Download the source code for all my videos here: patreon.com/vercidium If you have any rendering or game dev questions, ask them here!

The greater skill here isn't code optimization, but rather how you break it down with tasty metaphors to make for easily digestible content. Kudos.

123 optimization videos later: This game runs so smooth you can get 4k graphics with 120 frames a second on a potato clock even without the potatoes.

Vercidium trying to explain game engine optimisation to an American: "So imagine a burger."

This is why you can use double buffering, which is kind of like what you are doing but instead of for the framebuffer, you are doing it for the mesh. Normally GPU operations queue up in the buffer, they aren't meant to be completed by the end of the refresh. If you use a double buffered system, it allows everything to render without a stall, because when a stall would have occurred, it just shows the back buffer like normal.

I just binged all of your optimisation videos. The metaphors are awesome and easy to follow, the code and graphics are clean, the voiceover is easy to understand. This is what I wish school/university was like. Instant subscribe, you have earned a place among my favorite KZfaqrs. Thank you for your work.

Semaphores could also be ideal as fences can cause the CPU to stall waiting for GPU work to finish. Semaphores would allow for exclusive synchronization of GPU tasks ensuring the CPU is constantly writing commands and the GPU is constantly processing them. Great video nonetheless.

This is standard behavior in all major game engine used in production for the last 10-15 years.

I love this man's food analogies as much as this guy loves his optimization. Amazing video, I'm so in love with the animations!

It's worth noting that for some games, the additional input delay might be considered unacceptable. Most games it won't matter but in those rare cases it may be necessary to explore alternatives

It really was worth buying this 16,000 fps monitor---this is the first video to really use it to its fullest, but boy is it glorious.

Videos like this are rare to find But It has to be my favorite type of video by far, making code run faster, better gaming experience, GPU goes vroom I love it!

Bro i need more optimization videos this is amazing af

This is a highly optimized dev tutorial. Jokes aside I think you have a true talent for game dev and teaching. Subscribed.

This kind of sounds like double buffering, but for the geometry instead of the actual framebuffer. Very cool idea!

I'm not coding games and stumbled here by chance, but this is a really, really well done description which can be appreciated by anyone with at least a bit of coding experience. Thanks.

I really loved the video and the analogy. I think you should increase the size of the code and animations, because seeing them on mobile is kind of hard. Great video nonetheless!

The master chef has an infinite number of pans, but doesn't want to overwhelm his apprentices, so he keeps only the amount of pans necessary in the kitchen until a greater or lesser amount is needed.

yooooo i love that not only did you make a cool game i enjoyed, but that you're sharing all the knowledge you learned from making it with everyone. Good luck dude! I hope you continue this, It's really interesting and I love it.

"Explain double buffer like I'm five" Great video Vercidium!

I like how you're showing alternatives to the "just add more cores/memory" mentality.

Now I'm hungry.

not even 30 seconds in and it's already being broken down for the americans

the biggest thing this video taught me is optimization =/= fewer lines

Best programming analogy is always about food and cooking.

My favorite series on youtube, thank you for doing it!

Fantastic overview. Right to the point and thorough enough to show off everything without getting bogged down in the details.

Thanks for this, I'm really enjoying watching your videos. When I come to optimizing I'll be circling back to make sure I've covered the things your have. All the best 🙂

Great video and I am sure it will help me in the future 👍At the first moment I thought the sorting by swapping with the first element only works with 2 meshes but after thinking about it I realized it also works well with longer lists too if the transfer time is equal for every mesh. If not, then a circular array/ring buffer might be a better alternative to avoid rendering fast-to-transfer future meshes before slow-to-transfer past ones. That would cause the future mesh to render 2 frames while skipping the past one with your algorithm.

Isnt this just asyncronous compute?

This guy just added Promises to a game engine

something about this video made me super excited to watch, maybe the thumbnail/title.

Hey small feedback: could you please zoom in when displaying code? Small screens cant see the logic you're trying to showcase (even if the exact text doesn't matter). Awesome video as always❤️

I had a similar issue in my 2D game made in Game Maker Studio. When it tried to render some high resolution sprites I would get those stutters. However, I found a very useful command "texture_prefetch" which as you can guess loads a texture group into VRAM memory even if it's not being drawn yet. And "texture flush" clears a group from VRAM. By being manually proactive about texture loading, I not only removed all the stutter, but I also cut level loading time down to under half a second. I'm pretty sure all engines have asynchronous buffered data loading / rendering, but I think some developers don't utilize it to it's full extend and just try to render a bunch of unloaded data at once. Your explanation is very good, and I hope it reminds other devs to think more carefully about asset loading.

Great video, loved all the animations. Even as someone who isn't particularly experienced in coding I feel like I understood everything!

the major props for making my brain understand things i wouldnt have fathomed even trying to understand before

When you order a burger and the chef accidently adds extra bacon ;)

Got it, I just need enough memory for 15 thousand meshes.

Love the analogies and especially the visuals, you have such a knack for presenting problems in an easy to understand light! Great vid :)

That is not really seen as a real optimization, more like a technique that (as you said) has been done. It is really interesting to see your approach compared to approaches game studio's make. A GDC talk that is on a technical level really interesting is this one "Marvel's Spider-Man: A Technical Postmortem". I also want to ask a question about your previous video, you mention that indeed triangle strip is way faster than triangle list. The problem that I seem to face, is that there are no tools, as far as I know of that convert triangle list to triangle strip. This is not a problem when you make all other models yourself and create a script that does it for you. But how did you do it in your previous video?

This man spent 4 years of his life fixing the little freezes you get every once in a while. Real dedication!

There's actually an optimization mod for Minecraft that more or less does this where the renderer will keep using an older mesh until the new ones are ready. You did a nice job with the analogies!

This video has made me realize that I still don't understand coding at all

I can't believe the closest analogy to a bottleneck was a burger. A *bottleneck*. But seriously, great video!!

Great video, I learned a lot! Small comment for future endeavors: watching this on mobile was difficult because I assume you optimized the text size for fullscreen desktop viewing. Other than that, I loved it!

I don't code yet but it is something that I find fascinating. I really enjoyed how you made this problem make sense even to me. Thanks for sharing

Very cool! Time to stop asking where my burger is every five minutes to increase my fps.

The animations are hilarious :D Those little hands of a video card :)

Long freezes in modern games can be attributed to a variety of factors, often relating to resource-intensive operations or inefficiencies in handling game assets and rendering. Here are some of the common causes: Asset Loading: When games load large assets (like textures, models, or sound files) from the disk into memory, it can cause a noticeable freeze, especially if the game is not using asynchronous loading techniques. Garbage Collection: In games developed with languages that have automatic memory management (like Java or C#), garbage collection can sometimes cause freezes or stutters. This happens when the garbage collector runs to free up memory, temporarily halting other processes. CPU/GPU Synchronization Issues: If the CPU is waiting for the GPU to finish rendering (a scenario known as a GPU bottleneck), or vice versa (CPU bottleneck), it can result in freezes. Efficient parallel processing and synchronization are crucial to avoid such stalls. (the one this video covers part of, and far from the only possible cause) Inefficient Resource Management: Poorly managed resources, such as repeatedly loading and unloading the same assets, can lead to performance issues and freezes. Complex Calculations or Scripts: Intensive computations, like complex AI calculations, physics simulations, or extensive world updates, can cause freezes if they are not efficiently managed or offloaded to separate threads. Network Latency or Hiccups: For multiplayer games, network issues can cause freezes or lag if the game's state is tightly coupled with the timely receipt of network packets. Driver or Hardware Issues: Sometimes, the problem may lie outside the game itself, such as outdated or buggy graphics drivers, or hardware that is overheating or malfunctioning.

this is what wgpu does behind the scenes, I love wgpu

Why this works: modern GPU has a lot of bandwidth but with a high latency to transfer data from main memory to gpu.(a lot of car lanes but long way to drive). If we could have 0 latency then we wouldn't need this because any transfer would be finished very quickly (high likelihood of finishing transfer before being drawn). Instead of doing nothing and waiting for the transfer to complete, we effectively stack multiple extra transfers for future frames within the first transfer's wait time (we can do this because we have a lot of bandwidth. aka we are sending more trucks on different car lanes simultaneously instead of waiting for the first one to come back and then send it out again). This is a form of parallelism and this idea of "doing other things or doing more of the same thing while waiting for the first one to finish" is everywhere in both cpu and gpu programming and let's you make more efficient(full) use of your hardware.

Feel like I've seen a couple of these coding youtube channels run by fellow Aussies. Makes me happy knowing we're doing our part for the world.

Wow, I really like that underlaying background simulation at 1:51 I'm trying building something similar myself using Openprocessing but this looks like some great end result! Thanks for the inspiration.

You lost me at burger

Really excellent explanation for an issue a lot of players, and even developers, incorrectly associate with optimization. Also, I noticed your mic is picking up a lot of room reverb. A cheap mic shroud behind the mic will help a lot. The poor man's shroud is a closet full of clothes with the mic partially inside. If you can't fix it at the mic, try a "de-room" or "de-reverb" plugin. Your visuals are fantastic, your audio can be just as good without much more effort.

I really really really like this kind of contents, It's like an art. I wonder what it'd be like if there's a game engine that has all optimization techniques known today and implemented efficiently. I'd really love to try that game engine or it's reels.

I once implemented in GL4.5 similiar thing, for the uploaded uniform constants (UBO). It used fences to avoid remapping/reallocating the GPU memory each time when possible. It allocated N slots for the blocks in a single UBO. For each slot index, a fence is used tell if CPU is allowed to reuse it yet. If no slot was ready, a new (bigger) UBO would be allocated. After a few initial frames, the UBO's capacity would stabilize.

I really appreciate this format of video from you. Please keep up the good work:)

didn't realize you were sector's edge dev until the end!

I was hungry before watching this, now I know what I want but it's in the middle of the night and no place is open! :o

My god it's so hight quality. I subscribe immediately!

Hey love the video man but the code font is a little small for a pc and almost microscopic for my phone. The animations and everything are also amazing but if you could enlarge the code it’d make the content more accessible.

Wouldn't there be a chance that each new mesh clogs up the transfer pipeline even more, so you would end up with drawing really old meshes at some point?

Bruv you are the single most valuable resource I have as an indie dev. My games will run buttery smooth because of you and those like you. I've had these stories in mind since i was a child and didn't have the tools or skills to do them right. But with people like you and tools like blender and unreal, I will bring some really fun stuff! For the love of gaming!

Very humble explanation style for laymen to understand. Kudos!

Just a note. The code you show is a very tiny font size. It was difficult for me to read.

This is an epic video man, really liked the way you showcased your iterative design methodology to refine the problem into smaller chunks and solved them one at a time.

This is awesome! Managing stuff like this is still a real challenge and this is a great solution. Thank you for the great easy to understand explanation too!

Should you really be using a list? Wouldn't a pool be much preferable if we're considering optimization?

Very good, thank you for informing the community

Lots of people seem to be mad at the fact that these optimizations are not groundbreaking and already exist in some engines. But as a game designer making solo projects on several engines, I'm super thankful for the amazing explanations of processes I would otherwise be unaware of. I love to see how the technologies we used are built from thousands of smart decisions like these. Also, good job and good luck on your engine project :D

The first part also explains why some fast food drive thru allocate an employee taking orders instead of letting customers order at the fixed kiosk.

Reminds me of a swap chain, only for geometry instead of fully rendered frames. Very cool!

My guy explains double buffer so my 5 year-olds can actually understand it, kudos to you!

This is the first time I hear the recommendation of using a list instead of an array for a performance boost! I imagine that the list is actually an dynamic array under the hood. The performance boost of caching contiguous data in memory is just way over the benefit you can get from delegating freeing some memory of a linked list to the garbage collector.

Man, I was watching through this and I looked at your name and was like "Huh, his name reminds me of the developer of Sector's Edge." Then a moment later: "Wait... that is really similar..." And then I checked and you actually were the dev for Sector's edge lol. Cool! Man I miss that game, it was my favourite fps! I can't wait for your next project!

From now on, my toaster can run crysis.

Stuff like this is why i love graphics engineering, the interwoven understanding of what the hardware does and how you can get it to do what you want it to in the most efficient manner is just fun. Cool stuff!

I’m addicted to your channel! \o/

"a flashing red light, changes to green when your burger is ready" ... WHAT?!? i feel like Doc from BTTF "what year is it?"

Explaining optimization to an American: "So Imagine a burger..."

I’m curious: do you ever look at Minecraft optimization mods to get ideas of how to make your own voxel game engine better? People have tried many methods and there’s a lot of source code out there

I hope more game devs from indie to AAA take into account stuttering more, people seem to only be phased by FPS and are completely uneducated regarding frametime and how a consistent frametime is more important than FPS the majority of the time.

This specific technique is for when you're **dynamically building a mesh** to represent a voxel world. At first, I misunderstood and thought each block in the world was a separate mesh and the solution didn't make much sense. The title and Intro examples makes it seem applicable to all games. If you're making a game without generated meshes, this specific fencing technique won't be very useful. However, double buffering is generally useful in many scenarios.

Would sorting still be viable if the number of meshes is very big? Also, I loved this video. I never even realized this. Or rather, I knew of it but never was consciously aware of it while coding

Does this apply to game engine devs, or also to game devs that use game engines like unreal or unity?

Changing the thumbnail in the name of optimisation on a video about optimisation is pure optimisation #ISawTheBurger

Wasn't expecting to get burgered here

at the end the performance was 15k fps, what was the performance of the mesh rendering without any optimizations at the very start, how many fps?

Wow this is incredible, those this also work in Unreal Engine 5?

the anecdote between burgers/restaurant pagers and rendering is hilarious and useful

I''m kind of stunned that in the impressive history of modern gaming this hasn't been done before. Kudos for finding that out where others didn't.

0:02 I have 2K+ hours in NMS and never (on my computer which has over minimum spec) had a freeze like that.

Fun fact: Having a taste of this source code is cheaper than having a taste of an actual burger (at least where I live)

This is genuinely interesting. I love it. Keep it up!

Amazing video, one tip: The code font size is a tad too small for phone screens. I think 10-20% larger would fix it.

Bro spend 6 years to create a Game Engine just to told us, What a hero!

youtuber with 30k subs vs multimillion game companies

We finally know how to sell more burgers. Stonetoss would be so proud.

In my current engine project I've made an upload manager who reuses a set of staging buffers to upload data to the GPU at a capped rate (wouldn't want to choke out the bandwidth), it uses a counter fence to keep track of the latest job completed. When work is submitted to it, the submit function returns a job number and the dependent objects can check with the uploader if the work is finished (via the fence under the hood). Then, the majority of objects used in rendering hold their data in a variety of buffer types (swapped out as needed) with the more common being the swap buffer, which the uploader knows how to handle right away. I've been considering making a swap chain buffer which allows multiple jobs to be queued at once instead of just the 2, I think I'd need to make another system for recycling buffers when their done. I might also want to expand it for texture streaming solutions wherein they could produce work for both uploading and consolidating buffers, but that might be a waste of space if too many jobs are incomplete at once. I love stuff, its a wonderful puzzle, and entirely why I entered graphics programming and game engine development!