[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

[@SebAaltonen](/creator/twitter/SebAaltonen)
"@swooshclick It's an early idea. Maybe SDF maybe voxels. Both are easy to modify. Main difference would be that all players are in the same world that has persistent state modifying it together. And physics of course. We need more physics. Think Teardown but simpler. Must be fun too"  
[X Link](https://x.com/SebAaltonen/status/1972304747291091094) [@SebAaltonen](/creator/x/SebAaltonen) 2025-09-28T14:17Z 45.5K followers, XXX engagements


"@PurePlayerPC Yes but you need a bit more complex tech to make it efficient. Teardown is using dense voxels (bounding boxes with rotations). More detail would require a SVO/DAG to make it memory efficient"  
[X Link](https://x.com/SebAaltonen/status/1974062061710610544) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-03T10:40Z 45.5K followers, XXX engagements


"Why are people showing the backsides of phones in comparisons People look at the front side when they use the phone. Display is what matters. Front side improvements: Brightness: 1200 nits - 3000 nits Refresh rate: 60Hz - 120Hz iPhone 8-X was the biggest improvement"  
[X Link](https://x.com/SebAaltonen/status/1975441830906110154) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-07T06:03Z 45.5K followers, 15.5K engagements


"Imagine people saying this XX years ago: Such a lack of innovation in the new JVC VCR versus the Toshiba. Backside looks pretty much the same. Same ports and same black plastic. Why nobody innovates"  
[X Link](https://x.com/SebAaltonen/status/1975442788289565124) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-07T06:07Z 45.5K followers, 1988 engagements


"Performance was also the main reason Unity's Weta Digital acquisition failed. You can't just buy a movie company and believe that real time game engine + movies = real time movies. A massive amount of optimization and architecture refactoring work is required to make it happen"  
[X Link](https://x.com/SebAaltonen/status/1976922036082823297) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-11T08:05Z 45.5K followers, 7571 engagements


"This is the core principle of modern social media: Misinformation and arguments get more replies. Algorithm boosts these threads. They get more replies. Makes it look like everybody is fighting. In reality X% are fighting but we only see their posts"  
[X Link](https://x.com/SebAaltonen/status/1977396786575598045) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-12T15:31Z 45.5K followers, 7817 engagements


"@csubagio @CustomWetware I wouldn't create a game that feels like work. I bought Shenmue long time ago on PS2 and it felt too much work for me. Not my genre of games"  
[X Link](https://x.com/SebAaltonen/status/1979274495287341366) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-17T19:53Z 45.5K followers, XX engagements


"What do you guys do when you get a sudden urge to build a 100000 player MMO with fully destructible world You do the math that it could run on a single XXX core Threadripper server (everybody in the same world) with your crazy netcode ideas and super well optimized C code"  
[X Link](https://x.com/SebAaltonen/status/1972209558677606637) [@SebAaltonen](/creator/x/SebAaltonen) 2025-09-28T07:59Z 45.6K followers, 107.3K engagements


"It would be a nice idea to have a virtual place for people who want to build together and show their creations. A bit like a giant MagicaVoxel sandbox. A custom high quality compute shader raytracer to render it. Has to look good of course"  
[X Link](https://x.com/SebAaltonen/status/1972285005167726663) [@SebAaltonen](/creator/x/SebAaltonen) 2025-09-28T12:59Z 45.6K followers, 5310 engagements


"@skywind3000 That is possibly but can be detected easily and the opposing client can autoreport the activity to the server. If one player gets too much reports punitive action can be taken automatically"  
[X Link](https://x.com/SebAaltonen/status/1972356964178182387) [@SebAaltonen](/creator/x/SebAaltonen) 2025-09-28T17:45Z 45.6K followers, XXX engagements


"Voxels are so easy. uint64 stores a 4x4x4 brick. uint64 a b; // two 4x4x4 bricks Union (add): a b Intersection (cut) a & b Both are a single full rate instruction on most CPUs and GPUs today. As @mmalex said a decade ago: I don't like polygons :)"  
[X Link](https://x.com/SebAaltonen/status/1973266236881690872) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-01T05:58Z 45.6K followers, 26.3K engagements


"For most people a car is a tool for getting from A to B. You don't need fancy light strips or glass roofs (that's also gone in cheap Model Y). People care about the range and driving costs. 500km range for $40k is a great. EVs offer cheap miles and are cheap to maintain"  
[X Link](https://x.com/SebAaltonen/status/1976239039587254696) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-09T10:51Z 45.6K followers, 3743 engagements


"Sure it would have been great if Tesla would have shaved the price a bit more. The price difference is not that big. However if you compare to the EU car manufacturer trim levels this is pretty much what people expect for a 5000 price difference in trim level"  
[X Link](https://x.com/SebAaltonen/status/1976240622467154180) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-09T10:57Z 45.6K followers, 3323 engagements


"@AgileJebrim @mike64_t Yeah. Video codecs are such a massive code bloat. Integrating these encoder/decoder libs in your project is a lot of work and their compile times are massive. Their core compression loops are properly vectorized but the rest of the ecosystem is not optimal at all"  
[X Link](https://x.com/SebAaltonen/status/1976569159149314331) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-10T08:43Z 45.6K followers, 234K engagements


"This was also the main reason I left Unity. I got tired of pushing performance to middle/top level managers who didn't seem to understand that we NEED to prioritize high performance tech such as DOTS and GPU-driven rendering if we want to make real-time movies happen"  
[X Link](https://x.com/SebAaltonen/status/1976923309062471987) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-11T08:10Z 45.6K followers, 8421 engagements


"Spent an extra hour on my WebGPU/WASM prototype. Compute shader fills the pixels to offscreen storage texture then there's a blit pass (full screen triangle)"  
[X Link](https://x.com/SebAaltonen/status/1976974577336304103) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-11T11:34Z 45.6K followers, 15.7K engagements


"You need reference to sense scale. Above tiny humans are a good reference of scale. Birds offer a good reference for tall structures. In this legendary Jurassic Park scene the trees are a good reference. The dinosaur pulling/snapping the trees shows a physical response"  
[X Link](https://x.com/SebAaltonen/status/1979163919563284777) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-17T12:33Z 45.6K followers, 1757 engagements


"If we generate a voxel terrain with some fancy perlin noise how do we know how big it is If we add trees or rocks we start sensing some scale. If we add 100000 players (see my previous MMO thread) we can use them as reference of scale. Animals vehicles etc also work well"  
[X Link](https://x.com/SebAaltonen/status/1979165275015590100) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-17T12:39Z 45.6K followers, 1693 engagements


"In a first person game the height of the camera also gives us information about the scale. Our eyes are roughly 1.5m above the ground when we walk. If we put the camera closer to the ground it looks the same as scaling up the world"  
[X Link](https://x.com/SebAaltonen/status/1979165734858084438) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-17T12:40Z 45.6K followers, 1872 engagements


"Fog is also a good way to give sense of scale. Far away surfaces start to look washed out due to various molecules in the atmosphere reflecting (sun) light and blocking light from behind. Shadows of big objects also reveal details. Sharp shadow edge goes along surface details"  
[X Link](https://x.com/SebAaltonen/status/1979166780166746402) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-17T12:45Z 45.6K followers, 1825 engagements


"The next topic is edge of the world. The problem with world edge is that there's nobody beyond the edge and we want to maximize the presence of others. What if we do a wrapping world. A bit similar to our globe but in 2d. Works similarly to texture tiling"  
[X Link](https://x.com/SebAaltonen/status/1979174305779851330) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-17T13:15Z 45.6K followers, 1115 engagements


"@CustomWetware @csubagio I liked the original Master of Orion. You had a couple of sliders to control everything at planetary level. Orion X was a spreadsheet game in comparison. I never liked games that feel too much like work. I want to get paid when I do work"  
[X Link](https://x.com/SebAaltonen/status/1979276692272353498) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-17T20:01Z 45.6K followers, XXX engagements


"Apple M5 GPU has tensor cores similar to Nvidia and AMD's professional CDNA. Older Apple GPUs already had tensor intrinsics but these intrinsics ran using the vector pipelines similar to AMD RDNA3 WMMA. AMD UDNA is also expected to have tensor cores"  
[X Link](https://x.com/SebAaltonen/status/1979469715933462927) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T08:48Z 45.6K followers, 20.6K engagements


"Utilizing "GPT5 Thinking" model for graphics data model optimizations (adjoint vs inverse transpose). It's quite common that I have an idea and I use "GPT5 Thinking" model to validate my thoughts and calculate rough estimates for ALU cycles etc"  
[X Link](https://x.com/SebAaltonen/status/1979574710435279285) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T15:46Z 45.6K followers, 10.4K engagements


"@WuYixiang4084 @tinivule @henningsanden These things are usually done in decaling/polishing passes. If the game world is big (GTA X = biggest game project ever) then it's going to take time and an army of artists. Fingerprints in one pole is not a lot but achieving similar level of polish everywhere is significant"  
[X Link](https://x.com/SebAaltonen/status/1979609450399862796) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T18:04Z 45.6K followers, XX engagements


"In a 4x4x4 tree our uint64 gives the ray at least X steps (DDA) worth of data using a single memory load. In best case XX steps. These extra steps allow the ray to get further from a surface faster. When you exit the 4x4x4 brick you are further away"  
[X Link](https://x.com/SebAaltonen/status/1979627358840930772) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T19:15Z 45.6K followers, 2688 engagements


"For the 4x4x4 tree the children often span multiple cache lines (up to 6) since they are continuously allocated. This is great for cache access pattern especially in cases of nearby voxels representing a continuous surface which is a common pattern for real content"  
[X Link](https://x.com/SebAaltonen/status/1979631144066023594) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T19:30Z 45.6K followers, 1801 engagements


"Of course we have to ask: Why not a single cache line aligned nodes. That's iffy because A) are storing sparse data (surface near band is not a tile) B) GPU SIMD branch/loop coherency issues (homogenerous tiles are important). Allocating children contiguously is almost as good"  
[X Link](https://x.com/SebAaltonen/status/1979648280570311038) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T20:38Z 45.6K followers, 1556 engagements


"@JoaoBapt Link is just a 32-bit aligned offset if your data is 4-byte aligned then a 32-bit link it can address up to 16GB of voxel data. 16GB address space is fine for gaming purposes"  
[X Link](https://x.com/SebAaltonen/status/1979879500071702901) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T11:57Z 45.6K followers, XX engagements


"Asserts are a double-edged sword. One should only use asserts for code bugs never for data. Data should be validated at load/receive time before inserted into runtime data structures. Error handling should not be in the low level processing/transform code"  
[X Link](https://x.com/SebAaltonen/status/1979230610435481991) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-17T16:58Z 45.6K followers, 34.1K engagements


"Each brick has uint64 voxel mask (4x4x4) and XX bit shared child data pointer (can address 16GB of voxel data due to X byte alignment). A standard brick is XX bytes. Leaf bricks are just X bytes they don't have the child pointer (postfix doesn't cripple SIMD coherence)"  
[X Link](https://x.com/SebAaltonen/status/1979480028959121597) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T09:29Z 45.6K followers, 4829 engagements


"@DubDublin @TechOperator EU passed a law forcing EV chargers to accept credit cards and not discriminate price based on payment method. Previously you had to download charger network app + register. I had X different apps on my phone. Now I don't need to do that anymore. Much better"  
[X Link](https://x.com/SebAaltonen/status/1979624738273251809) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T19:04Z 45.6K followers, XX engagements


"This is the same reason why 4x4x4 tracing ends up being faster. It's touches less different cache lines and transitions faster to near/far trace (fast empty space skip - near surface root finding)"  
[X Link](https://x.com/SebAaltonen/status/1979628423976034541) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T19:19Z 45.6K followers, 2241 engagements


"2nd reason why 4x4x4 tree is better is the amount of metadata compared to payload. For each XX bits of voxels we have XX bits of link (offset). That's XX% payload / XX% link. 2x2x2 tree has XX% payload / XX% link. In data structure design you always want to minimize metadata"  
[X Link](https://x.com/SebAaltonen/status/1979629227873063129) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T19:22Z 45.6K followers, 2149 engagements


"Thus even if we have all X children present the octree still doesn't utilize the cache lines fully. Partially used cache lines will be evicted from the tiny L1$ and you pay for data you didn't use. In realistic case we have less than X children so this issue is even worse"  
[X Link](https://x.com/SebAaltonen/status/1979630681396482143) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T19:28Z 45.6K followers, 1943 engagements


"And most important last: Do not mix material/texture data in your ray-tracing data structure. Keep it minimal. Find the ray hits first. Then do material/texture stuff. You don't want to load material/texture data to L1$ for voxels that you miss"  
[X Link](https://x.com/SebAaltonen/status/1979650498321055902) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T20:47Z 45.6K followers, 3394 engagements


"@Jonathan_Blow I have to admit that I didn't fully understand integrals before I did prefix sums based memory allocation on the GPU. Prefix sum is the discrete integral (sum) of all previous elements. Integrals are just addition with all the nice addition qualities (order composition etc)"  
[X Link](https://x.com/SebAaltonen/status/1979893782171763159) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T12:53Z 45.6K followers, XX engagements


""It's much faster performance and to load things" Zuck's reasoning started with performance. Performance matters. Google maps won because of performance Nokia lost because of Symbian OS not being designed for real-time systems (touchscreen needs it). Unity has to wake up"  
[X Link](https://x.com/SebAaltonen/status/1976919371823169765) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-11T07:54Z 45.6K followers, 68.7K engagements


"Sense of scale plays a huge role in perception. 81923 map at 1m3/voxel is 8km3. 12.5cm3 voxels = 1km3. Why do they look different I've been traveling seeing natural and architectural wonders. Would be nice to build a game with this kind of scale. Thread"  
[X Link](https://x.com/SebAaltonen/status/1979163135815692594) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-17T12:30Z 45.6K followers, 16.7K engagements


"@henningsanden It's not technically hard. But it is labor intensive for artists to add all the small details. Labor intensive = adds cost = is hard to fit in the budget"  
[X Link](https://x.com/SebAaltonen/status/1979475068959838455) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T09:10Z 45.6K followers, 78.3K engagements


"There's also a lot of competing ways to spend the energy. AI training and inference is hitting a power wall. AI companies are buying all the cheap electricity they can. More demand = more expensive electricity. Hydrogen car dream relies on almost free electrify"  
[X Link](https://x.com/SebAaltonen/status/1979896923185356851) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T13:06Z 45.6K followers, XXX engagements


"To electrify all cars in Finland we would need one additional nuclear power plant. If we do the same with cars based on hydrogen electrolysis we need three additional nuclear power plants"  
[X Link](https://x.com/SebAaltonen/status/1979897471523230038) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T13:08Z 45.6K followers, XXX engagements


"It's hard to see politicians and voters choosing to build 3x more nuclear plants for hydrogen cars and a massive effort for hydrogen fuel delivery network on top of that. The existing power grid + small extra effort is fine enough for standard EVs. Much less hassle"  
[X Link](https://x.com/SebAaltonen/status/1979897924793327778) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T13:10Z 45.6K followers, XXX engagements


"What about the fluctuation caused by wind and solar There are couple of days in Finland every year with free electricity. Can't we use it to run a small batch of hydrogen cars The problem is that heavy industries NEED hydrogen and it's expensive and produced with fossil fuels"  
[X Link](https://x.com/SebAaltonen/status/1979898848160997401) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T13:14Z 45.6K followers, XXX engagements


"SDF tiling equivalent:"  
[X Link](https://x.com/SebAaltonen/status/1979174411782459672) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-17T13:15Z 45.6K followers, 2519 engagements


"People often think voxels take a lot of storage. Let's compare a smooth terrain. Height map vs voxels. Height map: 16bit 8192x8192 = 128MB Voxels: 4x4x4 brick = uin64 = X bytes. We need 2048x2048 bricks to cover the 8k2 terrain surface = 32MB. SVO/DAG upper levels add 10%"  
[X Link](https://x.com/SebAaltonen/status/1979476961010520207) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T09:17Z 45.6K followers, 20.6K engagements


"Let's discuss why I think 4x4x4 tree is better than 2x2x2 (oct) tree for voxel storage. It all boils down to link overhead and memory access patterns. L1$ hit rate is the most important thing for GPU performance nowadays. Thread"  
[X Link](https://x.com/SebAaltonen/status/1979625344748728652) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T19:07Z 45.6K followers, 32.5K engagements


"@AddictiveColors @ZorgMonkey @AMD @LisaSu Isn't that the whole idea behind the UDNA It takes tensor cores and multi-chip scalability tech from the CDNA and the modern compute units schedulers compressors and gfx bits from the RDNA"  
[X Link](https://x.com/SebAaltonen/status/1979904881071697944) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T13:38Z 45.6K followers, X engagements


"@AddictiveColors @ZorgMonkey @AMD @LisaSu Volta was Nvidia's professional chip but they chose to dump all the new tech to Turing. Everybody got full tensor cores and ray-tracing (and all the big architectural improvements). The rest is history. CUDA won. Nvidia is a $3T company. Availability matters"  
[X Link](https://x.com/SebAaltonen/status/1979905446958792985) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T13:40Z 45.6K followers, XX engagements


"@AddictiveColors @ZorgMonkey @AMD @LisaSu DLSS was the killer app. Otherwise there wouldn't have been any business sense to give tensor cores to consumers. Slow AI adaptation in games didn't matter since Nvidia driver was using the tensor cores in all games. As a bonus researchers got CUDA tensor cores in all laptops"  
[X Link](https://x.com/SebAaltonen/status/1979906096379691119) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T13:42Z 45.6K followers, X engagements


"@AddictiveColors @ZorgMonkey @AMD @LisaSu FSR4 and Sony PSSR are both AI based. They benefit from tensor cores. Apple also added tensor cores. They also need them for similar purposes. AMD can't afford to be left behind. And they need to focus on tensor cores anyways for AI business. They just made a big OpenAI deal"  
[X Link](https://x.com/SebAaltonen/status/1979906530020413685) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T13:44Z 45.6K followers, X engagements


"The above estimate is optimistic. If we have a rough terrain we end up having two bricks on top of each other in most places. Thus we have 64MB worth of leaf bricks. SVO/DAG upper levels don't increase much (as we use shared child pointers). Total is 70MB. Still a win"  
[X Link](https://x.com/SebAaltonen/status/1979477529141407962) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T09:19Z 45.6K followers, 3909 engagements


"The interesting realization is that octtree stores bits recursively in tree levels. If each level stores 4x4x4 bricks you get X bits of precision per level. Thus X levels = XX bits of precision in height. But the high bits are shared between lots of voxels"  
[X Link](https://x.com/SebAaltonen/status/1979519419060818034) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-18T12:06Z 45.6K followers, 4986 engagements


"@minitbnn @simplex_fx I use the formatting rules that the project uses. I am just a single person the team always has much more engineers. I use my own preferences in my own hobby projects"  
[X Link](https://x.com/SebAaltonen/status/1979913628653343086) [@SebAaltonen](/creator/x/SebAaltonen) 2025-10-19T14:12Z 45.6K followers, X engagements