tag:blogger.com,1999:blog-7277449027963623452.post8083054916476324258..comments2023-12-07T05:43:10.401-08:00Comments on Ray Tracey's blog: GPU path tracing tutorial 3: GPU-friendly Acceleration Structures. Now you're cooking with GAS!Sam Laperehttp://www.blogger.com/profile/05688552048697970050noreply@blogger.comBlogger57125tag:blogger.com,1999:blog-7277449027963623452.post-78592929495652794182017-01-18T09:35:26.470-08:002017-01-18T09:35:26.470-08:00Thanks for the great post! It has really helped me...Thanks for the great post! It has really helped me understand BVHs. I am building a small path tracer using an openGL compute shader. I was wondering what you though about a stackless traversal algorithm, for each node in the (depth first) array of bvh nodes, storing the index of its sibling node. then to traverse the tree you could simply iterate over each node in the depth first array, incrementing the nodeindex if the ray intersect the node, and setting the nodeindex to the sibling index of the current node is it does not. Thanks!Anonymoushttps://www.blogger.com/profile/04787789374640509271noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-83414308903941094842017-01-18T09:25:47.119-08:002017-01-18T09:25:47.119-08:00Thank you for this post! It has really helped me u...Thank you for this post! It has really helped me understand BVHs. I have been building a small path tracing using an OpenGL compute shader. I was just wondering what you though about a stackless traversal algorithm by storing the index of the each nodes sibling (in each node structure) and simply iterating along a depth first array of nodes incrementing the node index if the ray intersects the node, and setting the node index to the sibling index if it does not. <br /><br />Anonymoushttps://www.blogger.com/profile/04787789374640509271noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-55135991942319536662016-09-22T01:29:30.875-07:002016-09-22T01:29:30.875-07:00Good to know it's working :)Good to know it's working :) Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-33004886581352559392016-09-22T01:04:31.414-07:002016-09-22T01:04:31.414-07:00Ah, yes, I found that my project was in debug mode...Ah, yes, I found that my project was in debug mode (debug symbols enabled for CUDA). Now it's working fine! :)Anonymoushttps://www.blogger.com/profile/07371208059637455178noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-65948019913584418472016-09-21T13:48:43.106-07:002016-09-21T13:48:43.106-07:00Hi Mykolas, that doesn't sound right, you shou...Hi Mykolas, that doesn't sound right, you should get over 60fps with a gGTX970. Are you building it in Release mode? Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-73161518904444979882016-09-21T12:55:10.541-07:002016-09-21T12:55:10.541-07:00Hi, loving your blog!
I have a question regarding...Hi, loving your blog!<br /><br />I have a question regarding performance in this example. When I build it myself, everything's running so slow, like 1 frame every 4 seconds, and this is even with smaller model (bunny) and on 640*480 resolution. On GTX970. However, when I run your binary (https://github.com/straaljager/GPU-path-tracing-tutorial-3/releases) , I'm getting great performance. Can't figure out what's causing this, any ideas?Anonymoushttps://www.blogger.com/profile/07371208059637455178noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-86348458431253262132016-07-20T12:25:57.606-07:002016-07-20T12:25:57.606-07:00Is this acceleration structure the one used in the...Is this acceleration structure the one used in the path traced quake 2 for dynamic objects?<br />If not can you tell me it's name?<br /><br />PS: great post and blog.Anonymoushttps://www.blogger.com/profile/18100422342482298939noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-65465663454312022932016-07-20T12:23:36.499-07:002016-07-20T12:23:36.499-07:00Is this method the one used in the path traced qua...Is this method the one used in the path traced quake 2 for dynamic objects?<br />If not can you tell me it's name, I would like to try something new.<br /><br />PS:Great blog ;)Anonymoushttps://www.blogger.com/profile/18100422342482298939noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-36578044869813753622016-05-24T07:14:27.719-07:002016-05-24T07:14:27.719-07:00That paper on reduced precision hardware seems rea...That paper on reduced precision hardware seems really interesting.<br />How many spp/s can Brigade handle?<br />6% of one FPU of a GPU? What exactly is the FPU of a GPU and how big is it?<br />Is there in every shader core a FPU? <br />Seems almost like an april fools joke to me.<br />A few mm² at 1 Watt? + the add operations.....<br />Where does all the power go on a GPU if 5 TFLOPs only need 27 Watt.<br />Off -chip access to DRAM? INstruction unit? <br />Would there also be a path to accelerate shading operations for a Monte Carlo simulation?<br />I mean 1 Watt..... that is looking like path tracing would converge towards mobile integration one day.<br />A Tegra X1 yields 1 TFlops at half precision and 4 Watts.<br />The start up Parallela wants to produce a 64,000 core processor by 2018 executing 1TFLOP/Watt. 64GB of scratchpad memory, 1MB per core.<br />Where is path tracing going ?<br />Could it be that it will be one good day(2022?) executed at 5-10 Watts on a mobile device with a 20-40Whours battery?<br />Is it just a question of circuit organization ?<br /><br />Retinahttps://www.blogger.com/profile/07837338662579054049noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-65338348833940879822016-03-04T02:52:57.967-08:002016-03-04T02:52:57.967-08:00Of course they won't ever outmatch more raw si...Of course they won't ever outmatch more raw silicon area for computing or your power socket.<br />The best batteries we'll have will have a energy density of about 1180 Wh/liter (Sakti3,incorporated by Dyson). So 118 Wh for a very portable battery weighing 250 grams.<br />But there is also another constraint. How hot can a device get hold by your hand? 10-20 Watts max.<br />There are still lots of improvements that can be done beside Moore's law.<br />one example https://www.youtube.com/watch?v=wU0bEnQZAF4<br />Even exascale supercomputers are going to be 20 times more efficient. <br />Retinahttps://www.blogger.com/profile/07837338662579054049noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-71696376090170187532016-03-03T11:40:56.809-08:002016-03-03T11:40:56.809-08:00Mobile processors are and always will be capable o...Mobile processors are and always will be capable of less than a desktop-class processor without the restrictions on power, heat, and size. I don't think NVidia's TK1/X1/etc. line of processors will ever outmatch their deskop offerings, and their is no plan for them to be able to do so.<br /><br />Desktop-class processors will continue to lead the bleeding edge. Even when we reach some theoretical point where pathtracing is possible in realtime, we still have more need for compute in the area of physics (position-based dynamics, etc.) to get realistic movement. <br /><br />As for AR, this is a more subjective opinion and I could definitely be wrong, but I don't think AR will ever take off. If I were an investor I would not pump any money into it. I still see AR as a "solution without a problem". Contrary to this, I think VR headsets have unlimited potential, and that is where I would put my money.Brianhttps://www.blogger.com/profile/06687531419755063817noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-32111188793168068892016-03-02T06:38:54.681-08:002016-03-02T06:38:54.681-08:00Although.....
https://www.youtube.com/watch?v=Gy3n...Although.....<br />https://www.youtube.com/watch?v=Gy3nhYPUd7I Retinahttps://www.blogger.com/profile/07837338662579054049noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-4024041038635416352016-03-02T06:14:55.551-08:002016-03-02T06:14:55.551-08:00I think there will be a revival of the PC platform...I think there will be a revival of the PC platform, especially since Microsoft announced to abandon the console business yesterday.<br /><br />I think the PC platform will be new defined. A technological tectonic shift. haha<br />But it is not going to happen overnight.<br />Every year hundreds of millions of high end smartphones get sold. How do they get used?<br />Well, i don think i need to answer that. XD<br />Once those superprocessors receive a worthy visual interface and real keyboard things are going to change.<br />Backpain, headaches , all caused by sitting at a desktop, destroying entire social relationships will be gone once you carry the visual interface on your head. <br />Then the PC will bend to us rather than we to it.<br />I expect those goggles to get lighter and lighter(matrix Style) unlike the Rift which now weighs a whopping 470gramms. ^^ <br />I have backpain, so i want the desktop platform to go extinct. haha<br />Retinahttps://www.blogger.com/profile/07837338662579054049noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-67542049715889394742016-03-02T05:42:34.087-08:002016-03-02T05:42:34.087-08:00>> "The desktop PC will go extinct like...>> "The desktop PC will go extinct like the CRT." <br /><br />I think there will be a revival of the PC platform, especially since Microsoft announced to abandon the console business yesterday. Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-2124479138437596382016-03-02T04:42:34.403-08:002016-03-02T04:42:34.403-08:00EU project Ions4Set working on Single-Electron Tra...EU project Ions4Set working on Single-Electron Transistors with the promise of using 10x less energy. <br />5 component halfadder was realized in 2012. http://phys.org/news/2012-11-smallest-logic-circuit-fabricated-single-electron.html<br />The SET had multivalue reRAM on the device(SET) itself. <br />Although it sounds pretty crazy right now i am pretty sure that the future of path tracing will be mobile.<br />Nvidia X1 achieves 250GFLOPs/W. <br />I am so sure because i believe/know that Magic Leap's device will be again a new motivator and driver for the whole mobile chip industry. <br />My prediction therefore is that in 10 years there are going to be mobile chips with 10TFLOPs/W efficiency. Faster non-volatile memory included.<br />The desktop PC will go extinct like the CRT. :P<br /><br />Retinahttps://www.blogger.com/profile/07837338662579054049noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-89143835770178769202016-02-08T18:18:17.984-08:002016-02-08T18:18:17.984-08:00I am sure the rays made of Christ spritual charge ...I am sure the rays made of Christ spritual charge are super fast and super coherent.<br />Now it all makes sense Jenson, it really does...<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-69779034166950003502016-02-06T18:15:15.779-08:002016-02-06T18:15:15.779-08:00Anon, you can download the executable (Win 32) her...Anon, you can download the executable (Win 32) here: https://github.com/straaljager/GPU-path-tracing-tutorial-3/releasesSam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-9775064673996336362016-02-06T06:43:40.544-08:002016-02-06T06:43:40.544-08:00I just can't get it to compile and run correct...I just can't get it to compile and run correctly - any chance you build a windows binary with the suitable dlls files inside so I can run the demo please?Anonhttps://www.blogger.com/profile/14946022996350928259noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-71301985801210645642016-02-05T19:58:55.278-08:002016-02-05T19:58:55.278-08:00Interesting, looking forward to know more. I'm...Interesting, looking forward to know more. I'm also researching a novel acceleration structure which is unlike any of the traditional ones. Even though a BVH (with or without spatial splits) currently offers the best performance for GPU ray tracing, it's not flexible enough. Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-42152008537476731902016-02-05T18:39:50.784-08:002016-02-05T18:39:50.784-08:00@CPFUUU: Unreal has a lot of big customers and are...@CPFUUU: Unreal has a lot of big customers and are high volume, I don't think I could get away with a royalty-only license like they do. I'm just going to set a reasonable price for small studios ($400 or so) - the hobbyists are going to torrent it or pass it around anywayChrishttps://www.blogger.com/profile/09966938012901731212noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-57163499018432010162016-02-05T18:16:54.437-08:002016-02-05T18:16:54.437-08:00Sam: Yes, that's an average over multiple inco...Sam: Yes, that's an average over multiple incoherent bounces. I don't want to detail the acceleration structure here, but it is a homebrew and it isn't anything like a BVH. If you want, I'd be happy to detail it for you through e-mail sometime after release.Chrishttps://www.blogger.com/profile/09966938012901731212noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-31042588446120401572016-02-05T16:55:46.750-08:002016-02-05T16:55:46.750-08:00Jenson Button, >> "so that even Christ ...Jenson Button, >> "so that even Christ was born only from the seventh attempt". Looks like driving a McLaren has impaired your cognitive abilities. <br /><br />Chris: 1 billion rays/sec on a GTX680 is not bad. Are the rays incoherent? And what traversal scheme are you using? Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-45305418061332538802016-02-05T14:17:14.767-08:002016-02-05T14:17:14.767-08:00Chris something like a plug in would be ideal to s...Chris something like a plug in would be ideal to start. <br />Of course you should earn money with it, for "free" i thought about business models<br />like UE4 or something like that. Just cheap enough for people without big<br />investment money. <br /><br />Noise wouldnt be so important because its new and experimental. <br />Would be fun to see if simple indie games look much better than current aaa titles.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-30431499641494037222016-02-05T12:42:59.593-08:002016-02-05T12:42:59.593-08:001. CMOS incredibly high power dissipation simply s...1. CMOS incredibly high power dissipation simply stuns <br /><br />2. Albeit the most urgent task for everyone living in conditions of Earth's psychosphere is to kill the Core Evil (this is serious what is happening on the planet, so that even Christ was born only from the seventh attempt), Microsoft was unable only to make Wave files 64-bit. <br /><br />Johnhttps://www.blogger.com/profile/08154705424835302415noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-6264131886511026072016-02-05T10:21:31.397-08:002016-02-05T10:21:31.397-08:00@CPFUUU - We'll get there. For the past 2 year...@CPFUUU - We'll get there. For the past 2 years I've been working on a realtime pathtracer that I am releasing for Unity and Unreal Engine, in 1-2 months. I just recently broke 1B rays/s on a variety of scenes with the traversal scheme (no shading, nearest intersection) on a gtx 680, so it will be very competitive. Supports fully dynamic geometry and all platforms/backends: multi-gpu, windows/linux/mac, AMD/NVidia, and OpenGL/DirectX. <br /><br />Now the bummer is, I went completely broke in the process so I will be charging for it. Something on par with offline pathtracers but reasonable for the indies. Of course Sam will get one for free if he is interested, been lurking this blog for years :)Chrishttps://www.blogger.com/profile/09966938012901731212noreply@blogger.com