Scali wrote:

I think these two statements contradict eachother somewhat. Early 486SX CPUs actually did contain the FPU, so they were technica […]
Show full quote

mpe wrote:

I believe in 1991 it was simply a good time to have a cheaper 486 product.

I think these two statements contradict eachother somewhat.
Early 486SX CPUs actually did contain the FPU, so they were technically the same die as a 486DX. Which means they weren't cheaper to produce.
Given that they'd also require new motherboards, with an FPU socket, where is the cost saving?

"Cheaper" in the meaning of lower sale price (hundreds of $$$), not cost of production (a fraction of the former).

mpe wrote:

"Cheaper" in the meaning of lower sale price (hundreds of $$$), not cost of production (a fraction of the former).

I know, but why would Intel do that? They could just as well lower the price of the 486DX, if cost of production is equal.
That's my point. It doesn't make sense to introduce a new CPU where you artifically laser off the FPU and require new motherboards with an extra socket (which is MORE cost of production).
There must be some economic advantage to Intel, else there's no way they would have introduced the 486SX. They would have just lowered the price of the 486DX, that is common sense.

And in fact they did both. In July 1991 they lowered the price of DX-33 from the original $1056 to $445. Surely the chip cost them just as much to make in June 1991 as in July 1991.

And they introduced SX at the same time. Why? Because of competition. 1991 was the year of introduction of Am386 (it was ready sooner, but Intel kept it in court). It would be very hard for Intel to compete in mainstream/lower-end segment without a low-cost offering.

This situation is a reason why I don't think SX couldn't be a way of recycling DX chips when there was such commercial threat.

Scali wrote:

I know, but why would Intel do that? They could just as well lower the price of the 486DX, if cost of production is equal. That' […]
Show full quote

mpe wrote:

"Cheaper" in the meaning of lower sale price (hundreds of $$$), not cost of production (a fraction of the former).

I know, but why would Intel do that? They could just as well lower the price of the 486DX, if cost of production is equal.
That's my point. It doesn't make sense to introduce a new CPU where you artifically laser off the FPU and require new motherboards with an extra socket (which is MORE cost of production).
There must be some economic advantage to Intel, else there's no way they would have introduced the 486SX. They would have just lowered the price of the 486DX, that is common sense.

Actually it made total sense.

Instead of selling all of the chips for less, they were selling some of them for less and some for more.

A fraction of those who bought the "lesser" chips would also buy a 487 in the future, which drove additional profit for Intel.

This is a classic example of upselling.

mpe wrote:

This situation is a reason why I don't think SX couldn't be a way of recycling DX chips when there was such commercial threat.

Let's not get carried away, the 386DX-40 was not THAT big of a threat.
AMD was a small player, and 386DX-40 was a niche budget product.

Intel could obviously also have opted to introduce their own 386DX-40.

ShovelKnight wrote:

Actually it made total sense.

Apparently we disagree on that.

ShovelKnight wrote:

Instead of selling all of the chips for less, they were selling some of them for less and some for more.

Sure, but there would be some kind of economic advantage.
Either they validated the FPU and sold the broken ones... or they skipped validation on a certain percentage of the chips, so they saved money that way.

ShovelKnight wrote:

A fraction of those who bought the "lesser" chips would also buy a 487 in the future, which drove additional profit for Intel.

A very VERY small fraction, which probably wouldn't have made any kind of dent in Intel's sales figures whatsoever.
Virtually no software required an FPU in the 486-days, so there was also no reason to upgrade. Apart from the fact that the main reason why people would get a 486SX instead of a DX was to save money. Getting a 487 doesn't really fit with that market.

Well, Am386 (and for a moment forget if SX, DX or -40) wasn't a niche product. AMD sold tons of them. One milion units between March - October 1991, which was a lot back then. And I bought one too! 😀

Don't forget the context. In early 1991 hardly anyone was buying 486. It was ludicrously expensive at $1k. 486 sales were slow. Like 88.000 units in 1990 compared to 3.7 millions of 386. Mainstream buyers were deciding between 386SX and 386DX, if not a fast 286.

So the market in early 1991 was about selling 386 chips and the AMD eating up to the 386 created a big threat for Intel.

Scali wrote:

ShovelKnight wrote:

Actually it made total sense.

Apparently we disagree on that.

It doesn't make sense to us, but maybe you have to think like an MBA.

The reason I say that is because at various points after the time period in question, we know that vendors pulled exactly this type of nonsense. They artificially cripple and rebadge parts JUST to create a low-end part to fill a perceived gap in the market. Celeron CPUs stuck with low FSB, STB Velocity version of Voodoo 3 with a texture unit disabled in the drivers, CPU cores and GPU shaders disabled (that turn out to work). They've also floated the idea of disabling features that would later be unlockable through software by purchasing a keycode...

Giving the customer less just for the sake of not giving them too much is a valid strategy in the minds of some.

"Binning" parts makes a lot more sense though, and wasn't unique to Intel. Motorola 68040 chips came in different versions with either the MMU or FPU disabled, or both.

mpe wrote:

Well, Am386 (and for a moment forget if SX, DX or -40) wasn't a niche product.

Apparently we disagree on that.
AMD didn't have deals with large OEMs, so all the big companies (Dell, Compaq, HP, IBM etc) sold Intel, easily outnumbering whatever AMD sold.

mpe wrote:

Don't forget the context. In early 1991 hardly anyone was buying 486. It was ludicrously expensive at $1k. 486 sales were slow. Like 88.000 units in 1990 compared to 3.7 millions of 386. Mainstream buyers were deciding between 386SX and 386DX, if not a fast 286.

Well, in my area I didn't see the 386DX-40 until the 486DX2-66 was already becoming a common computer.
I had a 486DX2-66 before my neighbour bought a 386DX-40, which was the first time I saw one.
So that's how the 386DX-40 was over here: the budget choice for people like students etc, who couldn't afford a 486.
People with a 'normal' income could afford a 486 by then.

I never saw many 386DX-40 machines. 486 on the other hand...

mpe wrote:

So the market was about selling 386 chips and the AMD eating up to the 386 created a big threat for Intel.

Don't forget there was only the 386DX-40 that was a unique selling point for AMD.
Intel had the 386DX-33 and lower to offer. And as I said, it was a deliberate choice of Intel to not scale up their 386DX to 40 MHz, but instead to scale down their 486 to the SX series. That choice must have made business sense, else Intel wouldn't have made it.

Well, let's assume the chip costs $50 to make.

If you sell it for $1056, you end up with a very healthy profit margin. Even if you sell it for $445, you're still making a lot of money on each sale.

But you have a problem that a lot of people are not buying this chip because it's expensive. You create a lower-end part and price it at 1/2 of the full-spec part so it starts moving in larger quantities (which probably also drives the costs down, but let's disregard that). It doesn't matter if the lower-end part still costs $50 to make, because you're selling them for $220. But if somebody needs a full 486, they're still buying them for $445.

Now let's imagine that you simply lower the price of the full-spec part to $220. You may gain traction in the lower end of the market, but you lose the higher end of the market because nobody would buy a chip for $445 when you can buy exactly the same thing for $220.

You also have to remember that the high end market is about profit margin, not market share. Just ask Apple which has about 20% share of smartphone market but captures about 80% of all profits - in other words, Apple sells 4 times less smartphones but makes 4 times more money than the rest of the industry combined. Intel has always operated on the same basis (except that unlike Apple, they have the whole market covered and not only the top end).

mpe wrote:

Isn't great that in 2019 people are still being fascinated that some 486SX's may contain a real FPU :) […]
Show full quote

Isn't great that in 2019 people are still being fascinated that some 486SX's may contain a real FPU 😀

Personally I think that although there is enough evidence that the FPU is physically there in early SX units (and absent in some later chips) I am not so sure it was the way of dealing with broken DX chips.

In 1991 I doubt they had problems with 486DX yields. A simple connection removal by laser wouldn't change CPU identification (and no microcode update capability on chips before P6). I believe in 1991 it was simply a good time to have a cheaper 486 product.

A great article about the topic:
http://www.os2museum.com/wp/lies-damn-lies-and-wikipedia/

Wow, I definately learned a lot from that, thanks!

ShovelKnight wrote:

Well, let's assume the chip costs $50 to make. […]
Show full quote

Well, let's assume the chip costs $50 to make.

If you sell it for $1056, you end up with a very healthy profit margin. Even if you sell it for $445, you're still making a lot of money on each sale.

But you have a problem that a lot of people are not buying this chip because it's expensive. You create a lower-end part and price it at 1/2 of the full-spec part so it starts moving in larger quantities (which probably also drives the costs down, but let's disregard that). It doesn't matter if the lower-end part still costs $50 to make, because you're selling them for $220. But if somebody needs a full 486, they're still buying them for $445.

Now let's imagine that you simply lower the price of the full-spec part to $220. You may gain traction in the lower end of the market, but you lose the higher end of the market because nobody would buy a chip for $445 when you can buy exactly the same thing for $220.

You also have to remember that the high end market is about profit margin, not market share. Just ask Apple which has about 20% share of smartphone market but captures about 80% of all profits - in other words, Apple sells 4 times less smartphones but makes 4 times more money than the rest of the industry combined. Intel has always operated on the same basis (except that unlike Apple, they have the whole market covered and not only the top end).

Even more important - if you sell a $220 part, your competitor doesn't sell a part (at whatever price the AMD 386's were).

So if you had the choise of buying a 486DX at a premium price vs. an AMD 386 at a bargain, if you didn't have the money, the choice would be simple - but - if a pricepoint in between existed - aka the 486SX, you might consider if the extra investment was worth it. Also - I would've thought that by buying a 486SX system, I would be ready for a DX upgrade when prices got down ..

Scali wrote:

Well, in my area I didn't see the 386DX-40 until the 486DX2-66 was already becoming a common computer.

486DX2-66 a common computer? When was that. Like in 1993, 1994, 1995?

In "my area" I hardly remember seeing a 386 chip that was made by Intel. Perhaps even more so for 286 chips maket which was quickly conquered by clone makers (AMD, Harris, IBM, Siemens).

On the flip side. Back then it was common for many OEMs to not advertise AMD brand even if they actually used AMD chips. So it could well be that people in your area were using AMD chips without knowing. As we know AMD's 386 and early 486 chips were in fact almost identical parts, but cheaper, higher clocked or with SL features. Customers were buying 386 or 486 PCs. Not Intel CPUs.

In 1991 Intel initialised "Intel Inside" campaign to promote brand loyalty just because of this.

They might have been king of the upmarket, but definitely not without competition in lower segments.

First Intel CPU I had was a 486DX2 66. And that was only because I got it used to install in a motherboard I pulled out of a dumpster. Rode my bicycle around an hour or so each way to get that CPU.

I didn't have another Intel CPU until a C2Q Q6600.

Everything else was AMD besides the 486 up to the Q6600.

I worked at a computer store in 2000-2001 and a great percentage of the CPUs we sold were AMD.

It really wasn't until the Pentium 4 era and Intel's dirty/illegal tactics that they really gained the upper hand. AMD had them beat performance-wise very easily with the Athlon 64 CPUs. Even the Athlon and Athlon XP CPUs were very, very good at the time. AMD was way cheaper and offered similar performance.

mpe wrote:

In 1991 I doubt they had problems with 486DX yields. A simple connection removal by laser wouldn't change CPU identification (and no microcode update capability on chips before P6). I believe in 1991 it was simply a good time to have a cheaper 486 product.

My understanding is that silicon yields are still a problem today, so I don't see why it would be hard to believe that 1/2 the die might have a defect. If I were a product designer, I might count on that and have a simple ID selector fuse that can be burned based on test results. Not saying this is the case, since I don't have an ounce of knowledge one way or another, just a potential way to close that loophole.

mpe wrote:

486DX2-66 a common computer? When was that. Like in 1993, 1994, 1995?

Probably... See, the thing is that both 386 and 486 were very expensive computers, so around 1990-1992, people would get 286 or 386SX machines.
After that, prices came down very quickly. As a result, I've rarely seen any 386DX machines with consumers. Most consumers made the jump to 486 right away, as did I (from a 386SX-16 to a 486DX2-66).

mpe wrote:

In "my area" I hardly remember seeing a 386 chip that was made by Intel.

That would be very odd, given that Intel has been supplying 386 CPUs since 1985, and AMD only from 1991 onwards.

mpe wrote:

Perhaps even more so for 286 chips maket which was quickly conquered by clone makers (AMD, Harris, IBM, Siemens).

Those aren't clones. Those are second-source Intel CPUs.

mpe wrote:

On the flip side. Back then it was common for many OEMs to not advertise AMD brand even if they actually used AMD chips.

For 286 and earlier it was irrelevant, since they were second-source Intel CPUs.

mpe wrote:

In 1991 Intel initialised "Intel Inside" campaign to promote brand loyalty just because of this.

The full story there is that Intel never second-sourced the 386 and later CPUs. AMD felt that they were entitled to clone these CPUs based on the earlier second-source contracts. So AMD reverse-engineered Intel's CPUs, and wanted to sell clones.
Intel sued, managed to block AMD from selling their clones, and AMD actually lost that claim: court ruled that the second-source license did in fact NOT entitle AMD to clone Intel's CPUs (more specifically: their clones violated Intel's copyright on the microcode).
However, the court also ruled that because x86 was such an important ISA worldwide, that Intel must allow third-party licenses.
In 1991 this court case came to an end and AMD could finally start selling their clones (after replacing the microcode with their own reimplementation).

From then on, AMD could actually market their CPUs as Am386SX/DX and Am486SX/DX, as these were products independent from Intel (where the second-source were just Intel CPUs manufactured in third-party fabs, much like how AMD and NVIDIA do not make their own GPUs, but have them manufactured by companies such as TSMC. This was actually not Intel's choice, but IBM forced Intel when they chose the 8088 for their PC, because IBM wanted to avoid the risk of being dependent on a single supplier. By the 386, Intel was much larger, and IBM had become irrelevant amongst all the clone builders, so Intel no longer honoured IBM's second-source requirement).
This led Intel to introduce the Intel Inside campaign, to distinguish their CPUs from the clones. Intel also moved from using model numbers to using model names, starting with the Pentium. The reason for this is that they could not trademark 386 or 486, but they could trademark Pentium. So that meant that clones could no longer use the same product names.

cyclone3d wrote:

It really wasn't until the Pentium 4 era and Intel's dirty/illegal tactics that they really gained the upper hand. AMD had them beat performance-wise very easily with the Athlon 64 CPUs. Even the Athlon and Athlon XP CPUs were very, very good at the time. AMD was way cheaper and offered similar performance.

Actually, Intel was the only choice for the inaugural years of the IBM PC. So they had 100% of the market from the start.
The first successful clone maker was NEC with their V20 and V30 CPUs (but never had more than a few percent of the market).
There were various other clone makers back in the days of the 386 and 486, such as Texas Instruments and Cyrix.
AMD was not the largest player in those early years. They are just the only ones to survive today (give or take some weird VIA chips and such), and AMD is famous because of the 386 lawsuit.

But even in AMD's heyday of the Athlon, they were still a small player, and never had more than about 20% of the entire market. Intel had the other 80%. So AMD never had the 'upper hand', not even close. Intel is called Chipzilla, for good reason: they have always been incredibly much larger than any of their competitors.

Only in recent years has this started to change, mainly from the onslaught of ARM chips, and hugely successful electronics companies such as Samsung.
AMD is still nowhere near as large as Intel is.
In fact, I believe AMD is about the same size as NVIDIA.

In my circles (broke students early-to-mid 90's), we couldn't afford AMDs too. Everybody was running Cyrix on Tomato boards, good times 😁 )

ShovelKnight wrote:

Well, let's assume the chip costs $50 to make.

That assumption is flawed.
The majority of chip costs is actually in R&D and tooling up for manufacturing.
A single CPU may take years to develop and can cost billions of dollars. That is the reason why the selling price is not in relation to the actual cost of manufacturing (as in manufacturing alone, not taking into account the design and investment in manufacturing facilities).
The prices on new chips are high because they want a return on investment.

AMD rarely if ever actually turned a profit on their CPU department. And they certainly weren't selling their chips close to manufacturing costs.
Their problem was poor return on investment because they had to price their CPUs lower than the competition.

Which brings me back to why it doesn't make sense for Intel to just sell fullblown 486DX CPUs as 486SX. There will be cost-saving involved, else they'd cut into their return-on-investment quite significantly.
Better to continue producing 386DX CPUs, which had been around since 1985, and ROI was already covered.

mpe wrote:

I was wondering if there is a regular (non-OEM) motherboard that can drive these at 16 or 20 MHz. All my 486 motherboards start at 25 MHz.

FIC 486-GIO-VT can do it if it has a VIA clock generator chip. It can go even lower though it'll have problems with HDD access eventually (but a 8MHz 486 is fun).
Some hybrid 386/486 board can do that too, like DataExpert OPTi 495SLC/XLC. There's even a topic somewhere here @ vogons that shows the (undocumented) settings.

Scali wrote:

mpe wrote:

In "my area" I hardly remember seeing a 386 chip that was made by Intel.

That would be very odd, given that Intel has been supplying 386 CPUs since 1985, and AMD only from 1991 onwards.

That may be odd in the UK but certainly this is the case here in Eastern Europe. When 386 was cutting edge we still belonged to the dreaded Communist World and it was prohibited to sell such a secret treasure to us. Plus we were (are) poorer too so it added even more delay, even after the collapse of the old system when the "gates" to the West got opened. First the spoils of Capitalism trickled in slowly. So the typical 386 you can see here is a late 386DX-40 soldered to a tiny board. There are exceptions but that's the most common form.