Modern Generic SVGA driver for Windows 3.1 | Hacker News Reader

Modern Generic SVGA driver for Windows 3.1

github.com

72 pointsuserbinator19 comments

Incidentally this is the same driver that someone else used with an RTX 5060 Ti: https://www.tomshardware.com/software/windows/enthusiast-ins...

It's good to see that the latest GPUs can still be used in "dumb framebuffer" mode, are mostly VGA-compatible, and have VESA VBE support. I suspect AMD / NVIDIA might still have some sort of DOS-based factory tooling when bringing up new GPUs for the first time. In sadder news, I've read that the latest Intel integrated GPUs no longer have a VBIOS and are UEFI-only; although it might only be a matter of time before someone vibe-codes (vibe-ports?) one based on those from an older model.

BIOS support has indeed been dropped from modern video cards. I was hoping something like https://github.com/CSMWrap/CSMWrap might work, but the README says that MS-DOS games don't work under that.

That has a very generic VBIOS which would probably work on an original IBM VGA, but the problem is that each GPU needs a specific VBIOS to switch it between VGA and hi-res modes by writing vendor-specific registers. VBIOS is effectively a driver, so given the necessary documentation or existing OSS driver source it should be possible to write a suitable one even for GPUs that are UEFI-only.

Now I’m wondering if someone could write a 3dfx Glide driver for hardware acceleration of late-period DOS games on RTX 5060 Ti.

you can already play all Glide games accelerated in emulators.

Just like you can run Windows 3.1 in emulation. But the OP and this comment thread are about running old software directly on modern hardware.

> Running Windows 3.1 in True Color Full HD

People from the time would be astonished by the hardware we have now yet bloated software globs up every ounce of performance. What a waste! </granny mode=off>

I’m sure my privacy conscious setup would STILL on average transmit > 56kbps of telemetry, despite my best efforts

Things are slow, aren't they? I feel there was a lot of less lag in old operating systems and software.

I use two editors now. VS Code as full IDE when I want to code heavily. And a homemade FLTK based editor with just basic syntax coloring for writing notes and doing quick things.

I use a 27 year old Pentium 2 laptop with Windows 98 for a hobby project. And I keep asking myself: why does this thing feel so fast? And it could be even faster if I replaced the HDD with a CF card.

I believe Win9x (and the rest of the DOS-based Windows) has lower input latency than the NT-based ones, largely due to a simpler architecture with shorter codepaths.

Here's a related article: https://news.ycombinator.com/item?id=16001407

But there's a big different qualitative UX interaction latency delta between Win 2k and XP. 2k appeared to have a background thread do all UI stuff while XP and later did not. NT 3.1, 3.5x, and 4.0 (without Active Desktop) all appeared pretty responsive too. ME, XP, Vista... felt really slow.

In terms of apparent responsiveness, Win 3.1x, NT <4, and 2k felt the fastest.

I've a Compag Armada E500, and it runs Windows 98 fairly swiftly with its PIII processor and 256MB RAM. I've also a 2009 MacBook and it runs Snow Leopard like a dream, yet with "only" 2GB of RAM. And either of these machines could do nearly anything I ask of a PC today - programming, web browsing, comms, gfx edits, even some gaming, while feeling snapper, with less shite flying in my eyes ("notifications" and their wretched noises) as I work.

Someone will explain to me the business and economic reasons, but that just flies over my caveman brain that asks "why does bashing rock feel slower?"

The problem with these comparisons is often that the old OS doesn't actually do the same thing modern software does. Smoothly rendering a GIF/mp4/webm in a chat channel will bring that Windows 98 machine to its knees. Even complex software like web browsers on these older machines do a lot less work. They were also often a lot slower, as load times for modern SSDs are closer to old RAM than to the hard drives at the time.

I can imagine that your particular workload doesn't require all those bells and whistles, and I think it's probably true that only running the bare minimum software like you would back in the day is horrifically inefficient on modern operating systems. But, at the same time, kernels don't crash as often, disks encryption is actually a thing now, file downloads are no longer expressed in kilobits per second and the much prettier screens render much smoother media for a fraction of the performance impact.

Of course there are inefficiencies that could be fixed (like how chat apps are skins around browsers now) but a lot of efficient software from back in the day cost an arm and a leg to build. In the end, the software industry found out that customers are happier to pay when you deliver new features faster than when you deliver new features later (which still run on the old hardware, though the customer may have already replaced said hardware at the time you release your feature).

With current prices for RAM and other system components, I hope companies will once again feel the pressure to build for limited hardware. Then again, when I look at the hardware developers are lugging around, I highly doubt things will change quick enough.

> “Smoothly rendering a GIF”

Animated GIF is a format that was designed for playback on late 1980s PCs with a 20 MHz 386 and VGA graphics…

If anything, this example proves the point that we’ve made the simple stuff much too complex. The GIF format hasn’t changed, but somehow getting those indexed color frames to screen on time now requires a GHz core.

GIF playback should be efficient but...

About twenty years ago I was generating long animated GIFs. They worked fine in Firefox. In Internet Explorer they started fine but became jankier as playback progressed. I realised that every time IE displayed a frame it was rereading the entire file from the beginning to get to the current frame. Which took longer and longer as the current frame advanced.

It's just so easy to squander performance without noticing.

The reason you need a GHz core is that modern GIFs stretch the file format to its limits, by doing 30 or even 60fps in extremely-coloured files with resolutions that easily beat the render resolution of 1980s PCs just in a little corner of the screen.

GIF is an awful format for its modern usage that will easily waste tens of megabytes for even a short and small file. That's why many services secretly convert GIF files and serve them as video files, or other animated files that are more efficient (such as WebP).

The difference in opinion between "the simple stuff" and "missing the bare basics" seems to come down to what year you were born and what kind of services you grew up with. I don't need 90% of what Discord has to offer me but when reading along with discussions of Discord users looking for alternative platforms, fleeing their age verification and such, I find that most Discord users will absolutely demand the features I didn't even know chat apps support.

Can you install Office 4.3 from floppies faster than installing the latest Windows updates? Not being a recent Windows users, I'm actually kinda curious. I'm sure the CD version would install faster.

> The NVIDIA GeForce 1050 GTX has some issues with windowed DOS prompts, and presumably newer NVIDIA hardware has the same problem. The issues include potential freezes when using the mode command to change the size of the prompt, and font corruption when switching to full screen (which also affects blue screens).

Someone with connections to NVIDIA support should really file a bug about this!

Reply on news.ycombinator.com