WDStudios wrote on 2021-06-19, 07:06:

A small, low-cost, low-power computer, like the Raspberry Pi, but 100% compatible with Windows 98 and heavily optimized for parallel floating-point operations.

WDStudios wrote on 2021-06-22, 01:51:

The target of this project would be Win98 compatibility.

WDStudios wrote on 2021-06-22, 01:51:

The goal is out-of-the-box compatibility with the largest possible number of DOs and Win9x games, at a reasonable price.

The goalposts seem to move a bit. Let's simplify: You're proposing a Windows 98 computer. I have 2 or 3 of them already. The "parallel floating-point thing", the DOS gaming (beyond what's already in Win98), that's all pie-in-the-sky stuff which there's little point discussing while there's no detail to be had.

Shreddoc wrote on 2021-06-21, 22:31:

And the crux of the issue here is that the Stated Goal, and the Proposed Way Of Achieving It, are not currently aligned. That is, we enthusiasts know that a Windows 98 system based on ~Socket A era hardware can only cover a limited range of what we, a community of experts in the subject, consider to be "retro PC gaming".

WDStudios wrote on 2021-06-19, 07:06:

Limited, but not significantly so. I've already acknowledged that there will be the occasional games, like the Win95 ports of C&C and Red Alert, that were just programmed stupidly or based on certain failed predictions of what future hardware and software would support. And there will always be the occasional person who insists that "it's not retro gaming unless it supports Glide and Adlib", but that person does not speak for everyone. I think it would be a mistake to assume that self-proclaimed "experts" at Vogons are representative of the average person who just wonders why their copy of Doom 3 won't run on their Windows 10 computer.

If a person wants to buy something on which to run Doom 3, they have several capable options. A Windows 98 PC being among the least of them. Nonetheless your proposal to bring a new Windows 98 PC to the market is interesting. But it's still unclear how you'd be adding value over-and-above what we already have.

WDStudios wrote on 2021-06-21, 19:30:

That is correct; a product like this would primarily find use in applications that demand massively parallel floating-point operations, but where spending $10,000 on a computer with Geforce 3080s in Quad SLI would be severe overkill. Something along the lines of "teaching kids in third-world countries how to use CAD software", for example. Different embedded systems are going to find different niches depending on their capabilities.

I think you are underestimating how much silicon you would need to reach 4 x 3080 FP performance. (and keep them feed with data)

It would be almost as much silicon as 4 x 3080 chips, which is going to be hella expensive. 3080 dies are already enormously large (> 600 mm^2). 4 x 3080 GPUs is near 26 square cm of silicone die.

This could never be done the sane side of $10,000.

Now maybe this is not what you intended, but exactly what is your performance target? In any case, you are really going to have to nail what you want exactly, what you leave behind, if you are going to have any chance of making this.

I don't know much about this topic but it sounds like it may be easier to convince AMD/Intel to touch up the old designs to work in whatever FAB is most suitable and just produce those again.
Similarly with the auxiliary chips required to run these.
Then, you'd have like an "Intel Retro Platform" with, say... A 486, A Pentium 133 and an MMX233 - all installed on the same board... Corresponding auxiliary chipsets also installed for each CPU... User will have a slider they use to switch which CPU boots...

I imagine any game that needs more than a socket-7 will be written with more modern code that runs well on P2/P3/i7-1165g7...

That's how I imagine you save development from scratch... I mean the CPU design is there and just needs to be adjusted so it's manufacturable (I imagine the FABs for P55C are long ago dismantled).
The southbridge etc for those has been developed already, too.
It all needs to be plopped on a motherboard, which is the easier part... And you don't design this motherboard to have all chips working together, just a simple switch to get the chosen set working. Maybe with some common components, this isn't that expensive of a motherboard.

I mean what sounds even easier is if you just pick up these games and rewrite them to work on modern hardware. Some devs might give you the code and then it's relatively trivial to port to a modern PC. Some might not give you the code cause everyone involved and their computers have by now returned to mother nature... But most of these games are elementary to program. The beauty in those games is the game design and nostalgia - which you already have.
Some may object the use of their trademark/intellectual property - you just make small modifications and you are "original" in the eyes of the law

Or invest this time and money in fixing emulators. I struggle to imagine something that a 486 can do that a 64-bit, 10 core CPU can't do, even with a virtual machine in the middle.

How many million or billion$ in business does one require, in order for AMD to bother getting out of bed?

Funny thing, I lived 20 years in Israel. Such a small place, you are bound to know someone in Intel, Motorola, Tower Semi, Havana Labs, Marvell, Qualcomm, Broadcomm, ARM or the myriad of other chip designers and producers in that tiny country.

I know lots of chip people and while that doesn't make me an authority on semi-conductors, I know the VLSI field is one of the best paid and talent-starved fields in engineering right now. You are not going to get someone to stop designing a chip for Amazon's servers to make something that runs Commander Keen in 200x100 resolution or whatever it is 😀

Easier to hire some C programmers and just add functionality to the emulators...

I worked for Micro Focus, a company that (among other things) writes emulation software. We pretty much managed to emulate everything from 1976 to 2010 mainframes to run on RedHat-7 and we were not a big team at all... And yeah, some on that team are brilliant but not Design-a-chip-from-scratch-brilliant!

Shreddoc wrote on 2021-06-22, 02:54:

The goalposts seem to move a bit.

No, they're exactly where they were when the conversation started.

Shreddoc wrote on 2021-06-22, 02:54:

If a person wants to buy something on which to run Doom 3, they have several capable options. A Windows 98 PC being among the least of them. Nonetheless your proposal to bring a new Windows 98 PC to the market is interesting. But it's still unclear how you'd be adding value over-and-above what we already have.

If you wanted to build a high-performance Win98 machine, like literally the best hardware that Win98 can run on, it'll cost quite a bit. As I mentioned earlier, a Radeon 9800 XT alone will cost at least $150 on eBay. And It'll suck up as much space and electricity as a much more powerful modern machine. I'm talking about something that delivers comparable performance to the Win98 machine, but which costs much less and takes up much less space and power.

canthearu wrote on 2021-06-22, 02:57:

I think you are underestimating how much silicon you would need to reach 4 x 3080 FP performance. (and keep them feed with data)

That's not the performance target.

canthearu wrote on 2021-06-22, 02:57:

Now maybe this is not what you intended, but exactly what is your performance target?

Radeon 9800 XT if that's the best we can get. A few times that, if the 9800 core can be scaled up easily enough without breaking driver compatibility.

DraxDomax wrote on 2021-06-22, 02:59:

Then, you'd have like an "Intel Retro Platform" with, say... A 486, A Pentium 133 and an MMX233 - all installed on the same board... Corresponding auxiliary chipsets also installed for each CPU... User will have a slider they use to switch which CPU boots...

I don't know much about electrical engineering but I'm pretty sure that would be an engineering nightmare. Among other issues, each of those Northbridges would have a different kind of memory controller, so you'd be wasting tons of board space on not just 3 different CPU sockets and chipsets, but 3 times the RAM slots... and if the Southbridges can't share the IDE controller, you'd need 3 different IDE controllers and therefore 3 hard drives...

How about just a single CPU with an unlocked multiplier than can scale all the way down to 200 mhz? Or 100 mhz?

DraxDomax wrote on 2021-06-22, 03:16:

You are not going to get someone to stop designing a chip for Amazon's servers to make something that runs Commander Keen in 200x100 resolution or whatever it is

I'm not talking about a new chip design. I'm talking about, AT THE MOST, adapting existing old designs to newer fabrication processes. And even that might not be the direction in which this would go. It could be modern, already-existing, 100% off-the-shelf hardware with newly written Win98 drivers.

DraxDomax wrote on 2021-06-22, 03:16:

Easier to hire some C programmers and just add functionality to the emulators...

If a programmer were to be hired, it would be to write Win98 drivers, not dick around with emulation.

keep us posted on your negotiations with AMD!

Good luck getting any original source for the drivers except for Linux source. You could always port the Linux drivers.

Thing is, the newer hardware is not even going to necessarily work in Windows 98... possibly not even with the R.loew patches for such things as RAM and VRAM limitations.

The old Direct3d , Glide, etc. support is not in hardware so you are going to have to do some sort of emulation either on the video card or the CPU anyway.

You'll also need drivers for the motherboard devices such as SATA, the chipset, networking, etc.

This will get very expensive very fast unless you do it yourself.

It really just makes more sense to have wrappers and an emulator / VM that can interface with the wrappers.... Or just have the wrappers be part of the emulator / VM.

This type of setup would also eliminate the need for special hardware to slow down enough for speed sensitive games as it could all be controlled through software.

I think indeed the discussion may have betrayed the OP, by going into refuting the idea technically (which the low hanging fruit but not what he was asking) and not answering the question.
I will respect the initial question with an answer: No. I won't be.

Let's face it, Doom isn't THAT great compared to what's out there right now.
A huge part of the fun is shopping for rare, decaying, bits, trying to get them to work, trying to get games to work on them, fixing, collecting stuff.
Think how crazy it is that I am specifically looking for a Voodoo1... While I already have a pretty decent 2070S on my gaming PC...

At least for the people on this forum... Maybe in the dos-box forums you'd have more success, as those would be people who actually just care about the games... Although then you'd be arguing against a free software product that honestly works VERY well.

If you churn out a product that just works, than what you are selling is purely the games it can run... Well, if people wanted to play 2.5D games or PC-Speaker games, etc... Than the new hot titles coming out right now would THAT kind of conceptually and technically obsolete stuff... But they are not...

Actually, such games are being made all the time and that's your competition... Things like Pixelships Retro (sorry, that's the only new-retro game I know), which people do mainly for fun, definitely not as a full time job.

Yeah, I'd spend a year or two writing a new-retro game (I actually do that! But nothing to share as of yet)... Playing such game? Maybe a couple of hours... And I would definitely not buy a system just to play such game.

Now... If your system comes in a silly color... And you go out of business... And it's a PITA to get working... You might get actual cult following (although you don't care about this financially, at this point) 😁

Face it, you'd be doing something like this ONLY for the journey, not the end result!!! And, sorry, a company like AMD or a guy writing drivers for you are NOT going to share your passion - they'd want to make money... A LOT of money...

Learn coding, write your w98-Ryzen-9 driver and THAT will be your reward, not people paying you for them, as there will simply not be enough (IMHO)

WDStudios wrote on 2021-06-22, 04:19:

If you wanted to build a high-performance Win98 machine, like literally the best hardware that Win98 can run on, it'll cost quite a bit. As I mentioned earlier, a Radeon 9800 XT alone will cost at least $150 on eBay. And It'll suck up as much space and electricity as a much more powerful modern machine. I'm talking about something that delivers comparable performance to the Win98 machine, but which costs much less and takes up much less space and power.

Frankly I doubt your target market's big enough to achieve the economies of scale you'd need to reach those price points, or even to gain more than the most passing of interests from corporations on AMD's level, but I applaud the ambition.

x86nerds like us - an already minority market - don't really need it, because we have and build our own full Windows 98 systems, including CRTs and MIDI modules and other specialist gear which makes the retro experience premium. "Techies do their own gear", as a banal bumper sticker could and probably does say somewhere (sorry). On the other hand, non-x86nerds don't tend to be into Windows 98 or old computers. So it seems to leave your proposal in a kind of limbo whose target market I can't quite fathom.

It seems to me that it all hinges upon what value you would add, and how. "Very clever software, and the ability to pay it's writers for two years while they develop what you need, and pray that a competitor doesn't make you obsolete in the meantime", for example. That method can often produce some quite cool stuff. Less commitment on the other hand, fills many a neglected repo.

cyclone3d wrote on 2021-06-22, 04:47:

The old Direct3d , Glide, etc. support is not in hardware so you are going to have to do some sort of emulation either on the video card or the CPU anyway.

I'm pretty sure that both the Radeon 9800 XT and modern integrated graphics are both Direct3D-capable.

DraxDomax wrote on 2021-06-22, 05:33:

I think indeed the discussion may have betrayed the OP, by going into refuting the idea technically (which the low hanging fruit but not what he was asking) and not answering the question.
I will respect the initial question with an answer: No. I won't be.

Thank you very much for your feedback.

Shreddoc wrote on 2021-06-22, 05:51:

x86nerds like us - an already minority market - don't really need it, because we have and build our own full Windows 98 systems, including CRTs and MIDI modules and other specialist gear which makes the retro experience premium. "Techies do their own gear", as a banal bumper sticker could and probably does say somewhere (sorry). On the other hand, non-x86nerds don't tend to be into Windows 98 or old computers. So it seems to leave your proposal in a kind of limbo whose target market I can't quite fathom.

There are a lot of people who aren't "techies" but just want to play their favorite games from the '80s, '90s, and early 2000s without having to spend hundreds of dollars on 20-year-old hardware from eBay or hundreds of hours Googling for third-party hacks and workarounds.

But if such a product were to be successful, its success would mostly come from precisely the markets that we can't fathom. Sony did not make the Playstation with the expectation that it would be used to simulate the gravitational waves generated by black hole collisions. The makers of the Raspberry Pi had no idea that anyone would use it outside an educational setting. Retro gaming just happens to be the only market that I know for a fact exists for this product, and would be the first one that I'd try to sell it to, if I were the one to make it.

WDStudios wrote on 2021-06-22, 06:08:

Shreddoc wrote on 2021-06-22, 05:51:

x86nerds like us - an already minority market - don't really need it, because we have and build our own full Windows 98 systems, including CRTs and MIDI modules and other specialist gear which makes the retro experience premium. "Techies do their own gear", as a banal bumper sticker could and probably does say somewhere (sorry). On the other hand, non-x86nerds don't tend to be into Windows 98 or old computers. So it seems to leave your proposal in a kind of limbo whose target market I can't quite fathom.

There are a lot of people who aren't "techies" but just want to play their favorite games from the '80s, '90s, and early 2000s

What you are proposing, does not offer that capability. A significant proportion of PC games from the 80s and 90s will not natively run as intended on a late Windows 98 machine. I see you have denied the prevalence and importance of this, earlier in the topic, but regardless it stands as an important truth which many here will support. I suspect you would soon become very frustrated with the constant pushback, if you were persist in representing a Windows 98 machine as being capable of playing all games from the 80s 90s and early 2000s, because it simply isn't true, and to imply that the exceptions are trivial, is likewise not true.

So a more accurate description of the target market, in my opinion, is "x86 people who want to play their favorite software which runs on Windows 98". And it's that market which I believe is significantly smaller than the "overall retro gamers" market, much of which had nothing to do with the PC at all. You show your average retro gamer a packed Retropie setup for $50 total, or a Windows 98 PC for $100, and that's a hard sell in the broader market. In the broader gaming context the competition is immense and the options (including low price or free) are many.

If your mission is to help people play their games, I'd focus on a cloud service where everything is sorted out and you just log in and play.
That will be an emulation-based solution, BTW 😀
And a relatively straight-forward one. Hire someone to tune the existing emulators to get the absolute best experience in each game. This will be the path of least resistance. The trickier part would be how the game licenses/copyrights are handled in such a business model...
I'd happily pay for a service where I can log in and play some of the great games of my childhood, minus the headaches. Especially if it's a really rich library and I can discover old stuff.

If you insist on being a hardware company, I'd go with cloning some of the motherboards that are clearly running out. It's kinda similar to what you suggest only that you focus on one, relatively low-tech part.
CPU's and RAM are available and quite stable.
Sound cards are getting dire too but it seems that that gauntlet has been picked up already.
Just reimplement good known standards.
While this is unlikely to yield some serendipitous use case like calculating black holes... You'd have the base for crating the rest of the idea you presented here initially, if you think there's money in it

WDStudios wrote on 2021-06-22, 06:08:

cyclone3d wrote on 2021-06-22, 04:47:

The old Direct3d , Glide, etc. support is not in hardware so you are going to have to do some sort of emulation either on the video card or the CPU anyway.

I'm pretty sure that both the Radeon 9800 XT and modern integrated graphics are both Direct3D-capable.

There is stuff that isn't supported hardware-wise in the newer cards such as table fog and palletized textures.

You would need to emulate stuff that is no longer present in the current hardware.

DraxDomax wrote on 2021-06-22, 07:19:

If your mission is to help people play their games, I'd focus on a cloud service where everything is sorted out and you just log in and play.

That's an interesting idea for old games, streaming a 320x200 game could probably be lossless too. As always input lag would be an issue though, at least for action games.

Oetker wrote on 2021-06-22, 07:24:

DraxDomax wrote on 2021-06-22, 07:19:

If your mission is to help people play their games, I'd focus on a cloud service where everything is sorted out and you just log in and play.

That's an interesting idea for old games, streaming a 320x200 game could probably be lossless too. As always input lag would be an issue though, at least for action games.

I have a conjecture that a round trip time can be kept under 20ms and that would be good enough for anyone (except speedrunners) no? 😀

Oetker wrote on 2021-06-22, 07:24:

DraxDomax wrote on 2021-06-22, 07:19:

If your mission is to help people play their games, I'd focus on a cloud service where everything is sorted out and you just log in and play.

That's an interesting idea for old games, streaming a 320x200 game could probably be lossless too. As always input lag would be an issue though, at least for action games.

I had an idea kind of like this. Of course it's just be a proof of concept and just for fun, but I imagined taking a real pc (no emulation), hooking a raspberry pi to each input/output of a windows 98 computer to it and stream it.

But the cost of that and the programming required to do just a proof of concept was too important for me to bother.

Another thing that'd scare me is that win 98 just isn't secure and I'd fear some people trying to put viruses on it or trying to damage my hardware.

I remember someone doing this with an 8088 and there were tons of limits due to this.

In the meantime, this thread at least shows something : running Windows 98 games for cheap isn't easy these days.
Emulators are either too slow or just not good at that, virtual machines lack proper support and modern hardware isn't compatible anymore.

Moreover, as the years passes, we may witness the end of x86 architecture (I know it's been said countless times in the past, but ARM is becoming a real threat and other architectures could arrive too, whereas x86 seems hard to maintain).
I think it's time for a new PC emulator. Not PCem which is great for accurate emulation, not DOSBox which is great for DOS emulation, but rather something more targeted for running Windows, and running it fast.

Pardon my emulation programming unculture, but when I see we can now emulate things like the Wii or the PS2 which are far more powerful than, or comparable to a pentium 2, all of this with just a potato, I think it may be possible to do this with a PC emulator.
However unlike current pc emulators, they have the liberty of making crazy optimizations because the hardware is the same through all the consoles.

However if having diversity for DOS games and accurate pc emulation is relevant, is it necessary for running Windows ? There are some windows 98 games that are speed sensitive, but not as many as DOS games I think (and those should be old enough to run on current emulators).
What could be emulated would be a PC that doesn't exist, with a video card that doesn't exist with drivers written by the emulator's team.

I don't have the technical knowledge to write such thing, but maybe it'd give ideas to some people around here 😀

DraxDomax wrote on 2021-06-22, 08:08:

I have a conjecture that a round trip time can be kept under 20ms and that would be good enough for anyone (except speedrunners) no? 😀

It depends where the server is located. At bare minimum you have to contend with the speed of light. New York to LA is 13 milliseconds at the speed of light travelling as the bird flies, disregarding routing and processing latencies, as well as the fact the connection wouldn't be a perfectly straight line. As such, round trip (packet goes NY->LA->NY with no modification) could not possibly be faster than 26ms.

going back to the question, I'd be interested in an x86 based SBC but I'd be surprised if it was able to play everything from DOS to Doom3 with ease for reasons already stated, but I suppose it might be possible

I probably wouldn't buy it though, even if it existed as envisioned, at least not unless it was comparable to Raspberry Pi in price, which it almost certainly cannot be

people who love to play older games (not necessarily us lot, who also love to tinker with old machines) will for the most part download from steam and gog and rely on those companies to 'make it work' on their current PC

others with a little more technical interest will learn about DOSBOX and hunt down often quite simple patches for quirky windows games and then play them on their current pc

there will be games that wont run - and guess what, they will just be forgotten - all the big titles are taken care of by steam & gog, thousands of other titles will just slowly fade away, excepting a few stalwart vogons ofcourse, finding a way for the most ardent fans of those games 😀

so in short - nice idea, technically difficult to make it work, probably won't be cheap enough, not as large a market as you might think (coz steam, gog etc)

WDStudios wrote on 2021-06-21, 19:30:

Again, you're getting bogged down in the technical details. I know people in real life who can worry about those things for me. And yes, there will always be compromises, in the design of literally everything that has ever been designed in the history of designing things. The fact that I'd make different compromises than you would, does not indicate a lack of understanding.

I don't aim to be "another one of those that just contradicts you", but you really have to take into consideration the importance and the breadth of the technical side. Take this thread for example, Micro ATX 486 - what's the degree of interest?, this guy is simply trying to make a modern equivalent for a 486 motherboard, with new (or NOS) parts. Even this one is a big challenge on CPU choice, the sound system, the video chip, costs, etc.

One other thing, trying to run games from the 80s all the way to early 2000s on a single machine (without emulators) is a very hard task.
You'd have to clock down the CPU from GHz all the way down to single digit MHz (maybe even lower if the architecture is better than a 8088, which it probably is).
What about the video chip? Some games have scrolling issues with ATi cards for example, have a look here.

The unicorn you're trying to achieve here, while it sounds great on paper, it just does not make sense, both technically and financially.
I'm not referring to your level of expertise, that's besides the point here. Nobody is going to help you design such a thing without spending some fat $$$, that's the reality of things.

Unless there is some level of software emulation involved, I don't see this idea plausible, my opinion.