Why haven't AMD or Intel tried older proc designs on new technolgy

thorin

Diamond Member
Oct 9, 1999
7,573
0
0
I'm curious why no-one has tried a Pentium 133 (or whatever) design in 0.13um technogy (running at like 5GHz er whatever), it would run amazingly cool and allow for some ultra compact applications (watches, PDAs, Ultra-Mini-ITX systems) etc.......

Thorin
 

AndyHui

Administrator Emeritus<br>Elite Member<br>AT FAQ M
Oct 9, 1999
13,141
16
81
The closest product I can see is the embedded low power Pentium MMX processors, built on the .25 micron process instead of the .35 micron process. Maximum speed is 266MHz (Desktop maximum was 233MHz, so not that much of an improvement), but the package can operate at extended temperatures of -40 to 115 degrees celcius.
 

Mingon

Diamond Member
Apr 2, 2000
3,012
0
0
Not only cpu's what about video cards, a kyro II on 0.13 @ 350mhz would be an excellent budget card. Or say a ti4600 on 0.13 with 256bit memory addressing, that would be pretty nippy still.
 

Lynx516

Senior member
Apr 20, 2003
272
0
0
The main problem would be the work involved in "porting" it for another proccess. I did some work at ARM and was "porting" an ARM7 core to a certain process. This proccess takes 6 weeks for the ARM7 core which is a tiny tiny core. For something as "big" as a P133 it woudl probably take the best part of a year to do. Also the P133 woudl not give as good performance/Power rating as the XScale or the Arm 11 core.
 

arcas

Platinum Member
Apr 10, 2001
2,155
2
0
Chip firms have and continue to do these types of experiments. One such experiment at Intel a few years ago involved re-implementing the Pentium I as an asynchronous/clockless design. According to techreview, this clockless Pentium performed roughly 3x faster than a clocked Pentium (no idea what the benchmark was). So while it was a success, the benefits from moving to a clockless design weren't enough to justify ditching their existing verification tools and designing new tools from scratch.

 

Fandu

Golden Member
Oct 9, 1999
1,341
0
0
Originally posted by: arcas
Chip firms have and continue to do these types of experiments. One such experiment at Intel a few years ago involved re-implementing the Pentium I as an asynchronous/clockless design. According to techreview, this clockless Pentium performed roughly 3x faster than a clocked Pentium (no idea what the benchmark was). So while it was a success, the benefits from moving to a clockless design weren't enough to justify ditching their existing verification tools and designing new tools from scratch.



Anyone have any linkage or other benchmarks run on a clockless cpu?
 

Shalmanese

Platinum Member
Sep 29, 2000
2,157
0
0
Why bother porting a 10 year old CPU with obsolete features when you can get something like an Xscale for lower power consumption and better performance.
 

AndyHui

Administrator Emeritus<br>Elite Member<br>AT FAQ M
Oct 9, 1999
13,141
16
81
Probably because an X-Scale processor doesn't run x86 code?
 

pX

Golden Member
Feb 3, 2000
1,895
0
71
Originally posted by: Lynx516
The main problem would be the work involved in "porting" it for another proccess. I did some work at ARM and was "porting" an ARM7 core to a certain process. This proccess takes 6 weeks for the ARM7 core which is a tiny tiny core. For something as "big" as a P133 it woudl probably take the best part of a year to do. Also the P133 woudl not give as good performance/Power rating as the XScale or the Arm 11 core.


I'll agree with Lynx, I have ported a few small ASIC designs from a .35 technology to .18, and it was no where near cut and dry, you can't just shrink everything, it's not that easy; and the work may not justify the result.

 

Sunner

Elite Member
Oct 9, 1999
11,641
0
76
Just an uneducated guess from an amateur, but I'd guess the P5 core wasn't meant to scale much further than it got(300 MHz Mobile IIRC).
5 stage pipeline, slower bus protocol, ancient support chipsets, etc.

Or no?
 

PrinceXizor

Platinum Member
Oct 4, 2002
2,188
99
91
As has been mentioned...they do, sort of...

But the main reason is that the application of such a chip is not really in the personal computer arena. Low power, low-cost embedded processors are used all the time in the PLC/microcontroller fields. However, the architecture of a Pentium 133 is probably not as conducive to the applications required in this field as the newly designed embedded parts that are currently serving this market.

MOST importantly, though, is cost. Development work is one cost, fab cost is quite another. Are we going to reduce the available .13u fab space to run low quantity, niche processors which (in the applications listed) would also have to be priced low, and take away fab space from our current gen CPU's?

While I'm sure porting the platform to other fab processes is probably a good R&D path, it doesn't seem to make a lot of sense from a normal, retail sense.

P-X
 

borealiss

Senior member
Jun 23, 2000
913
0
0
there's a lot of powersaving features that newer cores meant for the embedded market have that wouldn't make older cores that competitive, especially when you think about the cost of porting from one process to another. i doubt an ultra low voltage pentium would be that competitive with an offering from transmeta or motorola.
 

kpb

Senior member
Oct 18, 2001
252
0
0
As crazy as it might sound there are different levels of design in something as complicated as a computer chip. There's the highest level which probably best described as the logic level. This is where you design the high level features of the chip like how many pipe lines, how many processing units etc. At the lowest level you have the actual layout of the transisters on the chip. At this lowest level you can't just take and shrink the processor down to a smaller size. It might work it might not and if it does work it isn't all of a sudden going to let you clock it significantly higher speed. A good example of this type of thing is the athlons since they same basic chip has gone threw 3 or 4 difference processes. The latest one even has had 2 revisions to it that where functionally identical but the later added an extra layer to allow them to change the layout of things that allowed them to clock the processor at higher speeds. An original pentium redesigned for a current tech wouldn't just miraculously start running at 5 ghz. The processor it's self wasn't designed to run that fast. Just like athlons can't do the 3+ ghz that the p4s are currently doing. The p4 was designed with a long pipe line and even stages in the pipeline just to allow things to move around the chip specifically so that it could run at very high clock speeds. The pentium or althlon were not designed this way.

If you think about the complexity of a processor it's not too hard to see why this is the case. There are often times different ways to do the same thing with differing results as far as number of transiters needed, speed, reliability etc. Then you have to connect all these together so that all the appropriate components can talk to each other. All that gets very complicated.