@@milasudril Actually, I think the most TempleOS way to drive this card would be to reprogram the card to exactly implement the TempleOS graphics primitives.
Wait!? So he’s actually running games on this thing? I thought it was going to be a super basic GPU that could technically render but only low resolution images in a very basic engine. I didn’t expect it to be running real games. He’s actually running Quake 2 at 720p? That is incredible. Great work brother. I’m so proud of this guy!
It is "basic", hence why he's using Quake II to demo it, as it's not a 3d accelerated game engine or have dynamic lighting, shaders, etc.... But even building a basic GPU like this is an insane display of skill, knowledge and intellect. Especially given all the trade secrets around GPU development and it doesn't seem like he's involved with one of the top 3
maybe with more hard work and dedication from this guy we can get annother competing brand of graphics cards in the future. may not be any time soon, but seeing people with the skill and intellect that this guy has gives me hope for the future.
@@jarsky idtech 1 add a 3D accelerated render as well, yes it has a software rendering path, to run on the CPU alone but it was updated by idSoftware to support a OpenGL render that is 3d accelerated and he used vulkan and a GPU, so he had to use the hardware render part.
Open source GPU design?. It would be fun to look at it. Sounds like he implemented the 2d part of the drivers. The Bom part cost would be expensive. A working NV 1650 starts sounding real good. I wonder if he used the arm processor to preprocess 2d commands into direct lower level hardware operations.
@@MegaVidsMike it's never going to happen. do you have any idea how expensive the tools to make a GPU chip are? hence why the semiconductor hegemony has control over chip/graphics card manufacturing. good luck finding 7 billion dollars to build a photolithography plant lol
So I'm an electrical engineer, and it feels understated what exactly this man has accomplished. You can't simply read a book or watch some youtube vids and tinker to make a gpu from scratch like he did. This is an incredible outcome for a dauntless task. Extraordinary intuition and computational mindset. It's really incredible.
why is it so difficult though. because from the software side the rendering engine is well known what are its requirements and features. you just need to build a hardware around it.
@2DarkHorizon from an analog perspective, getting the physical components to relay information in a specific pattern, limiting the voltage and amps, tracing routes, designing various limiters, repeaters, etc etc. There is a LOT that goes into a "simple" circuit, let alone an actual graphics card. Not to mention, he made a custom engine. It's significantly simpler than the modern ones used in most games, however, it's not necessarily the software side that is difficult. I'd say programming is much easier to learn and understand that physical component binary and signaling. It's much less intuitive.
@@InnaciKorushka alot of people get into software these days but i figure the hardware and electrical isn't as hard people think once people are versed in it. Not saying electrical is easy it is properly harder than software but you know what i mean there is a standards to hardware circuits you follow to get things done
Really cool that its a fpga! Fpga's are soo usefull they are litteraly the playground for chip designers and people who like making their own designs of chips and have the ability to erase/patch them without it being burned into the silicon. Fpga's are used to "emulate" old game console hardware for example if its dead or propiertary and are in some game consoles. They are even used in osciloscopes to do the processing and to hold most of the scope logic. Although it depends on what fpga you pick, they can be pretty expensive
After doing all the hardware and software design and development… How do you not register at a university … then write it up, “defend,” then get a “free” Ph.D. out of this?
@@xBintu I actually do not tend to think of university titles as having much meaning anyway. I generally view tech university titles as both a trophy and a certification of an “original contribution to knowledge;” in other words, a union card and perhaps a “license to learn.”
Because PhD requires original research and none of what this guy did would fall into that category probably. Maybe tricking the windows driver would qualify, but making a 1990s graphics card with already known render algorithms probably wouldn't.
@@baomao7243Well, you'd be wrong as my university education taught me a lot of things including FPGA design. It bothers me a lot that people view university this way, especially when many of them probably didnt even go to university and just regurgitate what everyone else thinks.
I recognise that screenshot. Descent! A fantastic game from the mid 90's. I spent a lot of time playing that (with the sound off) when I was supposed to be revising for my GCSEs.
Do you think that was Descent 2 or 3? It looked like 3 to me, but I do remember having Glide for Descent 2. Edit: Just double checked the watermark, and it is Descent 3. I always preferred 2.
@@kstarler I had descent 1 at home, then bought 3 when I was at university a few years later. I'm assuming that screenshot was 2 as I recognised the HUD, but not the actual level being played.
The number of different skillsets and expertise this takes for one person to produce is just astounding and incredibly impressive. Dylan Barrie deserves so much recognition for such an accomplishment.
thank you guys for covering such a cool passion project! would love to see more coverage like this. would even like to see some interview footage if the person is comfortable with being on screen
Finally, these guy is getting recognized by bigger media. I hope someday, competition gets so high we come back to 1080 ti times of price to perfomance
If anything the crypto boom of 2020/2021 assured NVidia and even AMD that they can spit any inflamed number in your face and you'd eat it and happily buy the product.
@@QuackZack as i said the problem from the consumers, their will always be those who act with ego to buy whatever shit these companies make for any price.
I'm a little surprised he didn't take the Intel Larabee approach and use a software rasterizer without the FPGA... he really went above and beyond for this project. Very cool.
I wish he had opened sourced it, so if there's a company that would carry on his work, they would be obligated to open source their work, too. That way, he could be an entry point for companies that wish to rival in the GPU industry
I wonder if this could be a good solution for classic PC gaming. It's pretty easy to find old CPU's, but old video cards are sometimes hard to come by if you're wanting a specific kind. With an FPGA, it could just be a matter of loading the correct configuration file and have the card re-program itself!
It would be an amazing case studie for engineer students and the A+ would be given to the best fps/ electric consumption ratio on a specific bench for all the class.
This is pretty cool. There is a guy(MNT Research)that did this for the Amiga a while back since Graphics cards for the Amiga platform are almost impossible to find and when you do they are over 1K US in most cases. He uses the same FPGA. I admire the brain power that goes into something like that. I also admire the brain power to tie shoes though. Where did my Sketcher slip-ins go?
more broadly, the Zync is an MPSoC -multiprocessor system on a chip: FPGA, ARM processors (usually several cores), and a few RTOS (real time operating system) cores. FPGAs are not persistent and need to be flashed/programmed every time they are powered up. this is where the ARM running LInux comes into play: as part of its boot sequence, it can run the software to flash the FPGA
Honestly, it's not far off from the GeForce 256 from 1999. They didn't have the term GPU yet, so they called it a "single-chip processor with integrated transform, lighting, triangle setup/clipping, and rendering engines that is capable of processing a minimum of 10 million polygons per second".
Great video, you guys also mentioned limits of FPGA which is a real step up for this channel. Do you guys know if the Verilog is available? With that u could order an ASIC right? or is would that require silicon manufacturing that we no have
You are describing my day job. I have 20 years of experience in FPGA programming. It's nice as a hobby project, with that he will certainly be hired as an FPGA developer. (Not as GPU programmer for Nvidia or AMD, that's ASIC design and totally different ball game in the restrictness and procedure domain)
@@ThatJay283 Yes, totally. But the video makes it sounds that it is totally revolutionary and special. Everbody in my team should be able to pull the FPGA part. (The HW and SW part are handled by other teams)
he should document and distribute this info...imagine what would happen if he had help coders might clean and optimise the information, add more support hardware tech people could help it to be built better, maybe even going back and designing better traces and such
This is such a blessing for tinkerers and a surprise! Many years ago, there was a seperate attempt in which the guy gave up. He documented everything. It didnt reach 3d gaming. Theres a lot of lessons to be learned, and a potential uni project to graduate!
There was so much hate towards Linus and his projects (for only 1 real frak-up), but Tech quickie format and content is one of the best I've ever seen in this field. Grabbing interesting and useful stuff, explaining it for toddlers (let's be honest, guys, that's the only way we would understand anything) and making it short and fun too. Keep up the good work!
sometimes i find myself thinking "what if every tech company went bankrupt and lost their tech". well, i guess we'd be okay. while we wouldn't recover immediately, i don't mind going back to 90's level tech to be honest.
Wow, props to him for going the whole hog with this - closest I can get to doing some hands-on on electronics as an EE these days is knocking up the odd 555 timer circuit PCB 😂
Its an impressive feat honestly. Especially writing out custom drivers. Personally if i dont need massive performance id just use my phone, a windows container wrapper and emulate older games that way. Every smartphone today trounces even some the most high end of pcs from 2010.
I've been wanting to do something like this, but with either a modified risc-v based architecture, or something crazier like a custom implementation of non other than the 6502 architecture, then modify & optimize it for graphics proccessing to create my own custom gpu & it's respective architecture. Of course, I'm only going to be aiming for low powered devices such as, microcontrollers, custom retro style games consoles & handhelds, also old or retro computers in general as a starting point before my homebrew gpus can be taken any further.
"Since you cant just insert a bare FPGA into a bare motherboard" When I was a kid I stuck a famicom cartridge into an ISA slot in my dad's PC. I don't know what I expected the outcome to be. The PC didnt turn on anymore afterwards.
The performance is more like ~1999-2000 gpu level (if I am not mistaken) 60 fps for quake1 demo was pretty amazing I don't think there was ANY graphics cards in "mid 90s" that can pull that off... Most "I build me own GPU/VDP from scratch projects" are like late 70s to mid 80s level tech.
fpgas are mostly for prototyping, if he really wanted to he could get a cpu made from the architecture he designed on the fpga nvidia/intel/amd have the same work flow, they prototype architectures on fpgas and when they bench well they get the cpu/gpu made
I’m a computer engineering major and I remember reading about if it was possible to literally make a CPU on your own, and the overwhelming consensus was that those sorts of projects are always industry level investments so I shouldn’t bother. I had assumed that’s also the case for GPUs, but this is telling me there’s hope!
He's using a FPGA so it's mostly board design and designing and coding the circuits in the fpga. However the retro community is making lots of efforts to replace and recreate older custom hardware. Like cpus and special chips. I've seen people working on custom 8088 and 286 cpus. I also saw a video the other day of a guy trying to make an actual die, this was for a 16 pixel monochrome camera but the process is close enough to other types. It seems like a lot is possible now with enough drive and some luck, and maybe some creative investment/modification in resources and equipment.
"Дайте мне точку опоры и я переверну мир" (с) Архимед. Была бы спонсорская поддержка и доступность оборудования, проблем бы не было собрать видеокарту. Первый шаг - спроектировать. Второй - создать образец. Третий - создать драйвера. Самый сложный - это первый этап. Как самый начальный и зачастую не знаешь что надо сделать. Когда начинаешь понимать как можно спроектировать видеоадаптер, становится всё намного проще.
Picture this: an oppresive middle aged kingdom the kingdom has many elites opressing peasants who are defenseless, they own the weapons after all they can do whatever they want however one day peasants come up with idea of making pitchforks the peasants while much weaker with crude farming tools as weapons can defend themselves to some degree this is an analogy why making open source technology is not a waste of time as companies slowly but surely start to get stingier and buying hardware for the regular consumer becomes less viable an alternative on the way is always a good thing, "fine ill do it myself"
@@GoldenBeans I will admit that, while I'm not certain this is the white knight solution you're portraying it as, I will agree that the effort is less of a waste of time than I originally considered it to be. Touche`
It is an interesting idea, I was just searching for fpga and gpu. The problem is though he has spent several years making an early 90s gpu instead of just getting a 4090 or something. Realistically this would become useful if someone wanted to try to start optimizing a memory bus or something. Really if you want to start competing with the giants and it would need more effort. In a way ARC was doing that.
You can actually commision tsmc to make a wafer of your own design mind you it will not be cheap if I remember its like 1300 per wafer (depending on process node) and I think the minimum order is like 100 wafers
It can be said that timing is a part of signal integrity as there are several signals and they have to be processed in parallel. Now that's a bit frustrating as PCI-e uses serial data to avoid the problems that parallell signals have. But anyway you look at it the signals has to be available at the same time. Now signals traces can't be located immediately next to each other or there will be noise inducted in the neighboring traces. Same with the zigzag of traces to make them longer. Those can't be to tightly packed or there will be inductance generating ghosts of the signals in the same trace. As frequencies go up this just gets worse and worse. Even at pretty low frequencies this can cause problems if the traces are long enough. Once a long time ago I had to lengthen the ribbon cable between a POS terminal and it's keypad. Not particularly high frequencies here and decent voltages in the signals. And yet it was enough to occasionally corrupt the keypresses. Just split the cable into individual strands, bunched them upp a bit with zip ties and it worked every time. This was a temporary fix as the shop were going to replace their desks but at the time I had to make the keypad fit into an old pocket on the desk. It was an ugly hack but the customers couldn't see anything of it. At the time I worked with component level repairs on a huge multinational computer company. I spent a lot of time studying signal integrity and I had loads of oscilloscopes almost always wired to something that didn't do exactly what it was supposed to do.
Calling it the Fury GPU is confusing for a GPU which runs 90's games, seeing that one of the most popular GPU's from the era is the ATI Rage Fury, commonly simply called the Fury at the time. It took me almost half of the video to understand this has nothing to do with the old ATI product.
The thing that makes me sad is there's enough computing power in a standard modern GPU that it could run its own operating system and be its own computer but these chip manufacturers don't seem to want that
I may be interested for one on simple PCI to use it in a Pentium 3 that doesn't have AGP. With 128MB, Windows 2000, XP, maybe 98 drivers, and to perform similar with a FX-5600. I may be interested in buying one if really does the job in games of that period.
They reportedly made standard PCI(ie: not PCI Express, though they were also available) FX-5600 cards at the time, along with other fast(for the time) PCI cards. You may get CPU limited anyway with these faster cards, though it may depend on your specific cpu.
There are many states that want to develop such a thing in their country, China comes first, and Turkey is definitely looking for such a thing, even the former 3DLabs developer Turk is still in the country. ( Yavuz Ahıska and Osman Kent )
If someone told me that they would build their own GPU I'd say he/she was crazy, my man actually did it! Amazing! Great to see! Thanks for sharing it. I know FPGAs are incredible but I thought a GPU was impossible. I'm glad I was wrong
Graphic processing with FPGA's aren't really a new thing. There already exist several designs that emulates different Nintendo systems you can find online, and within specialized equipment it's used all the time for graphic processing
Well, an FPGA could be quite feasible as a base for implementing a GPU. All you need is tons of ALU:s, not many flip-flops. I guess that a performant CPU would be much more challenging, since it needs to keep much more state between instructions.
Thanks guys! Was a ton of fun sitting down and chatting with you for this!
Amazing work.
Really inspiring man well done!
Awesome
Berry superb work 😅 dude 😎 I am on my way to create my os just need some people on board
absolute mad man
gpu prices so high that this guy made his own
Lol
4 years of income for a single GPU? Sounds about right. Thanks Nvidia!
You obviously have no idea how much Xilinx FPGA development stuff costs
@@yaroslavpanych2067 he paied more that for off the shelf product all things custom what if it could render direct 3d y mean the arm x elite can
@@adriancoanda9227 "paied"
Derp.
I´m an electrical engineer myself also designing and building boards. It´s very inspiring what the guy did. Thx for the motivation !
Really you are these engineers to build motherboard too well do you build server motherboard too .
as someone who had recently build their very first (almost) designed from scratch SBC with 80s hardware that guy is a magician
@@clementpoon120CPU type?
I'm 15yo and i have my first designed board all by me
I mean he demonstrated what Terry A. Davis did with TempleOS. One man CAN do things like this. HE IS A MAN OF FOCUS AND SHEER FUCKIN' WILL.
A John Wick you might say 👀
A driver for TempleOS
RIP Terry Davis. Got too powerful so the government had to take him out.
@@milasudril Actually, I think the most TempleOS way to drive this card would be to reprogram the card to exactly implement the TempleOS graphics primitives.
I wager one singular man can do a lot of things, long as there sheer will and commitment, time, and resources.
Wait!? So he’s actually running games on this thing? I thought it was going to be a super basic GPU that could technically render but only low resolution images in a very basic engine. I didn’t expect it to be running real games. He’s actually running Quake 2 at 720p? That is incredible. Great work brother. I’m so proud of this guy!
It is "basic", hence why he's using Quake II to demo it, as it's not a 3d accelerated game engine or have dynamic lighting, shaders, etc.... But even building a basic GPU like this is an insane display of skill, knowledge and intellect. Especially given all the trade secrets around GPU development and it doesn't seem like he's involved with one of the top 3
maybe with more hard work and dedication from this guy we can get annother competing brand of graphics cards in the future. may not be any time soon, but seeing people with the skill and intellect that this guy has gives me hope for the future.
@@jarsky idtech 1 add a 3D accelerated render as well, yes it has a software rendering path, to run on the CPU alone but it was updated by idSoftware to support a OpenGL render that is 3d accelerated and he used vulkan and a GPU, so he had to use the hardware render part.
Open source GPU design?.
It would be fun to look at it.
Sounds like he implemented the 2d part of the drivers. The Bom part cost would be expensive. A working NV 1650 starts sounding real good.
I wonder if he used the arm processor to preprocess 2d commands into direct lower level hardware operations.
@@MegaVidsMike it's never going to happen. do you have any idea how expensive the tools to make a GPU chip are? hence why the semiconductor hegemony has control over chip/graphics card manufacturing. good luck finding 7 billion dollars to build a photolithography plant lol
So I'm an electrical engineer, and it feels understated what exactly this man has accomplished. You can't simply read a book or watch some youtube vids and tinker to make a gpu from scratch like he did. This is an incredible outcome for a dauntless task. Extraordinary intuition and computational mindset. It's really incredible.
why is it so difficult though. because from the software side the rendering engine is well known what are its requirements and features. you just need to build a hardware around it.
@2DarkHorizon from an analog perspective, getting the physical components to relay information in a specific pattern, limiting the voltage and amps, tracing routes, designing various limiters, repeaters, etc etc. There is a LOT that goes into a "simple" circuit, let alone an actual graphics card. Not to mention, he made a custom engine. It's significantly simpler than the modern ones used in most games, however, it's not necessarily the software side that is difficult. I'd say programming is much easier to learn and understand that physical component binary and signaling. It's much less intuitive.
@@InnaciKorushka alot of people get into software these days but i figure the hardware and electrical isn't as hard people think once people are versed in it. Not saying electrical is easy it is properly harder than software but you know what i mean there is a standards to hardware circuits you follow to get things done
@@InnaciKorushka he used an fpga, it greatly simplified things.
But can it run Crysis?
But can it run DOOM?
@@0thewings Doom could run on a potato, so probably yes.
Crysis uses DirectX 10, so no, it can't.
@@hammerth1421 🤓
What about Return To Castle Wolfenstein?
Dude has a big career at AMD, Nvidia, or Intel in the near future. Smart dude.
Edit: looks like he works at Respawn
lol he isn't a kid
or he can start his very own gpu company
@@smalltime0 who cares how old he is.
Thats extremly tuff, i mean even intel is struggling to make it.@@w4hid
@@matthewlozy1140i think he has 2 decades of experience as a software engineer. It doesnt mater how old he is but you wouldnt cal him a kid
Really cool that its a fpga! Fpga's are soo usefull they are litteraly the playground for chip designers and people who like making their own designs of chips and have the ability to erase/patch them without it being burned into the silicon. Fpga's are used to "emulate" old game console hardware for example if its dead or propiertary and are in some game consoles. They are even used in osciloscopes to do the processing and to hold most of the scope logic. Although it depends on what fpga you pick, they can be pretty expensive
After doing all the hardware and software design and development…
How do you not register at a university … then write it up, “defend,” then get a “free” Ph.D. out of this?
because university certificates don't mean anything when you're independently...
He could be Doctor Honoris Causa in some places for his work.
@@xBintu I actually do not tend to think of university titles as having much meaning anyway. I generally view tech university titles as both a trophy and a certification of an “original contribution to knowledge;” in other words, a union card and perhaps a “license to learn.”
Because PhD requires original research and none of what this guy did would fall into that category probably. Maybe tricking the windows driver would qualify, but making a 1990s graphics card with already known render algorithms probably wouldn't.
@@baomao7243Well, you'd be wrong as my university education taught me a lot of things including FPGA design.
It bothers me a lot that people view university this way, especially when many of them probably didnt even go to university and just regurgitate what everyone else thinks.
Nice you showcased a homebrew passion project of one hobbyist. They need all exposure they can get.
I recognise that screenshot.
Descent!
A fantastic game from the mid 90's. I spent a lot of time playing that (with the sound off) when I was supposed to be revising for my GCSEs.
Do you think that was Descent 2 or 3? It looked like 3 to me, but I do remember having Glide for Descent 2. Edit: Just double checked the watermark, and it is Descent 3. I always preferred 2.
@@kstarler
I had descent 1 at home, then bought 3 when I was at university a few years later. I'm assuming that screenshot was 2 as I recognised the HUD, but not the actual level being played.
Ah, yes. The original "DooM killer".
He should add native support for Direct Draw, that'd make it great for RPG Maker games!
The modern GPUs don't support it?
@@RandMV FuryGPU doesnt tho
Now were taking about "diy" computer
Only GPU without spyware
Ohh good point, I'll duplicate the github and make drivers with spyware. Can't go missing that feature!
The number of different skillsets and expertise this takes for one person to produce is just astounding and incredibly impressive. Dylan Barrie deserves so much recognition for such an accomplishment.
He could write a book how he made it it would be a massive important resource in teaching others.
youtube glitched and all i heard was "he had to install over 4 capacitors" and i was damn that sounds about right
thank you guys for covering such a cool passion project! would love to see more coverage like this. would even like to see some interview footage if the person is comfortable with being on screen
Finally, these guy is getting recognized by bigger media. I hope someday, competition gets so high we come back to 1080 ti times of price to perfomance
900 and 1000 series cards really were special in terms of value for money
This will never happen as much as greedy these companies for the future AI!!!!! . Prices will increase since people who have money will buy.
@@weil46 A.I. is a joke.
If anything the crypto boom of 2020/2021 assured NVidia and even AMD that they can spit any inflamed number in your face and you'd eat it and happily buy the product.
@@QuackZack as i said the problem from the consumers, their will always be those who act with ego to buy whatever shit these companies make for any price.
I'm a little surprised he didn't take the Intel Larabee approach and use a software rasterizer without the FPGA... he really went above and beyond for this project. Very cool.
I wish he had opened sourced it, so if there's a company that would carry on his work, they would be obligated to open source their work, too.
That way, he could be an entry point for companies that wish to rival in the GPU industry
I think you're confusing open source and GPL.
I wonder if this could be a good solution for classic PC gaming. It's pretty easy to find old CPU's, but old video cards are sometimes hard to come by if you're wanting a specific kind. With an FPGA, it could just be a matter of loading the correct configuration file and have the card re-program itself!
I like this format, it would be cool to see more of these. just deep diving to what other people are doing and making a short story
Im a EE and do circuit design, and gotta say.... bruh, ur killin it.
I love this so much! You can learn to do anything if you are determined enough!
It would be an amazing case studie for engineer students and the A+ would be given to the best fps/ electric consumption ratio on a specific bench for all the class.
0:42 oh my god I saw Ed sheeran finally 😢😢
A lot of the tedium could be taken out of the soldering by using a Pick and Place machine.
field programable gatorade? 0:34
Heard the exact same thing too
This dude deserves all the attention. It's so awesome to see this. Mind blown.
this must be parallel universe equivalent of the Shovel AK guy
5:01 "Acutal footage" 😂
But it is
So cool to see a project I saw this guy developing in a small discord make it to here
ZZ9000 is a home built fpga based graphic card for the big box amiga ^^
This is pretty cool. There is a guy(MNT Research)that did this for the Amiga a while back since Graphics cards for the Amiga platform are almost impossible to find and when you do they are over 1K US in most cases. He uses the same FPGA. I admire the brain power that goes into something like that. I also admire the brain power to tie shoes though. Where did my Sketcher slip-ins go?
more broadly, the Zync is an MPSoC -multiprocessor system on a chip: FPGA, ARM processors (usually several cores), and a few RTOS (real time operating system) cores. FPGAs are not persistent and need to be flashed/programmed every time they are powered up. this is where the ARM running LInux comes into play: as part of its boot sequence, it can run the software to flash the FPGA
Absolutely mad lad! Also, seeing Decent! reminded me I need to see if I can install that on the kid's computer.
This is so impressive!
maybe there's a future full of FOSS and Libre hardware :)
This is cool! we been led to believe that only big companies can produce products like these. so good to see you can DIY graphic cards!
Honestly, it's not far off from the GeForce 256 from 1999. They didn't have the term GPU yet, so they called it a "single-chip processor with integrated transform, lighting, triangle setup/clipping, and rendering engines that is capable of processing a minimum of 10 million polygons per second".
Awesome work by him 👏💯🎉
Watch out letting LTT borrow this to review it, they might 'accidentally' auction it off for charity
Been a while since ive seen someone design a video circuit outside of 8 or 16 bit computing
Field Programmable Gatorade? Didn’t realize how far they’ve come in hydration beverage technology.
6:07 and MONEY since you are not paid to do this
2:45 It doesn't have to, you can rebuild a 3D printer into a pick-and-place
I've had one of these paperwork films on my pad pro 2nd ed since day one. So 4 years. They are very good.
Great video, you guys also mentioned limits of FPGA which is a real step up for this channel. Do you guys know if the Verilog is available? With that u could order an ASIC right? or is would that require silicon manufacturing that we no have
Well done and great work!!
I wish he open sourced it because this would've been a big help to a project I'm going to start working on.
You are describing my day job. I have 20 years of experience in FPGA programming. It's nice as a hobby project, with that he will certainly be hired as an FPGA developer.
(Not as GPU programmer for Nvidia or AMD, that's ASIC design and totally different ball game in the restrictness and procedure domain)
altho it makes sense he chose FPGA instead of ASIC here, because FPGAs can be reprogrammed while ASICs can only be replaced, and this is a prototype.
@@ThatJay283 Yes, totally. But the video makes it sounds that it is totally revolutionary and special. Everbody in my team should be able to pull the FPGA part. (The HW and SW part are handled by other teams)
he should document and distribute this info...imagine what would happen if he had help
coders might clean and optimise the information, add more support
hardware tech people could help it to be built better, maybe even going back and designing better traces and such
This is such a blessing for tinkerers and a surprise! Many years ago, there was a seperate attempt in which the guy gave up. He documented everything. It didnt reach 3d gaming.
Theres a lot of lessons to be learned, and a potential uni project to graduate!
There was so much hate towards Linus and his projects (for only 1 real frak-up), but Tech quickie format and content is one of the best I've ever seen in this field. Grabbing interesting and useful stuff, explaining it for toddlers (let's be honest, guys, that's the only way we would understand anything) and making it short and fun too. Keep up the good work!
sometimes i find myself thinking "what if every tech company went bankrupt and lost their tech". well, i guess we'd be okay. while we wouldn't recover immediately, i don't mind going back to 90's level tech to be honest.
Dear Techquickie Team, Thank you for this awesome roll-your-own GPU video. I thoroughly enjoyed all the technical details that were included. John M.
Wow, props to him for going the whole hog with this - closest I can get to doing some hands-on on electronics as an EE these days is knocking up the odd 555 timer circuit PCB 😂
Its an impressive feat honestly. Especially writing out custom drivers.
Personally if i dont need massive performance id just use my phone, a windows container wrapper and emulate older games that way.
Every smartphone today trounces even some the most high end of pcs from 2010.
Imagine the knowledge required to make even such a basic GPU alone. Now imagine how much more complex actual modern GPUs are.
I've been wanting to do something like this, but with either a modified risc-v based architecture, or something crazier like a custom implementation of non other than the 6502 architecture, then modify & optimize it for graphics proccessing to create my own custom gpu & it's respective architecture. Of course, I'm only going to be aiming for low powered devices such as, microcontrollers, custom retro style games consoles & handhelds, also old or retro computers in general as a starting point before my homebrew gpus can be taken any further.
At this point someone is going to build gta6 before official release
He took "Fine, I'll do it myself" to the next level 🔥
good luck to barrie in improving this GPU
Some of FPGA devboards has HDMI output, right cooling, and almost looks like GPU but also have DIMM slots etc. but those usually cost 3-10 ke
“I mean, you can, you’re just not gonna have a good time.” 😂 lmao Riley cracks me up
It would be really great to see the zz9000 covered too, as I guess that was a similar engineering process, albeit with additional ARM cores and RAM.
"Since you cant just insert a bare FPGA into a bare motherboard" When I was a kid I stuck a famicom cartridge into an ISA slot in my dad's PC. I don't know what I expected the outcome to be. The PC didnt turn on anymore afterwards.
New guy into FPGA : I'm playing god right there.
FuryGPU guy : No, it's just my experiment to learn more about hardware.
The performance is more like ~1999-2000 gpu level (if I am not mistaken) 60 fps for quake1 demo was pretty amazing I don't think there was ANY graphics cards in "mid 90s" that can pull that off...
Most "I build me own GPU/VDP from scratch projects" are like late 70s to mid 80s level tech.
fpgas are mostly for prototyping, if he really wanted to he could get a cpu made from the architecture he designed on the fpga
nvidia/intel/amd have the same work flow, they prototype architectures on fpgas and when they bench well they get the cpu/gpu made
I’m a computer engineering major and I remember reading about if it was possible to literally make a CPU on your own, and the overwhelming consensus was that those sorts of projects are always industry level investments so I shouldn’t bother. I had assumed that’s also the case for GPUs, but this is telling me there’s hope!
He's using a FPGA so it's mostly board design and designing and coding the circuits in the fpga. However the retro community is making lots of efforts to replace and recreate older custom hardware. Like cpus and special chips. I've seen people working on custom 8088 and 286 cpus.
I also saw a video the other day of a guy trying to make an actual die, this was for a 16 pixel monochrome camera but the process is close enough to other types.
It seems like a lot is possible now with enough drive and some luck, and maybe some creative investment/modification in resources and equipment.
CPUs can be created now!
Darn! He like did (almost?) everything from zero!
That is really impressive what one man could do!
Thanks for the video!
"Дайте мне точку опоры и я переверну мир" (с) Архимед.
Была бы спонсорская поддержка и доступность оборудования, проблем бы не было собрать видеокарту.
Первый шаг - спроектировать.
Второй - создать образец.
Третий - создать драйвера.
Самый сложный - это первый этап. Как самый начальный и зачастую не знаешь что надо сделать. Когда начинаешь понимать как можно спроектировать видеоадаптер, становится всё намного проще.
Insert Jurassic Park 🏞️ meme about scientists thinking "could" instead of "should".
Except in this scenario the "should" is as certain as reality and the "could" is still unknown..
Can it run SGI Fusion?
Fair. Not every exercise in futility is a waste of time.
Picture this: an oppresive middle aged kingdom
the kingdom has many elites opressing peasants who are defenseless, they own the weapons after all they can do whatever they want
however one day peasants come up with idea of making pitchforks
the peasants while much weaker with crude farming tools as weapons can defend themselves to some degree
this is an analogy why making open source technology is not a waste of time
as companies slowly but surely start to get stingier and buying hardware for the regular consumer becomes less viable
an alternative on the way is always a good thing, "fine ill do it myself"
@@GoldenBeans I will admit that, while I'm not certain this is the white knight solution you're portraying it as, I will agree that the effort is less of a waste of time than I originally considered it to be. Touche`
It is an interesting idea, I was just searching for fpga and gpu. The problem is though he has spent several years making an early 90s gpu instead of just getting a 4090 or something. Realistically this would become useful if someone wanted to try to start optimizing a memory bus or something. Really if you want to start competing with the giants and it would need more effort. In a way ARC was doing that.
He could make it open-source.
"You are only truly 'epic' if your name is written lowercase"
becomes
"You are only truly 'epic' if Techquickie made a video about you"
You can actually commision tsmc to make a wafer of your own design mind you it will not be cheap if I remember its like 1300 per wafer (depending on process node) and I think the minimum order is like 100 wafers
I thought trace lengths were more about signal timing rather than signal integrity...
It can be said that timing is a part of signal integrity as there are several signals and they have to be processed in parallel. Now that's a bit frustrating as PCI-e uses serial data to avoid the problems that parallell signals have. But anyway you look at it the signals has to be available at the same time. Now signals traces can't be located immediately next to each other or there will be noise inducted in the neighboring traces. Same with the zigzag of traces to make them longer. Those can't be to tightly packed or there will be inductance generating ghosts of the signals in the same trace. As frequencies go up this just gets worse and worse. Even at pretty low frequencies this can cause problems if the traces are long enough. Once a long time ago I had to lengthen the ribbon cable between a POS terminal and it's keypad. Not particularly high frequencies here and decent voltages in the signals. And yet it was enough to occasionally corrupt the keypresses. Just split the cable into individual strands, bunched them upp a bit with zip ties and it worked every time. This was a temporary fix as the shop were going to replace their desks but at the time I had to make the keypad fit into an old pocket on the desk. It was an ugly hack but the customers couldn't see anything of it.
At the time I worked with component level repairs on a huge multinational computer company. I spent a lot of time studying signal integrity and I had loads of oscilloscopes almost always wired to something that didn't do exactly what it was supposed to do.
so, new code to emulate old hardware on old rigs for older games?
Back in my day, FPGA meant flip-chip pin grid array.
Calling it the Fury GPU is confusing for a GPU which runs 90's games, seeing that one of the most popular GPU's from the era is the ATI Rage Fury, commonly simply called the Fury at the time. It took me almost half of the video to understand this has nothing to do with the old ATI product.
1:03 within cells interlinked
Linus would never expect his staff to prank him or optimize his disc drive,
Does it run on TempleOS?
Fury GPU? A nice homage to ATI Rage Fury... back in the time of the very first GeForce launch, when Radeon was not even a thing!
The thing that makes me sad is there's enough computing power in a standard modern GPU that it could run its own operating system and be its own computer but these chip manufacturers don't seem to want that
I want a longer video about this
5:00 minor spelling mistake
This needs more attention!!!
4:04 skip ad
My respect to him. Not everybody has this mind to do it.
He should build a commercial version and compete with big boys as fully open design free open source GPU
I may be interested for one on simple PCI to use it in a Pentium 3 that doesn't have AGP. With 128MB, Windows 2000, XP, maybe 98 drivers, and to perform similar with a FX-5600. I may be interested in buying one if really does the job in games of that period.
They reportedly made standard PCI(ie: not PCI Express, though they were also available) FX-5600 cards at the time, along with other fast(for the time) PCI cards.
You may get CPU limited anyway with these faster cards, though it may depend on your specific cpu.
There are many states that want to develop such a thing in their country, China comes first, and Turkey is definitely looking for such a thing, even the former 3DLabs developer Turk is still in the country. ( Yavuz Ahıska and Osman Kent )
The board took him one month?? What the!
That's frikkin insanely fast!
He is an absolute 100x guy
At university I made a very simple gpu on a fpga evalboard as well. It was able to run a own version of pac man :D
That’s wild I was thinking if anyone could do this a couple days ago!
If someone told me that they would build their own GPU I'd say he/she was crazy, my man actually did it!
Amazing! Great to see! Thanks for sharing it.
I know FPGAs are incredible but I thought a GPU was impossible. I'm glad I was wrong
Graphic processing with FPGA's aren't really a new thing. There already exist several designs that emulates different Nintendo systems you can find online, and within specialized equipment it's used all the time for graphic processing
Well, an FPGA could be quite feasible as a base for implementing a GPU. All you need is tons of ALU:s, not many flip-flops. I guess that a performant CPU would be much more challenging, since it needs to keep much more state between instructions.
That's a SMD device, so less soldering iron and more reflow station.