backtop


Print 45 comment(s) - last by vignyan.. on Nov 26 at 3:49 PM

Samsung also is eyeing Cortex-M processors for its Smart TVs and appliances

Two of the smartphone market's biggest powers met this week to discuss processors.  UK-based ARM Holdings plc (LON:ARM), an architecture company whose titular instruction set is found in 95 percent of smartphones sold today, traveled to South Korea, the home turf of Android phonemaker South Korea's Samsung Electronics Comp., Ltd. (KSC:005930).  Samsung accounted for over 30 percent of global smartphone shipments in Q3 -- nearly three times more than any other OEM.
 
Samsung and ARM met to discuss the adoption of ARM's new ARMv8 64-bit instruction set, which Samsung's upcoming Exynos 6 processors for next year's Galaxy S5 is expected to use.  Antonio Viana, ARM’s executive vice president of commercial and global development, met with senior Samsung executives to talk about the roadmap to 64-bit and beyond.
 
An unnamed senior manager at ARM with knowledge of the meeting suggested 128-bit chips might also have been discussed.  While PC CPUs have yet to hit 128-bit, The Korea Herald quotes the senior ARM executive as saying there was a "possibility" that ARM would release a 128-bit instruction set in the next two years and push for smartphone/tablet adoption.  But the Korean publication says the ARM official said these plans were only a "possibility" and not a definite plan at present.
 
Exynos
Samsung wants to make sure that its Exynos chips for next year's Galaxy S5 have access to ARM's best instruction set -- the 64-bit ARMv8 instruction set.

The official is quoted as suggesting that smartphones soon may have more than 4 GB of memory and need 64-bit processors to address their memory.  The source stated, "As technology moves from, for example, shifting to face recognition on smartphones from the fingerprint scanner to unlock an iPhone, it requires more powerful memory capacity."
 
It also appears that ARM is incentivizing the transition to 64-bit ARM chips in the mobile market, in part, to fuel its server ambitions, as well as its PC push.  
 
While 64-bit chips offer some gains in graphics and I/O addition to the aforementioned memory addressing gains, processing is generally otherwise unaffected.  However, ARM is purposefully tilting the playing field in 64-bit's flavor by designing its new instruction sets -- which allow more registers -- to only work for 64-bit chips.  32-bit chips will be forced to use the older, less optimized ARMv7.  In other words, ARM is telling mobile chipmakers to make the switch or settle with last generation instruction sets.
 
This push will allow ARM to foster an ecosystem of compatible applications, which may help the chip designer to push into the PC and server markets, two key areas of desired growth.

Cortex M
Samsung may look to use the Cortex M chips in new appliances.

Samsung and ARM also reportedly discussed the Cortex-M processor, a low power core design which Samsung may look to use in its appliances and Smart TVs.

Source: The Korea Herald



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

128-bit?
By Samus on 11/21/2013 2:21:28 PM , Rating: 3
Little early for that...we should wait until we are at least using 1 PiB of memory before worrying about addressing beyond 4 PiB. Until then, the burdon on developing 128-bit code is completely unneccessary :\




RE: 128-bit?
By milktea on 11/21/2013 2:25:15 PM , Rating: 1
They're just trying to beat Intel to it. Totally marketing gimmick.


RE: 128-bit?
By vignyan on 11/26/2013 3:49:48 PM , Rating: 2
RE: 128-bit?
By JasonMick (blog) on 11/21/2013 2:28:43 PM , Rating: 3
There might be some advantages in terms of speed for addressing memory buses that are wider than 64-bit (i.e. graphical memory buses), or in addressing sensors data links (i.e. 4K cameras).

Wikipedia points out that this is also a common key size for symmetric ciphers in cryptography and the address side in IPv6. Hence 128-bit could speed up networking and cryptography (cryptography is one of the major areas of better performance with the A7 processor, according to AnandTech).

There probably other reasons as well...

I'm sure there's a technical reason why ARM thinks this is practical. A company that dominates its market isn't exactly likely to be technically ignorant.


RE: 128-bit?
By DanNeely on 11/21/2013 2:38:47 PM , Rating: 2
By those definitions of bittedness, PC CPUs with standard dual channel memory buses are 128 bit chips, as are CPUs with SSE support. AVX instructions are up to 256 bits wide, and Intel has proposed an even wider 512 bit version with an ETA in their 2015 successor to the Xeon Phi coprocessor card.


RE: 128-bit?
By kingmotley on 11/21/2013 4:23:07 PM , Rating: 3
First read this: http://en.wikipedia.org/wiki/64-bit_computing

As a recap, there are 3 parts of a modern CPU: General registers, address pointers, and data bus. In actuality, there is a few more, including external address bus. A 128-bit CPU typically refers to the first one of those, using 128-bit general registers. It would not be inconceivable, and actually highly likely that the external address bus of any 128-bit ARM would still remain 64-bit (or even less... perhaps even 36 or 40-bit), and yet it could have a significant on its speed for many different types of operations, including networking, video processing, and cryptography.


RE: 128-bit?
By vignyan on 11/22/2013 3:46:08 AM , Rating: 2
I think NEON extensions for ARM already have 128-bit data registers. It might be that new ISA extensions might introduce better instructions to use these registers?


RE: 128-bit?
By jamescox on 11/22/2013 4:12:40 AM , Rating: 3
quote:
As a recap, there are 3 parts of a modern CPU: General registers, address pointers, and data bus. In actuality, there is a few more, including external address bus. A 128-bit CPU typically refers to the first one of those, using 128-bit general registers. It would not be inconceivable, and actually highly likely that the external address bus of any 128-bit ARM would still remain 64-bit (or even less... perhaps even 36 or 40-bit), and yet it could have a significant on its speed for many different types of operations, including networking, video processing, and cryptography.


I don't know how much point there is to writing this. Even 64-bit is mostly marketing BS at the moment for smart phones, although 4GB devices will arrive relatively soon. Talking about a 128-bit processor is ridiculous and indicates that whoever was taking about it has no clue, or was massively misunderstood.

The bittness of a processor usually refers to the size of a pointer used to reference memory, not the size of the registers. 32-bits can be used to access 4 GB of memory. Going to full 64-bit pointers allows addressing up to 16 exabytes. An exabyte is 1024 times bigger than a petabyte. A petabyte is 1024 times larger than terabyte, which is more familiar since we have hard-drives this large (they are not byte addressable though like system memory is).

Most high end 64-bit (virtual address space) server processors do not even support using a 64-bit physical address space; they only support maybe 40-bits (36-bits in early models) of physical memory (1024 GB or 1 TB). Some newer chips have this up to 48-bit support (256 TB) though. This is still a huge amount of system memory; most high-end systems are only going to have something like 256 GB, which is 3 orders of magnitude less than the max 256 TB. We will not run out of 64-bit address space for a very long time, so there is no need for 128-bit pointers, it is ridiculous.

Every time you increase the pointer size, it degrades performance. Most 32 -> 64bit switches have included other enhancements with the switch to 64-bit to hide this. The performance degradation is due to the double size pointers using up more memory and cache space. Many data structures store pointers to other objects; Going 64-bit increases storage needed significantly over 32-bit. Going 64-bit requires higher bandwidth and larger caches also. This is why some applications are still 32-bit even on 64-bit capable systems. As far as I know, Chrome is 32-bit since it uses a separate renderer process for each tab. Memory of a single tab will not exceed 2 GB.

There are a few applications which make use of large integers which get a significant boost when switching to 64-bit native operations. This doesn't have to be connected to whether the processor is 64-bit or not though. Intel introduced SSE in 1999; it uses 128-bit registers on a 32-bit Pentium3. An SSE register is 128-bits, and it can be viewed as many different sized values (16x1B, 8X2B, 4x4B, 2x8B : Byte = 8 bits). The AVX extension extends the SSE registers up to 512-bits. If ARM does not have any legacy extensions, then it may be reasonable to extend the general purpose registers to 128-bit. There is no reason to change the size of a pointer though. Apple's 64-bit extension changes the size of a pointer and the size of the general registers to 64-bit also.


RE: 128-bit?
By Nortel on 11/21/13, Rating: -1
RE: 128-bit?
By Argon18 on 11/22/13, Rating: -1
RE: 128-bit?
By inighthawki on 11/22/2013 4:02:05 PM , Rating: 2
No, it's more likely because it's not true... The closest thing to a 128 bit CPU at that point was IBM's model 85 which was able to do a handful of 128 bit extended floating point operations and storing the result across multiple registers. Hardly what could be called a '128 bit processor.'


RE: 128-bit?
By icanhascpu on 11/21/2013 10:50:58 PM , Rating: 2
*Scoff*

You do not need 128bit in anything youd do with an ARM cpu.


RE: 128-bit?
By vignyan on 11/22/2013 3:41:35 AM , Rating: 2
You are mistaken!

Current CPUs access a lot more data per cycle than 64-bits. So your assumption of 128-bit ISA helping anything with memory speed/bandwidth is false. DDR width of 64-bits is a coincidence and nothing to do with 64-bit computing. Memory requests are usually in terms of 64 to 128 *byte* blocks.

Cryptographic applications can use the 128 or 256-bit SIMD registers if needed. They don't need 128-bit ISA.

Having said that, I am still intrigued why ARM/Samsung would even discuss about 128-bit ISA at this point. May be they mean something else by mentioning 128-bit machines. There are no OS/language discussions for this yet.


RE: 128-bit?
By Silver2k7 on 11/24/2013 7:42:24 AM , Rating: 2
Well there was some talk about a 128-bit Win8 or Win9... so perhaps there is some other uses for 128-bitness other than wast ammounts of RAM...


RE: 128-bit?
By Shadowself on 11/22/2013 1:18:05 PM , Rating: 2
There certainly are advantages beyond the addressable space for going to a true 64-bit processor as you point out.

But what is really a 64-bit processor? For that matter, what is a 32 bit processor? To see how confusing the naming can be you only need to look back at Motorola's 68000 series processors, back when processors were actually chip sets and not monolithic chips. The 68000 chipset was called a 32 bit processor, but in its first implementation it did 32 bit integer math internally, had a 16 bit data bus, had a 24 bit address buss and interfaced to a math co-processor chip that did 32 bit single precision, 64 bit double precision and 80 bit extended precision math. So was the 68000 chipset a 16 bit, a 24 bit, a 32 bit, a 64 bit or an 80 bit chipset?

People often like to focus on the address space -- hmmm... the original POWER chipset was considered a 32 bit chipset. However, IIRC it had a 48 bit address space. So was it a 32 bit chipset or a 48 bit chipset?

Right after Apple announced the "64 bit" A7 chip, Samsung said that early next year they would be shipping a chip with "64 bit features", but what "features" are those? Just as an extreme example, if it has a 16 bit data path, a 32 bit address space and does 64 bit floating point can they say it has "64 bit features"?

Go back to the first desktop chips with vector processing capabilities. Most of those had either 128 or 256 bit vectors, but they were still called either 32 or 64 bit chips.

I wish the naming "conventions" were consistent, but unfortunately the naming of the "bitness" of chips has been run by marketing types since the late 70s and not hardware or software engineers.

The bottom line is that until the chips Samsung is discussing are taped out and the ground rules of that tape out are made public (they often are not) we have absolutely no idea what Samsung means by "64 bit" or "128 bit". Anyone who says otherwise is just blowing smoke. Until then it's 100% marketing hype.


RE: 128-bit?
By inighthawki on 11/22/2013 4:04:46 PM , Rating: 2
The bittedness of a CPU is commonly associated with the native word size of the processor, which is also typically the same size as the main registers.


RE: 128-bit?
By inighthawki on 11/21/2013 2:42:36 PM , Rating: 3
A 128-bit CPU doesn't necessarily mean that it has a 128-bit address space. You can bump up registers to 128-bits in size and introduce an instruction set to take advantage of it and still use 64-bit pointers. Most CPUs already have 128-bit registers for SIMD instructions (and some even more, Intel has 256 bit YMM registers for AVX), though not to be confused with 128-bit processing.

The xbox 360 (and I assume PS3 did this too, but I have no experience with it) was actually a 64-bit processor but because the hardware was standardized with under 4GB of RAM, the compiler actually generated 32-bit pointers. You ended up with all the benefits of 64-bit processing without the huge disadvantage of increase binary size. You could do the same thing here. Enable 128-bit operations, but continue using 64-bit pointers.

128-bit processing could have some performance benefits for things such as cryptography and UUID manipulation, although I would really love to see what Samsung had in mind.


RE: 128-bit?
By coburn_c on 11/21/2013 4:33:23 PM , Rating: 1
Floating point math can benefit from huge address spaces, integer math... no

That's why the 128-bit registers are in the FPU

The article spells it out, ARM pushed 64-bit to phones to finance their desktop battle, wasn't needed on phones. 128-bit frontend isn't needed at all, it's just to grab headlines.


RE: 128-bit?
By inighthawki on 11/21/2013 6:07:24 PM , Rating: 2
Huh? What you said makes absolutely no sense. floating point operations do not have any special benefits from address space whatsoever. Memory is memory, just treated differently.

The 128 bit XMM SIMD registers on the x86 platform is for parallel data processing. It allows you to process 2x64bit or 4x32bit values at a time in a single instruction - hence the name Single Instruction Multiple Data (SIMD). As of SSE2 Intel already had integer versions of all of the instructions, even going a step further with 16x8bit and 8x16bit instructions for integers as well. But more importantly, those 128 bit registers do not do 128bit floating point math, nor do they have ANYTHING to do with address spaces.


RE: 128-bit?
By coburn_c on 11/22/2013 2:13:39 AM , Rating: 2
Vector processing is pretty much only used for FP math. SSE was actually all originally integer functions. Commands that perform the same function on multiple data sets all prefer FP math. That was not the case when CPU FP was on an outside chip.. but these days... If you look inside any modern processor the 128-bit registers exist in the FPU. The FMAC performs the SIMD instructions in fact, using FMA3. Two guesses what those Fs stand for.


RE: 128-bit?
By inighthawki on 11/22/2013 3:21:52 AM , Rating: 2
quote:
Vector processing is pretty much only used for FP math

Commonly especially in games and scientific work, but integer is not at all rare. You can use integer based SIMD instructions for processing things like strings. In fact I've used SSE2 based integer instructions to implement string length functions that are an order of magnitude faster than standard non SIMD based implementations. I would also kill for the ability to have 128bit integer instructions to do things like comparing a GUID with a single compare.

quote:
SSE was actually all originally integer functions.

No, it was not. SSE1 was all FP. SSE2 added integer. Try again. Maybe you're thinking of MMX, which was 64-bit and superseded by SSE2. (SSE had a couple integer instructions that operated on the MM registers)

quote:
That was not the case when CPU FP was on an outside chip.. but these days... If you look inside any modern processor the 128-bit registers exist in the FPU.

In most cases this is only because the SIMD instruction sets may not actually have integer instructions for them. So duh, of course they will be in the FPU.

quote:
The FMAC performs the SIMD instructions in fact, using FMA3. Two guesses what those Fs stand for.

FMA3 , you mean the ' F used' M ultiply A dd 3 instruction set?


RE: 128-bit?
By coburn_c on 11/22/2013 3:41:26 AM , Rating: 2
Fused, meaning single precision, as in floating point.

To bring this full circle, 128-bit SIMD registers does not a 128-bit chip make, or all current chips would be considered 128-bit... and outside of SIMD what purpose is there for 128-bit silicon?


RE: 128-bit?
By coburn_c on 11/22/2013 3:43:22 AM , Rating: 2
Oh, and ARM has 128-bit SIMD already, it's called NEON


RE: 128-bit?
By inighthawki on 11/22/2013 11:12:44 AM , Rating: 2
What is your point? This has nothing to do with the point I'm trying to make. I'm WELL aware of what NEON is, as well as SSE, AVX, PPC's altivec. They're all SIMD instruction sets... so?

128-bit SIMD is not the same as 128-bit processing. There is a difference between process 4 32 bit elements and a single 128 bit element at once.


RE: 128-bit?
By otherwise on 11/22/2013 10:40:45 AM , Rating: 2
Fused, meaning there is no intermediate rounding step; i.e. if you think of floating point error as a e() function -- the result is e(x(y+z)) instead of e(e(y+z)*e(x)). The fact the operation can be optimized to be faster than a=y+z; b=e*x is a bonus.


RE: 128-bit?
By inighthawki on 11/22/2013 11:11:10 AM , Rating: 2
I made it very clear in my initial post that 128-bit SIMD registers was NOT the same thing as processing 128 bit data elements.

quote:
and outside of SIMD what purpose is there for 128-bit silicon?

Faster processing of anything that uses 128 bit data. Cryptography, GUID processing, hashes, etc.


RE: 128-bit?
By inighthawki on 11/22/2013 11:22:50 AM , Rating: 2
quote:
Fused, meaning single precision, as in floating point.

It's become even more apparent that you don't know at all what you're talking about. Please don't post useless garbage unless it's actually accurate. First you think (or incredibly heavily imply) that the F stood for float, and now you think 'Fused' is synonymous with single precision floating point? A simple trip to Wikipedia to look at the FMA3 instruction set shows that it has double precision instructions lol. Oh but see, you probably have never used them so you wouldn't have known that the PS and PD suffixes on the instructions stood for 'packed single' and 'packed double,' so that's an easy mistake. (Or 'scalar single' and 'scalar double' for the SS/SD variants if you didn't know those either)

quote:
To bring this full circle, 128-bit SIMD registers does not a 128-bit chip make, or all current chips would be considered 128-bit...

I didn't say it did at all. Please go back and (carefully) reread my original posts. Not once did I claim that SIMD meant that these were '128 bit processors.' Quite the opposite in fact.


RE: 128-bit?
By SteelRing on 11/21/2013 5:00:46 PM , Rating: 3
quote:
....although I would really love to see what Samsung had in mind.


3D Holographic Angry Birds.... naturally


RE: 128-bit?
By YearOfTheDingo on 11/21/2013 8:31:24 PM , Rating: 2
A 128-bit pointer doesn't necessarily mean a 128-bit address space. The extra bits could be used for things like hardware-assisted bound-checking. Given the prevalence of such code, it could actually lead to some real gain in efficiency.


RE: 128-bit?
By inighthawki on 11/21/2013 8:48:05 PM , Rating: 2
Of course not, I did not mean to imply that. In fact quite the opposite, my point was to the OP who implied that a 128bit processor was only useful to have 128 bits of addressable memory.

I'm not sure I follow what kinds of improvements you'd end up seeing from 128 bit pointers though. Like the transition from 32-bit, it'll just increase binary size (this time by twice as much) which will inevitably lead to an increase in cache misses, so I'm not convinced that there's really good uses for 128 bit pointers that would drastically improve efficiency like that. Could you give examples of where that would be useful? You've intrigued me.


RE: 128-bit?
By YearOfTheDingo on 11/22/2013 12:25:19 AM , Rating: 2
Say you have a pointer pointing to a buffer that's 100 bytes in length. The compiler could encode the length into the pointer, so that if pointer arithmetics resulting in an invalid pointer, it'll be flagged as such. When it's accessed, the CPU will generate a protection fault.

Such a scheme could actually reduce binary size, since it gets rid of instructions from OOB checks. More importantly from a efficiency standpoint, it takes some pressure off of the branch prediction unit.


RE: 128-bit?
By retrospooty on 11/22/2013 9:30:35 AM , Rating: 2
It's all BS anyhow. 128 bit isnt happening, isnt being planned for any ARM CPU. Someone likely made it up just to see how the dullards react.

http://www.androidauthority.com/no-128-bit-cpus-ye...


RE: 128-bit?
By kingmotley on 11/21/2013 4:11:03 PM , Rating: 2
(sigh) And here I though perhaps someone might have learned something from the whole 64-bit discussion, but I was mistaken. Please, read up on the very basic concepts before making ignorant comments, thank you.


RE: 128-bit?
By inighthawki on 11/21/2013 6:09:14 PM , Rating: 3
Unfortunately the way a lot of people were convinced that upgrading to 64-bit was useful was by telling them "OMG >4GB of RAM!" and so this has become some kind of myth that this is *the* reason as opposed to *a* reason for the update.


RE: 128-bit?
By haukionkannel on 11/22/2013 9:15:17 AM , Rating: 2
Well this is just the way of life... Tablets gets first good and affordable 4K screen, and now they allso get 128 bit prosessors before old (nobody use) desktop computers...

This figures it out completely...

Hmmm... Next year I want to have a tablet with 4K screen 128 bit CPU, hyper fast AMD or Nvidia gpu, external mouse and good external NAS for storaging all the programs and AV materia that I can ever hope...
In reality this can be reality in tablets than in desktop... O tempora, o Mores...


RE: 128-bit?
By Argon18 on 11/22/13, Rating: -1
RE: 128-bit?
By inighthawki on 11/22/2013 4:36:46 PM , Rating: 1
quote:
Surface is a failure, already removed from nearly all retail store shelves.

Surface 2 is selling much better, and was only ever sold in like 1 retail store (Staples I think), so you are heavily exaggerating.

quote:
Surface is a failure, already removed from nearly all retail store shelves.

And constantly increasing year over year. It's still growing.

You point out online efforts but then don't point out Azure, one of the most successful cloud computing ecosystems in the world, whose only real competition right now is Amazon?


RE: 128-bit?
By inighthawki on 11/22/2013 5:09:51 PM , Rating: 2
Sorry not sure why it didn't quote correctly. The second quote should reference your Windows Phone statement


"Game reviewers fought each other to write the most glowing coverage possible for the powerhouse Sony, MS systems. Reviewers flipped coins to see who would review the Nintendo Wii. The losers got stuck with the job." -- Andy Marken














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki