Here is an image from renderosity by Superza
Rendered at 1200*750 with 0.75 quality in 170 hours on a 3.6 ghz
i cant imagine how long my machine - a 2.4ghz PC would take to create this
and its not even at 1.0 quality
is there ever going to be a point in purchasing this software to use for large scale posters etc
There are allot of other things to consider as to why that render took 170 hrs.
0.75 quality doesn't tell me much really.
There may be other quality settings that are set too high and are adding to the
render time. There is the atmosphere quality setting. The cloud quality settings (one for each layer used). There is the anti-aliasing setting. And the killer of all renders the GI setting.
Any of these can add to render time and when combined they can really kill you.
Keep in mind that this is still a "tech preview" and rendering will improve time wise with the final release :)
There will be lots of optimization in the final (and hopefully in the interim as well) releases. The image you posted has lots of reflective surfaces and water and these take a LOT longer to render at the moment. These will also be much faster upon final release. The trick with terragen is to find your own way to reduce render times and there are many areas that can be tweaked to speed thing up. The render quality at 0.75 could possibly be dropped, the GI functions can be limited (down from the default "2") or switched off altogether. If the atmosphere has a high sample quality then this could also be dropped. Also remember that Large scale posters are rarely rendered at 300 dpi, often this is reduced to a lower amount as the viewer is standing further away.
Richard
hey guys thanks
i realise there is a lot of water on the image i posted and i know there are tricks and tips to bring render times down - but surely that is always going to produce inferior images - i wouldnt use a lot of water in my images but i would want atmos and cloud settings as high as possible along with ray traced shadows
there is also the case of antiliasing etc etc
i also understand that the GI settings could be slowing the whole thing down
but can you honestly see there being a huge change in render time between 170 hours for this image right now and say 20 hours by the time we see a finalised version - remember it is only a 1200*750 image and has no trees/rocks/plants etc etc
Actually reflections are currently one of the single most demanding features in terms of render time - even complex populations like trees, grass, etc. will tend to be notably faster. So since this image consists almost entirely of rough and reflective water (the more rough the water, the slower the render time due to interreflection), it is almost a "worst case scenario".
The points that others made are also very valid and important to fully understand. It would help to know more about the particular settings used, but above all you don't want to assume that the settings someone decided on for their render are necessarily sensible. Just because it took him 170 hours doesn't mean that was what was necessary to achieve this level of quality, and TG can't be held responsible for someone else's poor detail decisions. There are many detail-related settings in TG2 that have little or no effect in certain situations. A fine example is GI - in this case GI was probably almost unnecessary and if it was above 1 for either Relative Detail or Sample Quality it shouldn't have been. It simply would not show any improvement. Likewise with cloud samples. I do see clouds in the image very faintly, but they are so subtle and lacking in volume that samples of 16 or at most 32 really ought to be fine.
Another extremely important and misused setting is the "ray-traced shadows" in the quality tabs of both Atmosphere and Cloud nodes. These almost *never* need to be turned on because they only affect the appearance of atmospheric rays *from* the terrain and shadows *onto* clouds *from* the terrain. In other words they are responsible for interactions between terrain and atmosphere. In 90% of scenes there simply is no interaction so these can be left off. They increase render time by a great deal so they should really be avoided unless absolutely necessary. Note that there is a current known bug where the sun can "shine" through the terrain when at a low angle and this can only be fixed at the moment by turning on ray-traced shadows in the atmosphere, however this will of course be fixed in the future and this setting won't be required as a workaround.
The bottom line is that there are so many variables and so many easy ways to setup unnecessarily long renders that one really shouldn't judge likely render time by someone else's results, especially without knowing all detail settings and evaluating the scene yourself for their sensibility. It will take some time for people to better understand the detail settings in TG2 and how to use them to maximum effect. The most important thing to keep in mind is that simply increasing all sliders to maximum and turning on all "quality" options is *not* the best approach and is usually not necessary. Start conservatively, do lots of crop renders to test your detail settings, and increase detail in reasonable increments. You'll save yourself a lot of render time on the final images.
Finally in regards to the CPU speed, your 2.4Ghz means a lot more if it's either an Athlon XP/64/X2 or better yet a Core 2 Duo. The only CPU that reached 3.6Ghz was a Pentium 4 variant, and they are unfortunately not the best for TG2 rendering speed. Not to mention if you do have a Core 2 Duo or other multi-core CPU you can expect a huge reduction in render times when the rendering system is multi-threaded and able to take advantage of both cores. Right now it only uses 1 so it's almost half the speed it could be from that optimization alone. There are also many other optimizations planned that should ultimately result in very significant speedups. Going from 170 hours to 20 hours is almost a factor of 10 and that's not really a realistic expectation, but I would not be surprised to see some scenes rendering 3-4 times faster on dual core systems by the time TG2 is released.
- Oshyan
Quote from: Oshyan on March 30, 2007, 01:10:18 PM
Another extremely important and misused setting is the "ray-traced shadows" in the quality tabs of both Atmosphere and Cloud nodes. These almost *never* need to be turned on because they only affect the appearance of atmospheric rays *from* the terrain and shadows *onto* clouds *from* the terrain. In other words they are responsible for interactions between terrain and atmosphere. In 90% of scenes there simply is no interaction so these can be left off. They increase render time by a great deal so they should really be avoided unless absolutely necessary. Note that there is a current known bug where the sun can "shine" through the terrain when at a low angle and this can only be fixed at the moment by turning on ray-traced shadows in the atmosphere, however this will of course be fixed in the future and this setting won't be required as a workaround.
Just to give a reference, I rendered a crop of an 800x600 image that was about 100x90 pixels. It contained a piece of a moon and some sky (barely any clouds, and they were 2D with almost no density). With ray-traced shadows on in the atmosphere and clouds, the render took about ~85 minutes. With them off, it took 97 seconds. That's about a 52.5 TIMES increase in rendering time for no change (at least for my sample, but if you have clouds below, say, a mountain, you'll want it on so that the mountain will cast shadows on the coulds below it). Something to think about.
thanks oshyan for the lengthy input
i am of course not the owner of the image and dont know any pitfalls he made when he rendered the image
i am still unsure whether T2 Deep can be part of my utilities within the coming months/year
my computer is P4 2.4ghz / 1gig ram
anyone out there with similar settings to my PC that would like to do some tests on images 2000 pixels plus and post the results
would definitely be grateful
Lee
Quote from: doneydonydone on March 30, 2007, 02:18:42 PM
thanks oshyan for the lengthy input
i am of course not the owner of the image and dont know any pitfalls he made when he rendered the image
i am still unsure whether T2 Deep can be part of my utilities within the coming months/year
my computer is P4 2.4ghz / 1gig ram
anyone out there with similar settings to my PC that would like to do some tests on images 2000 pixels plus and post the results
would definitely be grateful
Lee
Lee, Why don't you check this Benchmark Thread?
http://forums.planetside.co.uk/index.php?topic=350.0 (http://forums.planetside.co.uk/index.php?topic=350.0)
Follow the link to the Page, download the test tgd., run the render on your machine and see how your machine fairs against others.
What Oshyan says is right on. For instance: If you bought a corvette would you drive it at 160mph all the time just because the Speedometer goes up to 160? ;)
hey jay
yeh i'd probably do 160 and wrap it round a tgo imported tree haha
i'll get my hyperactive ass over to the benchmark area right now!!
@D3.....I share your pain.....have a read of this thread http://forums.planetside.co.uk/index.php?topic=462.0 (http://forums.planetside.co.uk/index.php?topic=462.0)
I see that Intel have released details about their upcoming chips - http://www.intel.com/pressroom/archive/releases/20070328fact.htm
Not being a programmer... will any of the new SSE4 specifications announced be useful for speeding up TG2? And if so, is their inclusion likely for the slated render optimizations? I only ask as apparently no chips currently support SSE4 and not knowing how early-on developers are able to start working with them, would hate to think that they'd be omitted should they prove useful. Basically, upon TG2's release, I (like many I expect) wish to buy the best machine that I can - and it'd be great if TG2 could fully take advantage of everything it offered.
Don't really know how these things work though, but just thought I'd ask in case you can give an answer...
:)
oh ya...QUAD-CORE baby!!!
when are they coming out???...
Quote from: dhavalmistry on March 31, 2007, 01:25:57 PM
oh ya...QUAD-CORE baby!!!
when are they coming out???...
There's some more info in these links:-
- Intel 45nm CPUs to clock above 3GHz - http://www.reghardware.co.uk/2007/03/28/intel_spills_penryn_beans/
- Intel Hyper-threading to return with 45nm 'Nehalem' - http://www.reghardware.co.uk/2007/03/28/intel_confirms_ht_back/
- Intel to complete AMD impression in 2008 - http://www.reghardware.co.uk/2007/03/28/intel_nehalem_deets/
:)
looks like "Nehalem" is gonna be a BIGGIE!!!
Yeah, I'm pretty excited for those new chips. Being a laptop person, I'm especially interested in some of the new power saving measures that Intel has come up with. Right now, I'm just trying to scrape together enough cash to upgrade my PowerPC, hopefully with one of these. I splurged for TG2 Deep+Anim, now I just need to get a computer that can do something with it ;).
I don't think anyone with render time as a concern can really make any serious judgment about TG2's applicability to their needs without testing the final release on their own hardware. For that you'll just have to wait until later in the year. But I do think that expecting a modern graphics application to perform well on a 3+ year old machine is asking a bit much. TG2 is a high-end program and will work best with high-end hardware, but even a modest amount of money could get you a significantly faster machine for TG2 and many other tasks. You can get an entry-level Core 2 Duo which will outperform your current machine by more than 2 times for under $1000 these days.
- Oshyan
I can't wait until polymorphic computers start hitting the mainstream market. Raytheon made one but apparently they aren't going to release it for a couple of years. If you don't know, a polymorphic computer is one that can dynamically change its architecture on the fly in order to provide more power for back end calculations or front end user needs. Apparently the CPU Raytheon built is 10x more powerful than an Intel Quad Core Xeon :o
10x? Wow, that's impressive :D. I'm really interested in all of these alternate computing ideas (quantum, biological, polymorphic), but I'm not sure how well they will take off considering the fact that current mainstream architectures are already so entrenched. It will be interesting to see how it all unfolds, though.
I wouldn't put too much faith in the value of polymorphic computing on the desktop. Almost inherently by design a polymorphic computer will never be as powerful at a specific task as a dedicated processor would be because you will always lose some power in exchange for flexibility. The very fact that reconfiguration is necessary means there must be some overhead in doing so.
The "10x" number is stated but completely without reference to the workload. 10x faster than a quad core Xeon is completely unsurprising for certain kinds of tasks where even low-cost stand-alone processors can do the same or better. So without knowing what kind of task(s) they're talking about it's largely meaningless.
The main value of such a processor is for applications where the workload changes over time or where significantly different types of tasks would be performed and where both power efficiency and space are a concern. The biggest advantage for these situations is that you can use fewer discreet processors - perhaps even only 1 - thus saving space and power. But performance is generally not going to be the driving force for using this sort of technology. There's some more discussion on this here: http://hardware.slashdot.org/article.pl?sid=07/03/21/2015245
Note that Cell already outperforms the stated abilities of their prototype chip and of course it's already in production. Although I wouldn't pin a lot of hope on Cell on the desktop either, you at least have a better chance of seeing that and getting better performance out of it than you do of polymorphic computers revolutionizing the desktop computer. They'd be great for palm or cell phone computing though, due to the lower power use and ability to perform for example GPU-type calculations when necessary...
- Oshyan