Real innovations on CPU

[thefumigator]

Hello, my name is Amalfi Marini, I'm from Uruguay, South America (yes, land plenty of those geode laptop for kids who are around in our schools) I'm new at this forum but not new at computers, I work as a PC reseller/repairman technitian and networking, and I'm studing computer science in University. I also work in the video editing area.

Just to discuss about the most innovative stuff introduced on CPUs in the last decades...
Taking out "faster processor", to me two of the most innovative technologies were MMX by Intel, and 32 + 64bit brougt by AMD (despite 64 bit took most its advantage in the server/workstation arena).

Why do I take the "faster processor" out of the matter, maybe because I think -if I'm right- that making a faster processor is just the path that any CPU manufacturer has to take, or do you consider it as a type of innovation? Well, maybe "the way to make a processor faster" would be the innovation to me (Athlon XP at the time, Core 2 these days), but its not a new idea overall. So what do you think?

Sorry to take this to -maybe- a philosophical level... but wanted to know the opinion of different master minds that may be around.

thanks for the replies and have a good day.
Fumigator.

 

[thefumigator]

Looking around... I've been on many forums, this one is quite the biggest of them all... to much violence on the threads to my liking tho...

Offtopic: Got almost the same config as yours, but my 1950GT is AGP. also got an (Creative) EMU-0404 sound card. I make music in my free time...

 

[starcraftfanatic]

to me the biggest innovation was probably AGP. Without it gaming wouldnt have gone anywhere.
On a sidenote, SSE was also important in my opinion.

 

[thefumigator]

Yes, I forgot SSE. well, its like saying MMX but from a different persepctive... it was intel's answer to 3dnow if I'm right.

 

[righteous]

I think rather than innovations, marketing tactic has played a larger role. Innovation has been rather stagnant in my opinion.

How many REAL innovations have there been in recent history? I think most of them have been covered already.
I think the biggest innovation is marketing and packaging.

They can take the same chip and enable/disable parts of it and call it something completely different, or rather shrink it, and do the same.
It's not innovation. It's marketing.

Just my take...

 

[zenmaster]

The only thing that matters about a CPU is how much data it can crunch. That is the job of a CPU.

You might want to add considerations such as power usage since the amount of work for the amount of power is also important.

A point to note: AMD did not "Invent" 64-bit CPUs.

Both Intel and other companies were building 64-bit CPUs for Windows long before AMD ever made one. AMD was the first to build 64-bit extensions for x86.

Even this was not really all that innovative but more marketing since there was little need or market demand.

 

[bc4]

I think the biggest innovation since I bought my P2 with mmx has gotta be the OS. Yes, windows Vista, Xp are so far beyond 95/98 (i forgot what it came with). Seriously, that P2 will run most stuff (yes, a lot slower) but having stuff like network file sharing to name just one

And I think that software is where most of the innovations will come in the future. Just me thoughts

 

[thefumigator]

@zenmaster:
No, I never said AMD invented 64bit cpus, but brought to us the first 32/64 bit one. It was like uhmm, a surprise maybe. And there was/is little market demand who really make full use of it, that's right. But, in the other hand common users will start to feel the 4GB limit in the next years. I had to upgrade to 2GB not just for gaming, but also the amount of ram I use when I'm editing video and images.

@bc:
I believe in software improvements and optimization, but as far as these times are rolling, i don't really think soft developers are focusing on that. BTW If software became more memory and CPU efficient then the [hardware] upgrade path would start to be lower.

 

[surrealdeal]

@thefumigator
honestly, how useful ARE those OLPC machines? what do you use them for?

 

[thefumigator]

actually, the kids feel the same way I felt when my 386SX came home which was my first computer and I was 12 years old. I started programming it in basic, or batch DOS, and it motivated at the time to a point where I had an answer to the "what will you do when you grow up?" question. Of course, gaming also took place and these kids have a lot of educational gaming

The situation in uruguay was quite bad specially after the 2002 crisis which was the worst in the whole history of this place. Our state schools are free and they also provide the basic needs in a more or less efficient manner, specially in the most poor places, where they had PCs even before the OLPC program.

Ok, you can't replace the parents of those kids, many don't even know them... but well, at least the computers will make them smile as long they learn how to use them, but as far as my experience goes, I didn't have any problem with my first computer, I think its the perfect age to start to use them and mess around.

The OLPC in uruguay have windows, as Microsoft Uruguay division gave them as gift, and the uruguayan gov accepted.

One Funny thing, the machines have a lever to charge the battery by hand and well, the wifi antenna is quite extraordinary. I've never had one in my hands but its known they sport geode processors, don't know at which speed.

 

[surrealdeal]

So, they play games, and train for a career in Software engineering?

 

[thefumigator]

Update:
This is the model http://en.wikipedia.org/wiki/OLPC_XO-1
And theres interesting info here: http://en.wikipedia.org/wiki/OLPC
Uruguay is nº1 purchaser with 100.000 units, and they will arrive this year as only 5000 out of 100.000 have been supplied in my place.

 

[bobwya]

Hi thefumigator,

Nice to see you aren't starting another AMD vs. Intel flame war... Like that is so yesterday!!

When I did Comp. Sci. back at University I loved the Computer Architecture books by Henessey and Patterson.. Well worth checking out if you can get hold of it!! My fav. class at Uni. by a long stretch!! I even did my final year project on a pipelined processor implementation (however it sort went out with sewage in the end )...

Link to interesting interviews with the author of the book I mentioned...
http://acmqueue.com/modules.php?name=Queuecasts&id=21
where they discuss in outline one of the main problems/challenges facing Comp. Sci. today (Parallel Programming of course).

What always sticks in my mind is the whole latency x bandwidth equation that can describe any computer component. Obviously the problem area being harddisks that are significantly slower than the memory they backup in the storage hierarchy. But obviously this is slowly being addressed by the various SSD devices that are on the market these days.

Strictly speaking the 64-bit extensions that AMD introduced are a bit of a kludge. The whole x86 architecture is a bit of a mess but I suppose the decoding logic size diminishes in relation to the massive on-die cache sizes these days... I always did like the good ol' Load-Store architecture but I guess it has fallen out favour a bit (as used in the Itanium and PowerPC for example). The cleaner the ISA the less likely you are to have horrible bugs in the initial CPU implementations and subsequent generations!! (Thinking of the Barcelona/Phenom of course!!)

Things like AGP, SSE, are not as impressive as the actual state machine/logic required to ensure the correct state is maintained at all times within modern x86 CPUs. Multiway instruction issue with x86 architecture is quite an achievement when instructions are executed out of order. This probably requires a Future buffer (or similar) to retire the state changes of pipelined instructions in order (and thus maintain processor state). As modern CPUs operate in multiprocessor/multicore environments then this state handling and interrupt handling logic becomes way more complex...

I do think the AMD on-die memory controller and inter-processor/core serial links are quite neat but these are strictly implementation issues... (Kind of like the old math co-processors, secondary off-die caches, etc.)

Bob

 

[thefumigator]

@surrealdeal
I guess... I wouldn't stop playing! but however, they use it for educational purposes and while at classroom they have to follow the teachers.
So its exactly the same way we used PCs at school. But this time it seems everyone will have that oportunity. Anyway, lookin the XO-1 you don't have much room for serious non-education gaming... counter strike anyone?
One point, software industry is nº1 exporter in Uruguay with several billions, I guess we need more programmers... but the gov and the opposition were all agree with the purchase of these lappys as "computer analphabetism" was considered a form of analphabetism. Maybe, when the rest of the laptops arrive and more schools starts to get them I could make a visit to see how do the kids behave.

 

[thefumigator]

@bobwya:
Thanks for the comment! I will try to search for those books as I will have computer architecture this year or the next. I'm actually quite late as I still have to take the modula2 course which I lost very stupidly this year. The modula2 projects are very hard to program, and they have to work well if you want to pass the course. The other reason I'm late is that I'm still having problem with some calculus... I find discreet math very interesting and to date, it was the most useful course after pascal.

But one thing is true, I love the career and I hope to get more into it.

LOL if you see the PCs I have, you will see that I'm both intel and AMD fan:

Old daddy: AMD Athlon 64 EE X2 4200 + ATI 1950GT AGP + single channel 2GB DDR2 + Asrock K8upgradeNF3 + Asrock AM2 module + EMU 0404 proffessional sound card + Encore PCTV capture card.

Old grand-daddy: Compaq EVO N400c laptop: Intel PIII 800Mhz + 256MB ram + 20GB HD etc

Old grand-grand-daddy: Compaq Presario 1245: K6-2 300Mhz + 160MB ram + 15GB HD etc

taken from a pharaoh tomb: Acermate 386SX/20 : 386SX 20Mhz + 2MB ram + 140MB HD etc...

 

[yomamafor1]

The adoption of using HK/MG for 45nm production.

 

[thefumigator]

which one (HK, MG) will remain is still in doubt I think... I will have to update in my readings
update: it seems both will be used http://www.pennwellblogs.com/sst/eds_threads/labels/HK+MG.php

 

[I800C0LLECT]

OoOE seems to be one of the most prominent recent progressions.

Otherwise, HK/MG is pretty great except it has nothing to do with the CPU logic, just it's execution.

 

[endyen]

OoOE seems to be one of the most prominent recent progressions.

Otherwise, HK/MG is pretty great except it has nothing to do with the CPU logic, just it's execution.

I dont think the op ment to limit to logic.
The truth of it is, that Intel is number one in process in the world. Even number 2 (you guessed it, AMD) have a good lead over most IC manus.
My personal feeling is that the ability to develop an image at smaller and smaller nodes, is the number one inovation in processors today. It is what allows things like transistors for branch prediction, IMC, large caches, etc. It also enables higher clock speeds.
At the OP. I thought mmx was a good concept, but a practical dog. It took 3Dnow to show Intel how to do it.

 

[yomamafor1]

which one (HK, MG) will remain is still in doubt I think... I will have to update in my readings
update: it seems both will be used http://www.pennwellblogs.com/sst/eds_threads/labels/HK+MG.php

HK/MG = High K dielectric and metal gate. Both have to be used at the same time in order to enjoy High K's benefits (10x less power consumption for transistors)

 

[mahoumatic]

Many many indeed...

In design level, there are so many new innovation for every single
micro-architecture: K7, Netburst, K8, Core. Not only in execution
instruction sets or the length of instruction they can process, but also
in many other aspect: integrated memory controller (K8), introduction of
different level of memory cache, hyper-threading, multi-core (in single
die), EIST/PowerNow! (or any other evergy consumption reduction
technology), Hyper-Transport inter-connector (K8), Macro-Ops Fusion
(Core), Rapid Execution Engine (Netburst), etc etc etc...

In production level: minification of production process, usage of high-K
material, design advancement of transistor, multi-core application,
LGA package technology, again, etc etc etc...

I agree that the addition of MMX (or other multimedia instruction set)
can be one important innovation, but other new technologies mentioned
above should not be overlooked...

I would not consider 32/64-bit execution itself as an innovation, but a
combination of 32/64-bit execution within a single processor/ process
thread can be...

 

[bobwya]

I dont think the op ment to limit to logic.
The truth of it is, that Intel is number one in process in the world. Even number 2 (you guessed it, AMD) have a good lead over most IC manus.
My personal feeling is that the ability to develop an image at smaller and smaller nodes, is the number one inovation in processors today. It is what allows things like transistors for branch prediction, IMC, large caches, etc. It also enables higher clock speeds.
At the OP. I thought mmx was a good concept, but a practical dog. It took 3Dnow to show Intel how to do it.

I think you are right - large dies affords new luxuries to the chip designers!!

All the things I studied while at Uni back in the early 1990's have now been implemented in more modern processors.. The flawed design of the P4 (northwood and prescott cores) with their deep Superpipelined design (28 and 35 stages). This then moved to the more modern superscaler designs of the Athlon and Core/Centrino architectures. The more cache you have on die the lower the size of the decode units - so you can throw hardware there again to improve performance. The Core architecture is sweet with the globing of both macroinstructions and microinstructions (although it should be noted that is only for 32-bit processes).

Of course the main innovation of having multi-core on a die is amazing - given the power of each core. Although one could argue that the new(ish) Sun Sparc chip is going more in the right direction for the long-term future (scalable massively multiprocessor architectures where the connection between cores is more important than their individual processing unit capability - the cores are only clocked at 1.4Ghz or so but there are 64 of them!!). All dependent on whether hardware and software developers can collaborate to design a reliable method that makes Parallel programming more mainstream!! This is so mainstream now that even Microsoft is taking a very active role in pushing new programming paradigms!!

I think that the whole HK,MG process is just an evolutionary change... We are just talking about a better way of implementing exactly the same old transistor function. Don't get me wrong this is an important development as performance per watt is the metric to beat. But scaling down feature sizes will inevitably lead to these developments to try and reduce the heating effect of having more transistors in a smaller area.

What might be more interesting on the process front in the near future is the Z-RAM:
http://en.wikipedia.org/wiki/Z-RAM
I believe AMD has licensed this technology and if they don't go bankrupt beforehand we will see it being implemented in their CPUs (next millenium?? )!! Would be nice to see whopping caches with a lower size cost!!

Bob

 

[thefumigator]

yes, AMD licensed the 2nd generation of z-ram some months ago I think, and then there were rumours that next AMD processors may include 4MB of caché using z-ram techno. Don't know when tho.

Hypertransport is a great innovation, even Nvidia uses it to connect NB to SB in their chipsets. It also made possible to mix chipsets brands like you can see in the hybrid asrocks motherboards. Also I have an K8upgradeNf3 which is a 754 mobo upgradeable to 939 and AM2 with a module (I have both) that is connected using HT.

There's a similar motherboard from ECS that allows you to use AMD or Intel CPUs.

 

[jonisginger]

On die cache was quite important, it used to be off die years ago, used to slow down CPUs a lot until someone had the idea of moving it on.


I have got the right idea of this thread - clever inventions around a CPU theme. I.e. die shrinking...


I'm ok aren't I?

 

[TechnologyCoordinator]

die shrinking...

Die shrinking is more evolutionary than inventive or revolutionary. You might have to use some innovative methods to achieve that shrink, but it's not like every 18 months someone says, "Aha! I've finally invented what we should do! A die shrink!"