noshutdown wrote:

i did some tests between intel and amd dx4-100, both are 16kb WB cache versions, and the conclusion is totally different:
intel is significantly faster(by 10-15%) in integer benchmarks, while amd is noticeably faster(by 3-5%) in floatpoint benchmarks.

Cool!

I definitely want to test more chips. I was lucky this weekend and found the enhanced version of these chips for both, Intel and AMD, with the 16 KB Cache!

With the DX4 I don't have comparable models yet though.

I might do the benchmarks again with several runs, I doubt it will make a difference, my (plenty) of DOS benchmarking has always shown me that there is no variation in results. But you never know, it might calm down some people out there, though I think this is one of those topics that no matter what you do, someone won't be happy 😁

Scali wrote:

That's not a cross-license. That was the original license for production of second-source x86 CPUs. It was about chip designs, w […]
Show full quote

gdjacobs wrote:

No, not quite. The cross license was originally in place as a condition of IBM using Intel technology in the PC.

That's not a cross-license.
That was the original license for production of second-source x86 CPUs. It was about chip designs, which is why it had to be renewed for every new CPU.
The cross-license is a completely different thing, and deals with the x86 on a more abstract level (intellectual property), such as instruction sets and patents required to implement an x86-compatible CPU.

That is exactly what I meant by saying "AMD never had a license for using Intel's 386 or 486 designs".
The original license was a condition from IBM for the original PC, for 8088, and later extended to 286 for the AT.
By the time the 386 arrived, there were plenty of clone manufacturers, and Intel didn't solely need to rely on IBM. As such, Intel no longer cared about IBM's conditions, and the license was never extended beyond 286 CPUs, and production of subsequent x86 CPUs from Intel was never outsourced. IBM probably didn't care either, since Intel had become much larger and more powerful, and at the same time IBM had become a smaller player in the PC market.

AMD used this original license to claim that the terms allowed them to manufacture any x86 CPU from Intel. However, Intel never handed them the designs for anything newer than the 286, so AMD had to reverse-engineer the CPUs.
And as I said, AMD's claim to the original license was invalidated by the court. Which instated the x86 cross-license as we still know it today (although the terms have changed somewhat in recent years. Originally, AMD was only allowed to outsource 20% of production. Under those terms, Global Foundries would never have been possible).

Edit: This looks like a good overview: http://jolt.law.harvard.edu/digest/patent/int … gal-perspective

To be precise, it was a patent cross license (extending the agreement of 1976) and technology transfer arrangement. In exchange for royalties and access to their semiconductor designs, Intel was in fact obligated to provide access to AMD under the terms of the 1982 agreement. This was the basis on which the arbitrator and California Supreme Court largely ruled against them. The fact Intel had soured on the deal by that point is irrelevant from a legal point of view.

gdjacobs wrote:

To be precise, it was a patent cross license (extending the agreement of 1976) and technology transfer arrangement. In exchange for royalties and access to their semiconductor designs, Intel was in fact obligated to provide access to AMD under the terms of the 1982 agreement. This was the basis on which the arbitrator and California Supreme Court largely ruled against them. The fact Intel had soured on the deal by that point is irrelevant from a legal point of view.

Well, I don't know where you got that information from, seeing as I see no sources to any of this.
Some of what you say is not supported by the link I presented.
So I stick by the original story, which is supported by that link:
Intel and AMD (as well as various other manufacturers) originally had a licensing agreement regarding the manufacturing of chips.
This goes beyond just the 8088 and the 80286 CPU, and also includes the various chips that form the chipset of the IBM PC (the 82xx series basically, including chips such as the PIT, the PIC and the DMA controller).
Apparently in theory this was a 'cross-license' in that Intel was also allowed to produce chips based on the designs of their licensees, but as far as I know, in practice it only went one way, and Intel never manufactured third-party designs.

However, it was based around the *designs* for manufacturing rather than the *IP*. That would be a fundamental difference.
The AMD wiki page has some info on it: https://en.wikipedia.org/wiki/Advanced_Micro_Devices

Intel and AMD entered into a 10-year technology exchange agreement, first signed in October 1981[50][56] and formally executed in February 1982.[39] The terms of the agreement were that each company could acquire the right to become a second-source manufacturer for semiconductor products developed by the other; that is, each party could "earn" the right to manufacture and sell a product developed by the other, if agreed to, by exchanging the manufacturing rights to a product of equivalent technical complexity. The technical information and licenses needed to make and sell a part would be exchanged for a royalty to the developing company.

(The 1976 agreement is apparently only about microcode, according to that page).

AMD tried to reinterpret the license as being about IP. Apparently they did not succeed in this, since eventually a new license was drawn up, after which AMD could finally start selling their 386 and 486 clones.
Had the original license allowed the use of the x86 instructionset and other IP, then the new license would not have been required, because the old license was already in place.

I guess the bottom line is that IBM forced Intel to allow AMD and other players on their turf. AMD then muscled its way into the x86 world permanently (not getting access to a chip's design when you think you have access to it is one thing, but to actually reverse-engineer and put unauthorized clones on the market is an entirely different story). It's a pretty ugly way to get a licensed architecture.

I mean, to put it in today's terms... Various companies outsource chip production to others. Eg, nVidia has its GPUs manufactured by TSMC.
So TSMC gets the chip designs from nVidia and can manufacture chips on this basis.
However, the small print in the nVidia/TSMC contract probably precludes TSMC from passing these designs on to other parties, or modifying them, and/or producing chips based on these designs under their own name.
We'd be pretty surprised if TSMC started marketing their own "TSMCForce" chips, at higher clockspeeds than the original nVidia ones, yet fully compatible with NV's chips, and using NV's drivers and software for running them. That's not what outsourcing production is about.

Edit: This is an interesting article: http://law.justia.com/cases/california/suprem … /4th/9/362.html
It indeed says that Intel was not interested in AMD's products... and apparently if Intel didn't choose to manufacture AMD chips, then AMD would not be entitled to the 386 either.
While they say that Intel was not showing goodwill, AMD was also not trying to develop 32-bit processors themselves, nor look for arbitration. So AMD wasn't really holding up their end of the bargain either. They just wanted to leech off Intel's R&D apparently.

Here's a couple of sources I consulted.
https://www.sec.gov/litigation/admin/3437730.txt
http://digitalcommons.law.scu.edu/cgi/viewcon … 8&context=chtlj

The term you used, IP, is a blanket term often used to encompass (not exclusively) patent rights, trademark, trade dress, and copyright. Specific designs usually fall under copyright law and sometimes trade dress. The underlying technology falls under patent law (unless it's kept as a trade secret where it has no legal protection).

The agreement was a cross license in practice as exchange of semiconductor designs as well as cash was intrinsic to it. I did perhaps overstate Intel's obligations somewhat. They were obligated to negotiate in good faith the specifics of each of these exchanges. Because they were interested in securing a monopoly at this stage, they no longer held up their end of the bargain. The court also agreed that the 1982 agreement also extended the earlier 1976 patent cross licensing agreement between Intel and AMD.

gdjacobs wrote:

The term you used, IP, is a blanket term often used to encompass (not exclusively) patent rights, trademark, trade dress, and copyright. Specific designs usually fall under copyright law and sometimes trade dress. The underlying technology falls under patent law (unless it's kept as a trade secret where it has no legal protection).

The thing is, the term 'cross-license' is defined specifically as being about IP: https://en.wikipedia.org/wiki/Cross-licensing
The agreement for second-sourcing is not referred to as a 'cross-license', but rather as a 'technology exchange agreement', and focuses specifically on manufacturing rights for specific chips, under payment of royalties.

This is very different from what a 'cross-license' is. For example, even though AMD has an x86 cross-license, they are not allowed to manufacture Intel-designed chips, nor do they have access to these designs.
They only have the rights to use various patents and technologies intrinsic to building an x86 CPU. They still have to create the actual chip design themselves.

gdjacobs wrote:

They were obligated to negotiate in good faith the specifics of each of these exchanges.

That is a very subjective criterion obviously.

gdjacobs wrote:

Because they were interested in securing a monopoly at this stage, they no longer held up their end of the bargain.

Intel always wanted a monopoly on the x86. Every CPU designer normally has a monopoly on their own CPU designs. What makes the x86 different is that IBM interfered.
Other CPUs of that era (eg 6502, 6800, 68000, Z80) were generally not licensed to third-parties, and Intel had no intention of doing so either. They just could not pass up the deal with IBM.
AMD however was the only one to abuse their deal to clone 386 and 486 CPUs. Other second-source companies made their own designs.
If the license was indeed granting AMD the rights to the 386 and 486, wouldn't you think that all other companies would have done the exact same thing? The popularity of these CPUs made it a no-brainer.

gdjacobs wrote:

The court also agreed that the 1982 agreement also extended the earlier 1976 patent cross licensing agreement between Intel and AMD.

That's not how I read it. The way I read it, AMD was indeed found guilty of using copyrighted microcode, and the 1976 agreement did not cover this.
*However*, as part of the settlement, AMD was granted the rights to use the 386 and 486 microcode (but no other CPUs).

Scali wrote:

The thing is, the term 'cross-license' is defined specifically as being about IP: https://en.wikipedia.org/wiki/Cross-licensing
The agreement for second-sourcing is not referred to as a 'cross-license', but rather as a 'technology exchange agreement', and focuses specifically on manufacturing rights for specific chips, under payment of royalties.

There's nothing specific about IP. That same article has sections on patent and non patent cross licensing.

Scali wrote:

This is very different from what a 'cross-license' is. For example, even though AMD has an x86 cross-license, they are not allowed to manufacture Intel-designed chips, nor do they have access to these designs.
They only have the rights to use various patents and technologies intrinsic to building an x86 CPU. They still have to create the actual chip design themselves.

Not every cross license is for patents.

Scali wrote:

gdjacobs wrote:

They were obligated to negotiate in good faith the specifics of each of these exchanges.

That is a very subjective criterion obviously.

Yes, nonetheless the courts ruled on it. They also clarified the agreement as part of the judgement.

Scali wrote:

Intel always wanted a monopoly on the x86. Every CPU designer normally has a monopoly on their own CPU designs. What makes the x […]
Show full quote

gdjacobs wrote:

Because they were interested in securing a monopoly at this stage, they no longer held up their end of the bargain.

Intel always wanted a monopoly on the x86. Every CPU designer normally has a monopoly on their own CPU designs. What makes the x86 different is that IBM interfered.
Other CPUs of that era (eg 6502, 6800, 68000, Z80) were generally not licensed to third-parties, and Intel had no intention of doing so either. They just could not pass up the deal with IBM.
AMD however was the only one to abuse their deal to clone 386 and 486 CPUs. Other second-source companies made their own designs.
If the license was indeed granting AMD the rights to the 386 and 486, wouldn't you think that all other companies would have done the exact same thing? The popularity of these CPUs made it a no-brainer.

What, IBM "interfered" by negotiating contracts with their suppliers? Stop the presses!

If the legal situation was so clear cut, don't you think Intel would have fast tracked the case so AMD would be enjoined from selling their 386 and 486 clones?

Scali wrote:

gdjacobs wrote:

The court also agreed that the 1982 agreement also extended the earlier 1976 patent cross licensing agreement between Intel and AMD.

That's not how I read it. The way I read it, AMD was indeed found guilty of using copyrighted microcode, and the 1976 agreement did not cover this.
*However*, as part of the settlement, AMD was granted the rights to use the 386 and 486 microcode (but no other CPUs).

Apples and oranges. The patent aspect of their agreement was not so much in dispute.

gdjacobs wrote:

There's nothing specific about IP. That same article has sections on patent and non patent cross licensing.
...
Not every cross license is for patents.

I'm not interested in some kind of useless debate about semantics.
As I said generally 'cross-licensing' is about IP, not specifically about manufacturing/second-sourcing. The original agreement was not referred to as a 'cross-licensing' agreement, and is not the same as the 'cross-licensing' agreement that has been in place since the settlement in 1994.

So the important part is that we don't lump the two together as a single 'cross-license agreement', but rather make the distinction.

gdjacobs wrote:

What, IBM "interfered" by negotiating contracts with their suppliers? Stop the presses!

It is very unusual for a customer to have such fundamental demands on their suppliers, such as forcing them to outsource their production to what are essentially competing companies.
Trying to brush it off as "negotiating contracts with their suppliers" is just ridiculous.

gdjacobs wrote:

If the legal situation was so clear cut, don't you think Intel would have fast tracked the case so AMD would be enjoined from selling their 386 and 486 clones?

What makes you say they didn't?
The 386 was originally introduced in 1986. AMD didn't present their clone until 1990. It was clear that AMD didn't have the rights to manufacture 386es, and AMD had tried to get these rights, but Intel declined. AMD had already sought arbitration, but was denied. They went on and developed their 386 clone anyway.
All this made the case very complicated, because aside from AMD wanting to get the rights to the 386, it was now also about copyright infringements, whether or not the contract was honoured etc etc. One extra aspect here is that since AMD didn't get any manufacturing package from Intel, they also weren't paying royalties for their 386 clones.
Everyone sued everyone about everything. That's why eventually everything was lumped together into a single settlement.

The case of whether AMD was entitled to manufacture 386es was quite clear-cut legally: they weren't.
The eventual settlement was of an entirely different nature, and answered the question whether AMD should be entitled to develop their own x86-compatible CPUs. Which they eventually were, under the new cross-license... which the Am386 (and the Am486, which had been developed by then) itself would have violated, so it got some exceptions, so it could be sold anyway.

gdjacobs wrote:

The patent aspect of their agreement was not so much in dispute

I think you missed my point then: The 1976 deal is not so much a patent deal, but a deal revolving around microcode.
So as far as I know, it would never have covered the patents for building an x86 CPU (only the 1982 deal would cover that).
AMD however wanted to claim that they could use the 386 microcode. Which is why they just copied it. Which violated the copyright, which was not covered by the 1976 deal (which again is a difference between the IP for certain technological concepts vs the actual implementation of that technology).

Scali wrote:

Oh really? That's more of a 'luck of the draw', or at least having a very late model, built on a mature process, more than anyth […]
Show full quote

kanecvr wrote:

What strikes me about intel's DX4 chips is how poorly they overclock - if at all. A 66MHz AMD DX2 will often OC to 100MHz

Oh really?
That's more of a 'luck of the draw', or at least having a very late model, built on a mature process, more than anything.
I had an early Am486DX2-66 back in the day, and it wouldn't overclock at all... in fact, it needed a heatsink and fan just to reach 66 MHz. I couldn't get it to 80 at all.
The thing eventually burnt out (at its stock 66 MHz speed no less).
I then replaced it with an Intel 486DX2-66, which was an OverDrive model with a heatsink glued on. It could run 66 MHz fanless, and easily did 80 MHz when I put the fan from the AMD one on. That CPU is still running today... at 80 MHz.

And I can also vouch for the benchmark results... In fact, in my case I didn't see any difference *at all*.
The only difference I ever noticed is that there's a bug in the Am486DX2-66 which prevented me from running OS/2, while the Intel one ran OS/2 fine in the same system.

Well from the benchmarks it looks like I don't have to feel discouraged that I own an AMD Am486DX2-66. 😁 My part is brand new and untested. If I were to find a 486 glue-on heatsink, could I avoid the premature death of what happened to your cpu? I never overclock anything. Also, how can I tell if my part is an unstable early release model or not? Can I tell from the part number alone? Thanks.

Scali wrote:

Oh really? That's more of a 'luck of the draw', or at least having a very late model, built on a mature process, more than anyth […]
Show full quote

Oh really?
That's more of a 'luck of the draw', or at least having a very late model, built on a mature process, more than anything.
I had an early Am486DX2-66 back in the day, and it wouldn't overclock at all... in fact, it needed a heatsink and fan just to reach 66 MHz. I couldn't get it to 80 at all.
The thing eventually burnt out (at its stock 66 MHz speed no less).
I then replaced it with an Intel 486DX2-66, which was an OverDrive model with a heatsink glued on. It could run 66 MHz fanless, and easily did 80 MHz when I put the fan from the AMD one on. That CPU is still running today... at 80 MHz.

And I can also vouch for the benchmark results... In fact, in my case I didn't see any difference *at all*.
The only difference I ever noticed is that there's a bug in the Am486DX2-66 which prevented me from running OS/2, while the Intel one ran OS/2 fine in the same system.

amd had many versions of dx2-66, some of which would almost guarantee overclocking to 100 with ease as they were made with newer process. have a look at this thread:
too many variants of amd486dx2-66

computergeek92 wrote:

Well from the benchmarks it looks like I don't have to feel discouraged that I own an AMD Am486DX2-66. 😁

Everyone's a winner IMO. If you got an Intel you get that warm and fuzzy feeling of being a tiny bit in front, and if you AMD you can relax and know that you're not missing out on anything.

Still I'm eager to see some more chips 😁

Scali wrote:

I'm not interested in some kind of useless debate about semantics.
As I said generally 'cross-licensing' is about IP, not specifically about manufacturing/second-sourcing. The original agreement was not referred to as a 'cross-licensing' agreement, and is not the same as the 'cross-licensing' agreement that has been in place since the settlement in 1994.

So the important part is that we don't lump the two together as a single 'cross-license agreement', but rather make the distinction.

I'm with you there.

Scali wrote:

It is very unusual for a customer to have such fundamental demands on their suppliers, such as forcing them to outsource their production to what are essentially competing companies.
Trying to brush it off as "negotiating contracts with their suppliers" is just ridiculous.

Unusual, maybe, but entirely within their rights.

Scali wrote:

The case of whether AMD was entitled to manufacture 386es was quite clear-cut legally: they weren't.
The eventual settlement was of an entirely different nature, and answered the question whether AMD should be entitled to develop their own x86-compatible CPUs. Which they eventually were, under the new cross-license... which the Am386 (and the Am486, which had been developed by then) itself would have violated, so it got some exceptions, so it could be sold anyway.

Actually, the case revolved primarily around Intel honoring the 1982 agreement (they were allowed to bargain stringently, not shut the door completely) and AMD violating the Chinese wall when reverse engineering i386 firmware.

Scali wrote:

I think you missed my point then: The 1976 deal is not so much a patent deal, but a deal revolving around microcode.
So as far as I know, it would never have covered the patents for building an x86 CPU (only the 1982 deal would cover that).
AMD however wanted to claim that they could use the 386 microcode. Which is why they just copied it. Which violated the copyright, which was not covered by the 1976 deal (which again is a difference between the IP for certain technological concepts vs the actual implementation of that technology).

The 1976 deal was about patents.
https://books.google.ca/books?id=i3MLYUjRWisC … license&f=false

PhilsComputerLab wrote:

computergeek92 wrote:

Well from the benchmarks it looks like I don't have to feel discouraged that I own an AMD Am486DX2-66. 😁

Everyone's a winner IMO. If you got an Intel you get that warm and fuzzy feeling of being a tiny bit in front, and if you AMD you can relax and know that you're not missing out on anything.

Still I'm eager to see some more chips 😁

The benchmark differences we see in your scores are down to stuff like silicon age, fab process quality, pin oxidation and stuff like that. I ran a few tests of my own:

Test platform:

Lucky Star LS486E - SIS chipset, 256k L2 cache
32GB FPM
S3 Virge 4MB

Right. So none of my chips perform exactly the same. Wolf3d seems to discriminate most between chips, while speedsys and PCP Bench results are constant across CPUs. Unfortunately I can't tell witch chips is newer / older, but the "slower" ones do have whiter, slightly oxidized pins. Heck, shutting the machine down, waiting 10 minutes and retesting (with the same CPU) yields slightly different scores in wolf3d.

After all the testing my mainboard seems to have gone a bit flaky - I think I might have done something to the socket (this is my socket 3 test board and It's seen quite a few CPUs). Out of curiosity I'll bench my 3 am5x86-133 chips underclocked to 100Mhz (i don't own any 16kb wb AMD chips) and my intel DX4-100WB - I'm curious if the gaps will be wider or smaller. That's if the board holds out. If it doesn't I have two more identical ones 😀

Nice! Great to see a range of CPUs being tested.

Scali wrote:

Oh really? That's more of a 'luck of the draw', or at least having a very late model, built on a mature process, more than anyth […]
Show full quote

kanecvr wrote:

What strikes me about intel's DX4 chips is how poorly they overclock - if at all. A 66MHz AMD DX2 will often OC to 100MHz

Oh really?
That's more of a 'luck of the draw', or at least having a very late model, built on a mature process, more than anything.
I had an early Am486DX2-66 back in the day, and it wouldn't overclock at all... in fact, it needed a heatsink and fan just to reach 66 MHz. I couldn't get it to 80 at all.
The thing eventually burnt out (at its stock 66 MHz speed no less).
I then replaced it with an Intel 486DX2-66, which was an OverDrive model with a heatsink glued on. It could run 66 MHz fanless, and easily did 80 MHz when I put the fan from the AMD one on. That CPU is still running today... at 80 MHz.

And I can also vouch for the benchmark results... In fact, in my case I didn't see any difference *at all*.
The only difference I ever noticed is that there's a bug in the Am486DX2-66 which prevented me from running OS/2, while the Intel one ran OS/2 fine in the same system.

What version of os/2? My amd dx4 ran os/2 warp 4 okay.

gdjacobs wrote:

Unusual, maybe, but entirely within their rights.

I never said it wasn't. I'm just saying it is quite unique, and placed Intel in a unique situation.
Hence, all the unpredecented licensing, stretching of license agreement terms and court cases that followed.

However you want to view it, bottom line is still that the x86 is Intel's creation, and other manufacturers took advantage of the situation that IBM created to get a piece of the pie.
In some cases, they pretty much made it their business model, and pretty much halted development of their own products.

gdjacobs wrote:

Actually, the case revolved primarily around Intel honoring the 1982 agreement (they were allowed to bargain stringently, not shut the door completely)

That's the same thing, isn't it?
I mean, the subject of Intel honouring the agreement or not was the i386, which Intel didn't believe they had to share with AMD.
Given the fact that AMD didn't offer Intel any products that were in the same ballpark, Intel was not required to do so. The court admitted this much.

gdjacobs wrote:

The 1976 deal was about patents.
https://books.google.ca/books?id=i3MLYUjRWisC … license&f=false

It is not specific as to *which* patents though.
Other sources say they are about microcode: https://en.wikipedia.org/wiki/Advanced_Micro_Devices

When Intel began installing microcode in its microprocessors in 1976, it entered into a cross-licensing agreement with AMD, granting AMD a copyright license to the microcode in its microprocessors and peripherals, effective October 1976.

And that is why AMD pointed to the 1976 deal when claiming they had the rights to the 386 microcode. Which they didn't.

kanecvr wrote:

The benchmark differences we see in your scores are down to stuff like silicon age, fab process quality, pin oxidation and stuff like that. I ran a few tests of my own:

It doesn't work that way. Remember that these are digital circuits.

The variations you're talking about don't change the number of cycles a given instruction takes to execute.

It would be normal, however, for the clock generator to drift by a few % as the board warms up.

And depending on how the benchmark calculate what "a second" is for "frames per second", that would add up, since the CPU clock and realtime clock would be drifting independently.

SquallStrife wrote:

The variations you're talking about don't change the number of cycles a given instruction takes to execute.

Indeed. If that were true, then cycle-exact code would be impossible. But cycle-exact code is possible, regardless of the age of the machine (we're still doing it on 30+ years old machines).

SquallStrife wrote:

It would be normal, however, for the clock generator to drift by a few % as the board warms up.

Depending on how the machine was designed, you might not even see this.
For example, on a classic PC/XT, the CPU and PIT are running from the same clock. So the time is derived from the same frequency as the CPU clock. Meaning that they both drift together, and you won't be able to measure it.
However, if you are using a VGA card, it will have its own clock. So if you are measuring frames by using the VGA vsync signal, then things may drift.

Of course, if you use the same system, and just swap CPUs around, then all this can be eliminated. All that is left is to repeat the same benchmarks a number of times, so you get min/max/average scores, giving you an idea of the margin of error.

Scali wrote:

Depending on how the machine was designed, you might not even see this.
For example, on a classic PC/XT, the CPU and PIT are running from the same clock. So the time is derived from the same frequency as the CPU clock. Meaning that they both drift together, and you won't be able to measure it.

I figured that anything derived from the CPU clock would drift "in sync", but anything that's measured versus actual time would be measured against the RTC, which is potentially using a separate 32.768kHz crystal oscillator. (Or is that only in use when the system is powered down?)

SquallStrife wrote:

I figured that anything derived from the CPU clock would drift "in sync", but anything that's measured versus actual time would be measured against the RTC, which is potentially using a separate 32.768kHz crystal oscillator. (Or is that only in use when the system is powered down?)

The RTC wasn't introduced until the AT. So on PC/XT the time is kept with the PIT. This is why you have to tell DOS the date/time when it starts up. It needs a starting point, then it counts from there, until you reboot. It uses a timer interrupt for that, which fires at 18.2 Hz. Software that takes over the timer interrupt, can mess up this timing (you will probably find that the time is 'frozen' before and after running 8088 MPH).

Afaik DOS always uses the PIT to keep date/time, even on systems with an RTC. On an AT (or XT with RTC add-on), the time is copied from the RTC at boot, then the PIT takes over for that DOS session.

Scali wrote:

All that is left is to repeat the same benchmarks a number of times, so you get min/max/average scores, giving you an idea of the margin of error.

So true, I was about to comment this until I read your post.

Let me add another parameter in the discussion: How do we know that in the Quake demo the FPS score is not restricted by the graphics card for example? I'm not saying it is, just thinking it as a possibility. How do we know that both cpus perform equally because they are equal and not because one or both wait for something else? We don't have an indication that throughout the whole game demo the cpus are in 100% load. This is not the case of course for benchmarks restricted to cpus, like the Norton's SI, I'm talking about the games. Again, I'm not trying to say that this is what happened here and the results are misleading, only discussing it.

Probably a not so good example (it's exactly the opposite actually), just to lay out my point: I once wanted to see the difference between a Voodoo3 2000 and a 3000. So I run the quake demo on a PII and got less than 1% difference. Does this mean that they're the same? No, when running the same thing on a late/fast PIII I got the real difference.