Modern PC processors are great. It is an engineering marvel, a piece of fused sand on which billions of miniature transistors and their interconnects are located. But, it seems to me, they are boring. There is no spark of revolution in them. No rapid change. Competitions in the processor industry between giants, Intel and AMD, now remind me of the Formula 1 race. It is interesting to follow, but they are too refined, there is always a clearly dominant team, breakthroughs and changes of leader do not happen too often, and technical regulations are very complicated. And lovers in the race road ordered.
But before that, the grass was greener, and the water was wet, and the processor races were more like an amateur rally. Yes, the processor industry has always been extremely knowledge-intensive, but newcomers often appeared and often disappeared.
I wanted to review the history of the 32-bit part of the x86-compatible processor line through the performance lens. It is 32 bit because, in principle, this architecture, although it is already not even at dusk, but behind a terminator, is still applicable with relatively modern operating systems, as opposed to 16 bit, and is quite old and interesting compared to 64x bit. What is not worth looking for in this article, so it makes any sense. The article is purely entertaining, with nostalgic pictures.
Important note about the results of tests SuperPi. What I have indicated in the text, as a result of 22x iterations of calculating 8M characters, is in fact the result of just the 1st iteration. I apologize for the inaccuracy and thank fedorro , keishi and lokkiuni for checking.
I want to immediately say that I will consider only the main desktop lineup from Intel, without Celeron (SX), Xeon (Pentium Pro) and Overdrive (RapidCAD, i487) lines, which themselves are also very interesting, but the material is too much for entertainment review.
3
So, it all began in 1985 with an Intel 80386 processor. For Intel, this processor was a breakthrough in several directions at once. In addition to what was obvious, namely the transition of the x86 architecture to 32 bits, and all the related improvements, it was the first x86 processor, the production license of which was not received by any of the clone processor manufacturers. If someone did not know, or does not remember, when IBM chose Intel as the supplier of processors for the IBM PC, one of the conditions of the deal was licensing the processor to several manufacturers. IBM did not want to depend on one supplier. Intel has distributed quite a few licenses, and clones of processors from 8088 to 80286 have produced many companies, from AMD to Siemens. Despite the fact that Intel was the developer of the processor, and the main costs of creating the processor were also borne by Intel, the company's share in the market was far from the biggest. In older IBM computers, for example, it is much easier to find an AMD processor than an Intel processor. This affected the financial condition of the company, and the management needed to do something. So it was decided not to license the 386th processor. Now the clone processor manufacturers had to invest time and money in the development of their chips. Many companies abandoned this market, producing an accelerated 286 for some time (Harris released a 286 processor running at 25 MHz, while Intel stopped at 12.5 MHz), and some decided not to leave the market and develop your design. However, now Intel has an advantage in time, for some time only Intel had a x86 x86 processor (quite a considerable time, for example, AMD Am386, due to litigation with Intel, came out only in 1991, after almost 5 years after the debut of 80386 and more than a year after the Intel 80486!).
And what about the IBM requirement of several manufacturers? Yes, Intel feared that IBM would refuse to use 80386, but the game was worth it, because by that time IBM PC-compatible computers were made by a huge number of companies. And the PC clone makers are playing Intel hands. Taking advantage of the delay by IBM, Compaq has released the world's first 32-bit IBM PC-compatible computer. Now the tone in the market began to set not IBM, but Intel. So the 386 is indeed a very important processor for the company itself. Revolutionary in many ways. Revolutionary so that the component supplier began to dictate its policy to manufacturers of end devices.
What was the 80386 processor?
If you believe Wikipedia, the processor was introduced in October 1985, and consisted of 275 thousand transistors placed on a substrate, an area of approximately 104 square meters. mm The processor was manufactured using the technology of first 1.5 microns, then 1 micron. Intel originally planned to debut a processor at 16 MHz, however, due to manufacturing problems, the processor debuted at 12 MHz. The 80386 processor did not contain any cache for commands or data; the common cache of the first level was located directly on the motherboard. The processor did not contain a block of floating-point operations; a separate co-processor 80387, released a little later, was used for this, because of which the very first Compaq had a socket for the previous generation co-processor, 80287, which worked asynchronously at a lower frequency and had a lower performance per cycle . The processor did not multiply the frequency and always worked at the bus frequency, from 12 to 33 MHz. The 80386 processor was produced in several packages, but the 132x PGA output package in a brown ceramic package was probably the most common PC at that time.
Let's play a little bit from 386-12386 / 387DX-12, 16 MB FPM RAM, 128 KB L1 cache on board S3 P86C801 1M ISA
Of course, finding a motherboard with a 386m processor rated at 12 MHz is quite difficult. It was a very rare chip in general, it was not planned, but was the result of a not very successful production process, which did not allow the 386 to debut at 16 MHz. Therefore, for tests, I used a processor at 33 MHz, slowing down its frequency. The motherboard I used for the tests itself is also very interesting. It contains the OPTi BUS, which is a kind of analogue (predecessor) of the VESA Local Bus. Very few devices were released for this bus, it was very quickly supplanted by the VLB. I only know for sure about the graphics accelerators based on the TSENG LABS chip for this bus, but I don’t have them, so I used an ordinary ISA card. The board is designed for 386 or 486 processors. In the case of using 386 in the socket for 486, you can install the coprocessor 387 (its “legs” fit inside the 486th socket), which I did. In addition, the board contains a frequency synthesizer, instead of removable quartz, used on earlier motherboards, and this synthesizer does not allow setting the frequency to 12 MHz. The least he can do is 20 MHz. I had to unsolder one leg of the microcircuit (it won't work out all of it, since it generates several more frequencies needed by the board) and put the active oscillator at 24 MHz (386 divides the external frequency by 2) in order to get 12 MHz on the processor. When loading the BIOS, it still writes about a 16 MHz processor, but all benchmarks define it as 11.9 MHz.
The board contains BIOS from AMI, quite common on 386 and early 486 computers. Personally, my setup screen for this BIOS setup is strongly associated with the 386 processor. 286, usually, had a BIOS setup simpler, and 486 I came across mostly either from AWARD, which later migrated to Pentium and went further through Pentium 3 and 4 to more complex systems on UEFI, or with graphical BIOS AMI, which imitated externally Windows
The computer BIOS contains a 2000 bug, since with the normal flow of time from December 31, 1999 at 11:59:59 pm, when going to the next second, the date changes to 1900. However, if you manually set the year after 2000, everything works well and no problems arise in the future.
The BIOS can see the hard drives up to 8GB in size, I did not use the overlay and just broke the 20GB disk into 1 partition, with a capacity of 504 MB: the maximum that this BIOS is capable of addressing CHS.
There are no peripheral controllers on the board, such as serial and parallel ports, floppy controllers and hard drives, so I used a multi-card with an ISA interface. The board does not contain a PS / 2 mouse controller, and the keyboard connector is an AT standard, which, however, can be easily fixed to PS / 2.
Installing Windows 95 took 2 hours and 33 minutes. Windows is extremely slow, despite 16 MB of RAM. The S3 video card with 1 MB on board allowed us to set the resolution to 800x600 with 16 bit color, or 1024x768 with 256 colors. Windows open rather slowly, all disk operations take considerable time, unarchiving is a pain. In general, Windows 95 is, of course, not about 386 at 12 MHz, but no one expected the opposite. I think installing even a very old Linux or FreeBSD will take forever.
I could not find an MP3 encoder program that runs under Windows 95. However, Audacity can convert wav to ogg vorbis. And this will do.
Track is 5 minutes long and 50.4 MB
Importing a track from a CD comes with a speed of 0.35x (CD audio), on a fairly modern high-speed ATAPI drive.
Open the file in Audacity. File import takes 14 minutes. Converting a track takes 17.5 hours. Obviously not the most successful machine for organizing a collection of compressed music.
MS-DOS Benchmarks:
Superscape 3D bench: 4.8 FPS
Chris 3D bench VGA 2,4 fps; SVGA 0.7 fps
cachechk: (there is no cache in the processor, 128KB SRAM on the board) read 8.3 MB / sec
Main memory: 2.9 MB / sec
Landmark 2.0: 19 MHz AT with 39 MHz 287
Landmark 6.0: 21 MHz AT with 31 MHz 287
Memspeed: 4Mx1 write: 8.3, read: 7.5, move: 6.8
Sysinfo 6.0: CPU: 13
DOOM -timedemo demo3 high detail: 2 fps
Quake timedemo demo1 320x200: 0.5 fps
Windows benchmarks:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: The game itself started, but the benchmark did not work.
SuperPI 8M digits, 22 iterations: 3 h 25 min 2.327 s
Let's play a little bit from 386-33386 / 387DX-33, 20MB FPM RAM, 256KB L1 cache on board CL GD 5428 1MB VLB
The processor ran at 33 MHz of bus speed. Did not contain built-in cache. Had a bus address and data width of 32 bits.
The board is very interesting (picture in the title). This is a late board for 386DX and 486 processors, on the OPTi 82C495SX / 82C206 chipset. With the installation of the 386th socket 486 can be used to install the 387DX math coprocessor. The fee came to me in working condition, albeit a bit sad. I had to remove the leaked RTC battery, clean the board from the leaked electrolyte and restore 2 tracks.
The board contains BIOS from AMI, quite common on 386 and early 486 computers.
The computer BIOS contains a 2000 bug, since with the normal flow of time from December 31, 1999 at 11:59:59 pm, when going to the next second, the date changes to 1900. However, if you manually set the year after the 2000th, everything works well and no problems arise in the future.
The BIOS can see hard drives up to 8GB in size, but I used the overlay to work with a full disk capacity of 20GB. The overlay works great with Windows 9x, and for NT and * nix it is generally not needed.
There are no peripheral controllers on the board, such as serial and parallel ports, floppy controllers and hard drives, so I used a multi-card with an ISA interface (for lack of a VLB). The board does not contain a PS / 2 mouse controller, and the keyboard connector is an AT standard, which, however, can be easily fixed to PS / 2.
The board contains expansion slots for ISA and VLB tires. Strictly speaking, the VLB appeared with 486 processors, and was a “continuation” of their local bus. However, the external buses of the 386 and 486 processors are almost identical, so in such a hybrid motherboard, the 386 processor can work with VLB devices.
Installation 95 took 1 hour and 12 minutes. The download does not last long, 1-2 minutes
Computer constantly crashes in Windows, unzipping is still not about him. All very long. Many programs were distributed in self-extracting archives, and they are self-extracting forever.
Track is 5 minutes long and 50.4 MB
Importing a track from a disk comes with a speed of 1.1x (CD audio), on a fairly modern high-speed ATAPI drive.
Open the file in Audacity. 16 minutes. Converting the track takes 5 hours and 10 minutes.
MS-DOS Benchmarks:
Superscape 3D bench: 14.7 FPS
Chris 3D bench VGA 6.9 fps; SVGA 2.2 fps
cachechk: (there is no cache in the processor, on the board - 256KB SRAM) read 23.3 MB / sec
Main memory: 7.7 MB / sec
Landmark 2.0: 50MHz AT with 105 MHz 287
Landmark 6.0: 58 MHz AT with 86 MHz 287
Memspeed: 4Mx1 write: 20.4, read: 20.9, move: 15.2
Sysinfo 6.0: CPU: 34.6
DOOM -timedemo demo3 high detail: 6.52 fps.
Quake timedemo demo1 320x200: 1.5 fps
Windows benchmarks:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 0.3 fps
SuperPI 8M digits, 22 iterations: 1h 13 min 06.761s
Internet under Windows 95 works conditionally. The network with TCP / IP is rising, you can put a bit of a stale opera, but modern websites still won't be able to see: 20 MB of memory, by modern standards, is nothing. Do not forget about security problems - Windows 95 has not been updated for a long time and there are a lot of holes in the system.
What can you use this system for today? Well, as a typewriter, it works. You can connect to ftp servers, and in the local network, theoretically, you can use it, especially if you install freeBSD or Linux instead of windows (although the latest versions of the 386 kernel do not support it). You can play old DOS games. Well, that's all. But the car itself is cool.
Of course, the 80386DX-33 was not the most powerful processor in the family. So, AMD was selling the 40 MHz version, successfully competing with the younger 486 (especially the 486SX, without the built-in coprocessor) in the budget segment. Cyrix (and, under license, several other manufacturers) released 386 with a cache, calling it 486DLC / SLC, IBM released 386 with frequency multiplication (up to 100 MHz) and a cache of 16 KiB, and Intel introduced the RapidCad upgrade to the market, consisting of two chips, the RapidCad-1 was installed in slot 386, which actually was a 486DX-33 processor package 386, and the connector 387 was a RapidCad-2 socket, in fact, just a socket plug, providing signal compatibility for the board. But for the main line of 386 Intel 80386DX-33 processors was the latest and fastest version.
four
The Intel 80486 processor did not bring such revolutionary changes in the industry as the 80386, but it was still really a breakthrough device. First, Intel abandoned the external math coprocessor. The coprocessor of the 486th was integrated directly onto the processor chip and, thanks to this, it worked significantly faster than its predecessor. The processor received a built-in low-latency cache with a capacity of 8 (later - 16) KiB, and a pipeline that significantly reduced the average execution time of instructions, although the indicator 1 indicator per clock was still far away. All this led to the processor expanding to 1.1 million transistors. With 1 micron technology, the processor chip occupied an area of 81 square meters. mm., against 39 square meters. mm at 1 micron version of the 386th. Also, due to the complexity of the crystal, the processor debuted in the 20 and 25 MHz versions, at about the same time as the 33 MHz version 80386 appeared. The processor version 486, calculated at 33 MHz, appeared a year later, a year later it debuted 50 MHz is a monster that did not work stably at all motherboards, due to the high bus frequency (the processors were still running at the bus frequency), which made Intel use frequency multiplication in subsequent versions, and a year later versions in 66, 50 and 40 appeared MHz with internal frequency doubling.
The final versions with triple frequency, branded IntelDX4 (without mentioning "486"), calculated at a frequency of 75 or 100 MHz, and also having 16 KiB of cache, debuted in 1994 through the write and write back versions, in the Pentium era, and were mainly designed for mobile use (Pentium chips had a 5V supply voltage, and consumed 15 watts of crazy power at the time, while 3.3V Intel DX4 consumed 5 watts, although they had more modest performance).
Let's play 486-20486DX-20 16MB FPM RAM, no onboard cache, Intel Classic E Expandable / VLB, CL GD 5428 1MB VLB
The 486 processor, later renamed 486DX due to the appearance of the “Celeron” 486SX, debuted at 20 MHz. I do not have such a processor, but there is a board with a sealed 486SX-25, which “can” 20 MHz, and a “coprocessor” 487SX for it. 487SX - a great product. Since the 486SX differed from the 486DX only in the absence of a mathematical coprocessor on a chip, the “coprocessor” 487SX was used to upgrade systems with 486SX. In fact, the 487SX was not a coprocessor. When installing the 487SX on the board, the old 486SX turned off completely and did no more work, and the 487SX was just re-tagged 486DX, and worked completely the same. So all is fair. So, the processor runs at a bus frequency of 20 MHz. Contains 8KiB of combined cache memory for instructions and data, however, additional level 2 cache memory can be installed on the motherboard. There is no such memory on this configuration, which was quite typical for budget systems of that time. The processor has 32-bit wide address and data buses.
CPU image taken from wikipedia
The motherboard is made by Intel, but the OPTi chipset is installed here: 82C495B1 / 82C392 / 82C206, unlike the 386/486 board, discussed above, the chipset contains an additional chip that performs the functions of a multi-controller, therefore the IDE, Floppy, COM, LPT connectors are soldered on the board. . The board only supports parity memory.
As already mentioned, the 487SX processor was no different from the 486DX, with the exception of one additional pin, which serves as a signal for unplugging the 486SX processor soldered to the board. In fact, this is one of the progenitors of Intel Overdrive processors. Part of the 486 Overdrive, released later, had exactly the same pinout, although they were no longer called 487. Such processors, unlike the original high-frequency 486, could be installed on old motherboards that did not support the high-frequency 486.
The board contains a BIOS from Phoenix, quite characteristic of Intel motherboards.
The computer BIOS contains a 2000 bug, since with the normal flow of time from December 31, 1999 at 11:59:59 pm, when going to the next second, the date changes to 1900. However, if you manually set the year after the 2000th, everything works well and no problems arise in the future.
The BIOS can NOT see hard drives larger than 1GB, however I used the overlay to work with a full disk capacity of 20GB. The overlay works great with Windows 9x, and for NT and * nix it is generally not needed.
The board contains integrated controllers for serial and parallel ports, floppy and hard drives, as well as PS / 2 keyboard and mouse.
Like the board for the 386DX-33, this board contains ISA and VLB tires. VLB works much slower here (this bus always works at the “external” processor frequency, that is, if in the case of the 386DX-33, the bus operated at 33 MHz, then in this case it works only at 20 MHz).
Installation 95 took 1 hour and 16 minutes. The download does not last long: 1-2 minutes.
Unzipping is still a pain, but overall, the computer is more responsive than 386-33.
So converting wav to ogg.
The same track with a duration of 5 minutes and a size of 50.4 MB
Importing a track from a disk comes with a speed of 1.05x (CD audio), on a fairly modern high-speed ATAPI drive. Which is a little strange, since, even considering that 386DX worked on a 33 MHz bus, and our 487SX worked on a 20 MHz bus, the 386 disk controller was still connected to the ISA bus, and it worked at 8 MHz. In general, the poor performance of the disk subsystem of this board is a mystery to me. But the facts are a stubborn thing. All disk operations on this machine take longer than on 386DX, but in general the performance sensations are more positive.
Open the file in Audacity.
File import takes 4 minutes 46 seconds. Converting the track takes 3 hours and 9 minutes.
MS-DOS Benchmarks:
Superscape 3D bench: 11.9 FPS
Chris 3D bench VGA 7.5 fps; SVGA 2.4 fps
cachechk: (in the processor 8KB cache for instructions and data, there is no cache on the card) read 16.4 MB / sec
Main memory: 9.8 MB / sec
Landmark 2.0: 54MHz AT with 142 MHz 287
Landmark 6.0: 76 MHz AT with 109 MHz 287
Memspeed: 4Mx1 write: 14.4, read: 15.1, move: 14.3
Sysinfo 6.0: CPU: 34.7
DOOM -timedemo demo3 high detail: 6.75 fps
Quake timedemo demo1 320x200: 1.9 fps
Windows benchmarks:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 0.4 fps
SuperPI 8M digits, 22 iterations: 0 h 48 min 47.407s
Total, despite the loss in speed of operations with memory, I believe that due to the lack of a cache on the board, and disk operations (which, probably, can be connected), in general, 486DX-20 is faster than 386 / 7DX-33, especially on floating point operations. It turns out that the 486th processor did 2-2.5 times more work per clock, thanks to the pipeline architecture and the integrated coprocessor. However, today, even such an impressive increase will not be able to change anything: the car is absolutely useless, although it is cool.
Let's play from 486-100486DX4-100 WT cache 32MB FPM RAM, intel lassic / PCI Expandable Desktop (Ninja), i420EX, 256K L2 Cache, S3 Savage4 64 bit 8MB, PCI
486DX4-100 WT cache 64MB FPM RAM, Soyo SiS 496/497, 256K L2 Cache, nVidia RIVA 128 PCI 4MB
This is a processor with an internal frequency multiplication. The external bus operated at 33 MHz, and the processor core - at 100 (33x3) MHz. Also, there were 2 processor modifications: with support for write-type cache memory, which does not cache the write to RAM, and write back type, which provided somewhat better performance due to caching write operations. The processor required 3.3 volts of power, which favorably distinguished it from the previously released Pentium 60 and 66 MHz, which required 5 volts and consumed a lot of power, while releasing a lot of heat (by that measure).
The board contains a second-level cache size 256KB.
The board contains expansion slots for ISA and PCI buses. PCI bus conforms to specification 2.0.
The board contains BIOS from AMI, which, in my opinion, is somewhat unusual for Intel motherboards, but perhaps this is the OEM version for some PC maker, AMI is more common on them (for example, Dell). Or maybe not.
I decided not to install Windows 95, especially since I do not have many licenses for this system (they are already quite difficult to find). Installing Windows 98SE took 53 minutes. Not bad.
The system does not fly, because of the built-in Internet explorer, the windows of the conductor are slowed down when opened, however, in general, everything is bright, the unarchiving is quite cheerful.
nVidia RIVA 128 did not start up with the Intel chipset, while working fine on another motherboard with the SiS chipset, so I ran the Windows benchmarks on the motherboard with the SiS chipset. According to the results of DOS benchmarks, there was no difference in performance between the i420EX and SiS 496/497 chipsets.
Windows 2000 refused to be installed on the Intel board, but it worked fine on the SiS board. At the same time, the installation of Windows 2000 lasts a terribly long time, the download takes a decent amount of time, but, interestingly, after loading the computer works somewhat faster than under Windows 98SE. And much more stable. Of course, Windows 2000 allows you to use more modern software than Windows 98SE, but it is still hopelessly outdated.
So converting wav to ogg.
The same track with a duration of 5 minutes and a size of 50.4 MB
Importing a track from a disk comes with a speed of 4.8x (CD audio), on a fairly modern high-speed ATAPI drive.
Open the file in Audacity. Importing a file takes 2 minutes and 16 seconds. Converting the track takes 45 minutes.
MS-DOS Benchmarks:
Superscape 3D bench: 66.6 FPS
Chris 3D bench VGA 46.0 fps; SVGA did not work.
cachechk: (in the processor 8KB cache for instructions and data, on the board 256 KB cache) L1 read 102.7 MB / sec, L2 read 40.0 MB / sec
Main memory: 28.8 MB / sec
Landmark 2.0: 360 MHz AT with 881 MHz 287
Landmark 6.0: 435 MHz AT with 682 MHz 287
Memspeed: 4Mx1 write: 30.9, read: 94.2, move: 30.9
Sysinfo 6.0: CPU: 216.7
DOOM -timedemo demo3 high detail: 40.75 fps
Quake timedemo demo1 320x200: 10.7 fps
Windows benchmarks:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 1.6 fps
SuperPI 8M digits, 22 iterations: 0 h 8 min 38.553s
Prime95:
under 98 earned version 25.9
786K: 17084.311 ms
896K: 20767.006 ms
1024K: 22809.029 ms
1280K: 32255.364 ms
1536K: 51413.929 ms
1792K: 90053.033 ms
IntelDX4-100 486, 386DX-33 386. AMD Am486DX5-133 ( Am5x86-P75 AMD-X5-133), Cyrix/IBM 586 100, 120 133 , Cyrix 6x86, Intel Pentium, Pentium Overdrive, 62 83 . , IntelDX4-100, Intel, 486.
P
486 Intel . , . Pentium- , , . 80501, — 80502 (3.3 7) 80503 (MMX-). Pentium.
Pentium "" Intel. , Socket 370 ( Pentium III) VIA , Cyrix, . , Pentium, Nx586 NexGen, , ( ).
Pentium 60 66 1993 , 3.3 80486, 75 100 , 80486DX2-66, , . Pentium 3,1 , 294 . mm 0,8 . Pentium 0,35 . 86, 2 , 80486, . 60 66 , "". , 3.3 (2.8 MMX-, ), , , . , Pentium , 60 66 . , FDIV-, , - ( , , ). Socket 5 (3.3 Pentium), . , Pentium Overdrive, "" Pentium II, , Pentium , , MMX-, . , "Xeon", Pentium Pro, P6, Pentium, Overdrive Pentium II Deschutes 333 333 -. Pentium II , Pentium, Pentium Pro ( Xeon). Pentium "Celeron"-, (8088 "" 8086, 386SX 486SX "Celeron" 386DX 486DX , 286, Pentium, ).
Pentium 60 66 , 100 486 , 486DX2-66. , 5 , . , .
, . Socket 4, Pentium 60 66 , Pentium Overdrive, 120 ( 60 ) 133 ( 66 ).
Pentium Socket 5, Socket 4, 3.3 . , Pentium MMX Socket 7, Socket 5 , (2.8 Pentium MMX 2.0 AMD K6-2+/III) - (3.3 ).
Pentium-60Pentium 60, 32MB FPM , Intel Premier/PCI (Batman), i430LX, nVidia Riva 128 8 MB PCI
( 486DX2, ) 16 , 8 .
256 - .
ISA PCI. PCI 2.0
Intel Dell BIOS Dell AMI.
Windows 98SE 50 . Not bad. , IO, , 486-100
, - , , . , , .
nVidia RIVA 128
, wav ogg.
5 50,4
2.8 (CD audio), ATAPI .
Audacity.
2 26 . 29 .
MS-DOS:
Superscape 3D bench: 62.5 FPS
Chris 3D bench VGA 49.6 fps; SVGA 15.0 fps
cachechk: ( 8 8 , 256 ) L1 read 82.8 MB/sec, L2 read 59.2 /
Main memory: 44.2 MB/sec
Landmark 2.0: 347 MHz AT with 1169 MHz 287
Landmark 6.0: 626 MHz AT with 1002 MHz 287
Memspeed: 4M1 write: 28.1, read: 75.3, move: 28.1
Sysinfo 6.0: CPU: 190.3
DOOM -timedemo demo3 high detail: 39,51 fps
Quake timedemo demo1 320x200: 16.6 fps
Windows:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 3.7 fps
SuperPI 8M digits, 22 iterations: 0 h 7 min 33.547s
Prime95:
98 25.9
786: 4579.911
896: 5400.328
1024: 6126.222
1280: 7912.804
1536: 9606.480
1792: 11466.055
? : . . , , - . , , . (, 60 ), , , , , , .
Pentium-233MMXPentium 233MMX, 128 SDRAM , 1024 KB L2 cache on board, Ali Aladdin V Chipset, S3 Savage4 64bit 8MB PCI
Chaintech BIOS AWARD. SDRAM 66 , . Acer Labs, Socket 7 AGP 2x. , AGP , , PCI.
Windows 98SE 20 .
wav ogg.
5 50,4
1 19 . 12,5 .
MS-DOS:
Superscape 3D bench:
Chris 3D bench VGA 131.8 fps; SVGA 34.4 fps
cachechk: ( 16 16 , 1024 pipeline burst ) L1 read 323.1 MB/sec, L2 read 186.8 /
Main memory: 137.3 MB/sec
Landmark 2.0: 1530 MHz AT with 4679 MHz 287
Landmark 6.0: 2439 MHz AT with 3927 MHz 287
Memspeed: 4M1 write: 84,9, read: 293,8, move: 84.9
Sysinfo 6.0: CPU: 794,6
DOOM -timedemo demo3 high detail: 79,71 fps
Quake timedemo demo1 320x200: 54.9 fps
Windows:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 8.7 fps. OpenGL — 30,5 fps.
SuperPI 8M digits, 22 iterations: 0 h 2 min 23.189s
Prime95:
98 25.9
786: 1366.171
896: 1621.092
1024: 1842.669
1280: 2244.453
1536: 2730.737
1792: 3308.046
, . Windows XP (?), FreeBSD/Linux - . , Raspberry Pi, , , - . , . .
Pentium 200 , 166 . MMX Intel SIMD , , MMX 233 .
Pentium 233MMX Intel . 300 , , , , Pentium MMX.
, , Intel. AMD K6-III, 256 , , 550 . Cyrix MII 285 (- 400), Rise MP6 250 (- 366), IDT WinChip 2 250 (- 300), , MMX, , 250 , .
2
Pentium II . Pentium Intel . , , ( ) . , . , Pentium Pro , , . , : . Pentium II Intel , , - . , , . , Pentium II Pentium Pro, 5,5 . , MMX 16 , Pentium Pro , 16 , 8 Pentium Pro. , - , , , , , .
, Intel. Pentium , 86 , . "" NexGen Nx586, 86 RISC- , . , Pentium Pro, 86 5 6 . , 86 , , , , 32 . 350 , — 250 .
Pentium II 1997 233 .
Pentium II , Pentium MMX. , Pentium II 233, Pentium MMX 200 . . , Pentium II P6, Pentium Pro, 150 , , Xeon, , SX-Celeron, .
Pentium II-233Pentium II 233, 128 SDRAM , 512 KB L2 cache on CPU card, Intel 440EX Chipset, ATI Rage3D IIC AGP
Mitac, OEM . BIOS AWARD, . 440EX, , 440LX. 256 , , 66 . LX , AGP, EX , . AGP ATI Rage3D IIC, , OpenGL Direct3D. , 3Dfx Voodoo, 2D , . .
. , . 440EX ATI Rage3D . , . .
Windows 98SE 20 . Pentium 233 MMX.
wav ogg.
5 50,4
30 . 4 42 . -.
MS-DOS:
Superscape 3D bench:
Chris 3D bench VGA 131.8 fps; SVGA 34.4 fps
cachechk , . L2 read 244.5 /
Main memory: 92.7 MB/sec
Landmark 2.0: 1457 MHz AT with 4522 MHz 287
Landmark 6.0: 2944 MHz AT with 3653 MHz 287
Memspeed:
Sysinfo 6.0: CPU: 585,5
DOOM -timedemo demo3 high detail: 79,62 fps
Quake timedemo demo1 320x200: 54.3 fps
Windows:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 13.2 fps.
SuperPI 8M digits, 22 iterations: 0 h 1 min 19.320s
Prime95:
98 25.9
786: 849.73
896: 1011.639
1024: 1123.401
1280: 1424.224
1536: 1726.712
1792: 2066.404
Pentium II Pentium 233MMX DOS, Windows. , , , Quake Quake 2 — , DOS, Windows, 32 ( Pentium Pro, Pentium II 16- ). , , . , DOS Pentium II , : .
Pentium II-450Pentium II 450, 512 SDRAM , 512 KB L2 cache on CPU card, VIA Apollo Pro Chipset, ATI Rage 128 Pro GL 64bit 16MB PCI
IBM , , BIOS IBM. BIOS , .
Klamath (Pentium II 233), Deschutes (Pentium II 450) . , Rage 128 , Rage3D IIC, VIA , 440EX.
Windows 98SE 10 . Windows XP — 1 20 .
wav ogg.
5 50,4
15 , . Quickly. 2 35 .
:
Superscape 3D bench:
Chris 3D bench VGA 322.1 fps; SVGA 63.7 fps
cachechk , . L2 read 441.4 /
Main memory: 127.5 MB/sec
Landmark 2.0: 2794 MHz AT with 8671 MHz 287
Landmark 6.0: 4342 MHz AT with 7023 MHz 287
Memspeed:
Sysinfo 6.0: CPU: 1122,5
DOOM -timedemo demo3 high detail: 101,48 fps
Quake timedemo demo1 320x200: 100.7 fps
Windows XP:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 24.01 fps.
SuperPI 8M digits, 22 iterations: 0 h 00 min 39s (XP)
Prime95:
XP 28.10
1024: 687.192
1280: 866.363
1536: 1053.788
1792: 1264.316
, . . Windows XP , . , .
Intel, GTL+, Pentium II, Cyrix, VIA, VIA 6 , , Pentium III, Pentium II , , ( 100 ) Pentium III, , , .
3
Pentium III (, Katmai) Pentium II (Deschutes). 1999 , Pentium II 450 . SIMD MMX, SSE. MMX , SSE Intel. , ( , ) , 2 . , Deschutes. Katmai 250 , Deschutes. , Pentium II, , , 350+ Pentium II, Pentium III. Pentium II 450 Pentium III 450 .
Pentium III can hardly be called the next generation, after the Pentium II processor, however, it is extremely interesting. The first Pentium III had even fewer innovations compared to the Pentium II than the Pentium MMX had compared to the classic Pentium. In Pentium MMX, besides the additional MMX instructions, the cache increased, which is why the processor on the old code, without using MMX, worked somewhat faster; In addition, the Pentium MMX was Intel's first desktop processor, in which core power and I / O power became separate. The Pentium III brought only a set of new SSE instructions in addition to the Pentium II, so it was supposed to be called the Pentium II SSE. Then, of course, everything became somewhat more interesting. With the transition to the core, the Coppermine Pentium III received a second-level cache integrated into the processor chip, which, although decreased by 2 times compared to its predecessor, was much faster due to a much wider bus and operating frequency. With the advent of the Tualatin core, the Pentium III increased the cache volume in the second level, returning to 512KB (although there were also 256KB models).
Let's play with Pentium III-450Pentium III 450, 512MB SDRAM RAM, 512 KB L2 cache on CPU card, VIA Apollo Pro Chipset, ATI Rage 128 Pro GL 64bit 16MB PCI
The board was made for IBM and contains a typical, very high-performance IBM BIOS. This BIOS allows you to allocate system resources of devices and shows conflicts, in addition to the usual functionality.
Installing Windows 98 took less than 10 minutes. This is probably the last system on which I will install Windows 98. Even here it doesn’t make much sense. Installing Windows XP took 1 hour and 20 minutes.
Convert wav to ogg.
The same track is 5 minutes long and 50.4 MB in size. Converting a track takes 2 minutes 36 seconds.
MS-DOS Benchmarks:
Superscape 3D bench: Value too high
Chris 3D bench VGA 324.5 fps; SVGA 63.8 fps
cachechk did not detect a second-level cache processor, and did not measure its speed. L2 read 442.0 MB / s
Main memory: 144.3 MB / sec
Landmark 2.0: 2794 MHz AT with 8671 MHz 287
Landmark 6.0: 4342 MHz AT with 7005 MHz 287
Memspeed: NOT EARNED
Sysinfo 6.0: CPU: 1122.5
DOOM -timedemo demo3 high detail: 101.75 fps
Quake timedemo demo1 320x200: 101.0 fps
Windows XP benchmarks:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 23.121 fps.
SuperPI 8M digits, 22 iterations: 0 h 00 min 36s
Prime95:
under XP earned version 28.10
1024K: 602.094 ms
1280K: 767.146 ms
1536K: 930.788 ms
1792K: 1126.599 ms
The findings will not differ much from those for the Pentium II, since the test results do not differ. Yes, Prime 95 sped up a bit (I guess because of SSE). But, this is probably the only measurable change. Perhaps, you should not pay attention to the results of the cachechk - the L2 cache, the RAM controller and the RAM itself are the same in both systems.
Let's play with Pentium III-1400Pentium III-S 1400, 512MB SDRAM RAM, 512 KB L2 cache on CPU, Intel 815T Chipset, ATI Rage 128 Pro GL 64bit 16MB PCI
Fee released for HP. I do not know by whom. Very similar to Intel, but maybe someone else. The board contains a Phoenix BIOS, however, customized for HP.
Installing Windows XP took 48 minutes. Quite normal. In general, installing an OS is more disk speed checking than the system as a whole.
Convert wav to ogg.
The same track is 5 minutes long and 50.4 MB in size. Converting a track takes 46 seconds.
MS-DOS Benchmarks:
Superscape 3D bench: Value too high
Chris 3D bench VGA 576.4 fps; SVGA 88.2 fps
cachechk did not detect any cache in the processor and did not measure its speed.
Main memory: 722.8 MB / sec
Landmark 2.0: 8520 MHz AT with 41820 MHz 287
Landmark 6.0: 18000 MHz AT with 24000 MHz 287
Memspeed: NOT EARNED
Sysinfo 6.0: CPU: 1605.4
DOOM -timedemo demo3 high detail: 112,147 fps
Quake timedemo demo1 320x200: 235.7 fps
Windows XP benchmarks:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 58.39 fps.
SuperPI 8M digits, 22 iterations: 0 h 00 min 14.125s
Prime95:
under XP earned version 28.10
1024K: 179.283 ms
1280K: 208.583 ms
1536K: 255.533 ms
1792K: 306.780 ms
The architecture of the P6 scaled out great. With the growth of frequencies from 233 (not remembering about 150 MHz Pentium Pro) to 1,400 MHz, that is, 6 times, the performance increased by about the same. Somewhere less, somewhere even more. Overall, impressive. Probably, such a large growth was not achieved by any other Intel architecture. And if you remember that Core2 are distant relatives of the P6, then this is perhaps the longest-playing architecture in the x86 world.
Yes, we missed the real Pentium III, Coppermine (picture in the section header). The fact is, the Pentium III is not one processor. It seems to me that the last Pentium III, on the Tualatin core (130 nm), differs from the first Pentium III on the Katmai core, much more than the Katmai differs from the Pentium II on the Deschutes core, or even from the very first P6, Pentium Pro. However, the name is the same, so we assume that this is the last and fastest Pentium III. It appeared already deep during the Pentium IV, and was far superior to the latter in performance at the same frequency. But the frequencies of the Pentium IV were much higher. But this is not about that, but about the wonderful core Coppermine (180 nm), despite its name, using aluminum, rather than copper interconnects. This was the real Pentium III, which came back into the socket and competed with the magnificent AMD Athlon processor, which, ironically, had copper interconnects. Athlon eventually won the battle abroad at 1 GHz, but Coppermine was still more massive. And it was at his time came the heyday of 3D accelerators on the PC. Of course, 3Dfx Voodoo Graphics saw the light in 1996, 3 years before Coppermine, but it was during Coppermine that a truly odious battle broke out in this market, worthy of the processor wars of the first Pentium era, but without a clear leader, represented by 3Dfx Voodoo3, nVidia RivaTNT2, Matrox G400, ATI Rage128, S3 Savage3D, 3DLabs Permedia, Rendition Verite and others. During Coppermine, GeForce and Radeon brands were born. It was fun. It was Coppermine turned the gatherings in the "computer clubs" in eSports. Of course, not himself, just technology ripened. But it happened in his time.
four
The Pentium 4 processor debuted at the end of 2000 at 1.4 and 1.5 GHz. Subsequently, after about 2.5 months, version 1.3 GHz appeared. This was, as far as I know, the first time that a slower processor was added to the mainline x86 Intel processors after the debut of a faster model. The processor came out a few months after the Pentium III Coppermine 1.133 GHz, which, however, was extremely unstable at the official frequency, and was recalled. Subsequently, a more stable core was released, but this is not about it. The Pentium 4 was first released for the Socket 423 connector and, as with the first Pentium 60 and 66 MHz, it was immediately clear that this connector is a temporary solution, and future processors for the 478 connector will be massive. In addition, the vast majority of 423 socket boards used RDRAM, fast but terribly expensive. (there were models on the VIA chipset for DDR SDRAM, but it was hard to find them because of the semi-legal position of this chipset, and by the time they appeared in mass sales, processors for Socket 478 were already being sold). However, the choice of processors for Socket 423 was still quite wide: from 1.3 to 2.0 GHz in 100 MHz increments.
The processor was manufactured according to the technology of 180 nm, contained on an area of 217 square meters. mm 42 million transistors, a built-in cache of the second level of 256 KiB and a long pipeline, which allowed to overclock the clock frequency, but had a huge overhead for reloading the execution units in case of incorrectly predicted branching. Under this processor, it was necessary to be able to program, the "usual" software was executed on it more slowly than on the equal-frequency processors of the previous generation.
But the frequencies could be driven, and with this Intel competed with AMD, forcing the latter to return to the rating in the processor name, since the Athlon could not catch up with the Pentium 4, but the performance was about parity.
Pentium 4 was, perhaps, a record holder in the types of memory supported by chipsets for this processor. The processor started with the i850 RAMBUS DRAM chipset. Although support for this type of memory was also available for the Pentium III with the i820 chipset, RAMBUS DRAM could truly unlock the potential only with a high-speed 4x100 MHz Pentium 4 bus. However, the memory turned out to be excessively expensive, and first the competitors, and then Intel itself, began to offer alternative solutions . Intel was bound by a contract with RAMBUS and for a long time could not sell chipsets with DDR SDRAM support, therefore, offered as an alternative only a rather modest single-channel i845 with support for 133 MHz SDRAM. After the release of RAMBUS, Intel offered a scattering of chipsets for DDR SDRAM, first single-channel i845D, then - two-channel i865 and i875. The i915 chipset, which supports PCI Express and DDR2 SDRAM, formally, was designed for 64-bit Pentium 4 in the Socket LGA 775 version, however some manufacturers offered boards for Socket 478 with such a chipset.
Let's play with Pentium 4-1300Pentium 4 1300, 512MB RDRAM RAM, 256 KB L2 cache on CPU, Intel 850 Chipset, ATI Rage 128 Pro GL 64bit 16MB PCI
Fee released for HP. The manufacturer is most likely ASUS. The board contains a BIOS Award, however, customized for HP. During operation, it is heated on the ALL board. The processor, the north bridge, memory, a little - the south bridge. A warm video card also adds entropy.
Installing Windows XP takes 64 minutes. To record Pentium 3 far away. But everything, again, rests on the disk.
Convert wav to ogg.
The same track is 5 minutes long and 50.4 MB in size. Converting a track takes 46 seconds.
MS-DOS Benchmarks:
Superscape 3D bench: Value too high
Chris 3D bench VGA 460 fps; SVGA 81.6 fps
cachechk did not detect any cache in the processor and did not measure its speed.
Main memory: 1188.4 MB / sec. Here it is, the power of RAMBUS DRAM.
Landmark 2.0: 14243 MHz AT with 4274 MHz 287
Landmark 6.0: 18000 MHz AT with 6215 MHz 287
Memspeed: NOT EARNED
Sysinfo 6.0: CPU: 1533.33
DOOM -timedemo demo3 high detail: 99.85 fps
Quake timedemo demo1 320x200: 184.6 fps
Windows XP benchmarks:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 52.6 fps.
SuperPI 8M digits, 22 iterations: 0 h 00 min 19 s
Prime95: This is where the Pentium 4 shines with new facets. The SSE2 command set was recognized and adopted by the benchmark.
under XP earned version 28.10
1024K: 69.80 ms
1280K: 92,469 ms
1536K: 114.418 ms
1792K: 182.09 ms
Let's play with Pentium 4-3400Pentium 4 3400, 2GB DDR400 SDRAM RAM, 1 MB L2 cache on CPU, Intel 865PE Chipset, embedded graphics
The latest Pentium 4 for Socket 478 is the latest purely 32-bit Intel processor for desktop computers. The processor was manufactured using the 90 nm technology on the Prescott core and supported the Hyper Threading technology, which made it possible to “load” the idle execution blocks using the “virtual” second core. The technology is still used today in multi-core 64-bit Intel and AMD processors. The Prescott core was further developed, down to 3.8 GHz, but these were already 64-bit processors for the LGA775.
In general, the classic 478 socket scheme is DDR memory and AGP bus, however, towards the end of the platform’s life, manufacturers limitedly offered motherboards with support for both DDR2 memory and new PCI Express bus with SATA interest (these innovations “debuted” on the LGA775 platform with i915 chipset ).
The processor is able to "devour" up to 115 watts of energy. And dissipate, of course, about the same in the form of heat. In general, Socket 478, despite its long presence in the market, left behind mixed feelings. At the beginning of his life, he was strongly hampered by the fast and cold AMD Athlon, at the end - 64-bit AMD Athlon64, which are very fast and much less demanding of power. However, this concerned not only the 478 socket. The entire Pentium 4 / D line was not the most successful for Intel, although according to the megahertz it is impressive so far - the best-selling processors still tread around the frequency of the test subject, despite his age : the old man was introduced in February 2004, almost 15 years ago!
Installing Windows XP took an hour. It turns out that they intended that the disk is as slow as it was on the previous system. Yes, he is actually the same.
Convert wav to ogg.
The same track is 5 minutes long and 50.4 MB in size. Converting a track takes 20 seconds.
MS-DOS Benchmarks:
Superscape 3D bench: Value too high
Chris 3D bench VGA 442,7 fps; SVGA 71.9 fps.
cachechk did not detect any cache in the processor and did not measure its speed.
Main memory: 3104.4 MB / sec.
Landmark 2.0: 3104.4 MHz AT with 33153 MHz 287
Landmark 6.0: 35000 MHz AT with 24000 MHz 287
Memspeed: NOT EARNED
Sysinfo 6.0: CPU: 1400
DOOM -timedemo demo3 high detail: 138.57 fps
Quake timedemo demo1 320x200: 254.9 fps
In general, the benchmarks under DOS lost their meaning somewhere at the level of the Pentium II, but some results, with a stretch of speed, showed up to the Pentium III 1.4. Now they are completely meaningless.
Windows XP benchmarks:
Quake 2 timedemo1, map demo1.dm2: software 640x480, full screen, NO stripe alpha: 102.3 fps.
SuperPI 8M digits, 22 iterations: 0 h 00 min 5 s
Prime95: This is where the Pentium 4 shines with new facets. The SSE2 command set was recognized and adopted by the benchmark.
under XP earned version 28.10
1024K: 27.085 ms
1280K: 35.73 ms
1536K: 44.21 ms
1792K: 53.076 ms
Pentium 4 survived several cores, the technical process was reduced from 180 to 65 nm, instructions were replenished with SSE2, SSE3, a pseudo-core appeared in the form of Hyper Threading, and finally, after the release of AMD Althon 64, the Pentium 4 was compatible with the EM64T compatible with 64-bit mode. Ah, yes, the frequencies have grown from 1.3 (1.4) to 3.8 GHz (for the 64-bit version, the “clean” 32-bit Pentium 4 stopped at 3.4 GHz). 3.8 GHz will long remain unbeaten bar for future generations of processors.
And then…
Well, that's all. Then the processors became 64-bit. Yes, they still correctly execute the 32-bit code, but ... First, a dual-core Pentium D appeared. It was just 2 Pentium 4 on the same substrate. And at first he gave some advantages, not so much desktop software, ready to use 2 cores, and the frequencies had to be reduced compared to single-core versions. In addition, Intel, after AMD, was forced to enter ratings in the names, instead of frequencies, since the frequency, even for Intel processors, ceased to uniquely determine the performance of the chip. Then came the awesome Core 2, which returned Intel's performance and their heirs, the first generation of the Core i. And somewhere, then, whether I finally matured, or it became objectively boring, but everything became somehow ordinary in the processor world. Without a rally.
Nevertheless, here are all the test results in the form of graphs (the results are normalized by the Pentium 4-3400):
DOS tests:
Windows tests:
