tag:blogger.com,1999:blog-7277449027963623452.post2882990755823283728..comments2023-12-07T05:43:10.401-08:00Comments on Ray Tracey's blog: Real-time photorealistic GPU path tracing: Streets of Asia with dynamic waterSam Laperehttp://www.blogger.com/profile/05688552048697970050noreply@blogger.comBlogger40125tag:blogger.com,1999:blog-7277449027963623452.post-66456214992865143252012-04-23T02:17:12.054-07:002012-04-23T02:17:12.054-07:00Thanks for the clarifications Jules, nothing bette...Thanks for the clarifications Jules, nothing better than first hand information :)Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-19126079416751171922012-04-23T00:52:54.229-07:002012-04-23T00:52:54.229-07:00Jules, Have you this video of your test of the Rub...Jules, Have you this video of your test of the Ruby Demo scene in Octane ? It will be wonderful to watch this. ))Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-10997570320789970192012-04-22T21:13:12.294-07:002012-04-22T21:13:12.294-07:00Nice to see Jules Urbach talking more about the ol...Nice to see Jules Urbach talking more about the old Cinema 2.0 tech demos.<br />I was wondering what really happened when it was MIA at Radeon's 4870 launch.<br /><br />I'm still eagerly waiting for more Gaiking media releases. <br />I really like the idea of making a film + computer game simultaneously with almost no apparent graphical differences between the two (although I'm pretty sure there still will be, realistically speaking).<br /><br />Will there be any updates about OTOY for the upcoming E3? Or maybe SIGGRAPH?FreDrehttps://www.blogger.com/profile/12542997300264966934noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-4149502142409222472012-04-22T16:29:57.788-07:002012-04-22T16:29:57.788-07:00"Anonymous said...
Sorry Sam, but there is no..."Anonymous said...<br />Sorry Sam, but there is no evidence at all that the Ruby video you are referring to is real time."<br /><br />I made that demo 4 years ago. JJ and his guys made a animated clip in Max+VRAY, and I made a real time demo from that dataset. <br /><br />BTW, the scope of the AMD cinema 2.0 demo just to show relightingof a video clip- and nothing else. I wanted to do more, so I pushed the demo further, adding the ability to navigate through the scene as well. I spent weeks at BLR in 2008 to get the data I needed to get that working.<br /><br />I exported the scene as voxels from MAX, the GI lighting info from VRAY per voxel, and did reflections/raycasting using the voxel data. I also exported lightfields (per material) generated by VRAY to identically match the way VRAY did reflections.<br /><br />One reason the live demo wasn't released by AMD was because it was over 3 GB. AMD didn't enable binary CAL GPGPU kernels by the time the 4870 launched in summer 2008, so we didn't include our GPU codec in the final exe. If we had, it would have reduced the demo by 50x or more. It was a valuable lesson in shaping our future GPGPU work. It led us to write our own OCL compiler+runtime (from scratch) for Cypress and Cayman GPUs.<br /><br />For anyone interested in what the 2008 voxel demo looked like, see @ 1:12 in this video (it's just a few seconds long - it is also streaming from the cloud):<br /><br />http://vimeo.com/22048037<br /><br />When we presented an updated version of the demo at Siggraph 2008, I had the city scene deforming. This was to show we weren't limited to tracing static data. There was a video made for this Siggraph presentation, although I can't find the link off hand. <br /><br />There was also an experimental version done using tri-meshes converted to voxel data in real time (not using any of the VRAY GI info in this case). It was just a test, but the post FX was based on the combustion layers JJ made for the beauty shots:<br /><br />http://techcrunch.files.wordpress.com/otoy15.png<br /><br />I have to second Sam's post regarding JJ's work. He's an absolute genius and an inspiration to me and many of us working at OTOY.<br /><br />-Jules<br /><br />PS. When Brigade is further along, we'll test the Ruby Demo scene with it. I have tested it at 1080p in Octane, and it looks great.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-38619342319232867052012-04-21T06:01:54.120-07:002012-04-21T06:01:54.120-07:00@Anonymous: frankly, I would never trust Microsoft...@Anonymous: frankly, I would never trust Microsoft with a GPGPU language. But it would be interesting to see how a complex piece of code runs on CUDA and Direct Compute, performance wise, and implementation-efficiency wise.<br /><br />Right now, I believe the cross-compile approach is good. The GPGPUs are similar on a hardware level, and the languages are similar as a result; just use what you prefer, or what gives you sufficient control over your target hardware.<br /><br />Regarding multiple levels of CUDA backwards compatibility: true, but not such a problem for Brigade: we simply need tons of performance anyway so anything before GTX470 is not that relevant... That's a case-specific argument, obviously.<br /><br />- Jacco.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-60478280452421446402012-04-21T04:26:42.550-07:002012-04-21T04:26:42.550-07:00Cuda is great, if you have a Nvidia GPU. And using...Cuda is great, if you have a Nvidia GPU. And using Cuda is great for Nvidia because it locks out the competition. And even using Cuda there is still like 4 levels of compute backwards incompatibility.<br />If you target a very niche application that may be all you need, for any other use case it is not a smart choice as you limit your market.<br />In my experience Direct Compute is similar good as Cuda, and is supported well by both Nvidia and AMD. It is still early days for GPU compute, and eventually a C++ language extension may be the definitive solution.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-53730370690966187002012-04-21T04:11:15.024-07:002012-04-21T04:11:15.024-07:00StaseG: CUDA is more flexible and powerful for gen...StaseG: CUDA is more flexible and powerful for general purpose GPU computing than GLSL + everything Jacco saidSam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-22734308784710951922012-04-21T02:34:02.213-07:002012-04-21T02:34:02.213-07:00@StaseG: In GLSL, inter-thread communication is no...@StaseG: In GLSL, inter-thread communication is not possible (unless you do multi-pass, which introduces different inefficiencies). CUDA currently provides the best mix between low-level control and raw performance. OpenCL abstracts things a bit further, which removes some low-level opportunities. I think CUDA is currently the best 'base platform'; ideally other platforms should be targetted using some kind of cross compilation (there are tools for that; didn't have time to seriously investigate those yet).<br /><br />- Jacco.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-4666577432923128552012-04-21T01:41:59.887-07:002012-04-21T01:41:59.887-07:00Thanks :)
And one more question.
I`ve seen some ex...Thanks :)<br />And one more question.<br />I`ve seen some examples of GPGPU computing, using GLSL.<br />http://www.youtube.com/watch?v=aJCcqa87ZsQ<br />And I know only one path tracer (Made by Evan), that use GLSL in calculation.<br />Why you don`t use GLSL computing in Brigade?stanislav marchevskyhttps://www.blogger.com/profile/17155572798020824191noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-18444343887301223232012-04-20T08:33:42.003-07:002012-04-20T08:33:42.003-07:00Then I shall wait for the Kepler optimized Brigade...Then I shall wait for the Kepler optimized Brigade code :)Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-30025665291138742072012-04-20T08:31:04.626-07:002012-04-20T08:31:04.626-07:00Samuel: not necessarily. Apparently, Aila & La...Samuel: not necessarily. Apparently, Aila & Laine's ray traversal kernels are 40% faster on GTX680 with the latest (minor) changes. We simply didn't apply this in Brigade yet.<br /><br />- Jacco.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-44291768337877768082012-04-20T07:04:56.171-07:002012-04-20T07:04:56.171-07:00StaseG: Brigade doesn't use double precision, ...StaseG: Brigade doesn't use double precision, there's no advantage in using Tesla. Moreover GeForce is faster than Tesla or Quadro at path tracing.<br /><br />As for the 680, it's worse than the GTX 580 for Brigade and CUDA apps in general.Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-31554241157806022022012-04-20T05:10:38.701-07:002012-04-20T05:10:38.701-07:00I know, GTX580 has limit for double precision, but...I know, GTX580 has limit for double precision, but GTX680 is worse in DP, but better in SP float.<br />Does Brigade Engine use DP calculations?<br />If yes, Tesla should be better for this?<br />If no, why you don`t use GTX680?stanislav marchevskyhttps://www.blogger.com/profile/17155572798020824191noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-85377898881708216452012-04-20T00:33:02.312-07:002012-04-20T00:33:02.312-07:00Jonatas, muito obrigado. Espero que o proximo demo...Jonatas, muito obrigado. Espero que o proximo demo vai ser ainda mais legal :)Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-78794644360725754082012-04-19T23:52:56.167-07:002012-04-19T23:52:56.167-07:00Possibly also a mix of CG with real video sequence...Possibly also a mix of CG with real video sequences, to be clear.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-61479372321549519882012-04-19T23:44:29.522-07:002012-04-19T23:44:29.522-07:00We built this clip...
What do you think clip means...We built this clip...<br />What do you think clip means ?<br />It is a CG prerendered video clip like all the other clips listed at the BLR portfolio. That's what a VFX company like BLR does, create clips with tools like Maya, Max, ...Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-52112430178661601742012-04-19T12:52:17.547-07:002012-04-19T12:52:17.547-07:00Keep the good work Sam!
each demo its looking more...Keep the good work Sam!<br />each demo its looking more realistic.<br /><br />Greeting from Brazil. :)Jonatashttps://www.blogger.com/profile/08665711661495178734noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-3163637074189420402012-04-19T11:40:49.036-07:002012-04-19T11:40:49.036-07:00I don't know where you've read on BLR'...I don't know where you've read on BLR's portfolio that it's prerendered. This is what that site actually says: <br /><br />"This is our first project for AMD. We built this clip to promote the new AMD's Cinema 2.0, a new and exciting real-time tech. <br />Otoy and AMD have use all our stuff to test a new and amazing real-time technology as part of the new Cinema 2.0."Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-50311830525650555032012-04-19T11:05:17.692-07:002012-04-19T11:05:17.692-07:00Sorry Sam, but there is no evidence at all that th...Sorry Sam, but there is no evidence at all that the Ruby video you are referring to is real time. Everything indicates it is a pre-rendered CG video clip by BLR as they tell on their portfolio page http://www.biglazyrobot.com/<br />Somehow this video got associated with AMD (cinema 2.0, never to be heard of again). The idea probably was to make use of 3D scanned surface information of faces, as is used in VFX relighting for movies (and this did not even make it into the Ruby movie).Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-82294532126224414682012-04-18T13:55:50.540-07:002012-04-18T13:55:50.540-07:00Cool :)Cool :)Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-27750167950225011652012-04-18T12:46:58.559-07:002012-04-18T12:46:58.559-07:00Congrats on getting featured in Gpuscience.com.Congrats on getting featured in Gpuscience.com.Reavenhttps://www.blogger.com/profile/08421187431956440223noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-55625097581182795682012-04-18T11:56:30.238-07:002012-04-18T11:56:30.238-07:00There is actually a raytraced game by OTOY that wi...There is actually a raytraced game by OTOY that will be released this year:<br />http://www.brightsideofnews.com/news/2009/9/21/worlds-first-ray-traced-pc-game-and-movie-to-arrive-in-2012.aspxAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-6622827278907981552012-04-18T09:32:20.481-07:002012-04-18T09:32:20.481-07:00Anonymous: the Ruby cinema 2.0 video is real-time ...Anonymous: the Ruby cinema 2.0 video is real-time and was made by OTOY in collaboration with Big Lazy Robot (3DBlasphemy)<br /><br />Anonymous (last post): yeah, the random parameter filtering looks interesting, but it's far from real-time. Maybe a GPU implementation (or better yet a hardware implementation) could run at 30 fps.Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-73310272414898953082012-04-18T08:54:50.625-07:002012-04-18T08:54:50.625-07:00Please look at this technique: http://agl.unm.edu/...Please look at this technique: http://agl.unm.edu/rpf/<br /><br />It allows you to smooth out the noise in paritally rendered frames using a dynamically sized blur filter based on how the variance of the sample rays propagates to various parts of the image.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-26408318179997319532012-04-18T05:42:57.224-07:002012-04-18T05:42:57.224-07:00What makes you think those videos are rendered in ...What makes you think those videos are rendered in realtime with OTOY?<br />Or are you simply linking to some neat rendered animations?<br />I know for a fact that they are rendered with vray...unless they were recreated/ported over later on.<br /><br />http://forums.cgsociety.org/showthread.php?f=139&t=242576&page=2&pp=15Anonymousnoreply@blogger.com