Speed 4.2.1373 vs 5.8

Hello,

I installed RT 4.2.1374 in 2016 on Ubuntu 16.10. It has been a stable workhorse sinced then.
The installation had been very complicated with “cmake” etc, see old thread:

->Q: Is there any special reason why the old RT 4.2.1373 on Ubuntu is almost twice as fast as my new RT 5.8 under WINDOWS 7 on the SAME machine (old dual boot HP z620 with 2 x E5-2680 - total 32 threads - and 64GB of non uniform addressed 1600MHz DDR3)??
(I based the comparison on the same profile and the same NEF-file - none of the newer processing tools used…)

Looks I 'll stay with the old version for now.

If you are processing the exact same images in exactly the same way, I don’t think 5.8 should be twice as slow. If anything, since v4.2 (which came out in late 2014!) there have been numerous speed-ups in various places in the code, so things should go faster… It may have to do with the fact that you’re comparing Linux and Windows. Some other programs in Windows may make file-access slower. So the question is: how accurately did you do your time measurements? And how busy is your CPU while exporting?

In terms of responsiveness of the GUI, the 4.2 release was still using GTK2 if I’m not mistaken. Currently we use GTK3 and that might be slightly less responsive. But I don’t think you’re talking about this type of speed.

I am talking about speed of the 100% preview. The MOST important speed, since it decides how fast I can work. (If I leave a batch processing job of 1000 RAWs overnight, I don t really care if it takes 1.5 hours or 6 hours). I can work almost fluently on 4.2.1373 in ubuntu, on 5.8 in Windows, 100% view is “laggy”…

The issue is ALWAYS the speed of the 100% view in 4K.
Everything else is lightning fast, no matter if Windows or Linux, i7 or dual Xeon, 32 or 64GB RAM… My stone-age old 32 threaded system has an edge here over the newer i7-system…

You have the 4k monitor on both systems? That’s a lot of pixels to push.

Right: 2 houses - 2 computer systems - 2 4K screens just for RT.
(plus one 1920 x 1200 screen on each system to show panel, desktop, taskbar, file browser and the like)

At the time I did the “cmake” install in 2016, there was a german guy from very western part of Germany near France in the #RT Hexchat room. Hed had an FX-8350 and was optimizing some RT code. He was helping me to fight against the many error messages. Thank You! According to what I remember, he told me, that cumbersome cmake install process can make RT faster through CPU optimization. I am not a programmer. It looks that this may be the cause why my ubuntu RT is faster than that on Windows, where I just installed using the installer. If so - will the Windows version ever be as fast as the Linux version? Or will I have to go through that painful error message fight again on my next machine, after the almost a decade old workstation has finally failed (due to some dry electrolytic on mainboard etc…)

Yes, it is quite possible. There are a few threads discussing how the different options may impact speed and optimization. There are several builds too, if you don’t want to do the building yourself.

For me, the main problem that I faced when first upgrading to the newer versions was GTK3 but it seems that they resolved the issue for the most part. However, that doesn’t seem to be your problem because you say that you only get the slowdown when you move preview at 100%.

Which brings me to this question: Could you make a short demo video demonstrating this problem? The more info you send our way the better, including debug logs, hardware info (SSD? driver issues?), etc.

That clearly sounds like I am this guy :slight_smile:

A native build will be faster than an installer build (which has to be build to run on every machine). For example, a native build can make use of SSE4, AVX, FMA and so on whereas a non-native x86_64 build is restricted to SSE2

If the 5.8 build is slower than the 4.2.1373 build on the same machine and same monitor in preview mode, then this is caused by gtk3 imho, as the image processing speed in RT has much improved since 4.2.x

1 Like

Hello, Heckflosse! I guessed you would be still around here, but I forgot your nickname. Saarland, isn’t it where you live? Did you upgrade from FX to Ryzen or T-Ripper in the meantime? I have a good memory - you helped me through that cmake build!!
I have no idea, what gtk3 is actually doing, and why it would be slow but I ll try to produce a video now of RT 5.8 in WIN7 being slow. Let’s hope I can attach it to that thread without hitting some MB limit or the like…

You could host it elsewhere…

I have now a 5 minute video. It is “only” 500MB. No idea where to host. I ll try to compress it using handbrake now.

VID_20200530_023423.mkv (29.0 MB)

1 Like