Paul comments?
IDT-C6 Processor. I have written this quick assessment to get through the companies hype in their press release and help people understand the facts. I will post this on AMD/CYRIX/INTEL threads.
The C6 processor is an achievement. They have designed the simplest chip possible to achieve Pentium MMX compatibility. They have done their homework analysing everyday programs to find out which instructions are most important and have made sure those instructions are done in hardware not microcode. This is how they have achieved the high Winstone 97 rating. I do expect that other benchmarks will reveal that for many other applications the C6 chip be much slower compared to the K6/M2/Pentium MMX.
The chip is very small and IDT's costs will be low. Its only 88mm2 compared to 90 for the classic pentium. It sounds impressive when you hear that it has 5.4 million transistors but when you look at it - its really not that impressive. Why?
4 Million transistors are used for the 64K cache and support. (SRAM cells)
1.4 Million other transistors (Including 320,000 used for the ROM microcode)
This is why the C6 is so small. Cache can be packed more densely compared to logic transistors. It is also another reason why the C6 is a low power chip.
The Business Winstone97 mark shows the 200 Mhz chip being faster than a classic pentium and about the same speed as the Pentium MMX 200. But I expect many other benchmarks will show the chip to be poor in a number of areas. IDT has partly achieved this rating with their large 64K L1 cache. The K6/M2 do beat the C6 in the Business Winstone however.
FUTURE OF C6:
IDT will have a very hard time keeping this processor improving in speed at the same rate as AMD/CYRIX/INTEL. Why?
One way to improve speed would be to add a second Integer execution unit. However if they did this they would need to handle out of order execution and other complexities such as pipeline hazards that they have not had to handle with only 1 integer pipeline.
They will attempt to get the chip running at higher speeds. But..
One way to raise the speed of a CPU is to break the pipeline into more-smaller stages. The idea is that the latency in each stage is lower and thus the processor can run at a higher Mhz rating. However if IDT used more stages in their pipeline they would then need to add better Branch prediction logic or they would take a bigger hit when an incorrect branch was taken.
The point I am making is that to improve the CPU by either executing more instructions per clock, or even simply running it at much higher Mhz ratings would require MANY of the complex technologies that AMD/CYRIX/INTEL have used in their high performance processors. This would give them a logic transistor count similar to the other processors and would mean their processors would use more power. IDT management have clearly stated that they are only going for the bottom end of the market and given the above analysis they could not achieve the kind of performance of the other high end x86 processors anyway.
It was interesting to note that in their roadmap they intended to include the L2 cache in the same package, something that AMD and Cyrix might also do.
Market:
The question is will the C6 have a market?
The C6 processor will start to become available probably in late Q3 (August/September). The chip will sell for $150. This is certainly achievable given their costs will approach $30 per chip once they obtain a high yield. They have not indicated whether the chip will ship at 200 Mhz on Day 1 or if it will start at 150 Mhz and work its was to 200 over a period of months. But lets assume it can run at 200 Mhz in September.
The K6-PR2-166 / M2-PR2-166 will probably also sell for about $150 by this time and I expect that these processors will compete with the 200 Mhz C6. Thus from one end there will be cheap M2/K6 and maybe Pentium MMX 166.
From the other direction there will be the MediaGX running at 166 Mhz by then, and later this year at 200 Mhz. The MediaGX includes a video card/Sound card and probably a modem by then. It is also less than $100 although not socket 7.
The best market the IDT chip will have is for very low cost systems that wish to use socket 7 to retain the option to upgrade to faster processors. The only other option would be a socket 7 motherboard with Video/Audio integrated onto the motherboard - this would allow the C6 chip to compete with the MediaGX on the low end to some extent.
** BELOW I HAVE INCLUDED IDT's COMMENTS ALONG WITH MY INTERPRETATION **
"Instructions are issued one at a time in program order. Instructions are executed and retired in order. Cache misses stall the pipeline until the data is available for the requesting instruction.
there is limited branch prediction: only a special 8-entry Call/Return stack"
Read: The IDT-C6 processor is the simplest implementation possible to give Pentium MMX compatibility. It has only 1 Integer execution unit compared to 2 for all competing chips. Instructions can only be executed In Order.
"The IDT-C6 processor issues only one instruction per clock but most integer instructions and most floating-point instructions can execute in parallel"
Read: Per clock cycle the IDT processor can only issue one instruction. But if, for example, a FPU instruction takes 2 or more clock cycles - on the second clock cycle an Integer Instruction could be issued. This means that at best only 1 Instruction or less per clock cycle will be achieved.
"Minimal hardware is provided for functions that are not heavily used or that are not critical to performance in the target environments. The IDT-C6 processor optimizes total system performance by generally optimizing for highest clock frequency, even if this means reduced CPI for
some functions versus more complex processor designs."
Read: Our Business Winstone mark is good, but FPU stuff and some other functions are probably bad. This is why IDT has only released the Business Winstone97 results. Also some applications would perform much worse than competing processors.
Other than Intel/AMD/Cyrix and IDT there are a number of other companies working on x86 compatible processors. I have included the names of these companies below:
* International Meta Systems Inc., soon expects to finish a chip design that could compete with the Pentium II microprocessors Intel just announced.. The Meta 6000, as it's called, will have about the power of the fastest Pentium II, Mr. Smith says. A later version will also have a more unique attribute: Besides Intel-compatible software, the new chips also will be tailored for new programs written in Sun Microsystems Inc.'s Java programming language.
* SGS-Thomson Microelectronics Inc.
* National Semiconductor: 133 Mhz Pentum Integrated Processor (Like MediaGX) sell for $30.
Secretive start-ups:
* Rise Technology Co.
* Transmeta Corp.
* Metaflow Technologies Inc.
-------------------------------------------
Regards,
David Taylor |
