Welcome to Fractal Forums

Hey,

I just set up a new system with an NVidia card and started running the NV fragmentarium verions.
My scripts won't compile, but what bothers me is that the compiler won't tell me why. On AMD it used to say something and I was able to fix my code.
Now the only thing I am getting is: Failed to compile script (219 ms).

Is that normal on NV?
I'm running on Win 7 Ultimate.

so... v1.0.10 runs nice?

the only difference for nV vers is GL 4.0 minimum and AsmBrowser enabled.

too many topics for the same thing, maybe a moderator can merge these into one topic "problems running Fragmentarium" to tidy up this board! or maybe when you have a problem post it in "problems running Fragmentarium" thread ???

start with the default frags you used to create your current frags from and add things one at a time until it breaks
look up the specs for your board re:shader program size, number of uniforms etc etc

Certainly not normal on Nvidia. I have never had anything else than Nvidia cards.

Just a stupid suggestion: could it be that the log-window scrolls, so that you are only seeing the last line of the output?

You nailed it!

so more log needed... 100 lines 200 lines?

The problem is that for some reason now (and I don't know if its a matter of moving to NVidia, new system or 1.0.10 frag) it just does not scroll by line, but rather as a block and default scrolls to the last line (i.e. revealing the last line as a first line of the log window). Earlier I think It was possible to scroll it by line and even if it was scrolled down, it was scrolled to last/last line.

So now I needed to copy-paste the block from fragmentarium to an external editor to see entire log line by line.

// Edit//
Actually it works even weirder.
Scrolling from the top of the log downwards allows me to scroll by line almost to the end.
However, around 95% of the log window it stops scrolling by line and jumps to the very end of the log making it impossible to read a bunch of lines somewhere in between.


Next thing - on R9280 I wasn't getting any implicit cast warning from the compiler.
Now on NV I'm getting a bunch of lines like "0(123) : warning C7011: implicit cast from "int" to "float""
Does it make any sense to check all these lines and explicitly cast to match types?

hmm... never used or needed more than 20 or 30 lines and it should only call scrollToBottom() function after writing new lines, so you can see them as they are produced, not sure why the scroller widget (on the side of the log window) would behave strangely, the other windows all seem to work as expected, I am seeing the same effect, jumps to last page/line... investigating...

edit: I see the same log scroll bug in github-master (compiled/runnung on xeon 4 core)

the warnings are just letting you know of the cast action taken by the compiler and if the math is ok, that is, it is expected, then explicitly casting the values would be doing the same thing as the compiler is doing and is only really needed if you don't want to see warnings, of course this is based on my limited GLSL knowledge.

That's what I thought. Thanks.

Any ideas on how to avoid bumping up compile time?
- Usually my frags compile in 500-1000ms.
- Sometimes Frag becomes unresponsive for a few seconds, then it compiles successfully within 500-1000ms, but the total time is like 5-7 seconds, most of which is the unresponsiveness time. Not sure if its already compiling when it happens or not. I think I did not see that on R9280.
- Sometimes it stalls and after say 30 seconds it fails to compile only leaving a message "failed to compile". No hidden log above the scroll. This happens rarely.
- Sometimes the compile time jumps to 5-10 seconds and it compiles successfully.

* It is all dependent on my frags, compiling the same code gives always more less the same compile time.
* All these are stats for more less the same project and code, sometimes only with minor changes.

Do you have any suggestions or experience what should I avoid to keep compile time reasonable?
What are the things that are likely to bump up compile time?

it really all depends on how much stuff the compiler has to sort through, uniforms, custom data structures, textures, mapping... it has to do a pseudo run of the code to figure out what routines are actually getting called from... say... MathUtils or SpaceTransforms and only compile those in an effort to optimize the limited program space, it also, I think, does some funky GPU warp/thread optimizing that could, given a lot of complexity, lead to the compiler getting confused on how to fit everything in.

here is the output from desktop info center, this machine will not compile _scene.frag using PK_DE :(

RE:logs

the log area is made up of a list of widgets, some of these widgets can be many lines, when the log view area is shorter than the height of a widget ( compiler output is one widget in the list that can have many lines ) it shows the bottom of the widget, you can see this by selecting an entry in the log area and using the up/down arrows to scroll through them, having found this I checked the "scrollMode" and it was at the default ScrollPerItem so it is now set to ScrollPerPixel , this should help with log viewing ;) will be in next source release. (still testing)

Thanks for having looked into that.

I never needed more than 20 or 30 lines, I think I found something that (probably) contributed to the extreme lag, re: OpenGLInitializeFunctions(), it never lagged on linux so I had no idea it might be causing a problem, next compile for windows and source release will have this patch but still no EXR for windows :(

have you tested v1.0.10 on AMD ?

Yes. See http://www.fractalforums.com/fragmentarium/problems-running-fragmentarium/msg86562/#msg86562 and previous messages.

Hi!
What will be better for rendering - nvidia gtx660 or radeon 7850? nvidia has higher gpu frequency, but amd more stream processors. What component of videocard most critical for Frag?

review...
http://developer.download.nvidia.com/opengl/glsl/glsl_release_notes.pdf
nVidia is my choice ;) but do compare real-world scenarios, benchmarks, and user reviews
don't make a choice based only on a particular hardware feature,

 overall, performance, support, compatibility and stability are the most important things.

I can strongly discourage you from using Radeons. I had lots of issues with these in many scenarios. I think that their firmware and drivers for Win are underdeveloped.
I had serious issues and never succeeded using 2x R9 280X on my system in parallel (neither SLI or separate worked). COuld not do it properly with Davinci Resolve for grading, did not work at all for Fragmentarium. Those cards had also serious issues on how you connect and manage screens. Tons of frustrations.

I moved back to Nvidia and all these stupid problems disappeared all of a sudden.

FYI, AMD's glsl compilers conform to the standard.
Nvidia's don't by default. It even lets you use some HLSL intrinsic functions in glsl. It will only issue warnings by default. In order to make your shaders conform to GLSL on Nvidia you need to add #version xxxxx preprocessor thing (or something like that).
For example:

It wasn't the case for me.
Float x = 1; worked on AMD and gave warnings on Nvidia.

That is strange. I must be outdated.  88)

If you're in Berlin again I'd gladly buy a Radeon or two off you :D

I can send them our way if you liked. 2x not much used r9 280x

For realz, lemme know next time you're in Berlin, let's meet at Martin's :) Besides that we should really talk about fractal stuff, I can also offer you a 750 Ti or two for display GPUs (since usually it's best if the high powered GPUs are dedicated / not used for display).