I'd be more interested on the savings in size, power and TDP.

A light, fanless laptop with a battery that lasts for days would be possible.

Zup wrote:

I'd be more interested on the savings in size, power and TDP.

A light, fanless laptop with a battery that lasts for days would be possible.

But with very limited capabilities in terms of modern computers.

A core 2 would be a better idea marketing wise. Think of a core2quad on 10 or 14nm for a netbook or tablet.

candle_86 wrote:

I was just thinking about this, what if say Intel re released say the Pentium Classic but at 14nm or AMD re released the K6-III @ 14nm. What kind of preformance and clock speeds could these old dogs reach if given just modern processes and nothing else.

While they're at it, Intel can get with gerwin and add support for setmul to change multipliers on the fly. Then we can finally have the perfect DOS machine: Pentium 500MMX with smooth speed steps from 500 down to 66, combined with ability to disable L1 and motherboard L2. Oh, and throw in those test register settings too. Thanks, Intel. Send me one when it's ready, along with a compatible motherboard. 😉

the vortex chip is a 486 done with newer production process;

mrau wrote:

the vortex chip is a 486 done with newer production process;

Tell us more

frankly i cant, im just bragging here :d
heres the info http://www.dmp.com.tw/tech/vortex86dx/

Frankly, it's one of the few chips on the market which could be used for a new build retro motherboard.

clueless1 wrote:

While they're at it, Intel can get with gerwin and add support for setmul to change multipliers on the fly.

Good idea! I am currently negotiating with these guys at intel. Technically it is now covered, but I am afraid they have to use one existing fab, so their i3/i5/i7 production will have to be reduced a little. 😉 It will be called Super-Tualatin+ and will run on Super Slot 1 boards, though it will be backwards compatible with Slot 1.

Seriously though,

Knowing that "The dynamic power consumed by a CPU is approximately proportional to the CPU frequency, and to the square of the CPU voltage".
Modern Processors run at just below 1 volt.
Super Socket 7: There is an embedded K6-III+ here running on just 1,7 Volts, which is the lowest the motherboard allows.
Slot 1: Coppermine and Tualatin both run reliable at 1,3 Volts (at conservative speeds). VIA C3 is in that same voltage range, it was produced up to 2006.

So, as far as power usage is concerned, these SS7 and Slot 1 processors are not that far from that 1 Volt. That is pretty neat, and quite a difference from 3 and 5 Volt processors. For DOS systems the processing power of the mentioned processors is fine. For Windows 9X some more speed may be desirable. But for Windows XP it does not really matter since you can run that OS on Sandy/Ivy Bridge systems.

someone said it had bugs that made it incompatible, so i dont know; but i sure would like to have soft o/c-ing 300mhz 486 :> also i have no idea how compatibe and fast the gpu is there;

The easiest solution perhaps would be to make some kind of new "Overdrive" chip, soldered to some voltage adapter.
Maybe even multiplier software controlled, which would make it more insanely usable 😁

The 'good' thing is that even an old obsolete fab could be used for these chips.

Heck, I'd love a Cyrix MII "Overdrive" chip on ss7 and lets not even get started about their 5x86 design 😁

In terms of clockspeed, they will probably be limited.
This is because transistors have more or less 'fixed' transition times. They are not affected by a smaller process. The trick is to design an architecture that is a good balance between the amount of transistors you can put on a chip, and the amount of IPC you can extract from them.
Namely, with modern CPUs, you pipeline operations in a number of different stages. If you can put more transistors into a chip, you can cut up the pipeline into more states, where each stage will have a shorter chain of transistors, resulting in a shorter processing time (the transition time is the same, but there are less transistors in a dependent chain, so the accumulated/worst case transition time of the entire stage is shorter).

This means there is more or less a 'hard limit' of clockspeed for every CPU architecture. And the older the CPU architecture is, the lower that clockspeed will be. Depending on how well the designers gauged the expected performance of the manufacturing process(es) used for the CPU, the original products may already be reasonably close to the limit (Pentium 4 is an excellent example of an overly ambitious CPU, where the architecture was designed for 5+ GHz, but manufacturing problems meant that these speeds never materialized... although with supercooling, overclockers have pushed them past 8 GHz).

The more interesting approach here would be massive parallelism: A 486 or Pentium core is very small by today's standards. You could put hundreds of them on a single chip with today's technology.
In fact, that is more or less what Intel thought with their Larrabee GPGPU architecture: They took a Pentium-ish core as the basis, with a reasonably short pipeline, and added very wide SIMD units to that core. Then they could put a large number of them on a single chip, to get a massively parallel processor, not too dissimilar from products from AMD and nVidia (as you may know, GPUs tend to also run at much lower clockspeeds than CPUs, because they also prefer many simple cores with relatively short pipelines over a few 'deep' pipelines like in CPUs).

candle_86 wrote:

I was just thinking about this, what if say Intel re released say the Pentium Classic but at 14nm or AMD re released the K6-III @ 14nm. What kind of preformance and clock speeds could these old dogs reach if given just modern processes and nothing else.

It wouldn't work - the 14nm dieshrink of Pentium or K6 would be so small it couldn't even fit the 64bit FSB bus. (not enough space for all the needed pads) 🤣

Also - what would be the point of 14nm Core2.... when we have 14nm Skylake whis is much better in everything?

The frequency design limits of any particular CPU architecture aren't really much of an issue here anyway, as we're talking about retro hardware which probably wouldn't even be able to fully make use of much higher frequencies (what good would for instance a Celeron 6666MHz do when it's on a 66MHz FSB?). The real benefits also lay in other possibilities that were (for whatever reason) not available back when chips for certain platforms were still being made.

I think this idea is quite interesting one to debate about, but I don't expect to actually see such a thing being released on the market any time soon (if at all).

Tetrium wrote:

The frequency design limits of any particular CPU architecture aren't really much of an issue here anyway, as we're talking about retro hardware which probably wouldn't even be able to fully make use of much higher frequencies (what good would for instance a Celeron 6666MHz do when it's on a 66MHz FSB?). The real benefits also lay in other possibilities that were (for whatever reason) not available back when chips for certain platforms were still being made.

I think this idea is quite interesting one to debate about, but I don't expect to actually see such a thing being released on the market any time soon (if at all).

How about 256 core VSA-100 chip? Voodoo5 5500 has two, Voodoo5 6000 has four. Now I'm talking 256 in a single die, available in AGP or PCI version, with a bridge chip to make it appear as Voodoo1 for hardcoded 3dfx games. The card is entirely controllable using environment variables, so you can use it with DOS GLide games like Archimedean Dynasty. And of course, you can activate 64x FSAA for such games.

Of course, the market segment would be severely limited, but were I a billionaire, I would definitely buy 3dfx IP's from nVidia and make the card for personal use.

What would you do for RAM on a 256 core 3dfx cpu? Voodoo 5 had 32MB of RAM for each CPU, so 256 x 32mb = 8GB and associated bandwidth from RAM to GPU.

For a CPU done on new processes you have the issues of wait states to the BUS , cache issues, voltage issues, and finally software that just dies from hard coded timing loops.

Unknown_K wrote:

What would you do for RAM on a 256 core 3dfx cpu? Voodoo 5 had 32MB of RAM for each CPU, so 256 x 32mb = 8GB and associated bandwidth from RAM to GPU.

Well, okay, the number is exaggerating, but you got the picture. You see, core miniaturization makes parallelism easier. On the other hand, 3dfx VSA-100 was designed with parallelism from the ground up, so the possibilities are interesting. Yeah, of course, no T/L, and no shader either, but you could get a massive multi-core rasterization processor if the ancient GPU is made on a new process.

Kreshna Aryaguna Nurzaman wrote:

Unknown_K wrote:

What would you do for RAM on a 256 core 3dfx cpu? Voodoo 5 had 32MB of RAM for each CPU, so 256 x 32mb = 8GB and associated bandwidth from RAM to GPU.

Well, okay, the number is exaggerating, but you got the picture. You see, core miniaturization makes parallelism easier. On the other hand, 3dfx VSA-100 was designed with parallelism from the ground up, so the possibilities are interesting. Yeah, of course, no T/L, and no shader either, but you could get a massive multi-core rasterization processor if the ancient GPU is made on a new process.

I don't know from top of my head how many VSA-100 chips max can work on a single board, but with the miniaturization of old designs that's possible with todays fabs, how about a 4x VSA-100 die with all the RAM on the same die as well? On-die RAM. Or at least RAM on the same package.

Kinda funny thought, a "neo-Voodoo 5 6k" the size of a PCI NIC 🤣

Dunno if it's known how to even create a new VSA-100 core though.

Tetrium wrote:

Kreshna Aryaguna Nurzaman wrote:

Unknown_K wrote:

What would you do for RAM on a 256 core 3dfx cpu? Voodoo 5 had 32MB of RAM for each CPU, so 256 x 32mb = 8GB and associated bandwidth from RAM to GPU.

Well, okay, the number is exaggerating, but you got the picture. You see, core miniaturization makes parallelism easier. On the other hand, 3dfx VSA-100 was designed with parallelism from the ground up, so the possibilities are interesting. Yeah, of course, no T/L, and no shader either, but you could get a massive multi-core rasterization processor if the ancient GPU is made on a new process.

I don't know from top of my head how many VSA-100 chips max can work on a single board, but with the miniaturization of old designs that's possible with todays fabs, how about a 4x VSA-100 die with all the RAM on the same die as well? On-die RAM. Or at least RAM on the same package.

Kinda funny thought, a "neo-Voodoo 5 6k" the size of a PCI NIC 🤣

That would be cute, wouldn't it? 😁

Tetrium wrote:

Dunno if it's known how to even create a new VSA-100 core though.

I wonder if nVidia kept the processor design in archive, or just discard it. How hard it is to reverse engineer from existing VSA-100?

Kreshna Aryaguna Nurzaman wrote:

That would be cute, wouldn't it? :D […]
Show full quote

Tetrium wrote:

I don't know from top of my head how many VSA-100 chips max can work on a single board, but with the miniaturization of old designs that's possible with todays fabs, how about a 4x VSA-100 die with all the RAM on the same die as well? On-die RAM. Or at least RAM on the same package.

Kinda funny thought, a "neo-Voodoo 5 6k" the size of a PCI NIC 🤣

That would be cute, wouldn't it? 😁

Tetrium wrote:

Dunno if it's known how to even create a new VSA-100 core though.

I wonder if nVidia kept the processor design in archive, or just discard it. How hard it is to reverse engineer from existing VSA-100?

I'm not sure, but I'd guess they threw away the designs along with all the actual cards?

Or maybe they kept them due to patents or something?

But reverse-engineering may be possible, but I really don't know how feasible that is. Didn't Cyrix used to reverse-engineer the 486 just so they could "rebuild" it without it being a direct copy? Iirc there were still lots of VSA-100 GPU packages sold in trays a little while ago?

I really don't know what's possible or even realistic here. But it sure could've be interesting. Maybe even to NVidia but I doubt they will take the bait 😜