The Serpent Rider wrote on 2023-09-30, 12:01:

Classical DX2 is multiplier locked or at least there are no obvious ways to tackle multiplier registers, because that CPU was designed as a direct replacement for DX33 parts. Rebadged DX4 can switch between 2X and 3X multipliers.

Too bad!
Thanks

aries-mu wrote on 2023-09-30, 13:46:

Wait what? Can you run 486 PCI motherboards' FSB at 60 or 66 MHz?? WOW! Can you please mention some examples? But wasn't PCI ine […]
Show full quote

CoffeeOne wrote on 2023-09-30, 09:42:

I saw you opened tons of threads about 486, which is my main retro interest topic. My opinion about 50MHz FSB: Forget it. Unless […]
Show full quote

I saw you opened tons of threads about 486, which is my main retro interest topic.
My opinion about 50MHz FSB:
Forget it.
Unless you have special SRAMs, 50MHz means you have to run the L2 cache with 3-2-3. That means read 3-2-2-2 and write 3 clock cycles.
With 33MHz you always can use 2-1-2 timing. So in that case L2 access for 33MHz is FASTER than L2 access at 50MHz. 60-30-60 ns for 33MHz, 60-40-60 nanoseconds for 50MHz.
Additionally with VLB but also PCI at 50MHz one can have addtional problems.

So I think for the Intel DX4 33*3 setting performance is very close to 50*2 setting. Assuming that one gets the system stable at 50MHz.
But a different story is 40MHz FSB. With many boards, you are able to achieve 2-1-2 with 40MHz. That is really fast.

So about the Am5x86, when you are a bit lucky you can ran it at 40*4 = 160MHz, when this works you can of course also use 50*3 (despite you may have more stability problems).
A lot of people here on Vogons claim, that in some benchmarks the 150MHz settings beats the 160MHz setting.
I cannot confirm that, for me 160MHz was better in every benchmark like doom, quake, Wolfenstein3D and so on.

A different story is 60MHz and 66MHz FSB on a 486. But only late PCI boards can achieve that setting, and you need either an AMD 486-120 (to run it 60*2). Or for a Am5x86 that means either 180 or 200MHz, which is hard to achieve. So I have no experience with 60, 66MHz FSB.

Wait what? Can you run 486 PCI motherboards' FSB at 60 or 66 MHz?? WOW! Can you please mention some examples?
But wasn't PCI inevitably stuck at 33 MHz and not increasable at all? What about PCI cards? Even if you manage to run the PCI bus on the motherboard at 60 or 66 MHz, how can controller and vga cards designed to run at 33 MHz even work? I'm confused...
Unless you're talking about those motherboards with 66 MHz 64 Bit PCI slots, but I don't think so, because you mentioned 486 motherboards. Those, on the contrary, were later Pentium motherboards... And still, good luck finding a 64 bit 66 MHz PCI vga card (hdd controllers maybe...)

Can you please tell me what motherboards can handle well 2-1-2 cache at 40 MHz?

Also, I see many times wait states mentioned on this forum, and you seem to have made some calculations when you extracted 60-30-60 ns and similar out of the wait states numbers. Are there formulas? Is there a tutorial or something? I'd like to be able to understand the meaning of these things.
Also, how would one change those wait states you mentioned? In the BIOS setup or physically via jumpers?

Thanks a lot!

A lot of 486 mainboards can do 2-1-2 cache timing at 40MHz.
For example Asus VL/I-486SV2GX4 and Asus PVI-486SP3.
For a double banked cache configuration it is easier to achieve. Double banked cache configuration ususally means 8 cache chips (+1 tag). 15ns chips are fast enough.
My "calculation" is just stating the cycle duration multiplied with number of clock cycles. The cycle duration (EDIT!) at 25MHz (it is just the reciprocal of the clock) => 40ns. 33.33MHz => 30ns. 40MHz => 25ns. 50MHz => 20ns, 60MHz => 16.67ns, 66MHz => 15ns and so on.
So that was not wait states, I was talking about l2 cache timing.
On the Asus PVI-486SP3 it is:
Cache Burst Read Cycle - fastest option: 2-1-1-1
and Cache Write Cycle - fastest option: 2 CCLK.
Other Bioses might use slightly different namings.

A lot of PCI graphics card can handle 66MHz PCI bus. A good example is a Voodoo3 PCI.

There are a lot of later 486 boards that can do 66MHz and 60MHz. Yes, I was talking about 486 boards only, so no 64bit stuff. Just search here on Vogons.
Also the Asus PVI-486SP3 has a undocumented 66MHz setting. But unfortunately no 60MHz.
See here:
Asus PVI-486SP3 v.1.22 and 66MHz FSB

aries-mu wrote on 2023-09-30, 13:41:

Wow interesting, thanks! Well, if you're so fond of the DX50 and if it performs so well on ISA-only mobos (it must be thanks to […]
Show full quote

wierd_w wrote on 2023-09-30, 07:25:

CPUs manufactured then were binned/graded based on in-factory testing. Some chips can handle 'incorrect' clocking better than o […]
Show full quote

CPUs manufactured then were binned/graded based on in-factory testing. Some chips can handle 'incorrect' clocking better than others, depending on multiple factors that were in play when it was manufactured (Billet of silicon it was produced from, what batch of UV sensative etching/resist was used, etc...)

In a more direct answer to your question, there were some "Oddball" 486 chips that had a 1x multiplier. There *WERE* DX-50 chips, for instance. (NOT a DX/2-50, no, I mean a straight up DX-50)
https://dfarq.homeip.net/486dx-50-versus-dx2/

These chips were either "Blazing fast!" or "Cause so much horrible headache that they are useless!!" depending on where and how you put them into service. VLB motherboards for instance, were NOTORIOUS for **NOT** playing nice with these chips, as 50mhz was pushing the extreme edge of what the VLB architecture could accomplish. VLB is designed with a nominal bus speed of 33mhz, and 50mhz is some 50% higher than that. Not all cards could handle being driven that fast, not all implementations of the VLB could tolerate being driven at that speed reliably, and then there were other factors that might come into play (like shitty capacitors with poor noise suppression, crappy wiring in people's houses, people not using grounded outlets, et al.)

The actual answer to your question, is "If the CPU's bus speed stays at or near the nominal 33mhz, there is no real need for a clock multiplier by the chip, except that you will have crummier performance for in-processor applications-- at least as far as IO Bus technologies of the era were concerned." Clock multipliers allowed the chip to do things quickly internally (EG, multiple processor cycles could be done per IO cycle, and software written to take advantage of this fact, could do amazing things) while "Doing the best they could" with on-board discrete components, and the delays and other issues that circuit design and component selection from the era imposed.

I had a 486 that was ISA only; The DX50 I popped in it worked GREAT.
I later got a board that could do VLB, and realized quickly that my DX50 caused me no end of grief.

I sorely missed being able to play MP3s on my 486 (The DX50 could do raw RAM IO fast enough to decode MP3, which was something normally reserved for Pentium species chips) but being able to have better video performance and other gains elsewhere were a fair trade, when "Demoting" to a DX/2-50.

I suppose on a rare PCI equipped 486, the DX50 would be a fine choice. (though that assumes bus-mastering is present on that ancient PCI chipset)

Wow interesting, thanks!
Well, if you're so fond of the DX50 and if it performs so well on ISA-only mobos (it must be thanks to its 50 MHz FSB), then how about this: Get an AMD DX2-80 if you wanna try VLB!
I mean, the FSB runs at 40, which still is not bad at all, closer to 50 than a 25 or 33 Mhz bus. Plus you get a whopping 80 MHz CPU, well 30 MHz above your DX50. At the same time, the VLB should be able to handle 40 MHz smoothly, and also, if you get a VLB controller card and a VLB VGA card that claim to be able to handle 50 MHz at 0 ws, they should work well together at 40 MHz 0 ws. I suspect those MP3s of yours would fly on such a system (perhaps with 60 ns FPM RAM modules and the best cache you can get). Plus, having VLB controller and vga, the other tasks would also fly.
How about that?

Why demoting to a DX2 50? If you were so struck by the speed of the DX50, why going 25 MHz on the FSB? You could have gotten a DX2 66, faster CPU and faster FSB than the DX2 50.

Very curious: how would the DX50 be a fine choice on a PCI mobo, since PCI is stuck at 33 MHz and cannot be increased??? And what would "bus-mastering" have anything to do with helping with that?

Thanks!

In this case, said 486 existed in the 90s, and is long gone.

The 'pci always drives at 33mhz' is precisely the point; the faster fsb only deals with RAM access that way, and as long as your cache chips are not cheap, slow ones, and your ram has an appropriate access rate, it would be just peachy. The source of the instability was devices trying to be driven at 50mhz, that were designed to run at 33mhz! Driving them at 33mhz, regardless if the fsb, removes this issue!

Bus mastering is a technology that permits addon cards to communicate directly with RAM (and ram-io mapped devices) without the CPU being involved; eg, not subject to the kinds of 'let's tie up the cpu and memory bus with io!' Slowdowns or instabilities. This includes video cards (via linear framebuffer or memory io range) and hard disk controllers. Since RAM would prefer to talk at 50mhz, the cpu would be gagging on 50% NOPs to stay synced without bus-mastering, but is free to do other stuff (like decode mp3s) if other io operations are off of its plate due to busmastering being present.

My vlb 486 was a short splash; i moved to a pci based pentium 90 fairly quickly anyway.

Actually, we also run PCI bus at 60/66 Mhz for better benchmarks results.

Re: The World's Fastest 486
https://www.youtube.com/watch?v=qPRSHI4JoWI
https://www.youtube.com/watch?v=LI1_RlVLhu8

The Serpent Rider wrote on 2023-10-01, 03:54:

Actually, we also run PCI bus at 60/66 Mhz for better benchmarks results. […]
Show full quote

Actually, we also run PCI bus at 60/66 Mhz for better benchmarks results.

Re: The World's Fastest 486
https://www.youtube.com/watch?v=qPRSHI4JoWI
https://www.youtube.com/watch?v=LI1_RlVLhu8

If the bus and devices are made for it, it can be so.

But early pci was 33mhz.

https://en.m.wikipedia.org/wiki/Peripheral_Co … nt_Interconnect

The first version of PCI found in retail desktop computers was a 32-bit bus using a 33 MHz bus clock and 5 V signaling, although the PCI 1.0 standard provided for a 64-bit variant as well.[clarification needed] These have one locating notch in the card. Version 2.0 of the PCI standard introduced 3.3 V slots, physically distinguished by a flipped physical connector to prevent accidental insertion of 5 V cards. Universal cards, which can operate on either voltage, have two notches. Version 2.1 of the PCI standard introduced optional 66 MHz operation.

Historical pci 486 will be 33mhz pci bus.

486 PCI is tied to FSB, just like VLB, but also has dividers, so it can be anything. Even meager 12.5 Mhz.

PCI dividers depend on the chipset

Some UMC 8881 boards I had (like PC Chips M919) will halve the FSB clock if it's over 33 MHz at boot time. So at 40 MHz PCI will become 20 MHz. ALi M1489 boards usually have a jumper to either divide the clock or not, and you can force PCI to be the same as FSB there. Then you'll have to cherry pick cards that can tolerate that clock. E.g. it helps to use videocards with chips that were used for AGP cards (like Permedia or some Matrox chips), those can be used with a 66 MHz AGP bus. Realtek 8139 network chips are actually rated for 40 MHz, too.

Back to the original point — I think in practice you'll have issues when using memory and cache at 50 MHz and you'll have to loosen up the timings. So you won't exactly get a performance boost on every front compared to 40 MHz, you'll win some and lose some. It's probably possible to pick very good parts though, but there's a reason why people didn't often overclock to 50 MHz back in the day. On the other hand, AMD 5x86 at 160 (4x40) can sometimes be found in computer shops' pricelists — they would actually sell those systems overclocked.

CoffeeOne wrote on 2023-09-30, 17:07:

A lot of 486 mainboards can do 2-1-2 cache timing at 40MHz. For example Asus VL/I-486SV2GX4 and Asus PVI-486SP3. For a double ba […]
Show full quote

A lot of 486 mainboards can do 2-1-2 cache timing at 40MHz.
For example Asus VL/I-486SV2GX4 and Asus PVI-486SP3.
For a double banked cache configuration it is easier to achieve. Double banked cache configuration ususally means 8 cache chips (+1 tag). 15ns chips are fast enough.
My "calculation" is just stating the cycle duration multiplied with number of clock cycles. The cycle duration (EDIT!) at 25MHz (it is just the reciprocal of the clock) => 40ns. 33.33MHz => 30ns. 40MHz => 25ns. 50MHz => 20ns, 60MHz => 16.67ns, 66MHz => 15ns and so on.
So that was not wait states, I was talking about l2 cache timing.
On the Asus PVI-486SP3 it is:
Cache Burst Read Cycle - fastest option: 2-1-1-1
and Cache Write Cycle - fastest option: 2 CCLK.
Other Bioses might use slightly different namings.

A lot of PCI graphics card can handle 66MHz PCI bus. A good example is a Voodoo3 PCI.

There are a lot of later 486 boards that can do 66MHz and 60MHz. Yes, I was talking about 486 boards only, so no 64bit stuff. Just search here on Vogons.
Also the Asus PVI-486SP3 has a undocumented 66MHz setting. But unfortunately no 60MHz.
See here:
Asus PVI-486SP3 v.1.22 and 66MHz FSB

Oh thanks for the info!

wierd_w wrote on 2023-09-30, 18:03:

In this case, said 486 existed in the 90s, and is long gone. […]
Show full quote

In this case, said 486 existed in the 90s, and is long gone.

The 'pci always drives at 33mhz' is precisely the point; the faster fsb only deals with RAM access that way, and as long as your cache chips are not cheap, slow ones, and your ram has an appropriate access rate, it would be just peachy. The source of the instability was devices trying to be driven at 50mhz, that were designed to run at 33mhz! Driving them at 33mhz, regardless if the fsb, removes this issue!

Bus mastering is a technology that permits addon cards to communicate directly with RAM (and ram-io mapped devices) without the CPU being involved; eg, not subject to the kinds of 'let's tie up the cpu and memory bus with io!' Slowdowns or instabilities. This includes video cards (via linear framebuffer or memory io range) and hard disk controllers. Since RAM would prefer to talk at 50mhz, the cpu would be gagging on 50% NOPs to stay synced without bus-mastering, but is free to do other stuff (like decode mp3s) if other io operations are off of its plate due to busmastering being present.

My vlb 486 was a short splash; i moved to a pci based pentium 90 fairly quickly anyway.

Thank you too!

The Serpent Rider wrote on 2023-10-01, 03:54:

Actually, we also run PCI bus at 60/66 Mhz for better benchmarks results. […]
Show full quote

Actually, we also run PCI bus at 60/66 Mhz for better benchmarks results.

Re: The World's Fastest 486
https://www.youtube.com/watch?v=qPRSHI4JoWI
https://www.youtube.com/watch?v=LI1_RlVLhu8

Oh wow! Thanks!

The Serpent Rider wrote on 2023-10-01, 07:37:

486 PCI is tied to FSB, just like VLB, but also has dividers, so it can be anything. Even meager 12.5 Mhz.

jheronimus wrote on 2023-10-01, 10:07:

PCI dividers depend on the chipset

Some UMC 8881 boards I had (like PC Chips M919) will halve the FSB clock if it's over 33 MHz at boot time. So at 40 MHz PCI will become 20 MHz. ALi M1489 boards usually have a jumper to either divide the clock or not, and you can force PCI to be the same as FSB there. Then you'll have to cherry pick cards that can tolerate that clock. E.g. it helps to use videocards with chips that were used for AGP cards (like Permedia or some Matrox chips), those can be used with a 66 MHz AGP bus. Realtek 8139 network chips are actually rated for 40 MHz, too.

Back to the original point — I think in practice you'll have issues when using memory and cache at 50 MHz and you'll have to loosen up the timings. So you won't exactly get a performance boost on every front compared to 40 MHz, you'll win some and lose some. It's probably possible to pick very good parts though, but there's a reason why people didn't often overclock to 50 MHz back in the day. On the other hand, AMD 5x86 at 160 (4x40) can sometimes be found in computer shops' pricelists — they would actually sell those systems overclocked.

Thanks guys!