In the last decade alone, IBM scientists have announced one semiconductor breakthrough after another:copper technology, silicon-on-insulator, silicon germanium, strained silicon, and low-k dielectrics. All of these technologies came out of IBM’s fertile in-house research community. This prowess in modern chipmaking know-how didn’t come out of a vacuum — rather, it came out of the hermetically-sealed clean rooms of the most advanced R & D department in the semiconductor industry.
POWER to the people: A history of chipmaking at IBM
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
IBM semiconductors are great. It would be even better if IBM split their semiconductor business off and issued stock for it….. and then they buy out motorola’s recently split off semiconductor business…. Sun should switch to POWER. SPARC is great but more people perfer POWER. IBM even dosent show their PowerPC in the front like sun shows sparc and POWER is still more popular. It’s crazy and since they recently opened it up im all for it.
Ever since I switched to the Mac OS X platform I felt it was a great operating system, but it deserved better PowerPC processors from someone else than Motorola, I wished IBM would take over development, as it turns out they were already talking with Apple.
IBM has been really good with chip development, their starting to get to the point where Intel & AMD can’t catch up unless something really good happens really fast which I doubt would happen. It’s good to see a good history of where we have come from and where we are going. You can’t deny IBM makes some really fast chips that consume less power than the x86 chips wich they are designed to compete against. All I can say is good for IBM! Keep it up!
I have a IBM PC-RT in the attic ( 6150 ?, I do not remember ). It’s a strange computer :
– It’s a workstation. A huge tower with a 5V 40A power supply.
– It runs AIX 2.?
– It has some special busses for CPU / processing cards
– It has ISA-16bits slots for compatibility with IBM PC-AT cards.
– It has a i286 based coprocessor card for PC compatibility !
– It is as fast as a i486 50MHz
That computer was conceived as a bridge between personal computers and workstations. A total failure economically for IBM, but maybe one of the first RISC based workstations.
i have never owned any powerpc machine. i did use it before.
since powerpc arch is so good. why most of the world still using intel based?
This is but one of many reasons I am always buying IBM. In the world of computing everybody else is just playing catchup to Big Blue.
IBM do seem to keep towards the front of the march of progress chip-wise.
Other wonder teams like alpha have fallen into lore, and amd have yet to prove they can keep so close to the face.
Now x86-128 will be interesting to see 😉
“i have never owned any powerpc machine. i did use it before.
since powerpc arch is so good. why most of the world still using intel based?”
Because Microsoft doesn’t suppoert powerpc anymore and microsoft has got most of the world locked onto the x86 (intel/amd) chip. It’s more of a business decision than a technical one, if people REALLY had a choice we’d be using powerpc, quite frankly it’s the best bang for your buck there. PowerPC’s provide fast chips and don’t use as much power at the same time.
Because Microsoft doesn’t suppoert powerpc anymore and microsoft has got most of the world locked onto the x86 (intel/amd) chip. It’s more of a business decision than a technical one,
That all could be changing with .NET, maybe Microsoft will want to have Windows everywhere. Longhorn for apple machines?!!
“since powerpc arch is so good. why most of the world still using intel based?”
Simple: because MS does not want to port their software to PowerPC – many historical reasons in that one.
However, whenever compatibility with x86 is no limitation, other architectures have always been preferred. Intel x86 is actually extremely rare outside its established base. Technology-wise, it just isn’t such a good choice, despite its high volumes. Ok, nowadays power comsumption might have a lot to do with it, but it’s been like that even in the good old “MC68k days”, when that wasn’t really an issue.
We’re looking into using solar panel technology on our datacenter due to the energy costs. Would rather have PowerPC to cut costs. 🙁
Transmeta is great but pricey. That CPU (which has software built in) has the ability to run multiple archetectures (but they dont market it or sell it to the public). I never see any articles about it on OSNews… 🙁 Dont know how sparc compares sun has been slow to release new chips and are always pricey since they are with TI. I dunno about energy costs. The new prescott is absolutely crazy! it consumes so much power its unbelievable that anyone would use it in a server. I’ll stick with opteron.
about the powerpc thing, isnt that archetecture open to the public now like sparc is? would be great of ibm offered power boards to the public for sale like sun does. looks like the people want power but ibm is cautious to provide, kinda like how OS/2 just will not die
The fact that the PowerPC architecture is now public ( any link ? ) means that anyone can [try to] build a PowerPC in a custom ASIC or a FPGA without needing to pay IBM for it. I am skeptical about any competitor builiding PPC chips competing with the Motorola & IBM chips ( Fujitsu made Alpha, Cypress made SPARCs, … ), except for very special purposes. For example, the ESA ( European Space Agency ) sponsored a free SPARC implementation ( Leon ) for space based applications.
> best bang for your buck
I don’t think so. My AMD64 laptop costs as much as low-end powerbook, and is obviously faster (not to mention the bigger harddisk and bigger screen).
The processor might be cheaper, but that what surrounds it is more expensive.
http://www.eweek.com/article2/0,1759,1558748,00.asp
says its opened up, i dunno if thats the chip and stuff.. but sun’s sparc is open and people have sparc-based embedded cpus not much use in the pc arena. there are alot of sparc boards and computer server people than power it seems.. but seems power is more popular
> best bang for your buck
I don’t think so. My AMD64 laptop costs as much as low-end powerbook, and is obviously faster (not to mention the bigger harddisk and bigger screen).
The processor might be cheaper, but that what surrounds it is more expensive.
You are talking about the entire laptop and not the chip itself. The arguements for PPC aren’t really consumer based but technical.
Apple prices their hardware at a premiuem because they consider themselves a premium brand. Now AMD is start to do the same since they are no longer an “other-guy” but a significant brand that some people will pay for irregardless of almost anything (same as Intel is now). Look for AMD’s prices to rise (because the know that even if the lose a few users they’ll still make more $ over-all) and Intel’s to rise a little more because they are the premium brand here.
This should start to bite into the price advantage of the x86. Now VIA could step-up but they are already making a name for themselves in mini-ITX circles.
Now where does this leave the rest of the hurdles to dominance?:
(1) Another supplier, admittedly even if it is a puppet, but just for posterity sake of ‘competition’.
(2) Convincing OEM (Dell et al.) which would perfer sending money to Intel than IBM since Intel doesn’t sell individual boxes whereas IBM does, though market conditions can overcome anything (money is money)
(3) some consumers have learned to ask for Pentiums, this could be a problem but general consumer ignorance here is an advantage
(4) Existing x86 software, especially poorly ported ones. Even w/o this there be the same problem as with Linux instead on Microsoft though.
(5) GCC needs to be more thoroughly tested and optimized for PPC.
(6) This is probably the most important right now (though once solved something else will because some ad nauseum until dominance) is mainboard manufacturers. There will need to be more R&D and what not and how many are willing to risk it? Still consumer demand can push this too. (But demand for a product that doesn’t exist? This pretty much amounts to hype.)
(7) A hobbyist edition of the chip that is unlocked and massively overclockable that performs well in some meaningless benchmark. This is purely for the hype factor (or arguably as proof-of-concept for future performance). This is really nothing suprising to anyone thats heard of the AMD FX-series. Now even if this hobbyist edition loses the main benefits of PPC such as low power consumption and becomes a 200W monster it doesn’t matter because the regular chips will retain it and those massive-overclockers will use liquid-ntrogen for coolant or something else extreme
Technologically it is no argument. With all the money spent on x86 R&D spent on sparc64 or ppc64 we’d be light years ahead of now. Bless the free market on it’s lack of foresight that delegates thinking to the comman person!
Pessimism aside, the only advantages for PPC are for adoption are:
(1)Tech-hype, tech-people will recognize this as superior and if the common-person hears this often enough they’ll accept it as fact.
(2) Death of the megahertz myth, RISC chips generally run slower but more effeciently.
(3)Fortunately IBM is also a premium brand (lesser degree than Microsoft or Intel though) as is Apple (ditto)and PowerPC sounds cool. This should aid in hype making.
(4) Upcoming limits on process shrinkage will force effeciency to come from elsewhere, chip-design is a good place to look. However I heard the PPC is also on-chip transformed into micro-code like Intel/AMD so this might end up being a double-edged sword.
Hopefully x86 will go the way of AOL. Still alive, but only a bastion for the poor and unwashed masses… not even ‘masses’really anymore.
(2) Convincing OEM (Dell et al.)
(4) & (5) yes! this plagues sparc too! needs alot of work!
(6) Motorola offered PPC motherboards in the past but failed. I think it was a marketing failure.
“RISC chips generally run slower”
Isn’t PPC a RISC? It’s it alot faster than the average x86?
“Fortunately IBM is also a premium brand”
There’s some uncertainty surrounding IBM. They abandoning their AIX users and screwing over OS/2 Users. Saying they will support it yet then go and say. Hard to tell if they are committed. Same with HP but you pretty much know they aren’t. SGI is the same. The only provider that seems committed is sun micro.
If I’m not mistaken, can’t a chip be designed to be low power yet keep x86? (without emulation like Transmeta’s Efficeon and Crusoe which are low power) Isn’t x86 just an instruction set, not the overall CPU design.
Wrong.
Years ago ( “palaiai” ), when the RISC and CISC terms used to mean something, RISC CPU were simpler and needed a higher clock rate than CISCs for the same performance level.
Today, PowerPC are somewhat NRISCs ( Nearly Reduced Instruction Set CPU ) and x86 are ACISCs ( Awfully Complex Instruction Set CPU ).
The other difference between PPC and x86 instructions is that PPC instructions are bigger ( on average, 32bits vs 1 to 9 bytes ) and more powerful than x86 instructions, thus the lower clock-instruction rate.
If I’m not mistaken, can’t a chip be designed to be low power yet keep x86?
Pentium M (Centrino).
Both P4 and athlon are nowadays VLIW RISCs at the core, but have a hardware decoder to interpret the messy x86 encoding to the internal format. Both of them also have a microcode “emulator” to run the more esoteric/rare x86 instructions. That’s why “weird” instructions like bsr are so damn slow nowadays.
Both P4 and athlon are nowadays VLIW RISC
Incorrect. P4 and Athlon have a RISC core. VLIW is RISC but taken to the extreme, hence the reason why the compilers are so crap, because it requires the compiler to do more work opimising the code than one would if one was using a CISC or RISC. Two chips that are based on VLIW are Itanium and MAJC (SUN Microsystems).
Traditionally, microcode and intermediate internal code are quite VLIW as all the computational units of the CPU are used in parallel : One bitfield for the alu[s], for the FPU[s], for LSU, for the internal muxes, …
As modern CPUs use caches for the instructions feeded to the core, there is a limit for the width of the microcode.
Rather than a RISC, I think that the internal processor is some kind of orthogonal set of features programmed by the microcode.
Some DSPs shows their internal architecture in the opcodes, for example the Analog Devices “Sharc” has 48bits instructions allowing simultaneous operation of the ALU, Data Adress Generators, … Maybe the hidden cores are more like these DSP than a real RISC.
I worked at the RS/6000 line a long time ago when the power line came out. The power line belongs to Apple, Motorola and IBM. On paper the chip is/could be/should be a better platform that the x86. Saying that and saying that OS/2 is way better than any windowsxp and below is true also(although XP is almost a good copy of OS/2). It might be that IBM tried to keep the high end line too tight and not allow the smaller people to try to mass produce the products that compete. Tech companies need wide open spaces to roam and try to make a buck. Rules and limits and fees stop risk takers.
“RISC chips generally run slower”
Isn’t PPC a RISC? It’s it alot faster than the average x86?
By slower I meant in in terms of Hz. The average comsumer thinks that ‘more is better’ and don’t realize that they’d be doing apples to oranges comparisson. The higher-clock is a benefit for Intel’s marketing department as can be seen even in th x86 world with regards to AMD and their subsequent switch to ‘Power Ratings’. In terms of performance this is obviously not true. Apple’s G5 benchmarks while may questionably call them the fastest 64-bit desktops, but unquestionably do prove that PPC at least provides decent performance
“Fortunately IBM is also a premium brand”
There’s some uncertainty surrounding IBM. They abandoning their AIX users and screwing over OS/2 Users. Saying they will support it yet then go and say. Hard to tell if they are committed. Same with HP but you pretty much know they aren’t. SGI is the same. The only provider that seems committed is sun micro.
I meant with the average consumer. They will recognize IBM as a brand with that carries prestige (they’ve never heard of AIX so in this respect that point is moot). Sun less so (but still some through Java and probably some more in future from the JDS as that spreads), and SGI will most likely be unheard of. (What do they do anymore? Just package Itaniums into clusters?)Remember commodities are a number game so just techies aren’t enough, ‘word has to hit the streets’. IBM as a brand has, more so than Intel, and maybe even more than Pentium.
If I’m not mistaken, can’t a chip be designed to be low power yet keep x86? (without emulation like Transmeta’s Efficeon and Crusoe which are low power) Isn’t x86 just an instruction set, not the overall CPU design.
The Transmeta solution is not optimum since it requires the code to be transformed by firmware which slows things down (bad) and adds complexity(really bad), but I suppose it is feasible. I think that x86 does not necessarily necessitate high power consumption (see Transmeta), but rather invites. I haven’t done x86 asm in a while so someone correct me if i’m wrong: almost every instruction (ie. add, sub, and) requires at least one of the operands to be in a register, but the x86 has only 4 general purpose registers (ie that you can use; EAX, EBX, ECX, EDX methinks). This means that you always need to swap variable from memory to the chip. This is slow since main memory is slower than clockspeed. On other architectures such as SPARC all instructions but load and store (fyi: RISC is often called a ‘load-store machine’ for this reason) require both operands to be in registers, and (this is key) the SPARC provides 31 registers that can be used. Intel’s general overcome this by: 1 adding cache to serve as faster alternative to main memory (aka RAM), 2 creating very deep instruction pipelines, 3 ramping up clock speeds. Raising clockspeed directly increases heat, and I believe it does in n^2. 2 helps up to a point but then after that branch-mispredictions annihilate any performance increases. That said the PPC vs x86 is still fundamentally a social issue.
I would get a PCC board but They cost like $900 for 1ghz (with the CPU included) ATLEAST.
I can get a athlon board and CPU for just $50.
The problem is volume, the power people just need some big time investors willing to take a risk.
an 8MHz 68000-based Atari ST was consider FAST? And you could run applications in 512K-1Mb of RAM?
That’s where I’d like to be again, when people optimized games and applications down to fit and work nicely in that kind space.
Man, as soon as we hit 1GHz, the bottom TOTALLY fell out and the GigaHertz just POURED out! And what’s it all come down to? Does Windows “feel” or act that much faster? Do apps of today process faster, relative to the multi-Gigahertz system we now own?
I mean, I’m running a 466MHz G4 Digital Audio, running MacOS X Panther and I honestly can say, this system is PLENTY fast for all I do! Oh, sure, I’m getting a Radeon to replace my Rage128 Pro, but…
Keep yer silly 3.5GHz PC’s runing Windows XP… I’ll stick with my “horribly outdated and slow” 466MHz G4 running Panther… and be a lot happier at the end of the day.
Luposian
well as long as you are happy, technology progress has to stop.
In a few words said: Centrino saves power by slowing down itself. which means that u have to live with less power (in term of computing power) too. This doesn’t count. I guess…
At least
The key is not saving YOU time, the fast systems save developers time. Just look at .NET: so much of that programming model is based on the assumption that memory and processor clocks are virtually free, so the developer can write relatively easy but inefficient code. Sure, applications are less efficient nowadays and all that hardware power is not fully exploited, but the software makers win, the hardware guys win, and the users get more software, so who’s to complain?
I have leaved through the original PowerPC 601 ( the first PPC ) book years ago and read tourougly the 603 and 755 books… In that litterature, you see plainly IBM and Motorola logos but there is no mention to Apple. I don’t know whether Apple had really participated in the PowerPC design.
The PPCs have also various power reduction modes : Nap, Sleep, Doze, Awake…
[ Actually, the x86 have 6 general purpose registers : EAX, EBX, ECX, EDX, ESI and EDI, 4 of them can be sliced into 16 and 8 bits registers. The ESP and EBP registers are generally reserved for stack handling. ]
“[ Actually, the x86 have 6 general purpose registers : EAX, EBX, ECX, EDX, ESI and EDI, 4 of them can be sliced into 16 and 8 bits registers. The ESP and EBP registers are generally reserved for stack handling. ]”
Not true as the EBP can be used as a general register too (exception: when used to access memory it will use the stack segment register, but modern systems ignore segmentation so that is no problem). If one really wants to (and know that the OS is sane) even the stack pointer can be used as a normal register.