Why is 4K and the new Playstation 4 so vague?

[Wayne Morellini]

Past experience, over and over again.

Without going into those articles there already seems to be a number of issues. Two different process technologies can be apples and oranges. Intel's is apparently renowned low powered, love to see a good arm on that. The last I saw, a long while ago now, the APU's only did clearly better on gpu hungry tasks, stripped off the gpu and chipset to compare cpus then. In no way are these naive comparisons out there are the same as comparing a mass multi core chip to a gpu less i core. Another thing is the benchmarks favour a x86 over ARM. The architecture of x86 was hooned to C Unix and windows long ago, big mistake. Windows and Linux compatibility and performance, and the performance of programs designed to run on them is the issue. x86 has the refinement for them. However, remember I was only talking about native machine code games on an Arm processor on a arm game machine versus, well you could say a x86 processor on a PC windows machine, but more fair would be against native machine code on x86 processor on a ps4, where you have both stable architecture to refine and optimise code to the machine. The x86 should get a boost, but much more a competent ARM. Now, you still have not considered what a hundred low powered and speed arms will do on every metric, but quoted apple metrics in an orange race. It is not just what you have, but what you do with it in processors.

Apple have been considering going to Arm, but the move to use it for servers is telling:
http://www.theverge.com/2012/10/30/3576560/arm-cortex-a57-cortex-a53-cpu-core

I remember when Arm came out, it was in the fastest computer in the world, out doing vax mini's and outdoing x86 by country miles. The other fastest processor was the Novix. The guy that designed that deigns 6mw cores at several hundred MHz native speed each, probably around 6-10k transistors per processor, on a horrendous process compared to the one Intel is using. If using that process I imagine it might surpass 1ghz per mw. I have been trying to talk him into doing a high performance 32 bit version.

 

[Gavin Greenwalt]

That's employing the "No True Scotsman" falacy though. "If ARM hardware had real software it would run a lot faster!" Well... unless you consider Android + Linux or iOS or Windows Phone or any of the other ARM optimized operating systems and software stacks "real ARM software" then I have no idea what to suggest. If anything x86 is poorly optimized for Android and it still smokes ARM on many benchmarks.

I think this is telling: http://www.anandtech.com/show/6695/microsoft-surface-pro-review/5
The i5 (which is available in a 7watt chip) vs ARM's 4 watt CPU positively dominates with a 5-10x performance boost.


Also ARM is running now @ 2.3GHZ with its latest chip. Intel is actually running slower clockspeeds for its new 7watt chips @ 1.5/2.3 GHZ(Turbo). As to the A57 core, that's not slated to launch until 2014. So yes I'm sure it will offer 3x the performance but by then Haswell will also be out. Also most of the RISC advantages of ARM were eliminated ages ago when Intel and AMD started integrating RISC like features to the x86 platform. SSE etc have eliminated the purity of x86 chips long long ago. A modern x86 chip is really a hybrid CISC/RISC chip using the best where applicable.

Lastly for something like the PS4 it makes perfect sense to go x86. Multi-platform games just became trivial.

 

[Wayne Morellini]

That's employing the "No True Scotsman" falacy though. "If ARM hardware had real software it would run a lot faster!" Well... unless you consider Android + Linux or iOS or Windows Phone or any of the other ARM optimized operating systems and software stacks "real ARM software" then I have no idea what to suggest. If anything x86 is poorly optimized for Android and it still smokes ARM on many benchmarks.
.

Lastly for something like the PS4 it makes perfect sense to go x86. Multi-platform games just became trivial.

I didn't dispute that Gavin, I agree with it, I've advocated going towards a PC like architecture to save development cost for manufacturer and developers for many years, and had similar sort of people knocking the idea their counterparts now support. What I did say is that we don't need another another closed system when an open android system could do with native code, or open PC system.

I'm sorry Gavin, but you guys need to step back and look. Trying to invalid my arguments with invalid arguments isn't reasonable. Take a step back, have a look at the problem space and what is actually said. These other things are not true comparison of maxinum potential, but hobbled examples. It is no fallacy, the problem space was optimised for arm game performance on stripped down arm machine large core count, small core, low speed low energy (drop the consumption maybe 10x) against the stripped down core of a high end Intel CPU, not some other things that doesn't answer it. Also, what is the i5 doing these days to get 7 watts, what happens in high performance mode at max speed? You are also comparing on the basis of sloppy virtual machines, operating systems and apps, not real optimised code, you need to consider for high end gaming. I've been passing over other things too because it is late, 5am.

Thanks for the arguments guys but night and a new day, hopefully containing useful work this weekend, follows.


-----------------------
Some are meant to follow, but be careful who they follow.

 

[Jeff Kilgroe]

Trying to look ahead to the distant future -- about 10~12 years in this industry -- is nearly impossible and most of us are usually wrong about what will happen. That said, I really don't see the current ARM type architecture making any more in-roads to the market than what it has now. In fact, I see it going the other way. IMO, the distant future is hybrid CPU architectures that contain primary CPU cores in the form of x86 processors (and in some cases, ARM) and I'm expecting a convergence of MIC tech or "GPU stream processor" approach onto the same CPU. Imagine if you will, a future Xeon CPU at 18nm process which operates at 5.5GHz out of the box and incorporates 12 cores. In addition to that, it will have an integrated MIC engine (Xeon Phi), adding another 3000+ stream processors to the party. We're looking at 12 to 18 GT/s and TDP around 85~110W.

Intel has already started down this path with the recent process shrinkage to 22nm and their increasingly capable integrated GPUs in the desktop class chips and the continued integration of other peripheral components and processes as we have seen with the last couple Xeon iterations. Won't be long before those integrated GPUs are graphics units tied to Xeon Phi or Nvidia Kepler style MIC for OpenCL, OpenGL, etc.. "CUDA" acceleration. Others in the market, like nvidia are taking a similar approach, but more ARM-centric for the moment. Such as nvidia's Tegra approach.

To sum up my feelings right now, the current power-workstation situation feels like a bad rehash of the late '80s all over again. Back then we had to install separate FPU coprocessors, it wasn't until the i80486 line did Intel start integrating the FPU. Socketed and slotted FPUs and accelerator cards were commonplace, larger workstations for CAD could sometimes use multiple FPUs. In the "big iron" workstation sector from SGI/ Cray, Sun and Dec, we had similar things happening. Multiple graphics and compute pipelines. Big hardware to accomplish tasks that made more sense to be incorporated right into the CPUs, and eventually were. I think we're seeing this same process all over again.

Over the years we have continuously heard all the arguments against x86 and how it's supposedly inferior, it can't do this or that, blah, blah, blah... So far every other major computing architecture that has promised to replace it has broke against it. ARM isn't going to topple it, no way, no how. GPU or MIC style processors are not going to overtake it, they're a different and complimentary technology and I see them becoming integrated.

Where the ARM approach is working for now is that with the current paradigms in small computing and mobile devices, the majority of users don't need any more power than what comes on the current ARM chips. They can deliver an adequately powerful computer in an attractive power envelope. The compromise to compute power doesn't feel like a compromise at all when the majority of users are consumers of distributed media.

 

[Jake Bastian]

Maybe they are just trying to figure out how to differentiate the fact that games aren't 4K even though video is.

http://www.ign.com/articles/2013/02/21/playstation-4-wont-support-4k-games-3d-not-a-focus

Seems they are also spreading the reveal over a longer period. I'd expect to see some more concrete stuff around E3 this summer.

 

[Antony Newman]

'Elite' was written on a 2 MHz 6502 with only around 20KB of RAM. An amazing immersive and pioneering game.
What has 1000 times more CPU performance, and 100,000 times more memory enabled game writers to do these days?

The answer - I believe - is the creation of generic, easily maintainable, sometimes interpreted code, slow running code, but code that takes less time to write.

The ARM code in the DryOs was an exercise in how not to write efficient code (well in the 10,000+ routines I looked at).
Rewriting sections of code directly in ARMv7 sped the code up by around 10 times (but probably took 100 times longer to write).

My view is that existing code (Games and PC Apps) coould actually run considerably faster - but that it won't do so until it is rewritten to multiprocess efficiently.

When the ARM A57 comes out, each core will still perform slower than Intels fastest core for the same nm process.

I think the big question is cost. A 32 core A57 vs a 15 core Intel. If the Intel chip performed twice as fast, but cost 10 times as much - do you think ARM would start biting into Intels market?

In 5 years time I think Intels profit margins are going to get squeezed, as it is forced to compete with ARMs colaborative offerings on the market. As the core count increases - more brainpower will move into the CPU to CPU fabric domain ... and eventually I'll even be able to edit 6k footage in real time

AJ

 

[Tom Lowe]

They are probably holding back the 4K announcements until they can announce a 4K delivery system, which they are obviously working on. Hopefully it will be file based, instead of optical-disc based. Many factors point toward this outcome.

 

[Wayne Morellini]

Speak for yourself, I've had a pretty good record, the thing that often gets in the way is human greed, and incompetence. I usually come up with what people eventually invent a decade or more afterwards. You would hardly believe what's coming.

I've been pursuing new processing strategies for years, and some mind blowing strategies (probably not for you though) to replace transistors, gates and processing units, that I now know how to do. I might have said this somewhere before, but recently I read about a discovery about how human thinking works, that fit in with my old ideas. I then remembered when in primary school I came up with a way to design a computer to emulate human thinking, and how the mind worked, which I have kept and expanded on other the decades, because it turned out to be highly accurate and in advance of the research, and they still have a way to go to catch up. When you can work out what can be done, you live in the future already. Doesn't mean I spend all my time there or get it right though.

Arm is just a stable ecosystem you can use to compete today, not the best. I'm looking at a better system to use for my parallel os for my distant ce computer platform for the 3rd world computer. Of course, I would like to see things improve for people, so am advocating dome Arm competition for now. When not even the arm parent company could see the future anymore in the 80's, I could. Somethings are just a good workable idea.

As for hybrid architectures, I have pursued it in the past. I proposed a hybrid parallel processing gpu array based on a minimal, efficient instruction set with a version of open cl to match and a host of new parralel circuit strategies.

Trying to look ahead to the distant future -- about 10~12 years in this industry -- is nearly impossible and most of us are usually wrong about what will happen. That said, I really don't see the current ARM type architecture making any more in-roads to the market than what it has now. In fact, I see it going the other way. IMO, the distant future is hybrid CPU architectures that contain primary CPU cores in the form of x86 processors (and in some cases, ARM) and I'm expecting a convergence of MIC tech or "GPU stream processor" approach onto the same CPU. Imagine if you will, a future Xeon CPU at 18nm process which operates at 5.5GHz out of the box and incorporates 12 cores. In addition to that, it will have an integrated MIC engine (Xeon Phi), adding another 3000+ stream processors to the party. We're looking at 12 to 18 GT/s and TDP around 85~110W.

Intel has already started down this path with the recent process shrinkage to 22nm and their increasingly capable integrated GPUs in the desktop class chips and the continued integration of other peripheral components and processes as we have seen with the last couple Xeon iterations. Won't be long before those integrated GPUs are graphics units tied to Xeon Phi or Nvidia Kepler style MIC for OpenCL, OpenGL, etc.. "CUDA" acceleration. Others in the market, like nvidia are taking a similar approach, but more ARM-centric for the moment. Such as nvidia's Tegra approach.

To sum up my feelings right now, the current power-workstation situation feels like a bad rehash of the late '80s all over again. Back then we had to install separate FPU coprocessors, it wasn't until the i80486 line did Intel start integrating the FPU. Socketed and slotted FPUs and accelerator cards were commonplace, larger workstations for CAD could sometimes use multiple FPUs. In the "big iron" workstation sector from SGI/ Cray, Sun and Dec, we had similar things happening. Multiple graphics and compute pipelines. Big hardware to accomplish tasks that made more sense to be incorporated right into the CPUs, and eventually were. I think we're seeing this same process all over again.

Over the years we have continuously heard all the arguments against x86 and how it's supposedly inferior, it can't do this or that, blah, blah, blah... So far every other major computing architecture that has promised to replace it has broke against it. ARM isn't going to topple it, no way, no how. GPU or MIC style processors are not going to overtake it, they're a different and complimentary technology and I see them becoming integrated.

Where the ARM approach is working for now is that with the current paradigms in small computing and mobile devices, the majority of users don't need any more power than what comes on the current ARM chips. They can deliver an adequately powerful computer in an attractive power envelope. The compromise to compute power doesn't feel like a compromise at all when the majority of users are consumers of distributed media.

Wouldn't you agree, the success of x86 is inertia, ecosystem, political and culture, rather than any pure technical superiority, Jeff? Many break for this reason. It did break on consoles and embedded electronics, because it was inferior, and more bs. It is for other reasons it was forced in these areas down peoples throats (more the embedded area) but now has an inertia there and now people think it is normal. Lets not forget tron, which windows had to adapt to, to compete in embedded.

I am in a community that has a proven history of showing up c, ms and x86, is it their fault if no one wants to accept it? What they need to do is strip back the x86 instruction set to a efficient minim and change the way it works, and instructions (I'm not going to say change it to an arm and move the legacy instruction set over to emulated eventually. You can then more readily integrate it into a hybrid array with more higher speed cores per space.

The thing you said about the fpu in the 80's is interesting, for coincidental reasons. At the time we eventually got enough integer power for present apps but not enough floating point, eventually we had floating point but not enough graphics, so we got gpu integrated. Now we are looking at hybrid processing units, HPU's. Not to forget dsp in this history. Large arrays of small processors is next too, and clearly have superiority in processing density over the x86, but the problem the industry is having is how to best use all these processors, that is where my work comes in. A lot of people have mind blocks, which often are locks, where they are locked into what they know rather than able to figure more out.

'Elite' was written on a 2 MHz 6502 with only around 20KB of RAM. An amazing immersive and pioneering game.
What has 1000 times more CPU performance, and 100,000 times more memory enabled game writers to do these days?

The answer - I believe - is the creation of generic, easily maintainable, sometimes interpreted code, slow running code, but code that takes less time to write.

The ARM code in the DryOs was an exercise in how not to write efficient code (well in the 10,000+ routines I looked at).
Rewriting sections of code directly in ARMv7 sped the code up by around 10 times (but probably took 100 times longer to write).

My view is that existing code (Games and PC Apps) coould actually run considerably faster - but that it won't do so until it is rewritten to multiprocess efficiently.

When the ARM A57 comes out, each core will still perform slower than Intels fastest core for the same nm process.

I think the big question is cost. A 32 core A57 vs a 15 core Intel. If the Intel chip performed twice as fast, but cost 10 times as much - do you think ARM would start biting into Intels market?

In 5 years time I think Intels profit margins are going to get squeezed, as it is forced to compete with ARMs colaborative offerings on the market. As the core count increases - more brainpower will move into the CPU to CPU fabric domain ... and eventually I'll even be able to edit 6k footage in real time

AJ

Those were the days, at last something worth talking about. Sorry AJ, and Jeff, I thought Jeff was talking to me before.

I remember elite, took me something like 6 days to reach elite on my commodore 64. But they canceled the championship competition because some others were cheating.

Your points are right AJ I've worked on a lot of new game concepts, most can fit in 4k. Processing power, apart from more realism, effects and strategy, allow some new game types. The good programmers will work on optimal code, but only high end types need it, the rest go the slow but easy path. However, that still works for an arm array, in it's favour. So there is still room for Arm to compete with the x86, it was always a case of not that it cant be done, but how to do it.

If I get my 3rd world computer running, maybe Ill contact Jeff Minter again, and invite him to come along. Imagine, a new machine coded game a month He would be worth giving the backlog of game ideas too, a legendary programmer.

A57 is a bit of overkill in mass multi core chips, as only a few controlling cores need the extra features, the rest can be pure light weight processing units running at higher speed at lower energy in less space, that is where the biggest advantages come in. When dealing with a game system only certain cores need security for programs, most cores can be small and given over and efficiently managed by the game itself.

Even if we have an array of processing elements, the bigger it gets, the more control processors issuing commands are needed (or/and more independent decision making) as the amount of co processors gets too much for the number of control processors to handle for a the situation.

 

[Wayne Morellini]

They are probably holding back the 4K announcements until they can announce a 4K delivery system, which they are obviously working on. Hopefully it will be file based, instead of optical-disc based. Many factors point toward this outcome.

Sony announced a built in 4k delivery on their uhd tv.

 

[Sergio Perez]

I do question the fact it doesn't play 4K games. The previous Playstation lasted 7 years . It was HD at the beginning of the movement, Nintendo avoided that gamble and went SD, and this made Nintendo lose market share on the last couple of years, making them introduce the HD Wii U. The new Playstation will allow 4K content but not games. With a Hisense 4K 65 inch tv going for about 3K or less in China, I can only see 4K capable panels being in the mass market and in homes in a vast number in the next 2 or 3 years. It would probably have been a better move if Sony delivered a 4K games and Content capable machine next year. Maybe Microsoft will one up them in this, and this could make a difference...

 

[Jake Bastian]

I do question the fact it doesn't play 4K games. The previous Playstation lasted 7 years . It was HD at the beginning of the movement, Nintendo avoided that gamble and went SD, and this made Nintendo lose market share on the last couple of years, making them introduce the HD Wii U. The new Playstation will allow 4K content but not games. With a Hisense 4K 65 inch tv going for about 3K or less in China, I can only see 4K capable panels being in the mass market and in homes in a vast number in the next 2 or 3 years. It would probably have been a better move if Sony delivered a 4K games and Content capable machine next year. Maybe Microsoft will one up them in this, and this could make a difference...

Games in 4K is really unnecessary even with a 4K panel. Until photorealistic graphics become a reality you're just adding resolution where it doesn't exist in the first place. I don't need a 4K panel to highlight the fact that the character in my game is made of polygons. 4K video is enough this generation - if somehow a technology jump happens that allows real time rendering of photorealistic images at 4K within the next ten years I'm sure they'll just release a new generation of consoles. I don't see that happening though.

 

[Wayne Morellini]

I invented a system to do real time photo realistic graphics many years ago, but could not pursue it due to the ever increasing quality of the competition, made the entry level too complex to start, then somebody familiar seemed to patent it. It is not making photo realistic images, but animating it interactively and flexibly that ultimately is harder. I've advanced the system to a life like VR level. It will take decades to catch up..

 

[Sergio Perez]

I invented a system to do real time photo realistic graphics many years ago, but could not pursue it due to the ever increasing quality of the competition, made the entry level too complex to start, then somebody familiar seemed to patent it. It is not making photo realistic images, but animating it interactively and flexibly that ultimately is harder. I've advanced the system to a life like VR level. It will take decades to catch up..

WOuld love to see that in a Heavy Rain like game. Or probably in the next Pixar masterpiece...

 

[Bob Gundu]

I'm sure the graphics will look better than this https://vimeo.com/51947886

I invented a system to do real time photo realistic graphics many years ago, but could not pursue it due to the ever increasing quality of the competition, made the entry level too complex to start, then somebody familiar seemed to patent it. It is not making photo realistic images, but animating it interactively and flexibly that ultimately is harder. I've advanced the system to a life like VR level. It will take decades to catch up..

 

[Wayne Morellini]

me and big mouth, yeah. lovely working out on that one. Rain is a lovely problem, as I said it is the animation, and I should also say amount of content, that ultimately taxes a sysrem. I've dreamt of 3d arrays of circuits, and how many it would require to model the entire planet, surprisingly small eventually. What you do in a situation like rain is approximate it's effect for the space you are looking at, but each rain drop refracts light etc which refracts off each other and so on. ultimately most of the time it probably would not matter too much, because the accumulated effect looks similar, white or wavery etc. But for most of us an approximation especially at distance can suffice. I should say that I aimed for near photo realistic, but over shot if you have the horsepower. So, going from a space you can approximate it's effect and the affect of other spaces on it bingo, something our minds will have trouble deciphering compared to real life. I figured long ago that the 80% rule principle was a good metaphor for these things, a few people would know it here, something like this, to get 80% of the results requires 20% of the work, but to get the remaining 20% of the results requires the remaining 80% if the work. In various degrees this comes into the equation. The above type of solution in the rain example be used in many effects. Do you need exact physics when a 90, 80, maybe even 50 % solution will do. I think that was what originally got me big time onto the 80/20 principle, how to do physics for fps games using existing game/graphics engines years ago simpler.


re-edit: Patent controls

It is a shame about software patents, they should at least say it is too crude/inaccurate to be patentable, and save patents for ideal used systems, or at least downgrade royalties based on performance (a hard one to manage) until you get to 50%. Would make actual creation easier. I wonder if anybody in the industry keeps a list of defunct patents that are useable freely and how. Would be a good project for the game programmers representative body, and game engine writers. I diverge here, but an idea occurs to me, it would decrease litigation, if all you needed to do was to show how your device fits a another previous patent or use fully in every way that it is claimed that it violates a newer patent, for litigation to cease before even getting to trial and for all expenses to be awarded to you. All you would need to do us publish said proof, and if threatened, harrased bluffed, or taken to court you automatically get damages if they can't prove over wise. The threat of illegitimate patents and litigation would dry up. It would also cost patent holders a lot less than trying to chase dead money. That would be a great solution, how come nobody thinks like that? It could get rid of most of the problems in the patent system, by putting in a proper check and balance. And it should be this easy to get parts of parents invalidated for free, by sending reference information to the patent office. Then the patent office could be entitled to request and sue for costs, or send out a ticket to say it was the patenter's mistake rather than deliberate, and they let them off this time. This would prevent them getting bogged down pursuing small fish, and allow them to concentrate on big ones. In matter of fact, why can't the patent office sue people for damages on invalid patents already?

I know somebody who will love to read this.

 

[Peter Proniewicz-Brooks]

Games in 4K is really unnecessary even with a 4K panel. Until photorealistic graphics become a reality you're just adding resolution where it doesn't exist in the first place. I don't need a 4K panel to highlight the fact that the character in my game is made of polygons. 4K video is enough this generation - if somehow a technology jump happens that allows real time rendering of photorealistic images at 4K within the next ten years I'm sure they'll just release a new generation of consoles. I don't see that happening though.

Errrmmm, if they create higher resolution textures and up the poly count then there is a point to higher resolution, the road to photo realism is paved with small steps not massive leaps. Hell more smaller polygons and a higher resolution will help obscure the fact its made up of them on your 4k panel that this videos going to.

 

[Wayne Morellini]

I'm sure the graphics will look better than this https://vimeo.com/51947886

Just changing the models would do wonders. This is the sort of render you would do on a PS2 maybe. I would have to do rendering and animation calculations, but without PS2 documentation and hardware it is hard to say if it would have Bern up to this good or better. The PS2 troubled me for a while, until I realised I was thinking wrong about the problem. A missed opportunity that, I was wishing to put my system on the PS2 when it came out, but though certain problems would prevent it. I could had exclusive on the PS2 and had it up past Xbox 360 level graphics years ago, a huge, huge advantage. The ps3 would have pushed the best i7 system with top end card, just one little issue with these things, resources. The thing is, that it is ideal for gpu processing, just simple hardware would gave greatly increased it's performance. Just adding such a circuit to a PS2 (if you can interface it) would have been a serious boost, then all you need to do us upgrade the circuit speeds and quantities of everything else to have a top system today. 90% of the next gen system development costs out the window. Pity I never got to design my fpga game system back when wii was about to come out, gone now. With commodity PC hardware in next Gen systems and general purpose processing gpu systems, and decades of work on graphics engines refinement it is barking up the wrong tree for somebody like me to do something competitive, even in handheld game to compete on the level of vista.


Better just to get a big processing array and slap the latest gpu on it.

Errrmmm, if they create higher resolution textures and up the poly count then there is a point to higher resolution, the road to photo realism is paved with small steps not massive leaps. Hell more smaller polygons and a higher resolution will help obscure the fact its made up of them on your 4k panel that this videos going to.

This thing is low resolution, I just looked at the qvga static picture, did not realise it was that bad. I down grade that now, easily outdone.

 

[Wayne Morellini]

My apologies, and ARM solutions.

My apologies, and ARM solutions.

Sorry to waste time, I'm here to apologise. I can't find the 100 core arm, I'm mistaken, my apologies. There is a 18 core research one to go into a million price sir brain simulation, and have come a cross many non arms. But the reality is that the nvidia chips quoted are very poor choices for a 100 core arm, many times mire than what you can get a simple low powered arm to operate at and cheaply. I don't have time to thumb through spec sheets to find an example now, it's gotten late, but been there before looking for a processor for an old project. But luckily I got the link to the million core machine before, but the 18 processors per chip is still 100 million transistors, compared to the 2.3 billion transistor upper end Intel, or the little 774 million transistor i5-750, and not suitable for the hundred core machine, compared to a arm design that could fit in less than 100k transistors up to 1 million with working memory, in other words, with even a 50% die space sacrifice for interconnects that us at least 387-3870 arms on the Intel process running at a fraction of the power of the i5 probably up to 1ghz each. Sorry for the delay in getting to this, I believe I have just illustrated my point, even if I am 50-90% out. Arm is big and clunky compared to what some others are doing. How much less a hundred core, not including the gpu I mentioned to go with it? In reality so many millions of transistors would go to higher end arm control cores, but still many simpler array cores, but these controls would take a chunk of power consumption.

Check out the performance:

http://www.arm.com/products/processors/classic/arm9/arm968.php

Did not realise they still ran at such low frequencies in low power, but considering the process size. Extrapolate the figures, and it is sipping juice. Still ten times more than the processor I'm looking at using in my third world system. I've seen lower figures years ago, admittedly they did not pan out, but I wonder if there is lower again in their range.

 

[Antony Newman]

Wayne,

(My understanding) Moving from 28nm to 14nm for the same chip design is expected to give a 50% increase in performance - but take 1/4 the size.

Contemplating what could be done with a large number of CPU cores is therefore a great question:
+) In the mobile space - developers already need to research what can now be achieved with 8 CPU cores.
+) In the desktop space greatest performance gains are going to be made to efficiently multiprocessing rather than waiting for transistors to shrink.

Intel
Intel's sales price for a multi-core chip is a function of the number of cores that chip has (ie performance) rather than manufacturing cost.
http://download.intel.com/newsroom/kits/xeon/e5/pdfs/Intel_Xeon_E5_Factsheet.pdf (~$2000 for an 8 core, $300 for a 2 core).

If the costs were more 'reasonable' - I think there would a huge demand for 4 socket Xeons, rather than the two socket versions out there (this is assuming that Quad socket is possible!).

ARM
The ARM 64bit core (A53) is the reportedly the smallest one on the planet.
If the license cost per core is minuscule - and the chip area for 16 cores means you get a reasonable performance for low cost - I would expect companies to migrate towards it, especially as the inter-core ARM comms (CoreLink CCN-504) already exists.

Initially ARM may make some inroads into software that already mutliprocesses well - like server software, or for tailored software for something like OpenCompute.

But there may be some enterprising company out there that builds ARM Muti-socket motherboards that allow 8 x 8 x ARM chips to be plugged in .. maybe one that is Windows-8 Compatible. Even if a 16 core ARM A57 chips ran a quarter the speed of the Intel 15 core chip ... A 64 chip ARM computer would definitely get my interest ... especially if CS6 was ported onto it with Nvidia GPU support! /<dream-mode-off>/

AJ

 

[Wayne Morellini]

Wayne,

(My understanding) Moving from 28nm to 14nm for the same chip design is expected to give a 50% increase in performance - but take 1/4 the size.

Sounds about right, depending on process.
..
> But there may be some enterprising company out there that builds ARM Muti-socket motherboards that allow 8 x 8 x ARM chips to be plugged in .. maybe one that is Windows-8 Compatible. Even if a 16 core ARM A57 chips ran a quarter the speed of the Intel 15 core chip ... A 64 chip ARM computer would definitely get my interest ... especially if CS6 was ported onto it with Nvidia GPU support! /<dream-mode-off>/

If they can move to a comparable process in processing and energy performance, it would dramatically turn the tables, especially when lighter arm cores are mainly used, that is what should be looked at. Take a look here: www.greenarraychips.com , if the 32 bit version comes out it would be great.