Moderátor: Baneshee
No ja jsem si v 1080P zahral last light plynule na max az s GTX1070 a ve 1440p by to pravdepodobne nedala ani onamouldy píše:Last light nijak zvlášť nebyl, tam byla pro radeony náročná akorát tesselace. Jinak to proti 2033 byla dost optimalizovaná hra, někdy až moc.
benchmark je dobrej jenom na srovnání výkonu karet mezi sebou, ale o hře nic nevypovídá, resp. můžeš přihodit 20 fps a dá ti to ingame výkon.webwalker píše:Jen jsem se chtěl zeptat, vyhovuje vám ten vestavěný benchmark, nebo raději testovat přímo v nějaké lokaci? Jsou tam koukám dost velké výkonnostní rozdíly.
A jak jsem psal, jde o to, jestli to hráč vůbec dokáže poznat při samotnym hraní, jestli je něco jinak, nebo ne... Minimálně v těch ukazovaných scénách je ten rozdíl tak markantní, že nevšimnout si toho snad nejde.Swat píše:Paul: na DF jsem koukal včera, já bych ty rozdíly nenazval brutální, nýbrž viditelný
Zase je ale krapet dementní když si na základě benchmarku nastavíš mnohem horší grafiku než by sis mohl 99% hry užívat.Swat píše:V tom videu říká, že ten in-game benchmark to stresne víc, než samotná hra, což mi přijde fajn, takhle by měl správnej benchmark vypadat, aby si to pak člověk mohl nastavit pro nerušený hraní. Ne jako např. GTA5, kde si člověk dá benchmark, nastaví hru a pak zjistí, že to třeba v lese padá na čísla, kam se benchmark ani zdaleka nedostal..
Tohle je už druhý post v tomhle vlákně, který zmiňuje textury v souvislosti s nasvícením scény. Můžete mi pls někdo vysvětlit o čem je řeč?Paul píše:... tak mohli všechno nabejkovat do textur AC:Unity-style a RTX by bylo k ničemu..
https://en.wikipedia.org/wiki/LightmapKrteq píše:Tohle je už druhý post v tomhle vlákně, který zmiňuje textury v souvislosti s nasvícením scény. Můžete mi pls někdo vysvětlit o čem je řeč?Paul píše:... tak mohli všechno nabejkovat do textur AC:Unity-style a RTX by bylo k ničemu..
BakingKrteq píše:Lightmap i shadowmap jsou ale stále počítané v reálném čase, není to nic co by bylo napevno "bejkovené" v textuře
- este mensi, lebo keby mas komplet ciernu miestnost, tak ten rozdiel bude nulovy. Dovod je velmi jednoduchy, ked priame svetlo dopadne na nejaky povrch, tak sa odrazi a cim svetlejsi povrch to je, tym viac svetla odrazi, preto je svetly. Ale aj pri svetlom objekte dojde uz pri prvom odraze k znacnemu utlmu toho odrazeneho paprsku a ten druhy odraz sa este viac utlmi, preto uz nema zdaleka taky vplyv na osvetlenie. A takto to pokracuje dalej a dalej a jas toho paprsku s kazdym dalsim odrazom prudko klesa az sa dostane do stavu ze nema na scenu ziaden badatelny dopad.Tronar píše:Michal3d: A jaky by byl rozdil 1 vs 2 bounce, kdyby materialem mistnosti bylo tmave drevo?
- to je nejak bezne u Metra, co si pamatam, tak 2033 mala tiez nerealisticky narocny benchmarkwebwalker píše:@Paul: Jj, právě na to jsem koukal, příliš to neodpovídá realitě ve hře. Tedy krom lokace Tundra, tam je to hodně náročné. Já jen, jestli se tady nad tím nikdo nepohoršuje
Aneb jak vytrolit sam sebe. Baked lightmaps je naprosty zaklad, bez ktereho je jakakoliv debata na tema GI a RT jaksi nemozna. https://www.youtube.com/watch?v=eGmXDOGjKPAKrteq píše:
a prislo RT, ktery hromady neduhu starsich metod resi a melo by zapekani svetelnych map uplne odstranit.I will need to bring some examples from UE4 that I did and I will use UE4 examples to explain this once and for all. In fact Lightmaps and shadowmaps have been used for a long time now in many engines to mimic the effects of global illumination. Here is how they do it. They use a static source of light which gives as result lit surfaces with indirect lights and lit ones with indirect lights and the resulting shadows. This light and shadow data can be stored in UV textures called lightmaps and shadowmaps and can be applied on surfaces and meshes' textures to give the impression of being lit. This has been done in almost every game from that last generation with old engines to make it run on low end machines (this has bene done even in Cryengine 2 with Crysis 2 and many UE3 games and so). Since the GI is basically textures, they have to be stored in the game files and disk and will eat more space just like regular textures. I take now the example of UE4 whihc has a a new and advanced ligthing tech compared to old gen. In fact UE4 uses what is called Lightmass. This tech is like a dynamic source of Light but instead it is either static or stationary (for example turning on and off but stayin in teh same position). This method creates indirect lighting from the direct light creating lightmaps and shadowmaps in real time on surfaces and meshes without creating UV lightmaps and shadowmaps that are stored as regular textures that are blended with surfaces and mshes UV textures to give the impression of being lit but instead the light data stays stored in static volumes instead of turning into textures files which saves much space. So let is say that is a semi automatic GI, it is still static but it doesn't involve the artists to store light data within textures. The benefit of such tech :storing the light data in volume instead of textures is that it can make movable meshes like charcters interact with it and get the diffused lights like here in Unity where Arno socks and shirt get lit by red light diffused from the red carpet, so if they were just simple lightmaps and shadowmaps, Arno wouldn't get such detail. Unity must be using sth similar to lightmass. The benefit of such tech is that it takes time to get built and rendred but the result is so accurate and without artefacts or flaws since the source of light is static. sadly this method has one disadvantage when it comes to surfaces and areas that are difficult to reach by light which gets special type of shades (no defined shadow shapes) which is caleld ambient occlusion that defines the edges of most things, that is why an extra effor or tool or program or option ahs to be added to put them in emphasis like here the HBAO+ which just gives approximative results by trying to guess the surfaces which are difficult to reach by the light without having a direct connection to the static light source, that is why it is just an approxiamtion.
yes UE4 dropped SVOGI because it is the most technically advanced algorithm of dynamic GI and it eats a lot of ressources even on highest end PCs. Thye secertly replaced with LPV (light propagation volumes) a very rudimentary way to get dynamic GI.
This technic stores moving light data in that propgates in to those big volumes and tries to diffuse them dynamically. Well you get dynamic GI but the results aren't that accurate. I will show how it goes (using examples I made using UE4). Here is the dynamic GI LPV enabled but the light source (the sun ) is not moving, you get awesome GI results with even blending colours: http://imgur.com/a/VGQnS
When you make the sun move thus making it a dynamic source of GI, you get colours changing and shadows too (here no lightmaps or shadowmaps stored at all), following the sun postion and day/night cycle you get the same wall getting different colour combination from orange to purple in real time which is very awesome and impressive. You can even get lit characters in real timeby lights switching colours accoridng to the sun position and time: here being lit eitehr by red wall thhen yellow wall and standing in the same position but in different time of the day (different sun position): http://imgur.com/a/I3spm
So techwise it is very impressive. Sadly this tech is not that accurate when it hits advanced meshes even static ones and you get archaic results not as accurate like in Unity here for example which has static GI: http://i.imgur.com/qiC6YMe.jpg
And it also only works with sunlight, directional dynamic light aren't supported in UE4, they only cast direct lights: http://i.imgur.com/6troVKM.png
A dynamic method of GI doesn't need extra work to mimic the results of ambient Occlusion since the light source is dynamic so it creates dark occlued (hardly reachable by light) surfaces. With LPV, occlued surfaces is horrendous they are very blocky and pixalated and huge and not subtle eventhough they are real time.
I come back to the method Epic dropped aka SVOGI, but let me begin with the method Nvidia are devloping aka VXGI which is Voxel Global Illumation. This method is just like LPV but wayyyy more precise , it almost can reach the accuracy of the static lightmass. While LPV uses big volumes and stores big chunks of diffused lights in one direction per big volume, VXGI stores light information per voxel (cubic piwel) which is very tiny and can propagte light in different directions, now sum all the data colelcted in those tiny voxels and you will get avery accurate approximation of diffused lights IN REAL TIME. So this can be applied to everything going from moving light sources to emissive materials or textures. You can read about it here and watch videos : https://developer.nvidia.com/gi-works and here : http://www.geforce.com/hardware/technol ... technology and you can download the recently released VXGI UE4 Apollo 11 demo here: http://blogs.nvidia.com/blog/2014/11/11 ... ollo-demo/
Enlighten is also developing an effective dynamic GI for UE4 with multiple dynamic sources and emissive materails that can interact with static and mocing meshes : https://forums.unrealengine.com/showthr ... pports-UE4
The SVOGI (Sparse Voxel Octree Global Illumination) is even more impressive tech since it uses a sparse voxel octree which can collect light data from voxels from different shapes in an area accessible to the user's eyesight and it can even collect data and render from light sources invsible light sources and display their diffused lights in the field of view of the user. Both VXGI and SVOGI have also dynamic reflections created by dynamic GI sources. Sadly both techs require too much ressources: The Apollo 11 VXGI demo needs at least a GTX 970 card to just render a very simple scene with just 3 meshes and no complex surronding environments. Not only that but both can only have just one bounce per dynamic light source and light only get diffused in few directions per voxel making deeper areas dark and that still needs an external intervention of a method of Ambient Occlusion. Here is an exmaple fo someone who tried to enable two bounces per dynamic GI light in cryengine. (you can download the demo in the description if you have a hefty PC) https://www.youtube.com/watch?v=8PCnkJBvIrY
As you can notice dynamic reflections created by dynamic GI sources are not accurate and are blocky : https://forums.unrealengine.com/attachm ... 1414981939 and the same blocky real time reflections are noticable on the helmets of the cosmonauts' helmets https://www.youtube.com/watch?v=RNt5znFLv9Y
And both demos don't uses PBR materials and simple scenes otherwise it would be an overkill.
The only best and accurate instance of dynamic GI is used in The Tomorrow children wihic uses Cascaded Voxel Cone Ray Tracing. This method is very good and gives very gorgeus results. Ok it is not the same method as the SVOGI or VXGI, since both are 100% automatic methods meaning that anyone can throw them on the go and work and can work with anything in th scene from static to dynamic meshes. Q-Games tried SVOGI but they found out it was very slow so they ditched octree hiearchy stores data and to determine distance from the camera to scale voxels and optimize for distance, much like LoD for a new approach aka 3D textures that developers would have to generate the textures, mostly by hand, prior to importing the models or assets into the game. That is why they are using ray-traced generated volumes instead of polygonal models. The advantage of this method is that not only it frees up ressources but it also allows desv to get better results and verty realtsic ones too. Raytraced generated volumes and models are very accurate and you will never find any hard edge like in polygonal meshes that need heavy tesselation to correct the edges. They can now go for 16 diffused light directions per voxel and they can even reach 3 bounces per light direction (even for invisble light sources) which even Pixar movies didn't reach (and the dev confirmed that). Reflections are as accurate as in real life and you won't get blocky reflections. They can even apply 3 types of AA and the list goes one. Ok, their method isn't as impressive as VXGI or SVOGI, ok their GI is still dynamic but it needs extra interventions from the the devs to fill data in each 3D texture and get their own approximations so it means only Q-Games devs cna use it sinc ethey ahve to interact with it, while with VXGI and SVOGI it is automatic and anyone can integrate it on the game by just dropping it in any scene with any object without extra work. So their method is dynamic is dynamic but semi automatic while VXGI and SVOGI is dynamic and automatic, yet Q-games approach has better and more accurate results on screen than SVOGI and VXGI just like I said before that the static GI in Unity looks better than the dynamic GI available now which are more impressive and demanding techwise.
RTX reflections are pretty cool, but honestly reflection probes and screen space reflections have done a pretty decent job at making things look okay in games. Proper realtime GI though, that excites me. Metro Exodus is already do a fantastic job showing off how good it can be, but more stuff like this in games is really going to be impressive. No longer having to compromise between static lightmap baking or dynamic shadows, not having to settle for awkward not quite right AO... but getting true realtime light bounces. Reflections are cool, but as a cinematographer I can't tell you how excited I am to see light and shadows in games finally have proper depth and complexity.
Souhlasim, co se tyce poctu paprsku, ale vysledny rozdil intenzity v tmavem prostredi muze by pro pozorovatele subjektivne vyraznejsi, nez u bilych sten. Nebot 1 bounce muze znamenat tezko citelnou tmu a uz 2 bounce muze pridat aspon obrysy?michal3d píše:- este mensi, lebo keby mas komplet ciernu miestnost, tak ten rozdiel bude nulovy. Dovod je velmi jednoduchy, ked priame svetlo dopadne na nejaky povrch, tak sa odrazi a cim svetlejsi povrch to je, tym viac svetla odrazi, preto je svetly. Ale aj pri svetlom objekte dojde uz pri prvom odraze k znacnemu utlmu toho odrazeneho paprsku a ten druhy odraz sa este viac utlmi, preto uz nema zdaleka taky vplyv na osvetlenie. A takto to pokracuje dalej a dalej a jas toho paprsku s kazdym dalsim odrazom prudko klesa az sa dostane do stavu ze nema na scenu ziaden badatelny dopad.Tronar píše:Michal3d: A jaky by byl rozdil 1 vs 2 bounce, kdyby materialem mistnosti bylo tmave drevo?
No a tmavy povrch odrazi uz pri prvom odraze velmi malo svetla, preto je tmavy a tak u neho ma viac odrazov este mensi vplyv.
Prilis jasne okno je zpusobene paprsky bez bounce, ne? Stacilo by snizit jas/intenzitu/whatever (nejsem odbornik) techto paprsku. Funguje takhle ta kompenzace prepalu?michal3d píše:
Toto sa v renderingu musi riesit podobne, lebo dynamicky rozsah realne nasvierltenej sceny je tak obrovsky, ze sa samozrejme neda vmastnat do normalneho 24-bit obrazku a je nutne ten rozsah komprimovat tak, aby bol zmysluplne zobrazitelny na obrazovke. Aj ten moj render nie je linear, lebo to by ste tam maly naozaj bud uplnu tmu alebo biele okno, takze je tam zapnute kompenzovanie prepalov aby sa mohla pouzit dlhsia expozicia.
Takze raytracing dokaze zakomponovat snizeni jasu odrazem, ale nemuze separatne zakomponovat snizeni jasu zdroje (okno) bez ovlivneni ostatnich paprsku?michal3d píše:Ked znizis jas tych priamych paprskov, tak tym znizis aj jas odrazenych, takze tym neziskas nic len plosne tmavsiu scenu. To je dosledok toho, ze slnko (ale aj obloha) ma obrovsku energiu, alebo jas ak si to tak chceme nazvat. Odraz tychto jasnych paprakov ma ale velky utlm a teda odrazene paprsky maju energiu ovela mensiu a to vytvara tu disproporciu. A s tym priamo nejde nic spravit, lebo tal to jednoducho v skutocnosti funguje a RT pocita svetlo podla tych istych zasad.
