SpaceRay
![]() |
I have just upgraded my computer from an Quad Core AMD X4 645 to an Quad Core Intel i7 2600K. Also have upgraded to 16 GB 14900 DDR 3 memory and now I am running Windows 7 and all the applications from a OCZ Vertex 3 120GB SATA 3.0 6GB/s SSD drive (which from the benchmarks I have seen is one of the fastest if connected to a good SATA 3.0 controller using a 6GB/s cable)
Making some speed tests I am VERY SURPRISED to see the results !!!
Please, Please, I want to ask if I am making something wrong, or have the wrong configuration, or there something I have to change to get better speed, or this is right and I am not making nothing wrong ??? FILTER FORGE VERSIONS USED FIlter Forge 2 (FF2) is latest 2.011 Filter Forge 3 (FF3) is latest 3.03 Using both in Standalone version (not using photoshop) I will explain what I have done to test the speed NUMBER 1(N1) Old Computer AMD X4 645 8 GB Value ram 5400 rmp Hard drive SATA 2.0 3GB/s Nvidia Geforce 9600 GT with latest drivers Windows 7 Home premium 64 bit NUMBER 2(N2) New Computer Intel i7 2600K 16 GB 14900 DDR3 Special RAM SSD SATA 3.0 6GB/s Nvidia Geforce 9600 GT with latest drivers Windows 7 Home premium 64 bit RENDER SPEED RESULTS TEST 1 - Render a 2000x2000 texture using Crysta filter by Vladimir Golovin N1 - FF2 - 21 MINUTES N2 - FF2 - 15 MINUTES (only 6 minutes difference ???) N2 - FF3 - 24 MINUTES !!! (24 Minutes on computer 2 ??? Higher than computer 1 ?) TEST 2 - Render a 2000x2000 texture using Creepy filter by Vladimir Golovin N1 - FF2 - 11 MINUTES N2 - FF2 - 8 MINUTES (only 3 minutes difference ???) N2 - FF3 - 12 MINUTES (12 Minutes on computer 2 ??? Higher than computer 1 ?) FILTER FORGE CONFIGURATION On both FF I have put the same configuration, and is the default configuration, I mean: Use ALL available CPU/Cores Use Direct X style Using output dithering Optimizing Blur Calculations Temporary folder in C: with 56 GB free Anti Alias OFF Not using double precision Gamma correction enabled COMPUTER ACTIVITY WHILE RENDERING I have seen that in both the CPU activity is between 92% and 98% with 4 cores active on the AMD and 8 cores (4 real+ 4 HT)active on the intel. The RAM consumed was only 2GB. CONCLUSION I am very surprised that there is little difference between computer 1 and 2, and I think that at least it should be on FF2 64bit around half the time than it was on FF 2 64 bit on the first computer. Also very surprised that FF3 is SLOWER than FF2 64 bit on computer 1 !!!! How is possible ? There is so much difference between 32 bit and 64 bit ? Or is it because is still a beta version ? At first I wanted to make the test with 4000x4000 but it was REALLY VERY SLOW so I decided to make it at 2000x2000. |
|||||
Posted: August 14, 2011 5:21 am | ||||||
Sphinx.
![]() |
Did you set FF3 multipass setting to legacy (i.e. FF2 "mode")? My tests show that the new multipass rendering slows down things a little. Also try enabling double precision output cache (it may improve performance).
Here are the numbers from my quick test (crysta, first preset, 600x600, old dual core, vista): 5:33 (multipass: 4 pass) 5:18 (multipass: off) 5:17 (multipass: legacy) 5:04 (multipass: legacy; double precision output cache enabled) |
|||||
Posted: August 14, 2011 7:16 am | ||||||
Skybase
![]() |
I wonder if this starts heavily depending on processor... I've done some testing since I was curious and had some time to spare.
I rendered out a 600x600 DOGS filter using the first preset with everything set to default. Antialiasing was on as usual. Method of testing: Do the render 10 times using the same preset with some time in between, then average results. This test was done using my macbook pro running MacOSX 10.6.8 (not Lion). The processor was Intel Core i7 2.66GHz with 8GB of RAM (DDR3). All multipass renders were rendered with 100% degradation. All results are in seconds -- Using Multipass 5 -- (1) 3.31 (2) 3.34 (3) 3.49 (4) 3.54 (5) 3.47 (6) 4.78 (7) 4.39 ( ![]() MEAN: 4.16 seconds -- Using Multipass 4 -- (1) 3.47 (2) 3.05 (3) 3.19 (4) 3.10 (5) 3.33 (6) 5.46 (7) 4.31 ( ![]() MEAN: 3.866 seconds -- Using Multipass 3 -- (1) 3.28 (2) 2.92 (3) 3.50 (4) 3.87 (5) 3.03 (6) 4.57 (7) 3.88 ( ![]() MEAN: 3.60 seconds -- Using Legacy -- (1) 3.43 (2) 3.91 (3) 3.87 (4) 3.88 (5) 3.81 (6) 4.01 (7) 3.81 ( ![]() MEAN: 3.722 seconds Now... let me just say that I don't think these numbers represent too much. For one thing I'm not a technician and I haven't done this with my comp at the purest point. In other words I do have some programs running and they will contribute to the numbers somehow. I don't know how exactly FilterForge's render engine runs and I don't know what processes go behind the render. Either way I did these tests in a manner to see if Multipass is any better than legacy. Turns out its almost no different for this filter at 600x600 pixels. Maybe it'll change significantly with higher resolutions. I'll let somebody else try that. Oh well. I really enjoy progressive renders however. ![]() |
|||||
Posted: August 14, 2011 11:11 am | ||||||
Sphinx.
![]() |
I suspect the differences btwn.the various multipass settings highly depend on the actual filter. The more components (implicitly: amount of instruction data) the greater the difference.
This suspicion is based how the CPU cache levels work. If you have a small amount of instruction data that is used repeatedly, there is a higher chance that it remains in the top cache level of the CPU. Since the multipass rendering starts off by rendering over a large area this could mean a greater change in instruction data, causing more cache flushes than if the filter rendered continuously. If this is correct we should see a greater difference btwn. the various multipass settings in larger complex filters (compared to that of small simple filters). But it very much depends on the actual filter construction... |
|||||
Posted: August 15, 2011 2:24 am | ||||||
SpaceRay
![]() |
Thanks Sphinx and Skybase for your help and test results, and perhaps using different multipass would increase the rendering speed, but I do not know if it would be really see a great difference in higher resolution, and not just A FEW SECONDS DIFFERENCE.
I will have to test with different multipass and see really if this improves the rendering speed. IS TRUE that the rendering time depends VERY MUCH on the filter construction, and what the filter does, and how much CPU intensive is the filter, for example Photo effect filter are really very fast, but others that modify and distort and change very much the image are really slow because must make a huge amount of calculations to get the result. I choosed to make the test with Crysta and Creepy because: They are very CPU intensive and is good to test the rendering speed. These filters do not depend on any image that must be loaded to work These have been created by Vladimir Golovin so is supossed that as he is the creator of FF he should have optimized these filters to work in the best way and get the most of FF. |
|||||
Posted: August 16, 2011 4:50 am | ||||||
GMM
Moderator
Posts: 3491 |
Spaceray, you should get the same results with FF2 and FF3 (or close to the same) if you set Progressive previews to Legacy and turn on preview cache preservation. These are the only two settings that make FF3 different from FF2.
(note that I mean preview cache preservation and not the double-precision thingy on the Rendering tab – leave that off). |
|||||
Posted: August 17, 2011 10:46 am | ||||||
SpaceRay
![]() |
OK, I will try to make many other tests and time each one and see what differences I can find by configuring it as you have said. There are only two settings that make FF3 different from FF2, BUT there is ALSO one perhaps very important difference, at least I think it is for me, and is that FF2 is 64 bit and FF3 is 32 bit, and I am using Windows 64 bit. ¿ There is NO difference in performance or render speed by using the 32bit or 64bit versions ? |
|||||
Posted: August 19, 2011 12:09 pm | ||||||
GMM
Moderator
Posts: 3491 |
We don't have (and don't plan) a 64-bit version of Filter Forge. It can work as a plugin under a 64-bit host application but it still remains 32-bit.
|
|||||
Posted: August 22, 2011 4:40 am | ||||||
SpaceRay
![]() |
I want to thank very much to Sphynx and Skybase that have taken the time to make the time test and have put the results here and I appreciate it.
I think I have solved the mystery and why FF was so slow and I want to sincerely apologize and say that I am very sorry I did not make more research and investigate more to see if the results from the first had been really true.
As the SLOW reason is NOT true. I am sorry that I had not time to compare yet FF2 to FF3 render speed, BUT I have made a benchmark of FF2 and found a very interesting thing that I did not know and I think is very important for thinking that FF could be slow I had a very wrong thought that if the time taken to render a 2000x2000 was 8 minutes, the render speed for a 1000x1000 would be 4 minutes, and really is a little more than 1 minute. So I am sorry that I had made this mistake and have wrong thinking. Here is the result to the benchmark done and have seen that instead of the 8 minutes from the first post, I have make it 3 times more in a different day and got that it really was around 6 minutes, so it really is nearly half the time of old computer 1
|
|||||
Posted: August 29, 2011 5:19 pm | ||||||
SpaceRay
![]() |
Please, read this thread here about this topic http://www.filterforge.com/forum/read...5&TID=8784 Thanks very much |
|||||
Posted: August 29, 2011 6:17 pm |
Filter Forge has a thriving, vibrant, knowledgeable user community. Feel free to join us and have fun!
33,712 Registered Users
+19 new in 30 days!
153,534 Posts
+31 new in 30 days!
15,348 Topics
+72 new in year!
25 unregistered users.