Pentium iii 400: x86 cpus’ Guide — intel pentium iii

30 декабря, 2022

G4 vs. Pentium III | Low End Mac

1999 – One little chip sure can set off a world of controversy. It started when Apple introduced the Power Mac G3 in November 1997.

Based on benchmarks from Byte, a highly respected independent magazine that covered many types of computers, Apple claimed, “The processor in the Power Mac G3 is up to twice as fast” as the Pentium II at the same clock speed.

Wintel magazines kept asking, “Where’s the speed?” And Apple fans kept saying, “up to twice as fast doesn’t mean always twice as fast.”

For certain types of operations, as the BYTEmark results showed, the PowerPC G3 processor is twice as fast as a Pentium II at the same MHz rating. But only for certain operations.

Still, it made for great marketing.

Twice as Fast, Round 2

Last week at Seybold 1999, Steve Jobs didn’t proclaim the G4 twice as fast at the Pentium III. He demonstrated it – using Intel’s own benchmarks (press release quoted above).

Then he spilled the beans: The 500 MHz G4 wasn’t twice as fast as a 500 MHz Pentium III. It was over twice as fast as a 600 MHz Pentium III. In fact, some benchmarks showed the 500 MHz G4 at 2.9 times the performance of the 600 MHz Pentium III.

Intel’s response: We’ll be making faster chips.

News flash: So will Motorola. When Intel hits 700 MHz, expect the G4 to be running at 550 MHz, maybe even faster.

On top of that, to match the capabilities of the Velocity Engine-equipped 500 MHz G4, Intel would have to make 1600 MHz Pentium III chips. And that isn’t likely until mid-2001.

Of course, by then the G4 will be running at speeds in excess of 1000 MHz, too.

Benchmarks aside, the 500 MHz Power Mac G4 system does things like Photoshop and 3D rendering about twice as fast as Compaq’s top of the line 600 MHz PIII system.

Foul?

Expect the Wintel world to cry, Foul!” over this. After all, nobody likes being in second place.

Here’s one example (from CNBC):

“It was good showmanship,” says Martin Reynolds, vice president of Dataquest, based in San Jose, Calif. “But it was not good benchmarking.” The Intel PC used during the demo, for example, didn’t have the same software plug-in used by the G4.

The plug-in in question is one specifically designed so Photoshop can use the Velocity Engine in the G4 processor. Of course the Pentium system doesn’t have the same plug-in. On the other hand, neither does the Mac version of Photoshop use the MMX support built into the Windows version.

Apple still lags far behind Intel in the desktop PC market. The company controlled just 4.8 percent of the dollar market share during the second quarter, according to Dataquest. Intel-powered PC’s, by comparison, held more than a 90 percent market share during the same period.

If you can’t question the facts, confuse the issue with other facts. Does G4 performance depend on how many units Apple sells or their dollar percentage of the US retail personal computer market? Does the overwhelming number of Celeron systems sold somehow undermine the power of the G4?

Market share is a straw man argument, not one that holds any weight here.

Several other factors, including the availability of compatible software, are at least as important as raw microprocessor speed when it comes to assessing the competitiveness of a desktop computer, analysts say.

Straw Man #2. Does it matter if the Mac has only thousands of programs available instead of tens of thousands? The simple fact is that the Mac can do anything PCs can do, either with genuine Mac software or by running an emulator and using Windows applications.

What do you want to do today? Can the Mac do it? If not, then you’ve got a reason to look at Wintel – but that reason has nothing to do with G4 performance.

“The really important comparison is the price-performance comparison,” says Charles Smulders, principal analyst at Dataquest. “In real terms, the industry is not as focused on performance right now. It’s more focused on sub-$1,000 PCs and smaller appliances.”

Straw Man #3. If we can’t confuse performance with market share and how many software titles are available, let’s look at value.

Although this again has nothing to do with the performance of the G4, price-performance is a good issue to raise. But you have to be willing to look at the whole picture.

First, sub-$1K systems are not being sold to the same market as G4 systems, so you don’t want to compare the Power Mac with a Compaq Presario or any other low-end Wintel box. You want to compare it, as Apple did, with the 600 MHz PIII powerhouses.

For argument’s sake, let’s assume the 500 MHz G4 offers twice the performance of a 600 MHz PIII across the board. It doesn’t – it’s faster at some tasks and slower at others – but we have to make some assumptions in this comparison.

The 500 MHz G4 is a $3,500 computer. That includes a 27 GB hard drive, 256 MB of memory, a DVD-RAM drive, a Zip drive, FireWire, 10/100 ethernet, and accelerated 2x AGP video.

I can’t find any information on the Compaq 600 MHz PIII system on Compaq’s site, so let’s go to the Gateway site and build a system. If anything, this should offer the same performance for less money than a comparable Compaq system.

The Gateway Performance 600 with 600 MHz Pentium III, 256 MB RAM, 27 GB hard drive, a 10/100 ethernet card, a Zip drive, and a CD-RW (no DVD-RAM option) comes to $2,324.

The Gateway includes a floppy drive, so we’ll add $100 to the G4 system for that. We’ll estimate a DVD-RAM drive can be added to the Gateway for $500 (Outpost.com currently has prices from $530-610).

Total price of the Gateway computer: $2,800. At a performance of 10, we’ll give this a value rating of 0.35 (rating/price). It would rate 0.43 without DVD-RAM.)

Total price of the Power Mac G4: $3,600. On the assumption of twice the overall performance of the Gateway, that’s a performance score of 20. This gives the 500 MHz Power Mac G4 a value rating of 0.56, or almost 60% higher than the Gateway.

Now assume we’ve been too generous in rating the G4 at twice the performance of the 600 MHz PIII. Reduce that to 50% and the value rating drops to 0.42 – equal to the Gateway computer without DVD-RAM.

The Bigger Picture

But that only looks at purchase price. Let’s assume the user is a graphics professional who spends 10 hours a week using Photoshop – which is twice as fast on the G4. The G4 user could either reduce that to 5 hours, freeing 5 hours for other projects, or get twice as much done in the same amount of time.

Further, let’s assume this graphics professional earns $20 per hour. By putting a 500 MHz G4 system on his desk instead of a 600 MHz PIII system, the business saves $100 per week based on load of 10 Photoshop hours with the Gateway. That’s $5,000 per year, allowing two weeks for vacation.

A more likely scenario is that the worker will be twice as productive, increasing billables per hour to twice that of the PIII user. Taking an arbitrarily low rate of $40 billable per hour on the PIII, the same user could do $80 worth of billable work on a G4. Instead of billing $400 per week for 10 hours of Photoshop work, twice to productivity means $800 of billings.

Whether the company saves $800 or $1,200 by going Pentium instead of G4, even a low Photoshop workload obviates that price advantage within two to three months. From the perspective of billable services, the Pentium system loses any price advantage within the first month of use.

A Still Bigger Picture

There are other factors. Wintel machines have more down time. Assume 1% down time for Windows and hardware issues, which is probably low. That’s $400 per year spent paying the employee to tinker or wait for a tech to fix things. That’s also $800 less billable time over the course of a year.

We’ll assume the Mac is twice as reliable, although it’s probably more than twice as reliable.

Now stretch all this over two years, about the maximum time before a high-end production computer will be replaced by an even more powerful one.

 

System Gateway PIII Power Mac G4
Purchase price $2,800 $3,600
Billable services -$4,000 -$8,000
Down time $800 $400
Lost billables from down time $1,600 $800
Net cost $1,200 -$3,200

Plug in your own numbers: hours, wages, billables, down time, years of use. The more hours the system is used, the greater the Mac advantage. The higher the wages, the greater the Mac advantage. The longer until the computer is replaced, the greater the Mac advantage.

Unless you’re simply going to buy the computer and your time has no value, the Power Mac G4 is the best computing value on the market. Even if you don’t bill your time, it leaves more of your time for billable work or for yourself.

Conclusion

Because the Velocity Engine makes the Power Mac G4 the first personal computer with GFLOPS (gigaflops) performance, it is the best computer available for graphics and video production.

The Wintel camp can try to confuse the issue by looking at market share, titles available, or initial purchase price, but from the perspective of the professional user, the Power Mac G4 provides more value than any Wintel computer available.

Even if it isn’t always twice as fast.

Keywords: #ppcg4 #powerpcg4 #macvalue #pentiumiii #twiceasfast

Short link: http://goo. gl/FGKjFG

searchwords: g4vspiii, twiceasfast

Intel Pentium II (P6) — CPU MUSEUM

Pentium Pro ◄ Pentium II ► Pentium III

Intel Pentium II © Intel

The Pentium II brand refers to Intel’s 6th-generation microarchitecture («P6«) and x86-compatible CPUs introduced on May 7, 1997. Containing 7.5 million transistors, the
Pentium II featured an improved version of the first P6-generation core of the Pentium Pro, which contained 5.5 million transistors.
However, its L2 cache subsystem was a downgrade when compared to Pentium Pros. In early 1999, the Pentium II was superseded by the almost identical Pentium III, which basically only added SSE instructions to the CPU.

 

In 1998, Intel stratified the Pentium II family by releasing the Pentium II-based Celeron line of processors for low-end workstations and the
Pentium II Xeon line for servers and high-end workstations. The Celeron was characterized by a reduced or omitted (in some cases present but disabled) on-die full-speed L2 cache and a 66 MT/s
FSB. The Xeon was characterized by a range of full-speed L2 cache (from 512 KB to 2048 KB), a 100 MT/s FSB, a different physical interface (Slot 2), and support for symmetric multiprocessing.

The Deschutes core Pentium II (80523), which debuted at 333 MHz in January 1998, was produced with a 0.25 µm process.

80523PX300512PE
Specification Details 		Show
80523PX333512PE
Specification Details 		Show
#
Specification Details 		Show

CPU-Z:

Intel Pentium II Deschutes 400 MHz SL357 View from the side: Pentium II 400 MHz

BX80523U400512E
Specification Details 		Show

The first Xeon-branded processor was the Pentium II Xeon (code-named «Drake«). It was released in 1998, replacing the Pentium Pro in
Intel’s server lineup. The Pentium II Xeon was a «Deschutes» Pentium II (and shared the same product code: 80523) with a
full-speed 512 kB, 1 MB, or 2 MB L2 cache. The L2 cache was implemented with custom 512 kB SRAMs developed by Intel.

CPU:

450 MHz
Specification Details 		Show

Sorry, but there is no specification for this CPU at this time due to missing SSpec.

Unlisted models:

 

  • Pentium II Klamath: 233 MHz, 266 MHz, 300 MHz
  • Pentium II Deschutes: 450 MHz

 

Sumbit a picture or contribute to the museum!

RETURN TO INTEL

RETURN TO THE MAIN MUSEUM PAGE

End of the «Golden Age».

History of Intel Pentium III processors. Part 1 / Habr

This article continues a series of articles about the history of processors and platforms for them, which began with stories about the Pentium Pro, Pentium (part 1 and part 2) and Pentium II. Intel’s «golden age» continues, the sky still looks cloudless, but clouds are already gathering over the horizon. Ready to take a trip back in time?

Our starting point is February 26, 1999. On this day, Intel introduced its new family of processors: the Intel Pentium III based on the «Katmai» core. From a marketing point of view — another breakthrough, the conquest of new heights of productivity and efficiency. But in technical terms, the new processors practically did not differ from the previous, not yet old Pentium II — the technical process (250 nm) was preserved, the frequencies slightly increased.

As for the deeper changes, the improved L1 cache controller and the updated FPU don’t really add up to an extra «one» in the name, do they? But the reason for which these executive changes were needed, according to marketers, was just right for the increase — a set of SSE extensions (Streaming SIMD Instructions), Intel’s answer to colleagues from AMD with their 3DNow!

Two from the casket and one wife

Indeed, Pentium II «Deschutes» and Pentium III «Katmai» are surprisingly similar to each other. Let’s tell you a secret — even the processor board in the cartridge is identical. Only the BGA chip soldered in the center of the board has become a little larger — 128 mm 2 instead of 113 mm 2 . Frequencies — Deschutes ended at 450 MHz, Katmai started with versions at 450 and 500 MHz. The same process technology and the similar complexity of the cores gave the same power requirements — 2.0 V (for older versions released later — 2.05 V). And, the crown trick of Intel — the new processor did not receive new chipsets when it was released.

As mentioned in the previous article, this resulted in a surprising situation — processors that are marketing related to different generations used not just one chipset (as is the case with 440FX and Pentium Pro/Socket 8 and Pentium II/Slot1) or even more and one connector (like the later Pentium and Pentium MMX), but did not require any hardware changes in motherboards! Only software BIOS update with new microcodes.

The Pentium III was easy to install in early Intel 440BX based boards such as Asus P2B, many OEMs released Katmai based computers where CPU replacement was the only change. There were even lines where Pentium II and Pentium III coexisted within the same model! At the same time, the frequencies of the latter have already reached 550 and even 600 MHz!

However, this story did not last long. First, Katmai appeared with the designation 533B and 600B, their difference from their predecessors was the support for a bus with a frequency of 133 MHz. This is where the first troubles await Intel, but the events of interest to us began to occur a little earlier.

In the «Digital Vintage» collection, early Pentium IIIs are represented by the ServerGhost Rotoscope P6/2 gaming station based on the Asus P3B-F board, one of the best representatives of the second generation of Intel 440BX boards. The system is equipped with a Pentium III 550 MHz processor, 512 MB of RAM and a 40 GB IDE hard drive. As a video accelerator, a 3Dfx Voodoo 3 3000 AGP with 16 MB of video memory is used — which has become a cult nowadays, but back then it was just a good video card that largely built on the success of its predecessors. Sound card — Creative SB Live! 5.1. Installed modem and network card. The computer is running Windows 98 SE and allows you to comfortably play any games of the second half of the nineties, including those specially created for the Glide API.

ServerGhost Rotoscope P6/2

Flash New

In early 1999, Intel released a highly controversial but truly innovative product, the Intel 810 «Whitney» chipset. It so happened that most users know it from inexpensive and not very productive computers of the early 2000s, when the chipset was already a frankly weak solution. But at the time of release …

Source
The top graphics solutions at that time were nVidia Riva TNT2, ATi Rage 128, Matrox G400 and 3dfx Voodoo 3 — the last truly successful product of the famous company. They carried on board from 8 to 16 MB of video memory (32 MB versions will appear later) and used a 128-bit bus to work with it.

Even then the cost of such video cards was comparable to the cost of a good motherboard. And then Intel released an integrated (the first in its history) solution with a built-in sound controller (it was necessary to add only a codec and an analog part) and a fairly serious 3D accelerator.

Its performance was at the level of the previous generation cards (there were no budget solutions as such at that time, the former leaders simply fell into this segment after the release of new ones) — approximately between Riva 128 and Riva TNT, depending on the implementation. The functionality is quite adequate at that time, including support for DirectX 6.0. The GPU itself (though there was no such term in those days) was based on the rather successful Intel 740 solution released the year before. The functionality was upgraded (i740 only supported DX5.0) and the core frequency — from 66 to 100 MHz.

The original used a 64-bit 100 MHz memory bus, giving up to 800 MB/s throughput using 4 or 8 MB of local video memory. The AGP 2x interface made it possible to access the system memory at speeds up to 533 MB/s, a little slower than the local one, which made it possible not to waste time transferring data to the video memory before processing — the card was optimized for this kind of operation.

In the integrated Intel 810 video core (it was designated i752, discrete video cards with this core were also produced in extremely small volumes), this idea was brought to the absolute! At system startup, the video core used 1 MB of system memory for the frame buffer, when loading drivers, this value could dynamically (hence the growth of Intel DVMT — Dynamic Video Memory Technology, which is used to this day) increase to 4 MB, allowing you to switch to higher resolutions.

Another 2 MB were reserved when starting the 3D application for the instruction cache and 4 MB for the Z-buffer. Total — up to 12 MB. All work with texture data was done using system RAM! In the Intel 810-DC100 version, video memory chips were installed on the motherboard, or rather «display cache» in Intel’s terms — they were used under the Z-buffer. The volume of such a «cache» was fixed — 4 MB, and access to it did not take up the bandwidth of the main RAM.

Working with memory resembles a simplified implementation of UMA (Unified Memory Architecture — unified memory architecture, as opposed to SMA — shared memory architecture in most other integrated chipsets). For the sake of memory access speed (and at the same time simplification, of course), the chipset lacks AGP support. The video core itself uses a kind of «Direct AGP», which operates at a speed equal to the speed of access to the memory system — the same 800 MB / s that the predecessor’s local video memory had.

The memory controller itself was significantly optimized — even when using the same memory as the video card, the version with the display cache was almost as good as the famous 440BX, which remained unsurpassed in terms of SDRAM performance. True, by that time the leaders of the graphics industry had switched to a 128-bit bus and a faster 200 MHz memory, which already quadrupled the bandwidth — despite a good start, it was no longer destined to keep up with them in the future.

Doesn’t sound very impressive? And now let’s imagine that, in comparison with the current situation, along with the motherboard, the user received a video card of the GeForce GTX1660 Super or RTX2060 level (for options with and without display cache) for free at the top level RTX3090. The Intel 810 made it possible to play modern games with medium graphics settings at a resolution of 640×480, which was quite popular at that time. True, such technological advancement had a negative impact on the creation of drivers for alternative operating systems, especially open source ones — up until the mid-2000s they were considered extremely unstable, but now video from Intel is the most stable option for Linux users.

But it wasn’t just the level of integration and embedded video that was unusual with this chipset. Intel has tested new technologies on entry-level chipsets before (for example, i430VX was the first to receive SDRAM support), so this time — Intel 810 became the first chipset with the so-called «hub architecture».

Basically, this meant the rejection of PCI as a bus for connecting the north and south bridges, which were now called hubs — GMCH / MCH (Graphics / Memory Controller Hub), in which, as before, the key components of the system and ICH were located (I / O Controller Hub) — I / O bus hub. Even for a flash drive with BIOS code, they came up with a new name — FWH (FirmWare Hub). The PCI bus for connecting bridges has been replaced by a proprietary bus with twice the bandwidth compared to the previous version. Accordingly, the PCI controller went to the south bridge, taking the place of ISA — it was completely abandoned.

The chipset was originally aimed at the market of low-cost computers and was seen as a pair to the Celeron in Socket 370. The main options were:

  • Intel 810 — the main version without a display cache and the «southerner» ICH with ATA/66 support.
  • Intel 810L — cheaper version with ICH0 supporting only ATA/33 and up to 4 PCI slots

    • The

  • Intel 810-DC100 is a variant with display cache support and a full ICH.

The chipset supported processors with a 66 and 100 MHz bus and SDRAM memory with a frequency of 100 MHz, while the memory controller became asynchronous — it became possible to clock the memory regardless of the processor bus. Most of the first generation boards on it were released in the MicroATX format and supported only Celeron (Socket 370) processors based on the Mendocino core. However, there were also boards, including full-size ones, with Slot1 for Pentium II/III and slot Celerons.

Soon a version of the chipset appeared with support for 133 MHz processor bus — Intel 810E/810E-DC100, while the memory controller tightly integrated with the video core remained operating at the previous 100 MHz.

Epic refusal

Intel planned to expand the use of the 133 MHz bus, the upcoming Coppermine were to be the main beneficiaries of its implementation. Fast bus processors also required a new chipset, the successor to the 440BX — this role was assigned to the Intel 820 «Camino» — with support for AGP 4x and a new type of memory — the infamous Rambus RDRAM (also known as RIMM — Rambus Inline Memory Module ).

Source
RDRAM itself was very good on paper — a narrow 16-bit bus at an extremely high frequency, from 600 to 800 MHz, should have simplified board layout and increased the speed of working with memory even higher. Intel was still counting on using system memory to store textures for video processing, so higher bandwidth was required than the processor bus required (1200-1600 MB/s versus 1066 MB/s).

In the process of fine-tuning the chipset and the memory itself, it turned out that due to high frequencies, the bus is extremely susceptible to various interference and interference, and the memory chips also get pretty hot during operation. To reduce the effect of interference, it was necessary to install special C-RIMM modules in empty slots. And anyway, instead of the planned 3 modules on one channel (and Intel 820 was developed as a single-channel chipset), at a frequency of 800 MHz, no more than two could work simultaneously — most of the boards ended up with 2 memory slots.

RIMMs were released in sizes up to 256 MB (later 512 MB modules were not supported by the chipset). As a result, with a maximum supported amount of up to 1 GB of memory, the actual amount was 512 MB at a frequency of 800 MHz or 768 MB at 600 MHz and in the «if you’re lucky» format. It was especially funny to see dual-processor boards with just a couple of memory slots and support for 512 MB versus 1 GB for the predecessor!

But the real problem was not the amount of memory, but its cost — the prices for RIMMs were a multiple, and sometimes an order of magnitude higher than the prices for the usual and slightly slower SDRAM. Already in the later stages of development, Intel had to create a classic «crutch» — the MTH chip, the Memory Translation Hub. Thanks to it, it became possible to create motherboards with SDRAM support and even hybrid ones — with RIMM and DIMM slots (without the possibility of using them simultaneously).

Finally, in November 1999, the chipset saw the light of day. It would seem that all his problems are solved. But no — the «crutch» turned out to be weak, errors were found in its work, which it was no longer possible to correct programmatically. All boards released using it were recalled — finding one of them these days is not an easy task. Besides, boards based on Intel 820 with RIMM didn’t gain any notable popularity either. The main application for them has become mid-range workstations with one or two processors.

Copper is not here!

Almost at the same time, in October, there was another very important announcement — new Pentium IIIs based on the Coppermine core were released. They were already produced according to the 180 nm process technology, which made it possible to reduce the core area and integrate the second-level cache on the chip. It received half the volume — 256 KB, but worked at the core frequency and had a wider bus for data exchange — 256 bits (Katmai and Deschutes worked with a 64-bit bus).

This cache acceleration perfectly compensated for the reduced cache size and allowed new processors to perform on equal terms or even outperform their predecessors. Coppermine came in 500 to 733 MHz options and used a 100 or 133 MHz bus. In addition to models in the Slot1 construct, models in the «seleron» Socket 370 construct were released.

Now it was a real Pentium III — finally, a critical mass of improvements worthy of an increase in the model number has been accumulated! Although, it is worth noting that the «large» cache on a chip has already been tested on the mobile version of the Pentium II «Dixon» (but the width of the bus connecting the core with the cache was 64 bits, not 256), which was the first to try on the new technical process. Another interesting fact is that at that time the transition to the use of copper on-chip connections instead of aluminum was discussed, and many thought that the name Coppermine (from English — copper mine) was chosen for a reason. But no — only chips manufactured according to the next generation process technology — 130 nm received a copper interconnect.

VIA takes over

According to the original plan, Intel 820 was supposed to occupy not only the upper, but also the middle segment. But due to the problems that arose, the Pentium III B/EB (that’s how the models with 133 MHz bus were designated) didn’t get a decent mid-range chipset. On the one hand, this fact undoubtedly made Intel’s life more difficult, and on the other hand, it gave rise to a variety of boards based on third-party chipsets, unprecedented since the days of the Pentium. SiS and ALi performed relatively modestly, and the most notable of them was, perhaps, SiS 630, an ultra-budget integrated chipset, in comparison with which even the Intel 810 seemed to be a very productive solution.

Source
By 2000, ALi had an interesting trump card up its sleeve — Aladdin TNT2, which had a full-fledged TNT2 M64 on board (even with the ability to install dedicated memory) and which could become the king of integrated graphics and discourage buyers from discrete mid-range video cards, but the company could not play this card could. Then VIA turned around in full — this time was the period of its heyday, its chipsets during the «reign» of Coppermine not only equaled, but also outperformed Intel in terms of market share.

Source
The first were Apollo Pro Plus (VIA 693) and Apollo Pro 133 (VIA 693A) — still budget, relatively slow chipsets. Both supported AGP 2x and ATA / 66, the first one worked only with 66 and 100 MHz buses and supported 1 GB of memory, the second one received support for the 133 MHz bus for the processor and memory and could already work with 1.5 GB of RAM. These chipsets proved to be very reliable and stable, but the speed of their work was depressing — the memory controller was let down. The controller itself, inherited from earlier models, was very flexible and could demonstrate good performance, but only a few board manufacturers gave access to these settings — by default, the boards were configured for greater compatibility.

Source
Apollo Pro 133A (VIA 694X/DP) became a breakthrough — most of the compatibility issues have already been resolved, AGP 4x support and the ability to work in dual-processor mode have appeared. Up to 2 GB of RAM was already supported, although most boards were equipped with three slots and could only work with 1.5 GB, respectively. Asus P3V4X (single-processor Slot1 board) and Asus CUV4X-DLS (dual-processor Socket 370 with 4 DIMMs and on-board network and SCSI controllers) are deservedly considered one of the best boards in this series.

Seeing the success of Intel’s integrated solutions and looking back at its own successful MVP4 solution for the Super 7 platform, VIA has released a whole galaxy of embedded video solutions. By that time, VIA had acquired a whole galaxy of chip developers — S3, Trident, IDT / Centaur, Cyrix. This even made it possible to provide 3D cores — a faster S3 Savage solution (VIA PM133) was available for budget home computers, and an ultra-cheap Trident Blade3D (VIA PL133 with AGP and PLE133 without AGP) for less demanding office solutions.

Despite the resolution of problems with the memory controller, chipset drivers remained a weak point. Moreover, the problem was not even so much in the drivers themselves, but in a very buggy installer, with an interface that allowed discrepancies. Especially many problems were caused by the AGP drivers, without which many interface functions did not work, for example GART, and the port itself worked much slower than required, and with the installed drivers it was possible to get an unstable system if you did not follow the sequence of actions strictly recommended by enthusiasts.

All these chipsets used VIA 596B/686A/686B southbridges connected via the PCI bus. They supported a standard set of functions — ATA / 66, USB1.1, ISA. Bridges of the 686 series received an integrated AC’97 audio controller and SuperIO controller functionality (monitoring, serial and parallel ports, etc.), version B was distinguished by support for the ATA / 100 interface. At the same time, it is believed that 596 bridges were more stable in operation, and for ATA / 100, the bandwidth of the PCI bus was a limiting factor, because other peripheral devices used it in one way or another, and already two ATA / 66 channels could use its bandwidth completely .

Source
In 2000, one of the most productive variants appeared — VIA Apollo Pro 266 — the first and only Pentium III chipset with support for DDR memory (up to 4 GB!). This chipset was already ideologically close to DDR chipsets for the AMD K7 platform; it used the proprietary V-Link bus for connecting bridges, which had a performance similar to the Intel interhub bus.

Old men are going to fight again

And so, in the camp of Intel, the flanks were adequately covered, but the center turned out to be inexcusably weakened. While preparations were being made for a solution that could have drastically corrected the situation, a veteran volunteer took the fight. It turned out to be the well-known Intel 440BX chipset. Intel never released the long-awaited version with official 133 MHz bus support, and certainly didn’t sanction the release of factory overclocked boards. Nevertheless, it was they who became the saving link, allowing them to wait for the release of a new platform.

Source
It has long been known that the 440BX overclocks perfectly — many second-generation boards announced in early-mid 1999 supported overclocking on the bus much higher than 133 MHz, provided the power required for Coppermine, and, most importantly, some of them already had the ability to install a PCI 1 bus divider :4, which ensured its normal operation. There was no possibility to add a divider for AGP to work correctly, the value 89 was the closest to the standard oneMHz (133/1.5) instead of 66 MHz.

On the one hand, not all video cards were able to work with a higher interface frequency, but the most popular and productive solutions from nVidia easily took these frequencies. On the other hand, due to the increased bus frequency, the lack of support for the AGP 4x mode was partially leveled out — the gain in bandwidth relative to the standard mode was about 30%, which added several percent to the overall performance in games and 3D applications.

In 2000, the third generation of 440BX boards appeared — as a rule, equipped with a Socket 370 connector, guaranteed to operate at a frequency of 133 MHz (the guarantee, of course, was purely on the manufacturer’s conscience). Often these boards were equipped with additional ATA/66 or even ATA/100 controllers, sometimes even with RAID support — from HighPoint or Promise. Often, high-quality sound solutions were installed, up to the integration of the basic models of SoundBlaster chips from Creative (CT5880), built-in network cards appeared. For fans of external expansion devices, basic models of boards were produced, for example, one of the best boards of its time, Chaintech 6BJM, the successor to the famous 6BTM, had neither sound nor an additional IDE controller on board.

Intel asymmetric response

Meanwhile, in the dark forges of Intel, VIA’s devastating response was being prepared — the new Solano chipset, which received the designation Intel 815 when it came out. The chipset turned out to be fast, reliable, but extremely controversial. Let’s look at the specifications and decide what was wrong:

  • Support for processors with Pentium III/Celeron with a bus frequency of 66/100/133 MHz, only single-processor configurations are officially supported.
  • Support for up to 512 MB of SDRAM PC133, up to 4 banks, only unbuffered memory without ECC, asynchronous memory operation.
  • Seriously redesigned memory controller (relative to i810) — the frequency can vary from 66 to 133 MHz, it can match the bus frequency or differ from it by 33 MHz up or down.
  • AGP 4x bus support (except 815G/GE).
  • Optional — integrated video based on the seriously improved Intel 752/754 video core, up to 8 MB of dynamically allocated DVMT video memory (Intel 815/815E/815G/815GE with video, Intel 815P/EP without).
  • Support for ATA/66 or ATA/100 (Intel 815E/EP with ICh3 Southbridge).
  • Supports 4 USB 1. 1 ports.
  • Built-in AC’97 audio controller.

Most users have decided that this is a slightly updated i810 and the main difference is the presence of an AGP slot. This majority, in fact, had enough chipset capabilities — everything you need is on board, there is AGP for a video card, there is enough memory (then mid-range PCs were often equipped with 128 MB of RAM, 256 MB was the lot of advanced machines).

Source
Advanced users noticed that the new chipset did not become the heir to the beloved 440BX, it could only replace the 440ZX! Of course, few people needed ECC support, and even more so register memory, but the ability to install 1 GB of RAM using up to 8 banks was a reasonable necessity for those who planned to use the computer for a long time and use it not only for basic needs. Many enthusiasts are accustomed to reasonable prices for dual-processor boards based on mid-range chipsets (remember — 440LX/BX were just mid-range chipsets, in the top methyl 440GX).

In reality, one small but very proud company — Acorp, in 2001 (when the Pentium 4 was already selling with might and main!) released a series of dual-processor boards based on the i815. These were the 6A815ED (built-in video and sound, IDE RAID controller), 6A815EPD (built-in sound and IDE RAID controller) and the most common — 6A815EPD1 (no additional controllers at all). The author of the article once had a computer with the latter as a home machine.

A server/industrial version of this board for installation in 1U rack cases is also mentioned, but they are not found on sale nowadays, perhaps only demo samples were released. Acorp 6A815ED is the only dual-processor board with video built into the chipset and using the SMA architecture with video memory allocated from the system memory (the second exception is the SGI Visual Workstation 320/550 systems mentioned in the previous article, which are not IBM PC-compatible).

But back to the chipset. Despite the similarity of characteristics with the i810, it differs from it almost as much as it differed from the i440 family! Most importantly, Intel abandoned UMA and returned to the classic SMA architecture, but with dynamically allocated video memory. This made it possible to use the standard implementation of AGP, but the performance of embedded video dropped somewhat, despite the improvement in functionality and improvements made. The possibility of installing a «display cache» was preserved — now it was video memory, it was executed in the AIMM format — a memory model for installation in an AGP slot.

Changes were also made to the memory controller, as a result of which it became somewhat slower than the i810, the difference was even more significant compared to the 440BX. Among the advantages of the memory controller is support for memory chips with a density of 256 Mbit, which reduced the complexity of choosing modules by 256 MB and … made it possible to set the maximum amount using only one module. Most i815-based boards had 3 memory slots, while among compact models there were often boards with two slots.

Motherboard Abit SA6
But there were also «four-slot» boards, for example, Abit SA6 — it started the rapid development of the «full-size» direction in the «Digital Vintage» collection. Another interesting board was released for HP Vectra VL ready systems in MicroATX format. This board used the full version of Intel 815 with integrated video, AC’97 codec and 3Com network controller are integrated on the board, there are two memory slots. A special feature is the optional sub-board, which is installed in an ATX-format case and connected to the main flexible cable. Together they form a full size ATX system. The subboard contains an ISA bridge manufactured by ITE and a couple of ISA slots for special boards, classic sound cards and modems.

The vast majority of boards based on the i815 series chipsets carried a Socket 370 connector. Only one board with Slot1 is known — Abit SH6 with 4 memory slots, integrated video and sound, and an optional additional IDE controller.

An interesting fact — some sources claim that the version of i815 without built-in video was not a rejection of the integrated version, but had a different chip with a smaller area.

Coppermine is represented by two systems. Classic entry-level workstation ServerGhost Rotoscope P6 / 3 based on the Abit SA6 board and inexpensive dual-processor system ServerGhost Catalina P6 / 3L TE based on the same Acorp 6A815EPD1. Both are equipped with Pentium III 733 MHz processors, 512 MB of RAM and a 40 GB hard drive (IDE). The video subsystems differ — a single-processor machine is equipped with a Matrox Millennium G450 DualHead with 32 MB of video memory, a dual-processor machine is equipped with an nVidia GeForce 2MX with the same amount of memory. Systems work under Windows Millennium Edition and Windows 2000 Professional, respectively.

ServerGhost Rotoscope P6/3

ServerGhost Catalina P6/3L TE

To be continued…

At this point we will probably break off. The history of the Pentium III turned out to be too full of events and it is impossible to fit them into one article without making it overly heavy. See you!

In the second part you are waiting for:

  • Chasing gigahertz
  • Stillborn Timna
  • Life after death
  • Journeys to Parallel Worlds

Pentium III 1400 [in 1 benchmark]

Intel
Pentium III 1400

  • Interface
  • Core frequency
  • Video memory size
  • Memory type
  • Memory frequency
  • Maximum resolution

Description

Intel started Intel Pentium III 1400 sales in December 2001. This is a Tualatin architecture desktop processor primarily aimed at home systems. It has 1 core and 1 thread and is manufactured in 130 nm process technology, the maximum frequency is 1400 MHz, the multiplier is locked.

In terms of compatibility, this is an Intel Socket 370 processor with a TDP of 31 W and a maximum temperature of 69°C

We don’t have any test results for the Pentium III 1400.

General Information

Information about the type (desktop or laptop) and architecture of the Pentium III 1400, as well as when sales started and cost at that time.

place in the performance rating The
Type Deck

294

Compatible

Information on Pentium III 1400 compatibility with other computer components. Useful, for example, when choosing the configuration of a future computer or to upgrade an existing one.

Please note that the power consumption of some processors can significantly exceed their nominal TDP even without overclocking. Some may even double their claims if the motherboard allows you to adjust the power settings of the processor.

  • Passmark
Passmark

Passmark CPU Mark is a widely used benchmark that consists of 8 different tests, including integer and floating point calculations, extended instruction tests, compression, encryption, and game physics calculations.