When PCBWay stopped accepting PayPal last month I tried JLC and I have to say, their quality was impressive and the price was a little over half the PCBWay quote. Certainly worth shopping around with your gerbers. Might be an interesting write up comparing quality and price from the leading Chinese fab houses for the same board.
Only done very small 2-layer boards for hobby projects, but it's crazy that you can get custom PCBs made and shipped from China for <£10
It seemed like it might be coming to an end when PCBWay suddenly had no reasonable payment options, but JLC has been great so far (and I believe PCBWay has credit card payments sorted now?)
(Just wish I had far more free time to spend on hobbies, there's so many possibilities with 3D printing, microcontrollers, and custom PCBs now all so readily available)
The site configured at this address does not contain the requested file.
If this is your site, make sure that the filename case matches the URL as well as any file permissions.
For root URLs (like http://example.com/) you must provide an index.html file.
Read the full documentation for more information about using GitHub Pages.
I found the project on YouTube[1] and wanted to share it - but decided to find something that's text for HN, and in the rush to post I failed to check if the post is even complete. I should've posted the video instead.
I find first version https://github.com/petrmikheev/endeavour much more impressive. Dude somehow managed to get 100MHz DDR1 ram working on 2 layer board with no ground reference :o Its one of those things you only attempt when crazy or dont know any better. Anyone with EE experience will tell you its impossible, like flying commercial grade SoCs in satellites :) Mad lad.
This is very impressive. How did you learn to design a real computer, not the toy ones a lot of people made? I read part 1 and part 2 and looks like you just “thrown in” Ethernet and other stuffs and it was done. Really hope to learn from the process, thanks!
This is really cool and impressive... but relatedly...
Has anyone figured out what the minimum specs for Quake are?
I feel like the first thing everyone does with a computer is to determine whether or not it can run quake, and I'm just wondering what the like, most simple computer that could exist is, that could run quake?
You can find a lot of discussion about what the minimum specs for Quake are. Famously, it needs a decent FPU, and the Pentium was a convenient early CPU with a decent built-in FPU. It was significantly faster than a 486.
…But people have managed to run Quake on the 486.
And the myth people tell about Quake is that it killed Cyrix, because Quake performance on Cyrix was subpar. But was that true? And if it was true, was that because the Cyrix was slower than a Pentium, or was it because the Quake code had assembly that was hand-optimized for the Pentium FPU pipeline?
Anyway. “Most simple computer that could run Quake” is probably going to include a decent FPU. If you are implementing something on an FPGA, you can probably get somewhere around 200 MHz clock anyway. At which point you can run Quake II.
The Cyrix story is actually well-documented. Quake's software renderer used hand-optimized x86 assembly with FPU instruction sequences specifically tuned for the Pentium's pipeline. Cyrix processors had a different FPU execution pipeline that stalled on those specific instruction orderings — the issue wasn't raw FPU performance, it was that the Pentium-optimized code ran slower on Cyrix than straightforward C code would have. It was hand-optimization that made things worse, not better, on a competitor's hardware.
The timing was brutal for Cyrix. This was right when "Intel Inside" was becoming a meaningful consumer brand signal, and game benchmarks were becoming the primary way consumers evaluated CPU purchases. Quake wasn't just a game, it was the benchmark everyone ran at CompUSA to compare machines. Being demonstrably worse at Quake, regardless of the cause, was a marketing catastrophe.
The real floor for running Quake is basically "does it have a hardware FPU." The 486 DX (with FPU) could do it at low resolution and low framerate. The 486 SX (no FPU, software float emulation) was genuinely painful. The Pentium was the first CPU where it actually felt good.
My perspective from being a teen doing lan party stuff at the time: Quake ran slow on them, but it was far from the only thing that ran slow. Cyrix was well understood to be the value brand for general office apps and such, but not up to it for more demanding computing, and for having random compatibility issues here and there.
Ultimately what killed Cyrix is they just couldn't offer enough of a discount vs intel to matter, especially with all the lock in stuff intel was doing with Dell, Gateway, etc.
Intel Inside was a successful marketing campaign as well. If you were around back then I bet you can imagine the jingle/chord immediately.
I had a Cyrix 6x86 when Quake first came out. My disappointment at how poorly Quake ran on it was significant, especially because pretty much every other game at the time ran well on the Cyrix. The FPU performance in Quake was doubly handicapped on the Cyrix: not only was its FPU slower than the Pentium's to begin with, Quake's code was indeed hand-optimized for the Pentium's FPU pipeline. Fabien Sanglard's writeup of Michael Abrash's optimizations for Quake goes into great detail: https://fabiensanglard.net/quake_asm_optimizations/
Yes but also no. The problem with fixed point arithmetic is a lack of dynamic range compared to floating point. Floats are great at representing both large numbers with limited precision and small numbers with high precision, but with fixed point you have to make a choice based on which kind of number you're trying to represent. Meaning you need to use a mixture of 8.24, 16.16 and 24.8 fixed point types (and appropriate conversions) depending on the context of the calculations that you're doing.
It's possible to write a game engine with that limitation, but there's no easy natural conversion from Quake's judicious use of floats to a fully fixed-point codebase. You'd have to redesign and rewrite the entire engine from scratch, basically.
I want to look at this from a different perspective… a single-precision floating-point multiply is pretty simple, no? 24x24 bit multiply, which is about half as many gates as a 32x32 bit multiply.
Maybe I would prefer to rip out the integer multiplication unit first, before ripping out the FPU.
The PS1 doesn't an FPU but got a version of Quake 2, so it's possible. That said, it was somewhat different from the PC version, so it could be argued that it's not the same game.
I can't speak on Quake, but I was a level designer on the failed effort to port Unreal to PSX.
My understanding from talking to the coders at the time was that Unreal's software renderer was a huge advantage as a starting point. They were able to reuse a lot of the portal rendering stuff as setup on the R3K cpu, but none of the rasterization. That had to go to the graphics core, which was a post setup 2D engine that in addition to the usual sprites, could do tris and quads.
We had a budget of about 3k polygons post clipping, and having two enemies on screen would burn about half of that. The other huge limit is the texture cache was tiny, so we couldn't do lightmaps. Our lightning was baked in at vertex level and it just was what it was.
I imagine the situation with Quake was comparable. The BSP stuff would carry right over, but I can't imagine they got lightmapping proper working at the time. They'd also need some sort of solution for overdraw, as Quake's PVS was a lot more loose than Unreal's portal clipping.
The PS1 version uses a custom engine based on technology built for the game Shadow Master, the previous title by Hammerhead Studios. It was a technical tour de force for the original PlayStation.
I honestly don’t remember what the frame rate was, but it definitely improved when I upgraded to a Pentium 100. I distinctly remember a buddy giving me some RAM (2x4MB) which allowed me to play on the 486. I was so happy!
The DX2s _were_ a significant improvement over the 486DX, but I’ll admit, I might be remembering the excitement of getting to play Quake at all! The framerate may have been 15-20 fps and I just dealt with it,
The minimum requirements, on the box, were apparently Pentium 75Mhz. 8MB (DOS), or 16 RAM (WIN95).
As other had said: the only 'deal breaker' here is FPU.
Q1 is playable but not on any modern understanding of that word on a system with Am486DX4/100 with 16MB RAM and S3 Trio64V+. You can disable sound effects for a couple FPS more.
Mostly you would be fine because the level design in Q1 heavily tends to the closed spaces and corridors with only a glimpses of the outside and a rare halls and caverns. That being said the Necropolis would be a test of strategic thinking for turn-based FPS and Ziggurat Vertigo is unplayable.
Quake 2 was the one with the clever approximate inverse square root code, right? I wonder (especially since there’s an instruction nowadays to draw inspiration from), can you implement it “in hardware,” so to speak?
DDR3 traces need to be length matched, because at 800MHz (the slowest "standard" rate, though I think you can drop to 666MHz safely) the value on the pins is changing every 1.25ns, and having traces of different lengths means you probably won't see the right values on all the pins at the same moment. Length matching produces the squiggles.
The diagonal orientation of the DDR3 chip and corresponding diagonal traces I suspect is a choice made by the author to ease the layout process - it's more likely that is hand laid out to get traces of somewhat similar length with a minimum of fuss, followed by a length matching tool. A non-standard orientation can cause issues with pick-and-place machinery, which usually will handle 90 degrees fine, and _often_ 45 degrees fine, but (AFAIK) _rarely_ anything else, but that's not a problem for the author because he's assembling it himself. A diagonal IC also usually results in wasted space, which you can see in the empty areas of the resulting board. A 90 degree orientation may have allowed for a few more decoupling capacitor banks, but since his board works, who am I to sit here and judge?
> A non-standard orientation can cause issues with pick-and-place machinery, which usually will handle 90 degrees fine, and _often_ 45 degrees fine
This sounds like nonsense. Pick and place machines don't pick up components perfectly deterministically. There is always a tilt and an offset when you are picking the part up, which is why a computer vision system has to account for part orientation and the center of the part. The machine must compensate the error by moving and rotating the part accordingly.
Why would you be a weirdo? Two-panel is the only somewhat sane way to navigate the file system, though I still try to Ctrl+\ and need to remember to Alt+S and only then to type the search pattern.
I have the album on my phone. When I get called in to put out a fire and save the day, I like to put on March Of The Stroggs in the car when arriving at the destination. It's a great soundtrack for two reasons - the first one is sweet, wasted youth and the second is it's a great soundtrack.
What's the deal with the diagonal squigglys
Cool write up, getting initial bill shock from 2 layer to the 4+ layer PCBs is a rite of passage :)
The next step up is HDI. I knew it was (still) expensive, but I did not realize it was a cost multiplier, not adder :-)
Went with wrong board house. JLC seems to run 6 layer promo for 2 years now. This board is $36 for 5, $100 gets you 50 pcbs.
When PCBWay stopped accepting PayPal last month I tried JLC and I have to say, their quality was impressive and the price was a little over half the PCBWay quote. Certainly worth shopping around with your gerbers. Might be an interesting write up comparing quality and price from the leading Chinese fab houses for the same board.
Only done very small 2-layer boards for hobby projects, but it's crazy that you can get custom PCBs made and shipped from China for <£10
It seemed like it might be coming to an end when PCBWay suddenly had no reasonable payment options, but JLC has been great so far (and I believe PCBWay has credit card payments sorted now?)
(Just wish I had far more free time to spend on hobbies, there's so many possibilities with 3D printing, microcontrollers, and custom PCBs now all so readily available)
Section 6 where you link to Quake II is 404. (at the time of this post)
URL: https://blog.mikhe.ch/quake2-on-fpga/part6.md
404 File not found
The site configured at this address does not contain the requested file.
If this is your site, make sure that the filename case matches the URL as well as any file permissions. For root URLs (like http://example.com/) you must provide an index.html file.
Read the full documentation for more information about using GitHub Pages.
Parts 5 and 6 are 404. At the end of the article there are words:
"More pictures in the next part.
Next part: coming soon"
I suppose the link came to HN a bit too early.
yeah, not the author here.
I found the project on YouTube[1] and wanted to share it - but decided to find something that's text for HN, and in the rush to post I failed to check if the post is even complete. I should've posted the video instead.
[1]: https://youtu.be/sioLAkNQC_I
I find first version https://github.com/petrmikheev/endeavour much more impressive. Dude somehow managed to get 100MHz DDR1 ram working on 2 layer board with no ground reference :o Its one of those things you only attempt when crazy or dont know any better. Anyone with EE experience will tell you its impossible, like flying commercial grade SoCs in satellites :) Mad lad.
Signal integrity is a myth propagated by big-PCB.
That's some mad dedication to go from kicad schematics to running Quake. Very impressive!
Cool! Have you considered offering this board on Crowd Supply or similar? There don't seem to be many boards available for Efinix FPGAs.
This is very impressive. How did you learn to design a real computer, not the toy ones a lot of people made? I read part 1 and part 2 and looks like you just “thrown in” Ethernet and other stuffs and it was done. Really hope to learn from the process, thanks!
Very impressive, and I'm sure satisfying. Kudos!
Demo: https://www.youtube.com/watch?v=sioLAkNQC_I
This is really cool and impressive... but relatedly...
Has anyone figured out what the minimum specs for Quake are?
I feel like the first thing everyone does with a computer is to determine whether or not it can run quake, and I'm just wondering what the like, most simple computer that could exist is, that could run quake?
You can find a lot of discussion about what the minimum specs for Quake are. Famously, it needs a decent FPU, and the Pentium was a convenient early CPU with a decent built-in FPU. It was significantly faster than a 486.
…But people have managed to run Quake on the 486.
And the myth people tell about Quake is that it killed Cyrix, because Quake performance on Cyrix was subpar. But was that true? And if it was true, was that because the Cyrix was slower than a Pentium, or was it because the Quake code had assembly that was hand-optimized for the Pentium FPU pipeline?
Anyway. “Most simple computer that could run Quake” is probably going to include a decent FPU. If you are implementing something on an FPGA, you can probably get somewhere around 200 MHz clock anyway. At which point you can run Quake II.
The Cyrix story is actually well-documented. Quake's software renderer used hand-optimized x86 assembly with FPU instruction sequences specifically tuned for the Pentium's pipeline. Cyrix processors had a different FPU execution pipeline that stalled on those specific instruction orderings — the issue wasn't raw FPU performance, it was that the Pentium-optimized code ran slower on Cyrix than straightforward C code would have. It was hand-optimization that made things worse, not better, on a competitor's hardware.
The timing was brutal for Cyrix. This was right when "Intel Inside" was becoming a meaningful consumer brand signal, and game benchmarks were becoming the primary way consumers evaluated CPU purchases. Quake wasn't just a game, it was the benchmark everyone ran at CompUSA to compare machines. Being demonstrably worse at Quake, regardless of the cause, was a marketing catastrophe.
The real floor for running Quake is basically "does it have a hardware FPU." The 486 DX (with FPU) could do it at low resolution and low framerate. The 486 SX (no FPU, software float emulation) was genuinely painful. The Pentium was the first CPU where it actually felt good.
> Don't post generated comments or AI-edited comments. HN is for conversation between humans.
https://news.ycombinator.com/newsguidelines.html
Your entire comment history seems to be AI generated.
My perspective from being a teen doing lan party stuff at the time: Quake ran slow on them, but it was far from the only thing that ran slow. Cyrix was well understood to be the value brand for general office apps and such, but not up to it for more demanding computing, and for having random compatibility issues here and there.
Ultimately what killed Cyrix is they just couldn't offer enough of a discount vs intel to matter, especially with all the lock in stuff intel was doing with Dell, Gateway, etc.
Intel Inside was a successful marketing campaign as well. If you were around back then I bet you can imagine the jingle/chord immediately.
I had a Cyrix 6x86 when Quake first came out. My disappointment at how poorly Quake ran on it was significant, especially because pretty much every other game at the time ran well on the Cyrix. The FPU performance in Quake was doubly handicapped on the Cyrix: not only was its FPU slower than the Pentium's to begin with, Quake's code was indeed hand-optimized for the Pentium's FPU pipeline. Fabien Sanglard's writeup of Michael Abrash's optimizations for Quake goes into great detail: https://fabiensanglard.net/quake_asm_optimizations/
can it be rewritten to use fixed point arithmetic instead?
Yes but also no. The problem with fixed point arithmetic is a lack of dynamic range compared to floating point. Floats are great at representing both large numbers with limited precision and small numbers with high precision, but with fixed point you have to make a choice based on which kind of number you're trying to represent. Meaning you need to use a mixture of 8.24, 16.16 and 24.8 fixed point types (and appropriate conversions) depending on the context of the calculations that you're doing.
It's possible to write a game engine with that limitation, but there's no easy natural conversion from Quake's judicious use of floats to a fully fixed-point codebase. You'd have to redesign and rewrite the entire engine from scratch, basically.
I want to look at this from a different perspective… a single-precision floating-point multiply is pretty simple, no? 24x24 bit multiply, which is about half as many gates as a 32x32 bit multiply.
Maybe I would prefer to rip out the integer multiplication unit first, before ripping out the FPU.
The PS1 doesn't an FPU but got a version of Quake 2, so it's possible. That said, it was somewhat different from the PC version, so it could be argued that it's not the same game.
The PS1 version definitely has its own engine, which is not just a port of the Quake 2 engine to the Playstation, but a new engine.
I can't speak on Quake, but I was a level designer on the failed effort to port Unreal to PSX.
My understanding from talking to the coders at the time was that Unreal's software renderer was a huge advantage as a starting point. They were able to reuse a lot of the portal rendering stuff as setup on the R3K cpu, but none of the rasterization. That had to go to the graphics core, which was a post setup 2D engine that in addition to the usual sprites, could do tris and quads.
We had a budget of about 3k polygons post clipping, and having two enemies on screen would burn about half of that. The other huge limit is the texture cache was tiny, so we couldn't do lightmaps. Our lightning was baked in at vertex level and it just was what it was.
There's a bit more info here: https://www.terrygreer.com/unrealpsx.html
I imagine the situation with Quake was comparable. The BSP stuff would carry right over, but I can't imagine they got lightmapping proper working at the time. They'd also need some sort of solution for overdraw, as Quake's PVS was a lot more loose than Unreal's portal clipping.
The PS1 version uses a custom engine based on technology built for the game Shadow Master, the previous title by Hammerhead Studios. It was a technical tour de force for the original PlayStation.
Sure, but then you need CPU that is twice as fast :). Playstation did it by pushing geometry calculations to GTE.
Quake ran like shit on 486dx33, a few fps at best.
I played Quake on a 486 66Mhz DX2 with 16MB of RAM in the 90s. On the lowest resolution, but it was fine.
I had a 486dx33 and it was unplayable, a few frames per second...
I honestly don’t remember what the frame rate was, but it definitely improved when I upgraded to a Pentium 100. I distinctly remember a buddy giving me some RAM (2x4MB) which allowed me to play on the 486. I was so happy!
The DX2s _were_ a significant improvement over the 486DX, but I’ll admit, I might be remembering the excitement of getting to play Quake at all! The framerate may have been 15-20 fps and I just dealt with it,
The minimum requirements, on the box, were apparently Pentium 75Mhz. 8MB (DOS), or 16 RAM (WIN95).
Quake DX2 framerate is 5-7 depending on rest of the setup.
That’s only because everything can run Doom now.
As other had said: the only 'deal breaker' here is FPU.
Q1 is playable but not on any modern understanding of that word on a system with Am486DX4/100 with 16MB RAM and S3 Trio64V+. You can disable sound effects for a couple FPS more.
Mostly you would be fine because the level design in Q1 heavily tends to the closed spaces and corridors with only a glimpses of the outside and a rare halls and caverns. That being said the Necropolis would be a test of strategic thinking for turn-based FPS and Ziggurat Vertigo is unplayable.
Quake 2 was the one with the clever approximate inverse square root code, right? I wonder (especially since there’s an instruction nowadays to draw inspiration from), can you implement it “in hardware,” so to speak?
In x86 that would be the SIMD rsqrtps [0]. So absolutely possible to do at the hardware level.
[0] https://www.felixcloutier.com/x86/rsqrtps
No that was Q3 Arena.
The diagonal traces and the empty spaces are throwing me for a loop. Is this the autorouter in action? (But… still, nice work.)
DDR3 traces need to be length matched, because at 800MHz (the slowest "standard" rate, though I think you can drop to 666MHz safely) the value on the pins is changing every 1.25ns, and having traces of different lengths means you probably won't see the right values on all the pins at the same moment. Length matching produces the squiggles.
The diagonal orientation of the DDR3 chip and corresponding diagonal traces I suspect is a choice made by the author to ease the layout process - it's more likely that is hand laid out to get traces of somewhat similar length with a minimum of fuss, followed by a length matching tool. A non-standard orientation can cause issues with pick-and-place machinery, which usually will handle 90 degrees fine, and _often_ 45 degrees fine, but (AFAIK) _rarely_ anything else, but that's not a problem for the author because he's assembling it himself. A diagonal IC also usually results in wasted space, which you can see in the empty areas of the resulting board. A 90 degree orientation may have allowed for a few more decoupling capacitor banks, but since his board works, who am I to sit here and judge?
Pick and place machines can place components very precisely at any angle - they really don’t care!
> A non-standard orientation can cause issues with pick-and-place machinery, which usually will handle 90 degrees fine, and _often_ 45 degrees fine
This sounds like nonsense. Pick and place machines don't pick up components perfectly deterministically. There is always a tilt and an offset when you are picking the part up, which is why a computer vision system has to account for part orientation and the center of the part. The machine must compensate the error by moving and rotating the part accordingly.
Good to see I'm not the only weirdo still using Midnight Commander.
hahaha earlier today I needed to move some files on my server.
Used Midnight Commander within an SSH client…. on my iPhone lol
Being able to click in Midnight Commander makes this surprisingly usable for quick jobs!
Why would you be a weirdo? Two-panel is the only somewhat sane way to navigate the file system, though I still try to Ctrl+\ and need to remember to Alt+S and only then to type the search pattern.
Quake II had the best fn soundtrack.
Yes, check out Radioactive Man (from Two Lone Swordsmen) - Trespasser:
https://youtu.be/Zdy9TtInX-c
Lots of Quake II samples.
Sonic Mayhem!
I have the album on my phone. When I get called in to put out a fire and save the day, I like to put on March Of The Stroggs in the car when arriving at the destination. It's a great soundtrack for two reasons - the first one is sweet, wasted youth and the second is it's a great soundtrack.
And Jer Sypult who made Climb (track 10) who is not in Sonic Mayhem!
Hey, routing your own length-matched traces, nice. Is this Altium?
Even modern Kicad does auto length matched tracks now.
But I'm lazy, and just used the Zynq 7020 (Dual ARM A9 + FPGA) MYIR Z-turn Board V2 for hobbies.
Best regards, =3
Very impressive project!
With Claude, a software engineer can now be a hardware engineer.
Let's see Claude's buck converter layout, BGA fanout, and routing.
Another board has become Frag complete. Important milestone!