Hacker News Reader

Comment by jzer0cool | Hacker News Reader

One of my very first jobs I worked to test VGA cards. They were 4mb and 8mb. Yes you read that right, that is "m" as in megabytes. My setup was a 25 megahertz open case motherboard, a few of these. It was part of the last quality control station and I would pop each vga card into the PCI slot and boot it up.

I would basically run a macro which cycle through the resolutions from b/w, 4 colors, 16, 256, ... and up to a million or so. I think there is a name for this, but basically watching a prism of rainbow colors. At a million+ the color tones are very smooth and you don't see an outline.

I could not imagine I would be able to differentiate a billion colors from its previous factor. At that point, I would stamp it and it goes off to shipping for packaging, and to the customers.

View on HN

> They were 4mb and 8mb. Yes you read that right, that is "m" as in megabytes.

mb is short for millibit, the abbreviation for megabyte is MB (or MiB if you mean 2^20 bytes). I know it's a minor thing, but "m" and "M" does not mean the same thing, neither does "b" and "B", and they're mixed up too often.

Sorry for the off topic.

Where does mebibyte come into play?

Mebibyte comes from pedants (like myself) that say that the Mega prefix in SI means 10^6, but in computing we often take it to mean 2^20. Mebibyte is unambiguous, and if everybody used it (unlikely. I don't think either MacOs or Windows use it), then Mega would also be. Right now, when you see mega in a computing context you're not sure if it means 10^6 or 2^20, although a lot of times it doesn't even matter.

Okay, this is second hand as I don't own a mac, but a friend who does noticed that on his macbook, macOS reports file sizes with megabyte being 10^6 bytes, whereas windows follos the 2^20 convention.

He was going crazy over a file being shown as 600Mb on Windows and 629Mb on mac. When he compared the bytes values, they were the same (around 629,145,600 bytes) and that settled it.

This is true, macOS reports it using decimal sizing.

I believe the reason is that all storage devices are marketed in decimal sizes anyway.

They changed the behavior of macOS a number of years ago (2012?) because disk manufacturers advertise storage capacity in 10^x, and the operating system displaying sizes in 2^y made the disks seem smaller and (in theory) caused confusion.

My MacBook Pro's internal "1 TB" disk is actually 1,000,240,963,584 bytes -- neither 1 TB nor 1 TiB.

I remember this. The packaging showing the calculation that 1MB = 1024 bytes

Possibly interesting: https://developer.apple.com/documentation/foundation/bytecou...

> He was going crazy over a file being shown as 600Mb on Windows and 629Mb on mac. When he compared the bytes values, they were the same (around 629,145,600 bytes) and that settled it.

To be fair: It make sense to measure RAM in MiB, because it is addressed by binary address lines so you always end up with a power of two.

Files are, however, as big as somebody decided to write into them (modulo some rounding to the next sector, depending on file system implementation details). It does not make much sense to measure them in MiB. Quite often it just more confusing to calculate file sizes in MiB instead of MB. I dont even know why Microsoft stated that MiB business.

Similar with hard disks by the way. They contain as many sectors as the manufacturer was able to put on them. There is no power of two involved.

It wouldn’t make sense for macOS to use the binary prefixes for file sizes, as it uses the decimal sizing method to show file sizes.

"Mebibyte" and similar units were only coined in 1998, a few years after the era of 4mb/8mb graphics cards.

While we're being pedantic: I agree with you about SI prefixes, but bits are indivisible. A millibit doesnt exist (only possible use would be if one was referring to something like a transfer speed as "19 millibits per second", but that's kind of lacking meaning - "0.019 bit per second" or "1.14 bits per minute" would be better)

> bits are indivisible

No, they're not. Sure, in more familiar scenarios where you're measuring the size of some amount of data it's an integer, but there's nothing mathematically wrong with having a not-integer number of bits.

See: https://en.wikipedia.org/wiki/Shannon_(unit)

If you do any sort of calculations with information theory you're quite likely to end up with a non-integer number of bits, and there's nothing wrong with that.

You can absolutely have two ranges which require 7.5 bits to store each (0-180) and write them to 15 bits of storage.

Just because physical hardware is more or less restricted to integer bit quantities doesn't mean fractional bits aren't a useful concept.

Sure, I accept I was wrong, and millibits may have some niche usage. That said, the comment I was responding to was describing hardware with "4 mb" of ram. There's no confusion in that case.

Also in older texts MB often refers to MiB. Even today there are context where people nearly always mean MiB even when they write MB.

kB = KiloByte mB = MegaByte

> I could not imagine I would be able to differentiate a billion colors from its previous factor.

Darker single color gradients exhibit very strongly noticeable color stairs. Tried to generate one for a background, locked horrible. It only looked good once I randomized the gradients a bit, i.e., even with 8 bit colors, you might need dithering.

Similar effects can be noticed in relatively dark photos of scenes with gradients, e.g., a sky.

Are you sure the panel you are viewing on is in fact also 8bits/color? The only time I have experienced what you are describing was when the panel turned out to be 6bpc - at the time I was working with raw data from a 12 bit sensor.

Of course my experience is only 1 test point...

> Are you sure the panel you are viewing on is in fact also 8bits/color

Yes, if you actually compare a 10-bit and an 8-bit panel next to another, it's immediately obvious.

Here's a photo of a 6-bit, 10-bit and 8-bit gradient on a native 10-bit panel without any dithering (Dell UP2718Q): https://i.k8r.eu/VolYbQ.png

Thanks, that is very informative. Could you give a link to the source, or even the grey level range used here? I am trying to reproduce the results, but I an not having any luck unless I change my brightness/contrast to unacceptable levels.

That’s the grey level test of the official VESA DisplayHDR Test application: https://www.microsoft.com/en-us/p/displayhdr-test/9nn1gpn70n...

Running on a Dell UP2718Q, in HDR10 mode.

That said, in person can be – depending on environmental lighting conditions – much more subtle. The test here was done with blinds closed in a mostly dark room.

In bright environments 8-bit is enough, and 6-bit with dithering can be close, but in dark environments (such as when watching a movie or gaming) you'll even notice banding in 10-bit content.

A rainbow is a rather bad test-case to distinguish color stops because it can vary two of the components at the same time. Monochromatic gradients make it easier to spot the quantization.

Additionally wider color gamuts also mean that each quantization step would cover a larger absolute difference if we kept things at 8bpc.

I don’t remember VGA doing millions of colors. I thought it did 256 colors, at best.

Wikipedia agrees with that (https://en.wikipedia.org/wiki/Video_Graphics_Array), but adds “The 640×480 16-color and 320×200 256-color modes had fully redefinable palettes, with each entry selected from an 18-bit (262,144-color) gamut.”. That, I don’t remember knowing, even reading that.

So, a quarter of a million, but only 256 at a time. I think the Macintosh II, doing 256 colors out of 16.7 million, looked better, especially with a Trinitron display (https://everymac.com/monitors/apple/classic_monitors/specs/a...).

a "vga" card with 8mb of ram was probably a svga card (24 * 640 * 480 is just under 8 mega). no real reason for it to need 8mb otherwise back then.

That's less than 1 MB.

you're right, I multiplied it by 10 in my head. but even more so for OP. It wasn't a plain vga card.

4mb? 4mb! We used to dream of 4mb when I were a lad. We had 1mb for the whole computer, and we got 160k for graphics, if we were lucky!

Walking miles to school every day through rain and snow, uphill both ways!

Matrix Millenium had 4MB of ram and was considered the beastiest of beast in its era. It's still peak in my brain :)

>It's still peak in my brain :)

that's neat that someone else does that.

i'm like that with the voodoo2, for some unknown reason.

don't get me wrong, the voodoo2 was great, but i'm not sure why it stuck in my head for that long.

Actually I exaggerated a bit, Matrox was peak 2D era. And yeah 3dfx voodoo 2 still had a massive glow in my brain. It represented so much magic and power in its day. It started the 3d ~high quality FPS... (riva128 and geforce3 followed).

Same goes for some symbols like 386 dx2. Our young brain were imprinted hard.

I was watching a guy on YouTube the other day talking about a computer he was building. He made reference to how he was keeping some component simple (the memory interface, I believe), which would compromise performance, but he wasn't worried about it because the 68000 he was using in it was so fast.

My first thought was "yeah, that makes sense" but my second was "what could it possibly mean when we talk about a 68000 being 'fast' in 2021?!"

It really made me think about how an attribute like "fastness" can get endowed and then still stick 40 years later, almost as though it were describing some spiritual attribute that transcends the fact that a modern $0.25 microcontroller is many times more performant.

Think not about 4 times more hues, but about 4x the brightness range. Think about a 300 or 400 nit screen that shows you a scene with a bright sun but details in shadows still very discernible.

4mb! I remember trying driver after driver on the SimCity 3000 CD until I found one that worked with my 512K Oak VGA card.

Now cards have more RAM than I had disk.

"showmethemoney" ;)

> Now cards have more RAM than I had disk.

Now CPUs have more L2/L3 cache than I had disk!

> I think there is a name for this, but basically watching a prism of rainbow colors.

It's called seizure test, aka intermittent photic stimulation ;) https://en.wikipedia.org/wiki/Intermittent_photic_stimulatio...

Millions of colors is most likely 8bit color with 16.7 million colors. Still in use today.

7 bit color would also count as "millions of colors", but I don't remember any GPU ever offering it as a display mode.

Unless I'm missing something (dithering?), 24bit color is 16.7M. 8 bit color is 256 colors, 7 is 128 colors.

you're correct. I assume they're referring to 8bit per colour (RGB) which is 24bit colour (True Colour) 16.7M

8*3=24, 8 bit per channel is a description they use in visual effects. Recently the rest of the computing industry started using it. Yes, confusing.

Ah, okay. I'm using old terminology I guess. Appreciate the gentle pointer.

I asked the same question on starting a vfx job many years back. 8-bit, 256 colors for a feature film? Impossible, that would make it look like .gif or Windows95, right?

They must have been fancy to use PCI. I would have expected ISA of AGP.

PCI was in fact pretty standard for graphics cards, in the period between ISA and AGP.

Edit: link https://www.tomshardware.com/uk/reviews/graphics-beginners,1...

PCI graphics was standard around the time of Windows 3.11 and into Windows 95. Major difference in UI speed vs EISA graphics.

I used to work as a hardware technician in a local computer shop as a summer / weekend job back then. I installed dozens of 4mb Cirrus Logic PCI graphics cards when building PCs.

If I remember, AGP did not come out until later.

The motherboard multiple ISA and 1 PCI slot if I recall right. In fact I was testing VGA and also sound cards on the ISA slot. So checking sound and playing some mpeg video file.

I can maybe write up a short post about this if anyone would be interested more.

At home, away from work, I owned a mac which was running the power pc processor. And that was also 25 mhz. What I remember was how everything seemed integrated (e.g. not upgradable. Young me could not understand why there was no graphics card. I could only upgrade here the hard drive, or simm memory)

I hate to be a stickler for this but please say MB, not mb, when you mean megabytes. There are contexts where the distinction between m and M, or b and B, really matters!

There are contexts where the distinction between m and M, or b and B, really matters!

Is this one of them?

No, but I do see people use them inaccurately in those contexts also, and I suspect that’s partially out of habit.

> I could not imagine I would be able to differentiate a billion colors from its previous factor

Dark scenes on a OLED might look better since your eyes can adjust to low light.

> They were 4mb and 8mb

640px x 480x x 3 bytes per channel = ~1MB. So yeah, 4-8MB is about right for a VGA card.

VGA could not do truecolor; the original VGA at this resolution 640x480 could do 16 colors, i.e. 4 bits per pixel and that means 150 kB framebuffer. It could do 256 colors with lower resolution (oficially 320x200, or "mode-x" 320x240).

VGA came with 256 kB RAM onboard. Then, in the early 90's, SVGA came with 512 kB and 1 MB onboard, and it could do 800x600, 1024x768 and 1280x1024. The main RAM would be around 4 MB for the midrange computer. It took until mid-90, that you could get 4 or 8 MB SVGA, that could also do hicolor and truecolor.

25 Mhz! Wow. Is the the GPU frequency or the CPU?

CPU, likely a high end 386 or a low end 486.

CPU, 25 Mhz. This is when I learned to over clock the cpu to run at a higher speed, at risk of burning it. You simply moved a jumper on the board board and it would boot up at 33 Mhz, yahoo! My boss showed me this and I was able to test much faster, not having to wait so long during the boot sequence.

Reply on news.ycombinator.com