486th from the Republic of China
The struggle in the modern market for x86-compatible processors has long been conducted exclusively between Intel and AMD. But it was not always so. Back in the days of the Pentium III, VIA competed (though not very successfully) with two giants, and during the time of the first Pentium and 486x, the “zoo” of x86 processors was quite diverse. Many will probably remember the name of a company like Cyrix, whose processors were sold including under the names IBM, Texas Instruments and ST Microelectronics. Someone might remember IDT, which, together with Cyrix, was later acquired by VIA, or Rise Technology, with their rather interesting mp6 chip. Often, geeks from the 90s recall the company NexGen, whose development formed the basis of the very popular at the time AMD K6 processor. But there was another name, by the way, quite common in Eastern Europe, including in the territory of the former USSR, which, nevertheless, is less often remembered - UMC.
UMC introduced a clone of the 486th processor shortly before the appearance of the first Pentium from Intel in 1993. In 1994, Intel filed a lawsuit against UMC for infringement of patents on a 486 processor, which is why the sale of UMC processors in the US was banned, which was marked directly on the processor case. As a result, after not too long disputes, the UMC company abandoned further production of x86 compatible processors, and the U5 processor in few modifications remained the only x86 UMC chip in history.
The processor itself was extremely interesting. In contrast to the practically indistinguishable from Intel 486 in performance clones from AMD, and the slower 486 from Cyrix, the UMC processor worked faster than the 486th from Intel at the same frequency. According to Wikipedia, UMC engineers concentrated on optimizing the microcode (although I can hardly imagine how to optimize the microcode as much as indicated in the article without touching the conveyor itself. Probably some new execution units were added, nevertheless) and achieved certain success. But more interesting is the fact that the company UMC provided assemblers with a “complex” product. At a time when AMD and Cyrix produced only the processors themselves, and Intel only recently began producing chipsets for their processors, UMC released almost everything that was needed to assemble a computer.
Some time ago I wanted to assemble a UMC computer. And this is what came of it:
The processor itself, UMC U5S-SUPER33. The processor contains 8 KB cache of the first level and does not contain a built-in block of operations with floating point, which has not yet become the standard in those years, but already present in expensive models of competitors. The processor without any problems earned at a frequency of 40 MHz, also at reduced voltage, 3.45 Volts instead of 5 Volts. In addition to this version of the processor, there were others: the U5SX was no different from the U5S, the U5SD had 486DX pinout, although, like the U5S, it did not contain a block of floating point operations, unlike the 486DX and U5D itself, from UMC. U5SF, U5SLV and U5FLV were produced in QFP for fixed mounting, the last two were officially designed for a supply voltage of 3.3 V. They say that rare U486DX2 occurs in nature with bus frequency multiplication, however, it is not possible to verify this statement.
The motherboard on the UMC UM8498F chipset has connectors for ISA and VLB buses, supports up to 128 MB of EDO memory, while the rest of the chipsets for 486x processors (with the exception of later revisions of SIS chipsets for the PCI bus) supported only FPM. The board has 256 KB of level 2 cache, which is also dialed by UMC chips. The cache is a funny story, albeit a fairly typical one for inexpensive motherboard for the 3rd socket. At that time, SRAM chips used for the cache were very expensive, and unclean manufacturers sold motherboards with a “dummy” cache. On the board, chipsets were installed, and the BIOS of the computer always showed the presence of a cache. Some boards, especially the ones made by PC Chips, did not even have tracks connecting the cache to the processor and the chipset, so it was impossible to even complete such boards with a cache. Others, like mine, were able to work with the cache, if it was installed, but the BIOS reported the presence of a cache in any case, even if the chips were not physically installed and the jumpers on the board were set to 0 KB cache.
The video card is built on the UM85C408AF chipset, it reports about 512 KB of onboard memory, although, judging by the soldered DRAM microchips, it is 768 KB. The card is designed for the ISA bus and is rather slow, its speed is very short for the 486th. When installing a more modern card for the VLB bus, for example, on the Cirrus Logic 5428 chipset, everything (games) works much more fun, however, Cirrus Logic is not a UMC. There are naturally video cards on the UMC chipset for VLB (on the UM85C418F chip), and even are found on ibey, but they ask a lot for them and often they are not working. I do not have such a card.
The 486 processor boards rarely had onboard I / O controllers, and my board is no exception. Therefore, to connect the disks, I use the multi-card on the UMC UM8672 / UM82C863 chipset. This card contains the controllers of all the main ports (RS232, IEEE 1284, the port for connecting the joystick, as well as the floppy disk controllers and 2 IDE ports) and is distinguished by rather strange delays in operation. For example, a similar card on the Winbond chipset detects a CD-ROM when MS DOS boots up in about 3 seconds. This board needs more than 10 seconds. In steel, there are no complaints about the board. The board is designed to connect to the VLB bus and supports 32-bit disk access, but not work in DMA mode.
The network card is built on the UM9003AF chipset. NE2000-compatible, 10 Mbps, “thin” ethernet. A popular choice at one time, and it works pretty well. There are versions of the board for twisted pair.
There is no sound card in the computer. I found a mention of the UMC chip synthesizer, a clone of some ancient Yamaha, but I did not personally see the chip itself, or even the datasheet. However, datasheets for UMC chips are a separate and rather sad story.
To understand whether the UMC U5S was as fast as they say about it, we need competitors. Unfortunately, I had no competitors without a block of floating point operations, so the U5S will fight against the Intel 486DX-50, which will be slowed down to 40 MHz:
AMD 486DX-40:
Cyrix 486DX-40:
In the case of Intel and AMD, the presence of a mathematical coprocessor does not affect the performance of integer operations (however, the SuperPi test cannot be used, at least I could not turn off the coprocessor under Windows), but in the case of Cyrix the situation is different: Cx486S-40 is different from Cx486DX-40 is also the size of the cache of the 1st level, and in general, is simply a “remake” Cyrix 486DLC, an upgrade processor for 386x motherboards.
This 486DLC, also at 40 MHz, albeit under the TI brand, will also take part in the “battle”:
However, it will work on a motherboard with another chipset, OPTi 495SX. OPTi chipsets were a very reliable solution (even Intel released motherboards with these chipsets for OEM), but rather slow, so 486DLC initially loses a bit, especially since in 386 the cache management was different: in 386x the cache was one , and is located on the board itself.
Well, for the sake of completeness, let's take the most common 386 processor at 40 MHz:
So it will be possible at the same time to understand how 486DLC justified the name "486". In the role of "the usual 386go" will perform AMD 386DX-40. Very popular was the processor in the days of the "early" 486x. They said that he could compete on an equal footing with the 486SX-25, being much cheaper (not only the chip itself, but also the motherboards for it) were much cheaper. There will be no 486 processor at 25 MHz in the test, but suppose that with increasing frequency in those ancient times when there were no frequency multipliers in the processors, performance also increased linearly. Thus, if the 386th shows more than 60% of the speed of 486, operating at the same frequency, we will assume that it was telling the truth.
There will be few tests, more precisely 3. Synthetic benchmarks are opened by the cachechk test, which tests the amount and speed of processor caches, and also demonstrates the speed of working with RAM:
This clearly shows the superiority of the UMC processor over its competitors. According to the speed of exchange with any level of memory, it is the fastest. Also, this is the only processor for which the benchmark was able to accurately determine the presence and speed of the 1st level cache. Although it is possible, the UMC chipset "wins" the UMC processor. But this is only an assumption, especially since the motherboards on UMC chipsets at the time of 486x were nevertheless considered to be one of the fastest, despite the processor. But the UMC processors were then, nevertheless, in the global minority. Cx486DLC-40 works very well with the Level 2 cache located on the board, however, paired with the OPTi chipset, it loses much of the “real” 486m in speed with working with RAM. But his “elder brother” behaved strangely: the difference in the speeds of the RAM and the 2nd level cache is so low that the cachechk did not “see” the cache.
In addition, only U5S has a line with the processor name: “UMC UMC UMC”.
And yes, cachechk determines the frequencies of all processors correctly. Exceptions: 486DLC (defined 45 MHz) and U5S (determined 849 MHz).
And the 3D bench test 1.0c, the predecessor of the 3D Mark, which does not use, however, not only 3D accelerators, but even a mathematical coprocessor, closes the “synthetics”. Only a fixed point, only hardcore.
And here the UMC is significantly ahead of its rivals, although in this test we can already see the “brake” video card. If you replace it with a normal VLB card, the result will grow impressively even for the 386th processor (yes, the later cards for the 386th supported the VLB bus. In general, the address and data buses on the 386DX and 486 processors are identical, and even “hybrid” cards exist where It was to install either a 386 or 486 processor, to choose from). 486DLC here is still significantly closer to the 486m processors than to the 386m, losing first only about 10%. Most likely, a similar result would have shown Cx486S-40.
And the most important thing for which everything was started - Doom. Used shareware version of the game 1.9. Test - demo3. Full screen, but visible bottom panel. High detail.
As it seems to me, here the U5S fell victim to a “native” video card. And in general, the 486 processor at 40 MHz may well play Doom. 10 FPS is still not level 486. And yet, even here, albeit a very modest one, U5S won. The 486DLC is exactly midway between the slowest 486x and the fastest (and only) of the 386x. Of course, you can put a normal video card and “retest”, but ... why? After all, it’s about UMC, and not about Doom at 486m.
So, the U5S was really an interesting processor. Really very fast 486m, and it is very well suited for a budget computer. The lack of a floating point operation block was not decisive: for the home user, the math coprocessor will become important after 2 years, with the release of the Quake game, for which the speed of any of the tested processors is still not enough (it will start even at 386m, yes, but play with pleasure will not work). And those users for whom the math coprocessor was important, at the time of the spread of the U5S, were gradually switching to the Pentium (yes, and there was a surprise in the form of FDIV BUG).
In general, it would be sad without you.
Good that you were.
The 386DX-40 was quite a competitor for the 486SX-25. From 50 to 70% from 486DX-40 it provided. Well, the 486DLC, nevertheless, is indeed the 486th processor in terms of performance, even the slowest, and is installed in the socket of the 386th.
UMC introduced a clone of the 486th processor shortly before the appearance of the first Pentium from Intel in 1993. In 1994, Intel filed a lawsuit against UMC for infringement of patents on a 486 processor, which is why the sale of UMC processors in the US was banned, which was marked directly on the processor case. As a result, after not too long disputes, the UMC company abandoned further production of x86 compatible processors, and the U5 processor in few modifications remained the only x86 UMC chip in history.
The processor itself was extremely interesting. In contrast to the practically indistinguishable from Intel 486 in performance clones from AMD, and the slower 486 from Cyrix, the UMC processor worked faster than the 486th from Intel at the same frequency. According to Wikipedia, UMC engineers concentrated on optimizing the microcode (although I can hardly imagine how to optimize the microcode as much as indicated in the article without touching the conveyor itself. Probably some new execution units were added, nevertheless) and achieved certain success. But more interesting is the fact that the company UMC provided assemblers with a “complex” product. At a time when AMD and Cyrix produced only the processors themselves, and Intel only recently began producing chipsets for their processors, UMC released almost everything that was needed to assemble a computer.
Some time ago I wanted to assemble a UMC computer. And this is what came of it:
The processor itself, UMC U5S-SUPER33. The processor contains 8 KB cache of the first level and does not contain a built-in block of operations with floating point, which has not yet become the standard in those years, but already present in expensive models of competitors. The processor without any problems earned at a frequency of 40 MHz, also at reduced voltage, 3.45 Volts instead of 5 Volts. In addition to this version of the processor, there were others: the U5SX was no different from the U5S, the U5SD had 486DX pinout, although, like the U5S, it did not contain a block of floating point operations, unlike the 486DX and U5D itself, from UMC. U5SF, U5SLV and U5FLV were produced in QFP for fixed mounting, the last two were officially designed for a supply voltage of 3.3 V. They say that rare U486DX2 occurs in nature with bus frequency multiplication, however, it is not possible to verify this statement.
The motherboard on the UMC UM8498F chipset has connectors for ISA and VLB buses, supports up to 128 MB of EDO memory, while the rest of the chipsets for 486x processors (with the exception of later revisions of SIS chipsets for the PCI bus) supported only FPM. The board has 256 KB of level 2 cache, which is also dialed by UMC chips. The cache is a funny story, albeit a fairly typical one for inexpensive motherboard for the 3rd socket. At that time, SRAM chips used for the cache were very expensive, and unclean manufacturers sold motherboards with a “dummy” cache. On the board, chipsets were installed, and the BIOS of the computer always showed the presence of a cache. Some boards, especially the ones made by PC Chips, did not even have tracks connecting the cache to the processor and the chipset, so it was impossible to even complete such boards with a cache. Others, like mine, were able to work with the cache, if it was installed, but the BIOS reported the presence of a cache in any case, even if the chips were not physically installed and the jumpers on the board were set to 0 KB cache.
The video card is built on the UM85C408AF chipset, it reports about 512 KB of onboard memory, although, judging by the soldered DRAM microchips, it is 768 KB. The card is designed for the ISA bus and is rather slow, its speed is very short for the 486th. When installing a more modern card for the VLB bus, for example, on the Cirrus Logic 5428 chipset, everything (games) works much more fun, however, Cirrus Logic is not a UMC. There are naturally video cards on the UMC chipset for VLB (on the UM85C418F chip), and even are found on ibey, but they ask a lot for them and often they are not working. I do not have such a card.
The 486 processor boards rarely had onboard I / O controllers, and my board is no exception. Therefore, to connect the disks, I use the multi-card on the UMC UM8672 / UM82C863 chipset. This card contains the controllers of all the main ports (RS232, IEEE 1284, the port for connecting the joystick, as well as the floppy disk controllers and 2 IDE ports) and is distinguished by rather strange delays in operation. For example, a similar card on the Winbond chipset detects a CD-ROM when MS DOS boots up in about 3 seconds. This board needs more than 10 seconds. In steel, there are no complaints about the board. The board is designed to connect to the VLB bus and supports 32-bit disk access, but not work in DMA mode.
The network card is built on the UM9003AF chipset. NE2000-compatible, 10 Mbps, “thin” ethernet. A popular choice at one time, and it works pretty well. There are versions of the board for twisted pair.
There is no sound card in the computer. I found a mention of the UMC chip synthesizer, a clone of some ancient Yamaha, but I did not personally see the chip itself, or even the datasheet. However, datasheets for UMC chips are a separate and rather sad story.
To understand whether the UMC U5S was as fast as they say about it, we need competitors. Unfortunately, I had no competitors without a block of floating point operations, so the U5S will fight against the Intel 486DX-50, which will be slowed down to 40 MHz:
AMD 486DX-40:
Cyrix 486DX-40:
In the case of Intel and AMD, the presence of a mathematical coprocessor does not affect the performance of integer operations (however, the SuperPi test cannot be used, at least I could not turn off the coprocessor under Windows), but in the case of Cyrix the situation is different: Cx486S-40 is different from Cx486DX-40 is also the size of the cache of the 1st level, and in general, is simply a “remake” Cyrix 486DLC, an upgrade processor for 386x motherboards.
This 486DLC, also at 40 MHz, albeit under the TI brand, will also take part in the “battle”:
However, it will work on a motherboard with another chipset, OPTi 495SX. OPTi chipsets were a very reliable solution (even Intel released motherboards with these chipsets for OEM), but rather slow, so 486DLC initially loses a bit, especially since in 386 the cache management was different: in 386x the cache was one , and is located on the board itself.
Well, for the sake of completeness, let's take the most common 386 processor at 40 MHz:
So it will be possible at the same time to understand how 486DLC justified the name "486". In the role of "the usual 386go" will perform AMD 386DX-40. Very popular was the processor in the days of the "early" 486x. They said that he could compete on an equal footing with the 486SX-25, being much cheaper (not only the chip itself, but also the motherboards for it) were much cheaper. There will be no 486 processor at 25 MHz in the test, but suppose that with increasing frequency in those ancient times when there were no frequency multipliers in the processors, performance also increased linearly. Thus, if the 386th shows more than 60% of the speed of 486, operating at the same frequency, we will assume that it was telling the truth.
There will be few tests, more precisely 3. Synthetic benchmarks are opened by the cachechk test, which tests the amount and speed of processor caches, and also demonstrates the speed of working with RAM:
This clearly shows the superiority of the UMC processor over its competitors. According to the speed of exchange with any level of memory, it is the fastest. Also, this is the only processor for which the benchmark was able to accurately determine the presence and speed of the 1st level cache. Although it is possible, the UMC chipset "wins" the UMC processor. But this is only an assumption, especially since the motherboards on UMC chipsets at the time of 486x were nevertheless considered to be one of the fastest, despite the processor. But the UMC processors were then, nevertheless, in the global minority. Cx486DLC-40 works very well with the Level 2 cache located on the board, however, paired with the OPTi chipset, it loses much of the “real” 486m in speed with working with RAM. But his “elder brother” behaved strangely: the difference in the speeds of the RAM and the 2nd level cache is so low that the cachechk did not “see” the cache.
In addition, only U5S has a line with the processor name: “UMC UMC UMC”.
And yes, cachechk determines the frequencies of all processors correctly. Exceptions: 486DLC (defined 45 MHz) and U5S (determined 849 MHz).
And the 3D bench test 1.0c, the predecessor of the 3D Mark, which does not use, however, not only 3D accelerators, but even a mathematical coprocessor, closes the “synthetics”. Only a fixed point, only hardcore.
And here the UMC is significantly ahead of its rivals, although in this test we can already see the “brake” video card. If you replace it with a normal VLB card, the result will grow impressively even for the 386th processor (yes, the later cards for the 386th supported the VLB bus. In general, the address and data buses on the 386DX and 486 processors are identical, and even “hybrid” cards exist where It was to install either a 386 or 486 processor, to choose from). 486DLC here is still significantly closer to the 486m processors than to the 386m, losing first only about 10%. Most likely, a similar result would have shown Cx486S-40.
And the most important thing for which everything was started - Doom. Used shareware version of the game 1.9. Test - demo3. Full screen, but visible bottom panel. High detail.
As it seems to me, here the U5S fell victim to a “native” video card. And in general, the 486 processor at 40 MHz may well play Doom. 10 FPS is still not level 486. And yet, even here, albeit a very modest one, U5S won. The 486DLC is exactly midway between the slowest 486x and the fastest (and only) of the 386x. Of course, you can put a normal video card and “retest”, but ... why? After all, it’s about UMC, and not about Doom at 486m.
So, the U5S was really an interesting processor. Really very fast 486m, and it is very well suited for a budget computer. The lack of a floating point operation block was not decisive: for the home user, the math coprocessor will become important after 2 years, with the release of the Quake game, for which the speed of any of the tested processors is still not enough (it will start even at 386m, yes, but play with pleasure will not work). And those users for whom the math coprocessor was important, at the time of the spread of the U5S, were gradually switching to the Pentium (yes, and there was a surprise in the form of FDIV BUG).
In general, it would be sad without you.
Good that you were.
The 386DX-40 was quite a competitor for the 486SX-25. From 50 to 70% from 486DX-40 it provided. Well, the 486DLC, nevertheless, is indeed the 486th processor in terms of performance, even the slowest, and is installed in the socket of the 386th.
Source: https://habr.com/ru/post/409695/