Dan Luu writes; Apple II responsiveness 30ms, modern 2016 X1 Carbon... 150ms

[ocanada_techguy]

Summary

 Reported in PCGamer by Alex Wiltshire among others https://www.pcgamer.com/the-latency-problem-why-modern-gaming-pcs-are-slower-than-an-apple-ii/ 
noticing former Google and Microsoft Engineer Dan Luu couldn't help but notice old hardware seemed much more responsive that newer, he decided to test and wrote on his website.  

 

Quotes

Quote

 Keyboards are often more powerful than entire computers from the 70s and 80s! And yet, the median keyboard today adds as much latency as the entire end-to-end pipeline as a fast machine from the 70s.

Dan Luu

 

My thoughts

 Though counterintuitive I can almost see some engineering reasons for this to be the case.  Still, nevertheless, on it's face it really is shocking and surprising. 

 

Sources

http://danluu.com/input-lag/

https://www.pcgamer.com/the-latency-problem-why-modern-gaming-pcs-are-slower-than-an-apple-ii/

 

[justpoet]

Every layer of complexity adds a layer of delay.  Chip speeds can help offset this, but only to a certain point.

The more object oriented, zero compile, library using, remote loading, etc that an app is, the worse it is for performance.  The more parallel, high bandwidth, converted, etc an IO device/port is, the worse it is for input delay.  The more layers of workarounds (such as most patches are) that are in place, the exponentially worse everything gets.  Input validation is insane these days, to account for security issues, and often didn't exist at all except in a text entry field or similar before.

 

Roughly when Moore's law took off with Ghz and quad core, each of those milestones yielded an extra layer of "let's make this the easy way, hardware performance is way more than we need" vs "let's make this performant with less overhead and less memory usage".  Then everybody did it and the whole system grinds to a halt fairly quickly with just a "few" things running, even on multi-core high clock speed systems with reasonable amounts of RAM.  Compilers today do a LOT of work to try and optimize what they compile…but it is still nowhere near as optimized and efficient as what most developers wrote in either assembly or C back when memory and clock speeds were scarce.

[HalGameGuru]

Apple II was from back in the day of interrupts. USB is polled. Plug a direct interrupt keyboard into a PC you are gonna get a snappier response

[BlueChinchillaEatingDorito]

Wasn't this always known in the PCMR space that PS2 keyboards were preferred for their zero latency? 

[S w a t s o n]

Honestly Lenovo is probably not the best choice to test keyboard latency on, they only just recently fixed the skipped input problem on many of their high end laptops for the 4th time.

[Master Disaster]

1 hour ago, HalGameGuru said:

Apple II was from back in the day of interrupts. USB is polled. Plug a direct interrupt keyboard into a PC you are gonna get a snappier response

Same can also be said of internal devices too. Back in the ISA & PCI days internal cards used IRQs and DMA to talk to the rest of the system. These days plug & play handles most of that for us.

 

As for his almost pointless exercise. He was measuring from keyboard to monitor which is a VERY skewed comparison. CRT displays are known for having very close to 0 input latency where as modern flat screens are very much not. Its like saying old cars are more responsive than modern cars because they used direct cabling to a carburettor where as modern cars have to talk to the ECU then it has to tell the engine what to do. While its technically correct it doesn't take in to account all the other benefits we get from it.

[RejZoR]

7 minutes ago, Master Disaster said:

Same can also be said of internal devices too. Back in the ISA & PCI days internal cards used IRQs and DMA to talk to the rest of the system. These days plug & play handles most of that for us.

 

As for his almost pointless exercise. He was measuring from keyboard to monitor which is a VERY skewed comparison. CRT displays are known for having very close to 0 input latency where as modern flat screens are very much not. Its like saying old cars are more responsive than modern cars because they used direct cabling to a carburettor where as modern cars have to talk to the ECU then it has to tell the engine what to do. While its technically correct it doesn't take in to account all the other benefits we get from it.

Anyone misses regular "IRQ not less or equal" BSOD's? I know I don't...

 

People need to understand that in the olden days, everything was "wired" more directly. Sure it had lower latency, but unless everything was aligned perfectly with the planets, you had bunch of really dumb issues. These days, it's all pretty much plug and play and it just works. But having that luxury, some of the layers in between sure create some latency. Burt is imo not that bad.

[jagdtigger]

27 minutes ago, Master Disaster said:

where as modern cars have to talk to the ECU then it has to tell the engine what to do

Actually most of that lag comes from the ECU computing how it can accelerate without going over its set emission target, which is so detached from reality that the ECU actually has a hard time with its low power cpu (plus the constraints of RTOS). Patch out that routine so the ECU will only use the sensors in the exhaust as a feedback and not as a control signal and it would be more close to old cars.

[Deli]

2 minutes ago, comander said:

Every time I see someone with a mid range video card worry about CPU bottlenecking (the i7 gets 2fps better than the 3600!!!!) I laugh to myself because they're probably playing over congested wifi and using a keyboard with 50ms delay. 

 

Input delay matters and it's not discussed enough. My personal take is to ps2 the keyboard and wireless the mouse (on the theory that a cord getting in the way matters) while ensuring 1000Hz polling. 

 

With that said the reason I am only low master league at sc2 is because I suck, not because of my computer. 

Don't worry. When new CPU(either from Intel or AMD) is 2 fps faster than his 10900K. He will think 10900K is bottlenecking his games.

 

So the new Corsair K100 gaming keyboard with the selling point of 4000Hz polling rate is just BS marketing gimmick? This BS is $230....

[Master Disaster]

6 minutes ago, comander said:

Every time I see someone with a mid range video card worry about CPU bottlenecking (the i7 gets 2fps better than the 3600!!!!) I laugh to myself because they're probably playing over congested wifi and using a keyboard with 50ms delay. 

 

Input delay matters and it's not discussed enough. My personal take is to ps2 the keyboard and wireless the mouse (on the theory that a cord getting in the way matters) while ensuring 1000Hz polling. 

 

With that said the reason I am only low master league at sc2 is because I suck, not because of my computer. 

Its weird because I'm ultra sensitive to input lag playing retro games on modern systems (like SMW on Snes9x, you can press the button then let it go fully before Mario leaves the ground) but on modern games, it really doesn't bother me at all.

[AndreiArgeanu]

Yeah, but machines weren't nearly as complex and as powerful back in the day. Frankly enough, the response time still isn't long enough that you can actually say that it's holding you back in any way.

[AndreiArgeanu]

3 minutes ago, Master Disaster said:

Its weird because I'm ultra sensitive to input lag playing retro games on modern systems (like SMW on Snes9x, you can press the button then let it go fully before Mario leaves the ground)

That's an issue with the emulation though as it isn't perfect, and there's probably added delay as a result. Personally I can notice the lack of responsiveness in emulation as well but in modern games I don't notice it because that lack of responsiveness isn't there to start with.

[Deli]

11 minutes ago, comander said:

Any part like that would need to be tested. There's likely multiple things to consider including the time to actuate a key switch. 

 

Diminishing returns kick in quickly. The big thing should just be avoiding very low polling rates. 

This is the quote from the article,

" And you should basically just go for one speed: 1000Hz. "Everything else that you read on keyboard packaging is mostly a gimmick," says Spinelli. 1000Hz is USB's full-speed rate, a figure set all the way back in August 1998 when USB 1.1 was released. It means that the USB controller is checking the keyboard or mouse for changes every 1ms. "Realistically, there's been no need to improve the data and report rate capabilities of HID [USB human-interface device] gear since then."

[porina]

I can't say I've ever noticed peripheral latency outside of two cases, both mice. One was a wireless mouse. It went to sleep to save power. Although it woke up quickly with movement, it wasn't quick enough, so the first time you moved it after a moment of rest, there was this slight initial lag I found annoying. The other was a corded USB mouse. I think that just had an awful sensor in it. You had to move the mouse a certain amount before it registered the movement, so that also felt laggy. I use "gaming" grade keyboard and mice now, but I find myself setting the polling rate to 250 Hz. Some games seem to show unpredictable behaviour when at higher speeds. Never got to the bottom of it, but I can't "feel" the difference between 250 Hz and 1000 Hz. We're talking about a best case improvement from 4ms to 1ms here (on average, half that), and I'm not good enough to notice it. G-sync is possibly the best thing to happen for perceived gaming latency. Even at lower frame rates, say around 50 fps, it still feels responsive, even better than say fixed 60 fps with V-sync.

 

In my previous job I found the digital communication world annoying compared to the analog days. Pass analog signal down a wire: pretty much speed of light transmission speed. Digital had you sampling, perhaps do some processing on it, encode it, send packets through some unpredictable route, decode it, do some more processing, convert back to analog. But, offsetting that, we get a bunch of new features. Hard to go back to analog now.

[BuckGup]

7 hours ago, BlueChinchillaEatingDorito said:

Wasn't this always known in the PCMR space that PS2 keyboards were preferred for their zero latency? 

I think anyone in PCMR won't have a clue what PS/2 is and think it's a console 

[Jito463]

11 minutes ago, BuckGup said:

I think anyone in PCMR won't have a clue what PS/2 is and think it's a console 

Imagine if we added AT into the mix.

[Sauron]

Yeah, no wonder...

7 hours ago, BlueChinchillaEatingDorito said:

Wasn't this always known in the PCMR space that PS2 keyboards were preferred for their zero latency? 

There's no such thing as zero latency but PS/2 does have lower latency than usb... at least in theory. In practice it's going to depend on a lot of other factors (e.g. if your keyboard does any sort of onboard processing that's likely going to dominate the latency you get from USB anyway).

[Doobeedoo]

Yeah old post. More complex modern systems definitely add up to it. Though really machines and their capabilities are not even compatible at this point. This kida reminds me like RAM total latency over gens. 

[Kisai]

11 hours ago, justpoet said:

Compilers today do a LOT of work to try and optimize what they compile…but it is still nowhere near as optimized and efficient as what most developers wrote in either assembly or C back when memory and clock speeds were scarce.

 

The problem here is that there is generally no point to coding anything in straight assembly because it makes it non-portable. It's not being written in assembly that makes it faster, it's being compiled from C++ or OBJC that makes it slow because the memory management is handled by the runtime. In C, the memory management is almost entirely manual. In Assembly, there is no allocation/deallocation at all, you just pick someplace to start writing and reading to, with the assumption you have the entire system available. It's fast, it's also completely unfriendly to having any other program resident in memory since it may arbitrarily overwrite it. Hence all the 8-bit and 16-bit consoles didn't have any kind of OS, or even a BIOS in some cases. You plugged it in and the CPU just started executing from the known starting position.

 

32-bit and later CPU's often have a BIOS that sets up some ground rules for their hardware, like the PC BIOS had the memory space above 640K to 1M reserved for hardware. If you had an EMS board, that expansion board was only accessible in 64K pages inside that 384K of hardware reserved space. If you had an XMS board (or later, just any installed memory over 1MB) it was literately located after that 384K. 

 

DOS, being the operating system on PC's up to Windows ME, provides services to everything running on top of it, which may include device drivers. Bad device drivers, lead to unstable systems. Windows NT, XP, Vista do a lot to protect the OS from bad drivers (though a bad disk or CPU driver can still explode a system) and likewise all the programs have their own memory space, so they can't arbitrarily overwrite other programs unless the program explicitly is hooked into it.

 

Nobody in their right mind would write 64-bit assembly programs. You need that portability because you may want to run that code on a x86-64 CPU, or an ARM 64-bit CPU, a GPU, or even some parallel CPU's. That's just hell to do in assembly, and takes very little to break it.

 

The reason there's so much latency involved with a modern PC, most of that can be laid at the feet of how "general purpose" a PC is. A game written in Unity will never have good latency, because games written with C# or LUA on top of a general-purpose game engine (eg Unity), which is then written to use the OS-native API's. Where as games written for the NDS or N64 were written in C directly against the hardware.

 

Where I think Dan Luu might have err'd here is that Apple II games were largely written in BASIC, and only some were written in assembly. Interpreted games were very slow, and even if your metric is input latency, the actual drawing time was not very fast, and no PC computer had fast drawing time until the Pentium.

 

Games written for the PC up until Windows 98, were never fast or highly responsive because the video hardware (CGA, EGA, VGA) never had enough memory to do page-flipping, they operated on the "update portions of the screen using rectangles" just like a game console did. 16-bit game consoles were very much ahead of PC's, when operating at similar resolutions, because the consoles weren't having to swap bitblt's from disk to video memory, when on the console they could just change a pointer and load the tile and palette from the fixed address on the game rom.

 

Which when put into context with the PS5 advertising and nVidia's RTX IO marketing, we're only NOW getting back to where we were on game consoles 30 years ago.

 

[bit]

12 hours ago, Doobeedoo said:

Yeah old post. More complex modern systems definitely add up to it. Though really machines and their capabilities are not even compatible at this point. This kida reminds me like RAM total latency over gens. 

I was gonna say "we've know this for a while" then I noticed the linked post is from 2018 so lmfao

 

Guess someone fell for the ol PC Gamer posting a link to an old story to get some more clicks onto it.