a question about ram and cpu stress

[rubeos]

When I use zbrush see in my task manager the cpu always does an hard work, although I have an I7 processor and CUDA.
While when I use 3dcoat the processor works at most 20%...this is looks good..but I can't see that, I suppose that the 3dcoat work charged is all to the RAM..My questions:
How many 3dcoat stressed the RAM?
May the continuous 3dcoat usage (no voxel but only high resolution surface) crash the ram memory hardware?
How can I check that does not happen?

[AbnRanger]

When I use zbrush see in my task manager the cpu always does an hard work, although I have an I7 processor and CUDA.

While when I use 3dcoat the processor works at most 20%...this is looks good..but I can't see that, I suppose that the 3dcoat work charged is all to the RAM..My questions:

How many 3dcoat stressed the RAM?

May the continuous 3dcoat usage (no voxel but only high resolution surface) crash the ram memory hardware?

How can I check that does not happen?

3D Coat is largely dependent on CPU multi-threading, but not all operations are multi-threaded. Most brush strokes are. RAM comes into play when you are adding a lot of layers or working with very dense objects, and large texture maps, etc. CUDA aids only Voxel Sculpting

[rubeos]

I understand ...my computer (looking at the task manager windows seven)  use the CPU (intel I7) not more than 20%, even when I sculpt objects with approximately 5.000.0000 visible triangles, in surface mode.
But navigation and especially the brushes become slower .... but the processor does not make much effort. I do not know what depends this, but it's good.
While in Zbrush, to make a comparison, with a 5 milion triangles object, the CPU runs at 100%

[AbnRanger]

You can't compare 3D Coat and ZBrush this way. 3D Coat is multi-threaded using Intel's TBB (Thread Building Blocks) library. ZBrush may be using something else. ZBrush is also constricted to 4GB because it is still a 32bit application. Using 100% of the clock cycles means it's putting a heavier load on the CPU. Why is that always a good thing? It's coded entirely different than 3D Coat, and uses a 2.5D mode, so there is a completely different engine under the hood.

 

Now, navigation performance is related to your graphic card, so CPU usage has practically nothing to do with it.

[ajz3d]

Two examples.

 

When brushing fast on a 15 mln tris sphere (surface) with Artman's LiveClay preset (radius=1.5), I get around 11-15% CPU load (usually - just one thread loaded up to 50%) and very choppy, jagged strokes showing up with a small delay. When brushing slow, the strokes are smooth, but the CPU load is the same. In both cases GPU load is minimal: around 3% and less than 10% after I finish the stroke (maybe CUDA smoothing kicks in?). If I choose a bigger radius of 4, the CPU and GPU load doesn't change when brushing, but the performance drops significantly.

 

Voxel brushes behave in a slightly different manner. Fast brushing with the radius of 1.5 Extrude brush, from the CPU side is the same as it was in the surface mode. Radius of 4 reaches 17-20% load. Performance is also almost identical. But if I pump the radius up to crazy number 15, the performance drops down almost to zero, but CPU load reaches only something around 50%.

GPU load in voxel brushing is practically the same as it is in surface mode.

 

Navigation in both cases was smooth.

 

Why some much CPU and GPU power is wasted?

Edited August 6, 2013 by ajz3d

[AbnRanger]

That's something you'll have to ask Andrew about. All I know is that it's MUCH faster now than it was several months ago. I'm sure hardware makes some difference. For example, you'll notice some slowdowns if you are near your RAM limit, and if I run memory at default 1066 speeds, instead of their rated 1866mhz there is a big difference.

 

http://www.youtube.com/watch?v=RIGMArh0myo&list=PL0614F2A03AD725CD&index=9

 

http://www.youtube.com/watch?v=2c5THKHI4_o&list=PL0614F2A03AD725CD

 

The new Surface mode brushes are ridiculously fast on my PC, so I don't know what is going on with your system. You haven't mentioned any system specs, what version you're using, DirectX or OpenGL, etc.

[rubeos]

You can't compare 3D Coat and ZBrush this way. 3D Coat is multi-threaded using Intel's TBB (Thread Building Blocks) library. ZBrush may be using something else. ZBrush is also constricted to 4GB because it is still a 32bit application. Using 100% of the clock cycles means it's putting a heavier load on the CPU. Why is that always a good thing? It's coded entirely different than 3D Coat, and uses a 2.5D mode, so there is a completely different engine under the hood.

 

Now, navigation performance is related to your graphic card, so CPU usage has practically nothing to do with it.

 

 

 

 

 

 

Why is that always a good thing?

 

 

Thank you for the answers, I wanted to say that maybe is a good thing that 3DCoat not overdrive the processor as it does Zbrush with mesh with the same number of polygons.

Perhaps 3Dcoat has a better engine of Zbrush as regards the processor management ?... although it should be taken into account that Zbrush is only 32bit.

My initial question was to know if 3Dcoat keeps stress the RAM memory, in a way that I can not see...while I can see (by task manger)  the processor it does not seem to make much effort .. just 20%.

Edited August 7, 2013 by rubeos

[rubeos]

Two examples.

 

When brushing fast on a 15 mln tris sphere (surface) with Artman's LiveClay preset (radius=1.5), I get around 11-15% CPU load (usually - just one thread loaded up to 50%) and very choppy, jagged strokes showing up with a small delay. When brushing slow, the strokes are smooth, but the CPU load is the same. In both cases GPU load is minimal: around 3% and less than 10% after I finish the stroke (maybe CUDA smoothing kicks in?). If I choose a bigger radius of 4, the CPU and GPU load doesn't change when brushing, but the performance drops significantly.

 

Voxel brushes behave in a slightly different manner. Fast brushing with the radius of 1.5 Extrude brush, from the CPU side is the same as it was in the surface mode. Radius of 4 reaches 17-20% load. Performance is also almost identical. But if I pump the radius up to crazy number 15, the performance drops down almost to zero, but CPU load reaches only something around 50%.

GPU load in voxel brushing is practically the same as it is in surface mode.

 

Navigation in both cases was smooth.

 

Why some much CPU and GPU power is wasted?

 

I do not know if I understand correctly, but I get the same performance, for example:

-Voxel: a sphere with 8.000.000 tris (I do not dare get up to 15 million!) using clay brush, it becomes very slow .... but maybe this is normal and happens to everyone.

-Surface: a sphere with 8.000.000 tris and using Artman's LiveClay preset: it is much faster than brush 'clay voxel'..... but only if 'stretching' option is disabled.

While navigation is slow, but not slow as in voxels.

In both cases, voxel and surface, my CPU does not go over 20% of work .I do not know if my pc does a good performance or not.

 

[rubeos]

That's something you'll have to ask Andrew about. All I know is that it's MUCH faster now than it was several months ago. I'm sure hardware makes some difference. For example, you'll notice some slowdowns if you are near your RAM limit, and if I run memory at default 1066 speeds, instead of their rated 1866mhz there is a big difference.

 

http://www.youtube.com/watch?v=RIGMArh0myo&list=PL0614F2A03AD725CD&index=9

 

http://www.youtube.com/watch?v=2c5THKHI4_o&list=PL0614F2A03AD725CD

 

The new Surface mode brushes are ridiculously fast on my PC, so I don't know what is going on with your system. You haven't mentioned any system specs, what version you're using, DirectX or OpenGL, etc.

 

The specs of my system are: - Win 7 64bit - RAM 6 GB - Intel I7 CPU, 2.7 GHz - graphic card: Nvidia GeForce GTS 250 - DirectX 11 - CUDA driver version 280.26

I did test only recent 3dCoat version, 4.0.06A CUDA (DX64) and (GL64) .. I do not know the differences between DX and GL.

 

[AbnRanger]

The specs of my system are: - Win 7 64bit - RAM 6 GB - Intel I7 CPU, 2.7 GHz - graphic card: Nvidia GeForce GTS 250 - DirectX 11 - CUDA driver version 280.26

I did test only recent 3dCoat version, 4.0.06A CUDA (DX64) and (GL64) .. I do not know the differences between DX and GL.

 

A lot also depends on how large the brush size is. Hit the W key on your keyboard, with that 8 mill poly sphere, and you can see that it's VERY dense. Way too dense to be practical. Remember, 8 million polys on a larger object like a character, is no big deal. But on a relatively small object, it's overkill. Try working on the bust or the Human Primitive from the Splash screen. That will be a better test. You can increase the resolution as you need it. Also, trying to compare performance in 3D Coat's Voxels vs ZBrush isn't a fair apples to apples comparison. Surface mode is. Why? Because ZBrush is only pushing polys...it's not having to carve/build volumetric pixels and remesh the outer skin simultaneously, the way 3D Coat is.

 

Pixologic tried to copy Voxels in ZBrush, with "Dynamesh," but fell far short. It has an extremely low limitation threshold and you have to hit a hotkey to remesh the object. No arbitrary limits in 3D Coat and it remeshes on the fly. With all of that said, DX tends to perform much better than OpenGl (3D Coat has those 2 different versions to startup with), and with that amount of RAM, I wouldn't try to push your system working with Voxels. If you need to create some booleans, it's better to do it early in your sculpt, and keep the density only as high as is needed. Then, you should switch to Surface mode for most of your sculpting. I think this is where much of Andrew's focus has been over the past few years. You can tell, because Surface mode brushes far outnumber the Voxel mode brushes.

 

http://www.youtube.com/watch?v=zxHVejgoLZw&list=PLlQ3JITh9bXOfcZqGuO8gClZisN6yUd1x&index=5

[rubeos]

Yes, in fact 8 million is a heavy number on my computer! Thanks for the link ... I'll try to understand it better. And also for the tip on the DirectX rather than OpenGL, I did not know and certainly will help me.
I would like so that voxels could preserve the local detail, as is the case when using liveclay, and that they were not so slow whit medium / high resolutions
The voxel give an incredible freedom, but their resolution, even while you work, must always be homogeneous: this would not be a problem, if the computer did not slow down when the resolution increase.

 

I state that I am not technical expert .... especially I do not know well 3Dcoat and the workflow: for example I have the impression that voxel is good to make some sketches, or hardsurface, but otherwise it is better to do all with current liveclay.
It may be that in the future liveclay can totally replace voxels?
 

I prefer to work while preserving the local resolution, denser in some places than in others, from the beginning of the sculpture .... and NOT FORCE to proceed gradually increasing the resolution, as in Zbrush / mudbox...e others who have taken this easy way.
With 3DCoat, Scuptris and Blender I can sculpt as I like, but at different levels of comfort and precision: Sculptris is incredible, but the main problem is that it does not sustain mesh over 1 million polygons.
The dynamic sculpting Blender is too primitive and brushes still too uncomfortable ...and development is stopped.
Instead 3Dcoat reached a big level, and sculpt with liveclay is now really very good.
I really hope that Pilgway still improvement LC  and refine it.
I agree when you say that Zbrush is not able to copy the voxels with Dynamesh. The problem with Dynamesh is always the same: you lose DETAIL, every Remesh .... or, if you want to keep the detail, you are forced to increase the number of polygons of the entire mesh. This slows the machine .... at least mine.
I do not know how PIxologic solve this methodological problem .... but this is not the ultimate freedom that a sculptor can have, liveclay or dynamic topology are more elegant methods than dynamesh....probably, sooner or later they also will introduce dynamic sculpting technology.
 

Edited August 8, 2013 by rubeos

[Tony Nemo]

Voxels, being 3D cubes through the whole volume instead pf 2d polygons on a surface, take far more memory than polygons. With your limited resources, it would be best to confine your voxel work to simple shapes and booleans. For details on the surface, switch to Surface mode and be sure not to return to Voxels.

Edited August 8, 2013 by Tony Nemo

[AbnRanger]

Voxels, being 3D cubes through the whole volume instead pf 2d polygons on a surface, take far more memory than polygons. With your limited resources, it would be best to confine your voxel work to simple shapes and booleans. For details on the surface, switch to Surface mode and be sure not to return to Voxels.

Agreed. Until recently, I always preferred to go as far as I can in Voxels...and on a beefy system, I can reach 70-100mill polys in voxel mode, easily. But now, I don't think I will be sculpting much in Voxels. Surface mode is where the sweet spot is. Booleans and merging take longer in Surface mode, so I try to do that in voxels early on. But beyond that, the Surface mode brushes are so good now, it only makes sense to do most of the sculpting there. This (dynamic subdivision) is the future of digital sculpting.

 

Raul is supposed to be back in Kiev, this month, so we should expect to see even more LiveClay goodness, soon.

[rubeos]

Voxels, being 3D cubes through the whole volume instead pf 2d polygons on a surface, take far more memory than polygons. With your limited resources, it would be best to confine your voxel work to simple shapes and booleans. For details on the surface, switch to Surface mode and be sure not to return to Voxels.

 

Yes, I think I'll do that, like this: sketches in voxel and details work on surface ... 90% of the work will be with brushes and surface liveclay

[rubeos]

Agreed. Until recently, I always preferred to go as far as I can in Voxels...and on a beefy system, I can reach 70-100mill polys in voxel mode, easily. But now, I don't think I will be sculpting much in Voxels. Surface mode is where the sweet spot is. Booleans and merging take longer in Surface mode, so I try to do that in voxels early on. But beyond that, the Surface mode brushes are so good now, it only makes sense to do most of the sculpting there. This (dynamic subdivision) is the future of digital sculpting.

 

Raul is supposed to be back in Kiev, this month, so we should expect to see even more LiveClay goodness, soon.

 

100 million polygons! incredible ... basically with your pc you can start a sculpture starting directly from the maximum resolution...like real clay.

I also believe that the dynamic division is the future ... and I hope to be improved even more ... for example maybe there is the possibility in the future to imitate even more classic hardsurface modelling, as architectural and mechanical engineering..

I did not know about Raul, and this news I really like .... I often look at his topic on this site, in the hope that there are updates!

However, really this latest version is already outstanding.

 

[Voxelapocalypse]

More good tips and info.  

 

...So one strategy could be to simply save an object while still in Voxel Mode, at a point where you realize

if you go any further, and jump into Surface Mode, you may find out you want to change somethings

only possible on a voxel level, like major tweeking or addition/subtraction of the object.

This way you can just import the previous voxel object and rework that, then jump into Surface Mode.

It would be like having an earlier stage of your model before it got out of hand unexpectedly

you could always have to fall back on and regroup with and try again.

Edited August 16, 2013 by Voxelapocalypse

[Tony Nemo]

More good tips and info.  

 

...So one strategy could be to simply save an object while still in Voxel Mode, at a point where you realize

if you go any further, and jump into Surface Mode, you may find out you want to change somethings

only possible on a voxel level, like major tweeking or addition/subtraction of the object.

This way you can just import the previous voxel object and rework that, then jump into Surface Mode.

It would be like having an earlier stage of your model before it got out of hand unexpectedly

you could always have to fall back on and regroup with and try again.

A very good idea!

[BeatKitano]

Agreed. Until recently, I always preferred to go as far as I can in Voxels...and on a beefy system, I can reach 70-100mill polys in voxel mode, easily. But now, I don't think I will be sculpting much in Voxels. Surface mode is where the sweet spot is. Booleans and merging take longer in Surface mode, so I try to do that in voxels early on. But beyond that, the Surface mode brushes are so good now, it only makes sense to do most of the sculpting there. This (dynamic subdivision) is the future of digital sculpting.

 

Raul is supposed to be back in Kiev, this month, so we should expect to see even more LiveClay goodness, soon.

Glad you finaly agree

[AbnRanger]

Glad you finaly agree

Well, we actually have agreed on a lot of things ....but it's the work Andrew did, in consultation with Artman, recently, and with LiveClay being much more stable/usable now, that has convinced me. Plus, I really only used about 2-3 brushes in Voxel mode. The number has grown considerably in Surface mode, so that seems to be where Andrew intends for the user to do most of their sculpting, now.