights but we have fixed things that have failed forever like facetting in anisotropy and other artifacts. Remember when it was not possible having instances with displacement together, or displacement with motion blur. In each release we try to get rid of these kind of limitations, for us this is even more important than new features.

But there is still no Hdri + sun. And a simple, flexible and robust dirt solution is also one thing that i'm still missing - for me a key feature to realistic renderings and standard in other engines. But maybe i was the only one submitting this in the survey...

I read all the survey entries and honestly I just remember a couple of requests about that, but now it seems there is more people who would really welcome it. We will try to get this done for an early 3.x update.

Thanks,

Juan

[choo-chee]

we also need to feed our kids, so we both need more SPEED heh heh ....
and let me offer another idea which could be easy to install :
came in this morning to find out that most of my weekend renders were destroyed because of power failure.
So how about an option to back mxi every hour or something, so if the "rendering" mxi is lost, I still have something that was written to disk not long before...

[photomg1]

Juan,
thank you so much for your post . You have thrown a bone for us to chew on : ) . It's given me some confidence that you understand what needs to be addressed and are working towards achieving that goal.
It would still be great if we could limit (dumb down) maxwell to get a speed up . Regardless of that thank you again.

That chip the 120 core , does it sit on a pcie lane? I'm sure you cant answer but it would be sweet to know.If it doesn't I'd bet we would have to sell a ferrari to purchase it

best regards

roll on v3 !!

[mojo]

hdri+sun and dirt would be very welcome for me, too.

great job by the way, im looking forward to play with the new v3 features...

Regards,
mojo

[numerobis]

Thanks for the reply Juan!
First of all i want to mention that i don't meant that there is no progress at all. There have been many important improvements within the last years and the results of mxi compression for instance are still impressing me. And i'm aware of all these free upgrades between the major releases. I didn't say that we don't get enough for our money. But there are still some limitations that make it impossible to use maxwell at all in some cases. (without getting wrong results which have to be fixed/hidden in post. I just think that a render engine must be capable to deal with any kind of scene to be fully usable. And i can say that really love to work with maxwell - otherwise i wouldn't have used it for such a long time - despite all these difficulties.

...massive scenes, with lots of geometry, ... grass, ... etc. These optimizations are huge

(...hopefully the ram usage of grass could be also reduced.)

A big CPU manufacturer just sent us a prototype with 120 cores and it renders benchwell in 68 seconds (we cannot publish these results because the machine has not been released yet), ...

Xeon Phi support for maxwell?!?

There are many limitations fixed that of course are not pointed out in the highlights but we have fixed things that have failed forever like facetting in anisotropy and other artifacts. Remember when it was not possible having instances with displacement together, or displacement with motion blur. In each release we try to get rid of these kind of limitations, for us this is even more important than new features.
...

I read all the survey entries and honestly I just remember a couple of requests about that, but now it seems there is more people who would really welcome it. We will try to get this done for an early 3.x update.

That sounds good

[choo-chee]

I read all the survey entries and honestly I just remember a couple of requests about that, but now it seems there is more people who would really welcome it. We will try to get this done for an early 3.x update.

Thanks,

Juan

hope you have seen my request about supporting hebrew fonts ...

[feynman]

Rendering speed is a non issue for me thanks to rendering farms. Even for test renders, if I need them fast. Pad your fee a bit and problem solved.

Lucky you, being able to pad any industrial design phase with $1000 for 10 high-quality test renderings. Unfortunately, in the academic and small studio environment, render farm usage by way of fee-padding is not an option.

[choo-chee]

Rendering speed is a non issue for me thanks to rendering farms. Even for test renders, if I need them fast. Pad your fee a bit and problem solved.

Lucky you, being able to pad any industrial design phase with $1000 for 10 high-quality test renderings. Unfortunately, in the academic and small studio environment, render farm usage by way of fee-padding is not an option.

+SL100 ....

[l1407]

Thanks for the reply Juan!
First of all i want to mention that i don't meant that there is no progress at all. There have been many important improvements within the last years and the results of mxi compression for instance are still impressing me. And i'm aware of all these free upgrades between the major releases. I didn't say that we don't get enough for our money. But there are still some limitations that make it impossible to use maxwell at all in some cases. (without getting wrong results which have to be fixed/hidden in post. I just think that a render engine must be capable to deal with any kind of scene to be fully usable. And i can say that really love to work with maxwell - otherwise i wouldn't have used it for such a long time - despite all these difficulties.

...massive scenes, with lots of geometry, ... grass, ... etc. These optimizations are huge

(...hopefully the ram usage of grass could be also reduced.)

A big CPU manufacturer just sent us a prototype with 120 cores and it renders benchwell in 68 seconds (we cannot publish these results because the machine has not been released yet), ...

Xeon Phi support for maxwell?!?

There are many limitations fixed that of course are not pointed out in the highlights but we have fixed things that have failed forever like facetting in anisotropy and other artifacts. Remember when it was not possible having instances with displacement together, or displacement with motion blur. In each release we try to get rid of these kind of limitations, for us this is even more important than new features.
...

I read all the survey entries and honestly I just remember a couple of requests about that, but now it seems there is more people who would really welcome it. We will try to get this done for an early 3.x update.

That sounds good

Windows platform can't support xeon phi......!

[l1407]

A big CPU manufacturer just sent us a prototype with 120 cores

There is Intxx e7-8xxx, 15 cores x 8, but 1 cpu is about us$ 6000, so one this computer may be about us$ 60000........! which is not for common users......!

[numerobis]

Windows platform can't support xeon phi......!

http://software.intel.com/en-us/article ... oprocessor
http://software.intel.com/en-us/article ... stack-mpss

[photomg1]

I'd love that to be the case numerobis, but i1407 is most probably correct those 15 core cpu's are due soon. Also I almost feel sure next limit would be shouting about if it was a xeon phi.

We can dream though , for the record I would have jumped on the phi even if it was just for linux . Thats a crazy increase in speed from where I am today although I'm 99 percent sure that bench speed Juan gave
is from a configuration that i1407 mentioned. Not sure where he is getting the prices from.

[numerobis]

I'd love that to be the case numerobis, but i1407 is most probably correct those 15 core cpu's are due soon. Also I almost feel sure next limit would be shouting about if it was a xeon phi.

We can dream though , for the record I would have jumped on the phi even if it was just for linux . Thats a crazy increase in speed from where I am today although I'm 99 percent sure that bench speed Juan gave
is from a configuration that i1407 mentioned. Not sure where he is getting the prices from.

yes i thought that too, but this would not even be worth mentioning... you could always build an E7 8 CPU system for a crazy amount of money and see a logical speed increase compared to a single or dual socket sys. But for this amount of money you could also build yourself a 50 CPU farm with 300 cores @4-4,5GHz and a much higher benchmark.

[juan]

Sorry, it was 60 physical cores with hyperthreading (and no, it was not Phi). I just wanted to point out that we have put enormous efforts in making Maxwell to scale almost linearly with high number of threads, and this is a kind of optimization that does not seem very sexy at first but is critical in the long term.