It’s a beautifull file, thankyou Agus3D.
I am a Cycles noob, how come I can’t see a low res version in the 3D view when I select type: Rendered? It caches all the dupes but only the sky is visible, all the land and water remains black.
??? what are you saying!?..680 has 1500 CUDA, three times more, the weird benchmarks are the openCL ones…
That’s exactly what i was going to say…570 FTW! 
Best speed/price ratio
2 minutes 11.59 seconds using GPU rendering with my 2GB GTX 560 Ti 
I’ve done some other tests, so I decided to configure the scene for animation, with some tweaks my render times dropped to 45sec/frame with:
- 10 passes
- Decomposition between diffuse/glossy/transmission and direct/Indirect/color in the layer panel
- Re-composition in the compositor applying some bilateral blur filter to avoid noisy zones
The node configuration (before the original compositing):
Some screens
Original without decomposition - 10 passes:
New decomposited and re-composited WITHOUT bilateral blur - 10 passes:
(The color il slightly different because of the de-re composition process)
Original - 10 passes (final node):
New decomposited and re-composited WITH bilateral blur - 10 passes (final node):
Results:
Original - 100 passes - rendering animation with 5 frames - GPU time: 7:39.33
Modified - 10 passes - rendering animation with 5 frames - GPU time: 4:28.64
Considerations:
-
This is just an experiment and the result isn’t obviously good as with 100 passes but is suitable for much fast render and animation
-
The heavy part that leaks a lot of time is the “Copy transformation to device” and “building BVH”, so with cache BVH and some other tweaks the time have to drop down significatively
-
Sorry for my bad english
Sparazza
Wow ! Beautiful blend.
GPU render with the updated blend file: 2’ 25" on
Linux 3.1
CPU i7 2.67GHz
Nvidia GeForce GTX 275
12 Gib memory
Used modified blend file from http://www.blendswap.com/blends/landscapes/cycles-performing-landscapes/ (updated 2012-03-24 02:11:25)
CPU render: 4:34.61
GPU render: 2:08.70
Using Blender 2.62.0 r44136 (no BMesh).
1st peak: CPU temperatures.
2nd peak: GPU temperatures. (yellow line)
Specs: i7-2600K, 16 GB, 120 GB SSD, 1 TB HDD, GeForce GTX 560 1 GB, Win 7 Pro.
You have your gtx570 overclocked? That might explain the discrepancies in the render times. So you say that you feel comfortable having it overclocked? Haven’t had any overheating issues? How long have you been running like this? Are you using standard cooling or do you have 3rd party cooling for it? I’ve been thinking about doing it but don’t want to trash my brand new hardware.
iMac 8,1 Core 2 Duo 3.06 GHz, 4GB RAM
time: 08:13:14
Mem:102.37M
peak: 326.93M
First file rendered in about 47min, but I was watching some tutorials online in the same time.
I have absolutly no concern there my temps stay very low but if you look at the card it explaines these low temps (it’s the asus direct cu ii) they max out at about 55° C which is damm cool for gtx570s. With the overclocking I went slowly. I read for hours in forums about how to overclock safe and what is possible with my card. If you go slow and benchmark the setup for an hour or so with furmark, you should be on the safe side.
I guess with reverence designs you will not be able to get these low temps, but if you got good cooling allot is possible.
I hope it helped and don’t haste the overclocking because as always haste brings waste.
Here are my render details.
One with Default 60% Time taken: 4:40.52
and the Other at 100% (1080) Time taken:11:42.28
Render Image Links
60% Link : http://www.pasteall.org/pic/show.php?id=29065
http://www.pasteall.org/pic/show.php?id=29064
1080 Link : http://www.pasteall.org/pic/show.php?id=29067
http://www.pasteall.org/pic/show.php?id=29066
My sys config
Intel core i7 2600 CPU at 3.40GHz
8GB DDR3 RAM
XFX Geforce 7600GT
Win7 64bit
Hi all ;);
MyConfig :
- Windows 7 x64
- Atlhon II x4 3,5
- 16 GB DDR3 (1600Mhz)
- Azus CrossefireX 2x HD 7700
Blender 2.62.2 r45133 Compiled MSVC++ 2008 x64 & scons
CPU: Time:10:57.83
- Mem:140.42M (221.48, peak 364.76M) Time:10:57.83
GPU OpenCL/Juniper: Time:03.27.16
Mem:141.07M (221.48M, peak 362.77M) Time:03.27.16
Interesting. Same card, so much different results. Sparazza, why don’t you post a screenshot? I’m curious about the memory usage. Anyway, you’re comparing pears with oranges.
You’re comparing a single overclocked GTX 570 card with a GTX 580 at stock frequencies. And beside this, Cycles doesn’t scale well on multi GPUs. Once the load balancer will be fine tuned and multi GPU support will be implemented properly, we can recompare properly. Anyway, post some LuxMark 2.0 screenshots with the room scene. That’s a test that use full GPU power.
Sparazza failed to mention details for his configuration. Frankly I don’t beleive to those numbers with stock frequencies. And, considering Blender measures even the time used to build the BVH tree and to copy the raw data to the card (which are CPU bound procedures), I guess his CPU is overclocked too. So “credits” don’t go to the GPU only. Before “LOLing”, post right facts.
Why, do you have CUDA benchmarks that are sane, instead? 
GPGPU on GTX 680 is capped by nvidia’s admission. You can have 100 times the cuda cores, the performance are capped.
Besides this, what does it mean? Is OpenCL less valuable than CUDA? For your information, nvidia’s OpenCL implementation is built on top of CUDA. They’re very similar, so it’s no surprise that bad OpenCL performance would mean bad CUDA performance too.
It’s no secret GPGPU computing on the GTX 680 is completely “broken” (by nvidia itself):
http://www.theinquirer.net/inquirer/review/2162193/nvidias-gtx680-thrashed-amds-mid-range-radeon-hd-7870-gpu-compute
People should just stop to read spec sheets. Hardware performance is much more complicated to achieve than reading 2 numbers on a spec sheet.
12:28 for a 100% render on a AMD 6 core at 3.8 ghz. I’ve got so much to learn from this file. Thanks for posting it!
Computer:
Windows 7 Pro SP1 x64
CPU QuadCore Intel i7 960
GPU GeForce GTX 550 Ti (1024 MB)
16gigs Ram
First Blend 5:55.27
Updated Blend
Total 2:49.42
BVH - 1:39.61
Render 1:09.81
Gadzooks! Thanks for the eye opener!
to mjordan:
yes, you are right, its time to full support OpenCL rendering just now! but there is still bug in AMD’s driver… then AMD will be hero of Cycles rendering.
Original blend file: 03:41:79 on my i7 Windows 7 8GB with GTX 560 ti using 2.62.2 version 45091 from graphical.org
latest blend as of 0:5:00 on east coast of US 3/25/12 was 00:43:02.
Is anyone else crashing Blender when checking cache BVH? This happens with the original and updated versions of this blend file. It locks up, the render screen saying “looking for cache BVH.”
StraLOL! (don’t mean challenge, I really laughed at the first and second post)
my CPU is stock i7 2600k (K is for overclockable but I don’t even know how! I need to gather some information about this)
And my video card isn’t overclocked at all!
Screens:
Believe it or not… i could do a screengrab video to prove it if you care!
The main difference between our systems, at this point, is the SO.
(I can’t use LuxMark 'cause it seems to bea sw only for windows)
This is the newest file (EDIT 3) without “copying” time.
CPU
GPU progress
GPU
So CPU: 2:08.17
GPU: 00:37.05
Cheers
Sparazza
PS: sorry everyone for my bad english