VOGONS


First post, by aries-mu

User metadata
Rank Oldbie
Rank
Oldbie

Hey guys,

just wanted to create a thread of appreciation for the glorious DX 50, not to be confused with the DX2 50 (specifying for possible novices).

With its 1:1 FSB:CPU clock ratio, and, potentially the bus running at 50 MHz, I was always fascinated by this CPU!

However, it was too long ago, when I was just starting with computers as a mid school kid... so I can't remember very well, but I'm almost sure the DX 50 lived in the pre-32 bit buses (VLB and PCI) era.

So, my guess is that the ISA bus had no problems with that fast FSB, as it is simply stuck at max. 16 MHz, no matter the FSB speed. Am I correct?

So, no issues of cards stability.

At that point, the arising question is: what if there were any branded DX 50 PCs with a mobo built-in good vga? How fast would the GPU have been? Downclocked?
And, ISA bus aside, I also wonder how good the CPU-CACHE-RAM performances must have been...!

I wonder if there would be any VLB or PCI mobo which slots could be set at 50 MHz stably, with good VLB or PCI SVGA and EIDE/UATA controller cards (and of course Solid State Drives that can use that speed without bottlenecking), that would also be stable at 50 MHz.... Assuming it's possible, I wonder what the overall performances of the computer would be, a clean:

50 MHz peripherals • 50 MHz FSB, RAM, Cache, etc • 50 MHz CPU
in a perfect
1:1:1 ratio

against a system with identical components except a DX2 66 CPU, and, subsequently, everything beyond the CPU clocked at 33 MHz instead of 50...

Any thoughts?

Have fun wondering and commenting! (and, hopefully, even telling possible real experiences!).

Good luck for finding a DX50 CPU still around!!!!!

PS: for the same reasons, just in reverse, I never liked too much the DX2 50, and I particularly disliked the DX4 75 and the Pentium 75 CPUs... Come on... I can understand an FSB & rest of the system running at 25 MHz during the DX2 50 times... okay that's fair enough... but, still carrying around the 25 MHz bus during the DX4 and even the 2nd generation Pentium era?????

They said therefore to him: Who are you?
Jesus said to them: The beginning, who also speak unto you

Reply 1 of 26, by dionb

User metadata
Rank l33t++
Rank
l33t++

When I met my significant other a bit over 15 years ago, she used a 486DX-50 on a VLB motherboard with VLB VGA. Unfortunately I came bearing a 486DX4-100 laptop (this was in AthlonXP era, mind you 😜 ), and we never looked back - so I can't really comment on how it ran stable or what hardware was involved. I strongly suspect the VLB did 50MHz with a lot of wait states...

Thing is about DX-50 performance that it wasn't that stellar, about on par with a DX2-66. Most demanding tasks were CPU-limited anyway, that was the whole rationale of double-clocking, and the reason that 3x and even 4x multipliers actually increased performance. The DX-50 had distinctly higher I/O and memory performance (when stable), but so long as neither was really a bottleneck, it simply wasn't efficient to keep it all in sync.

Different architectures had different bottlenecks though, so with say a Netburst (P4) system, the more memory bandwidth you threw at it, the faster it went, far more so than with competing Athlon architectures. Currently it's the other way round, Ryzen eats up bandwidth whereas Intel's Core series is far less dependent on it.

Oh, and you're not right about 25MHz bus in the Pentium era. The P75 ran at 1.5x50MHz, which was bad for performance, but still twice the 25MHz you claim. That makes it an interesting candidate to test your hypothesis on - there were early P54C-era motherboards with VLB support (some odd OPTi designs). I never had a working one, so I don't know how fast that VLB ran, but if you could set it up with a 1/1 divider combined with a P75, you could test that performance.

Reply 2 of 26, by aries-mu

User metadata
Rank Oldbie
Rank
Oldbie
dionb wrote:
When I met my significant other a bit over 15 years ago, she used a 486DX-50 on a VLB motherboard with VLB VGA. Unfortunately I […]
Show full quote

When I met my significant other a bit over 15 years ago, she used a 486DX-50 on a VLB motherboard with VLB VGA. Unfortunately I came bearing a 486DX4-100 laptop (this was in AthlonXP era, mind you 😜 ), and we never looked back - so I can't really comment on how it ran stable or what hardware was involved. I strongly suspect the VLB did 50MHz with a lot of wait states...

Thing is about DX-50 performance that it wasn't that stellar, about on par with a DX2-66. Most demanding tasks were CPU-limited anyway, that was the whole rationale of double-clocking, and the reason that 3x and even 4x multipliers actually increased performance. The DX-50 had distinctly higher I/O and memory performance (when stable), but so long as neither was really a bottleneck, it simply wasn't efficient to keep it all in sync.

Different architectures had different bottlenecks though, so with say a Netburst (P4) system, the more memory bandwidth you threw at it, the faster it went, far more so than with competing Athlon architectures. Currently it's the other way round, Ryzen eats up bandwidth whereas Intel's Core series is far less dependent on it.

Oh, and you're not right about 25MHz bus in the Pentium era. The P75 ran at 1.5x50MHz, which was bad for performance, but still twice the 25MHz you claim. That makes it an interesting candidate to test your hypothesis on - there were early P54C-era motherboards with VLB support (some odd OPTi designs). I never had a working one, so I don't know how fast that VLB ran, but if you could set it up with a 1/1 divider combined with a P75, you could test that performance.

Oh wow, this is all so interesting, thanks!
And, I missed the point of the 50x1.5 in the Pentium 75!
I don't know why, I always thought that it was a 25x3 system!!! Than it's much better! (then, my profound dislikes nor are all focused against the DX4 75 🤣).

Yeah, those would be interesting tests.

Maybe, even downclocking a Pentium 75 to 50 MHz (if possible) by setting the mobo multiplier to 1x, having a 50 MHz bus and 50 MHz Pentium CPU, and comparing against a DX50... interesting stuff!

They said therefore to him: Who are you?
Jesus said to them: The beginning, who also speak unto you

Reply 3 of 26, by Koltoroc

User metadata
Rank Member
Rank
Member
aries-mu wrote:

However, it was too long ago, when I was just starting with computers as a mid school kid... so I can't remember very well, but I'm almost sure the DX 50 lived in the pre-32 bit buses (VLB and PCI) era.

not quite. There were EISA nad MCA, but both were only about twice as fast as ISA ith 8 or 10Mhz clock speed

aries-mu wrote:

So, my guess is that the ISA bus had no problems with that fast FSB, as it is simply stuck at max. 16 MHz, no matter the FSB speed. Am I correct?

correct. The last time ISA was dependent on system clock was during the early years of the 286, but that abandoned when they got to fast.

aries-mu wrote:

So, no issues of cards stability.

card stability no. System stability on the other hand, that was a huge issue. Building 50Mhz motherboards wasn't easy back then and manufacturing issues as in much tighter tolerances were biting them in the ass hard. Cache chips, memory and other components were more expensive since they had to get the fastest possible to get the systems stable. That made them really expensive. They were mostly found in high end systems where cost was not worth considering because of that.

aries-mu wrote:

At that point, the arising question is: what if there were any branded DX 50 PCs with a mobo built-in good vga? How fast would the GPU have been? Downclocked?
And, ISA bus aside, I also wonder how good the CPU-CACHE-RAM performances must have been...!

integrated VGA at that time would have been a normal VGA chipset directly soldered to the bus, most likely ISA. There was no integrated graphics as we understand today.

aries-mu wrote:
I wonder if there would be any VLB or PCI mobo which slots could be set at 50 MHz stably, with good VLB or PCI SVGA and EIDE/UAT […]
Show full quote

I wonder if there would be any VLB or PCI mobo which slots could be set at 50 MHz stably, with good VLB or PCI SVGA and EIDE/UATA controller cards (and of course Solid State Drives that can use that speed without bottlenecking), that would also be stable at 50 MHz.... Assuming it's possible, I wonder what the overall performances of the computer would be, a clean:

50 MHz peripherals • 50 MHz FSB, RAM, Cache, etc • 50 MHz CPU
in a perfect
1:1:1 ratio

against a system with identical components except a DX2 66 CPU, and, subsequently, everything beyond the CPU clocked at 33 MHz instead of 50...

VLB Performance likely would have been somewhat disappointing. Lots of waitstates on VLB to get it stable would have eaten a lot of the performance advantage, while the pure CPU power would have been slightly beaten by a DX/2 66. The DX-50 would overall edge out the DX/2 but at the cost of stability and the need to get significantly more picky for the components you can use without running into trouble.

Faster yes, but not by enough to make the trouble setting it up worth it.

PCI runs independent of system clock so it would be 33Mhz (later revisions of the PCI spec 66Mhz, but uncommon in desktops) no matter what, but the extra bandwidth would have helped somewhat. Stability due to system clock speed would be less of an issue here but early PCI had its own stability issues independent of that so I would call it a wash on that front, however it would have been in theory easier to find compatible cards.

It would have been faster overall, but once again stability issues would make it a questionable proposition at best

aries-mu wrote:

PS: for the same reasons, just in reverse, I never liked too much the DX2 50, and I particularly disliked the DX4 75 and the Pentium 75 CPUs... Come on... I can understand an FSB & rest of the system running at 25 MHz during the DX2 50 times... okay that's fair enough... but, still carrying around the 25 MHz bus during the DX4 and even the 2nd generation Pentium era?????

as other already said, 50, 60 or 66Mhz were pentium bus speeds and the pentiums introduced multiplicators in 0.5x steps up from 1x

dionb wrote:

Different architectures had different bottlenecks though, so with say a Netburst (P4) system, the more memory bandwidth you threw at it, the faster it went, far more so than with competing Athlon architectures. Currently it's the other way round, Ryzen eats up bandwidth whereas Intel's Core series is far less dependent on it.

a small note here, Ryzen is not technically dependent on high memory bandwidth, it is dependent on high memory clock speed. The reason for that is that the infinity fabric that connects the CCXs with each other and the IO on the die is running at half the memory clock speed. It is not the memory bandwidth that actually results in most of the performance gain, it is the increase speed of the Infinity Fabric that does. The additional memory bandwidth is more of a welcome side effect.

Reply 4 of 26, by aries-mu

User metadata
Rank Oldbie
Rank
Oldbie
Koltoroc wrote:

[...] not quite [...]

Wow, great analysis, thanks! Lots of insights I had no idea about!

They said therefore to him: Who are you?
Jesus said to them: The beginning, who also speak unto you

Reply 5 of 26, by amadeus777999

User metadata
Rank Oldbie
Rank
Oldbie

I had the DX50 that I got from a friend running at 50, 60 and 66 mhz.
It's very nice but the "bandwidth" is not really being made use of due to the slow cpu. I may re/build the setup(with a GA-486AM/S) when I get the fast srams and provide some scores at 50 and 1:1:1.

Reply 6 of 26, by dionb

User metadata
Rank l33t++
Rank
l33t++
aries-mu wrote:
[...] […]
Show full quote

[...]

Oh wow, this is all so interesting, thanks!
And, I missed the point of the 50x1.5 in the Pentium 75!
I don't know why, I always thought that it was a 25x3 system!!! Than it's much better! (then, my profound dislikes nor are all focused against the DX4 75 🤣).

Better, but still not great. It hardly outperformed the original P60 in many tasks. Partly that was the slower system/cache/memory bus, but a big part was the effect of running PCI at 25MHz.

Maybe, even downclocking a Pentium 75 to 50 MHz (if possible) by setting the mobo multiplier to 1x, having a 50 MHz bus and 50 MHz Pentium CPU, and comparing against a DX50... interesting stuff!

Not going to work. All So5/7 P54C designs have a minimum of a 1.5x multiplier. The only P54C that doesn't is the PODP63/83 for use on 486 boards. Default multiplier is 2.5x, but if you remove the integrated heatsink fan, it automatically drops down to 1x for thermal protection. Never tried to run it faster than 33MHz FSB, but I imagine if you mask the detection pin (but still keep the fan running for thermal reasons!) you could run it head-to-head with the 486DX-50. The PODP would probably win hands-down.

Reply 7 of 26, by aries-mu

User metadata
Rank Oldbie
Rank
Oldbie
dionb wrote:

Better, but still not great. It hardly outperformed the original P60 in many tasks. Partly that was the slower system/cache/memory bus, but a big part was the effect of running PCI at 25MHz.

Right!!!! Now I remember, wow man you're making me dig in such old memories! That's why I have always been convinced that the P75 run at 25 MHz... it was because of the PCI! So, my rant against buses still running @ 25 MHz in the 2nd generation Pentium era still stands, just it only applies to the PCI and not to the FSB... so it's basically a PCI:FSB:CPU 1:2:3 ratio, to me it sucks.

dionb wrote:

Not going to work. All So5/7 P54C designs have a minimum of a 1.5x multiplier. The only P54C that doesn't is the PODP63/83 for use on 486 boards. Default multiplier is 2.5x, but if you remove the integrated heatsink fan, it automatically drops down to 1x for thermal protection. Never tried to run it faster than 33MHz FSB, but I imagine if you mask the detection pin (but still keep the fan running for thermal reasons!) you could run it head-to-head with the 486DX-50. The PODP would probably win hands-down.

Wow, I didn't fully get your last sentences, but for what I got they suggest a very interesting experiment. What's that detection pin thing? And how would it be "masked"?
The PODP63/83 is an interesting creature.... I wonder how fast would the VLB/PCI slots and the FSB run...

They said therefore to him: Who are you?
Jesus said to them: The beginning, who also speak unto you

Reply 8 of 26, by Garrett W

User metadata
Rank Oldbie
Rank
Oldbie

Don't have much to comment on, but I have some trivia that stayed with me throughout the years. Magic Carpet's minimum system requirements in early 1994 called for a 486 running at 50MHz, but they don't specify whether that's a DX-50 or DX2-50. I'm willing to believe it's the first, which would mean that it's essentially asking for a DX2-66, which always seemed absolutely crazy for early 1994. I'm fairly certain it runs like crap on such a system, kind of like how Doom runs on a fast 386.

Reply 9 of 26, by aries-mu

User metadata
Rank Oldbie
Rank
Oldbie
Garrett W wrote:

Don't have much to comment on, but I have some trivia that stayed with me throughout the years. Magic Carpet's minimum system requirements in early 1994 called for a 486 running at 50MHz, but they don't specify whether that's a DX-50 or DX2-50. I'm willing to believe it's the first, which would mean that it's essentially asking for a DX2-66, which always seemed absolutely crazy for early 1994. I'm fairly certain it runs like crap on such a system, kind of like how Doom runs on a fast 386.

I'd be curious to compare the game on the three systems, same configuration, just different CPU (and bus speed accordingly):
• DX2 50
• DX 50
• DX2 66

Hey, are you by any chance the Garrett Wang who was Ensign Harry Kim in Star Trek Enterprise?? 😲

They said therefore to him: Who are you?
Jesus said to them: The beginning, who also speak unto you

Reply 10 of 26, by sunaiac

User metadata
Rank Oldbie
Rank
Oldbie

Something like that ?

486.png
Filename
486.png
File size
52.19 KiB
Views
3352 views
File license
Fair use/fair dealing exception

R9 3900X/X470 Taichi/32GB 3600CL15/5700XT AE/Marantz PM7005
i7 980X/R9 290X/X-Fi titanium | FX-57/X1950XTX/Audigy 2ZS
Athlon 1000T Slot A/GeForce 3/AWE64G | K5 PR 200/ET6000/AWE32
Ppro 200 1M/Voodoo 3 2000/AWE 32 | iDX4 100/S3 864 VLB/SB16

Reply 11 of 26, by aries-mu

User metadata
Rank Oldbie
Rank
Oldbie
sunaiac wrote:

Something like that ?

486.png

Woooow man that's TERRIFIC!!! THANKS!!!!

Very very interesting....

Surprised to see how superior the DX2 66 is VS the DX50, despite the faster bus of the DX50... I wonder what the speed of the DX50's controller and ET4000 W32 were... VERY IMPORTANT DETAIL!

Surprised to see how important is and how huge the impact is of the 16 KB L1 cache VS the 8 KB one for same CPU. Maybe because of the BUS:CPU imbalance of the DX4 CPUs!

Surprised to see how faster the DX4 120 is VS the Pentium Overdrive 83 !!! Except... in Quake! Maybe (relatively) modern game had started to be optimized for Pentium instructions...

Would have been also interesting to see how a DX2 80 and a DX2 80 with 16KB L1 cache (if existed) would have behaved under those different settings.... And also, all those CPUs with cache in Write Back and same everything else as shown....

Wonder why the huge impact of Himem on Quake though....
Wow, there's so much to say about that table... so inspiring and interesting!!!

Thanks man!

They said therefore to him: Who are you?
Jesus said to them: The beginning, who also speak unto you

Reply 12 of 26, by nforce4max

User metadata
Rank l33t
Rank
l33t

I wonder how it performs at 60 and 66mhz should it be overclockable and the board can cope with such clocks, I do have a DX50 with a stable board but have yet to build a system with it due to being short on AT cases and supplies.

On a far away planet reading your posts in the year 10,191.

Reply 13 of 26, by sunaiac

User metadata
Rank Oldbie
Rank
Oldbie

Aries-mu, it is important with a limited set of data to not jump to conclusions too quickly. Else you can prove anything you want 😀

The difference between the two dx4s here is not only 8k vs 16k, but also intel Vs AMD. At same cache quantity, intel's is already quite faster than amd's.

As for the pentium OD, this mainly gives an idea of its performance in a motherboard that does not support it well.

You can refer to the table in the other thread where I used more CPUs and a mb8433 to validate that.

Basically this table is more representative of the DX and dx2 performances.

And yes, the dx2 66 is much faster than a dx50. I have no idea why ppl keep repeating that urban legend saying they're more or less equal. See how I had to lower timings for the 50mhz bus. And that's with a very fast and stable vlb video card.

R9 3900X/X470 Taichi/32GB 3600CL15/5700XT AE/Marantz PM7005
i7 980X/R9 290X/X-Fi titanium | FX-57/X1950XTX/Audigy 2ZS
Athlon 1000T Slot A/GeForce 3/AWE64G | K5 PR 200/ET6000/AWE32
Ppro 200 1M/Voodoo 3 2000/AWE 32 | iDX4 100/S3 864 VLB/SB16

Reply 14 of 26, by Garrett W

User metadata
Rank Oldbie
Rank
Oldbie
aries-mu wrote:
Garrett W wrote:
I'd be curious to compare the game on the three systems, same configuration, just different CPU (and bus speed accordingly): • D […]
Show full quote

I'd be curious to compare the game on the three systems, same configuration, just different CPU (and bus speed accordingly):
• DX2 50
• DX 50
• DX2 66

Hey, are you by any chance the Garrett Wang who was Ensign Harry Kim in Star Trek Enterprise?? 😲

No, not really.

Magic Carpet 2 lists the DX2/66 as a minimum and it's not that different to the first game, so I'm willing to bet DX50 and DX2/66 are very similar for these games. The original can certainly bring a fast Pentium or Pentium MMX to its knees when run at 640x480. There was also this odd option to enable some sort of blurring effect which kills performance.

Reply 15 of 26, by aries-mu

User metadata
Rank Oldbie
Rank
Oldbie
sunaiac wrote:

Aries-mu, it is important with a limited set of data to not jump to conclusions too quickly. Else you can prove anything you want 😀

The difference between the two dx4s here is not only 8k vs 16k, but also intel Vs AMD. At same cache quantity, intel's is already quite faster than amd's.

Right!!! I didn't notice the AMD/Intel difference!!! Thanks.
And, I didn't know AMD was slower than Intel per se (like, even if let's say cache is identical). That's too bad. Because Intel never made DX 40s, DX2-80s, and DX4-120s, which I like a lot. Actually, if we exclude overclocked CPUs and various 5x86 CPUs, the DX4-120 is the fastest pure 486, officially "486" labeled, ever released. If then it has to loose a little bit of performance because it's AMD, then I'd totally have wanted an Intel DX4-120 released!

sunaiac wrote:

You can refer to the table in the other thread where I used more CPUs and a mb8433 to validate that.

What thread? I'll search among the threads you created... just in case I won't find it, the link would be appreciated, thanks.
EDIT: I found it, thanks!
Wow man! S3 Trio and DX4s at 133 MHz and Quake is still stuck around 10 complaining??

sunaiac wrote:

And yes, the dx2 66 is much faster than a dx50. I have no idea why ppl keep repeating that urban legend saying they're more or less equal. See how I had to lower timings for the 50mhz bus. And that's with a very fast and stable vlb video card.

Yeah this sounds very surprising for me too.
What's the very fast and stable VLB video card? Are you talking about the Tseng ET4000 W32?

I'm surprised because we have +17 MHz on many things (FSB, RAM, Cache, Bus, cards, etc.) vs -16 MHz only on the CPU, and still the difference of performances is huge! Is that timing thing so important?

They said therefore to him: Who are you?
Jesus said to them: The beginning, who also speak unto you

Reply 16 of 26, by sunaiac

User metadata
Rank Oldbie
Rank
Oldbie

The speed difference between 486s from AMD and intel appears only at dx4 level.
The dx2 66s are pretty much equal.
AMD's dx4 is probably just a faster version of their dx2 while intel seems to have made some improvements.

As for the bus speed, remember results I show are video games. They should be less dependent on bus speed than more serious applications. Also, when you have 1 cycle of wait at 50mhz, that can be as bad as running at 25mhz without waiting between memory accesses.

The et4000 w32 are amongst the fastest vlb cards for dos gaming.

Now don't get me wrong, the dx50 is an extremely cool CPU, in my top 3 486s with the intel dx4 writeback and the 40mhz UMC super40 😀
But mundane CPUs like the dx2 66 and the AMD dx5 133 are faster.

You are safe with your dx4 120 choice. It is the fastest non overclocked non 5x86 CPU if not run on a 27mhz PCI bus 😁

R9 3900X/X470 Taichi/32GB 3600CL15/5700XT AE/Marantz PM7005
i7 980X/R9 290X/X-Fi titanium | FX-57/X1950XTX/Audigy 2ZS
Athlon 1000T Slot A/GeForce 3/AWE64G | K5 PR 200/ET6000/AWE32
Ppro 200 1M/Voodoo 3 2000/AWE 32 | iDX4 100/S3 864 VLB/SB16

Reply 17 of 26, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

If I recall correctly, when Intel moved to the DX4 they didn't just triple the clock, they also made some improvements to the ALU and possibly added a few Pentium instructions. AMD on the other hand simply added clock tripling and was done with it. They made up for this shortcoming somewhat by making their 8kb cache writeback. Intel also had a writeback version (with 16kb), but it was relatively uncommon outside of OEM machines.
I also seem to recall from what others on these forums have reported, that the IntelDX4 at 100 performs about the same as the am5x86 at 133. Of course the 5x86 can be easily overclocked to 160, whereas only some iDX4s can make it to 120.

The Intel 486DX-50 is indeed an interesting animal. (For whatever reason both AMD and Cyrix had DX-50s too). These chips were most commonly found in EISA systems, as EISA is not a local bus and not prone to stability issues like VLB boards. I think the DX-50 predated VLB by 6 months to a year. From what I understand most motherboards do not fully support the DX50, because they have to add plenty of wait states to the RAM and cache to make it stable. I think only Intel's EISA chipset fully supported it. I'm not sure if 50ns DRAM was required, but I believe it was, in which case upgrading would have been a pain.

On the motherboards that I own which use 60 DRAM and 15ns cache (12ns TAG), the memory performance with all the wait states added is similar to that of a well tuned DX-33 system, which in my opinion kind of defeats the purpose of a 50MHz bus. There were however VGA cards that could work at 0ws VLB, and were "hella fast". The cards which were just dumb frame buffers tended to be the ones that worked best. The ET4000W32P is one of the better ones, but think it tends to be more reliable if you have 45ns video DRAM. Cards which had issues at 50MHz tended to be the more expensive ones with VRAM and accelerated graphics. In my opinion, the far greater issue was the hard drive controller. Overclocking tended to result in data corruption of the drive, meaning you'd lose everything.

BTW, the am5x86 and cx5x86 and DX4 chips all work well at 50MHz with reduced mupliers. These chips are actually more fun to play with than the 486DX on a 50MHz system bus.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 18 of 26, by aries-mu

User metadata
Rank Oldbie
Rank
Oldbie
sunaiac wrote:

40mhz UMC super40

The heck is that????? 😕 😲 🤣

sunaiac wrote:

The et4000 w32 are amongst the fastest vlb cards for dos gaming.

Man, since you seem to be a living DOS-era encyclopedia, I gotta ask you THIS question!

I have this 486 with mobo-built-in ET4000W32 on a "local bus" (I'm quoting the manufacturer's specs). It's a Compaq Prolinea 4/66, an ISA-only slots system (not to be confused with the later Prolinea 466 without slash, which had a PCI slot).

Now, there is this tale since the 90s that this mobo built-in Compaq "local bus" is way faster than the ISA, but I found no certain data/facts around, nor anybody seems to be able to quantify how faster it would be.

1) So, first of all, what would your educated gut-feeling/guess be about it?

2) Second, I designed a thought-experiment to be able to measure that: run any kind of benchmark under both DOS and Win 3.11 with the built int W32. Then, disable via jumper the built in VGA, install an ISA W32, run all the same benchmarks, and compare the results. What do you think about this experiment? Would that allow me to quantify the difference? Or would the benchmarks not be "heavy" enough to "saturate" the ISA performance threshold, thus giving me similar results EVEN IF the built-in local bus would really be much faster than the ISA, just because it's not being pushed hard enough?

3) In the number 2, what array of benchmarks for DOS 6.22 and Win3.11 would you recommend?

4) Finally, have you got some lottery winning numbers for the next draws? Looks like the few ISA W32 available in the used computer components online market are quite expensive!

5) Any other suggestion to test it without going through the expensive purchase of the ISA W32 ? Like, just run certain benchmarks on my machine and then compare with public-existing results of some sort??? Or something??

Thanks so much!!!

They said therefore to him: Who are you?
Jesus said to them: The beginning, who also speak unto you

Reply 19 of 26, by aries-mu

User metadata
Rank Oldbie
Rank
Oldbie
Anonymous Coward wrote:
If I recall correctly, when Intel moved to the DX4 they didn't just triple the clock, they also made some improvements to the AL […]
Show full quote

If I recall correctly, when Intel moved to the DX4 they didn't just triple the clock, they also made some improvements to the ALU and possibly added a few Pentium instructions. AMD on the other hand simply added clock tripling and was done with it. They made up for this shortcoming somewhat by making their 8kb cache writeback. Intel also had a writeback version (with 16kb), but it was relatively uncommon outside of OEM machines.
I also seem to recall from what others on these forums have reported, that the IntelDX4 at 100 performs about the same as the am5x86 at 133. Of course the 5x86 can be easily overclocked to 160, whereas only some iDX4s can make it to 120.

The Intel 486DX-50 is indeed an interesting animal. (For whatever reason both AMD and Cyrix had DX-50s too). These chips were most commonly found in EISA systems, as EISA is not a local bus and not prone to stability issues like VLB boards. I think the DX-50 predated VLB by 6 months to a year. From what I understand most motherboards do not fully support the DX50, because they have to add plenty of wait states to the RAM and cache to make it stable. I think only Intel's EISA chipset fully supported it. I'm not sure if 50ns DRAM was required, but I believe it was, in which case upgrading would have been a pain.

On the motherboards that I own which use 60 DRAM and 15ns cache (12ns TAG), the memory performance with all the wait states added is similar to that of a well tuned DX-33 system, which in my opinion kind of defeats the purpose of a 50MHz bus. There were however VGA cards that could work at 0ws VLB, and were "hella fast". The cards which were just dumb frame buffers tended to be the ones that worked best. The ET4000W32P is one of the better ones, but think it tends to be more reliable if you have 45ns video DRAM. Cards which had issues at 50MHz tended to be the more expensive ones with VRAM and accelerated graphics. In my opinion, the far greater issue was the hard drive controller. Overclocking tended to result in data corruption of the drive, meaning you'd lose everything.

BTW, the am5x86 and cx5x86 and DX4 chips all work well at 50MHz with reduced mupliers. These chips are actually more fun to play with than the 486DX on a 50MHz system bus.

Very interesting points you bring up, thanks man!

Then, at this point, I think the mystical DX50 machine would have EISA slots, an ultra rare EISA accelerated SVGA good under both DOS and Win 3.xx, a good EISA EIDE controller (I think it doesn't even exist), better if caching, otherwise an EISA UWSCSI controller, with ACARD UWSCSI-to-IDE converter and then an ultra fast UDMA CF Card, 50ns RAM, 12 ns CACHE, and zero wait states whatsoever.
Possibly, with the EISA bus forced to run at around 12-16 MHz rather than the crappy standard 8 MHz which, being a 32 bit bus, would be not bad at all.

What about that eh?

They said therefore to him: Who are you?
Jesus said to them: The beginning, who also speak unto you