64-bit ARM coming

[jhu]

sometime in the future, around 2014.

 

[Edrick]

Meh, Give me 128-bit Intel and AMD CPUs

 

[nonameo]

what do you need 128 bit for? I mean maybe sometime a lot later than now, but 32 bit lasted for what... 25, 30 years? And its still not totally outdated yet? Give 64 bit some time

 

[Voo]

what do you need 128 bit for? I mean maybe sometime a lot later than now, but 32 bit lasted for what... 25, 30 years? And its still not totally outdated yet? Give 64 bit some time

Presumably Edrick didn't mean address space (well current Intel CPUs use what? 48bit? so not completely 64bit either) but register size for some computation.

If he was thinking about address space - no idea why we'd need more than 16k PT RAM in the next few decades..

 

[antisocialmunky]

I means you can crunch 2 64 bit numbers at a time, I think the FPU on SB does that.

 

[Edrick]

Presumably Edrick didn't mean address space (well current Intel CPUs use what? 48bit? so not completely 64bit either) but register size for some computation.

This. There were actually rumors last year that Win8 was going to support 128-bit. I do not imagine we see true 128-bit CPUs for at least another decade. My comment was more of a joke about ARM being late to the 64-bit party (2014).

 

[antisocialmunky]

Because they were too busy being in every embedded device.

 

[beginner99]

Meh, Give me 128-bit Intel and AMD CPUs

Compnay were I work will switch to Win 7 next year and the rumors I heard and the actually test system I could use: 32-bit. If they actually do that it would be retarded but then I guess there are some old, shitty stuff that might not run in 64-bit...

 

[TuxDave]

I means you can crunch 2 64 bit numbers at a time, I think the FPU on SB does that.

SB can crunch 8 at a time with AVX. (4 double precision FMUL and 4 double precision FADD)

 

[sangyup81]

sure 64-bit ARM in 2014 but by then, Intel should have a chip to compete with ARM

 

[DeathReborn]

How much of this ARM 64bit is from Project Denver is what I want to know.

 

[MisterMac]

Too late, id wager.


Haswell looks too promising.


Assuming intel doesn't fumble up - a Haswell based ULV /SoC chips look like a scary competition for any tablet or low usage Device.

It's an interesting war going on next few years.

 

[ncalipari]

Too late, id wager.


Haswell looks too promising.


Assuming intel doesn't fumble up - a Haswell based ULV /SoC chips look like a scary competition for any tablet or low usage Device.

It's an interesting war going on next few years.

too late for what? Competition where?


Keep in mind that haswell still is one order of magnitude (10X ) more power hungry than ARM. You can have 5 ARM CPUs and still consume half the power of Haswell at a lower price point.

Different CPUs for different markets.

 

[MisterMac]

too late for what? Competition where?


Keep in mind that haswell still is one order of magnitude (10X ) more power hungry than ARM. You can have 5 ARM CPUs and still consume half the power of Haswell at a lower price point.

Different CPUs for different markets.

Haswell mobile SKU's = 10w(The design goals atleast).

Cortex A9 Dual core = 2w.


You do that math.
So, no you can't.


At the same time, future ARM Mobile SKU's will most likely draw more than 1w per core, so the gap dies even further as ARM tries for performance.

And this is a Laptop variant not a DIRECT Haswell Embedded system design.


Get your facts straight.



Don't forget the Medfield SoC Atom chip either.

Intel looks more than ready to catch up the TDP of ARM chips, rather than ARM beginning to try hit low end x86 chips on their performance.

Wether or not ARM Mobile chips will be enough for a netbook user and so on, is different.

 

[Voo]

Haswell mobile SKU's = 10w(The design goals atleast).

Cortex A9 Dual core = 2w.

Does ARM now include cache and co in their power numbers? If not, well you can certainly add quite a lot to the 2w numbers..

And for servers performance/w is rather more important than just low power consumption - and even if performance/w is identical, space does cost money and who'd want 10 times the racks for the same performance?

 

[ncalipari]

Haswell mobile SKU's = 10w(The design goals atleast).

10–20W is the stated goal (at least according to wikipedia )

Cortex A9 Dual core = 2w.


You do that math.
So, no you can't.

10W is still an order of magnitude (10 times) greater than 2W.

At the same time, future ARM Mobile SKU's will most likely draw more than 1w per core, so the gap dies even further as ARM tries for performance.

how can you say that? So you dont take account of technological improvements? Intel can improve, while arm can only get worse?

And this is a Laptop variant not a DIRECT Haswell Embedded system design.

which most likely will increase power consumption to decrease cost of purchase.

Intel looks more than ready to catch up the TDP of ARM chips, rather than ARM beginning to try hit low end x86 chips on their performance.

x86 will NEVER EVER be as power efficient as ARM. ARM and X86 are both RISC, but X86 adds decode logic to accept CISC instructions.

In the 1-10 W range ARM is unbeatable, since it doesn't have the cost of the Decode step.

x86 is still alive only due to a very sad problem: Legacy compatibility.

 

[Ferzerp]

10W is still an order of magnitude (10 times) greater than 2W.

I've always wondered how people can think that arm architecture is an immediate threat to x86. Now I know! They have broken math centers in their brains!

 

[Edrick]

10W is still an order of magnitude (10 times) greater than 2W.

My math may be rusty, but I thought 20W would be 10X greater than 2W.

x86 will NEVER EVER be as power efficient as ARM. ARM and X86 are both RISC, but X86 adds decode logic to accept CISC instructions.

And you know this because? As soon as ARM starts adding 64-bit extensions and other instructions/cache in an attempt to increase performance, their power figures will increase. Also, if you are going to start making statements like that, then I can make a statement like this: ARM will NEVER EVER be as powerful as x86.

You really lose credibility with those statements.

 

[ncalipari]

I've always wondered how people can think that arm architecture is an immediate threat to x86. Now I know! They have broken math centers in their brains!

when you think that everybody else is crazy... you should give a look to your high school math


http://en.wikipedia.org/wiki/Scientific_notation#Order_of_magnitude

 

[CTho9305]

Haswell mobile SKU's = 10w(The design goals atleast).

Cortex A9 Dual core = 2w.


You do that math.
So, no you can't.


At the same time, future ARM Mobile SKU's will most likely draw more than 1w per core, so the gap dies even further as ARM tries for performance.

And this is a Laptop variant not a DIRECT Haswell Embedded system design.


Get your facts straight.

That's a completely bogus comparison for multiple reasons.
1. You're comparing the most power-hungry version of A9 to the least power-hungry version of Haswell. ARM also offers a 0.5 watt dual-core version of A9.
2. You're comparing a 40nm CPU you can buy today (last year?) with a 22nm CPU you can't even buy next year. If you want to make a remotely useful comparison, you should be comparing to 28nm A15 (possibly with A7 cores on the side that are used when high performance is not required) or possibly to a Qualcomm Snapdragon-like A15 derivative (which tend to dramatically boost performance in a similar power envelope relative to ARM's implementations).

 

[zsdersw]

x86 is still alive only due to a very sad problem: Legacy compatibility.

Completely incorrect. x86 is alive today because neither ARM nor any other platform has been used in any significant way to create the content we all consume with a myriad of devices (smartphones, tablets, etc.)

There will always be a need for devices that excel at creating content... and that demands a suitable nexus between processing power and cost, which isn't met by anything other than x86.

 

[Ferzerp]

Wow, that guy is now trying to defend that 2 x 10 = 10 with a link to wikipedia...

Here is a tip for you, guy who is bad at math, 20W is an order of magnitude (in base 10) greater than 2W. In fact, 2 x *5* = 10. Just an fyi. 10W is five times more than 2W. That's half what you claim. I think I get where your math center is broken, but your bizarro world logic would suggest that 10 is also an order of magnitude greater than 9 because it has a tens value. Sorry, no.

However, if you do want something that is an order of magnitude different, try to IO capability of a current arm chip vs a current x86 chip

 

[ncalipari]

Completely incorrect. x86 is alive today because neither ARM nor any other platform has been used in any significant way to create the content we all consume with a myriad of devices (smartphones, tablets, etc.)

There will always be a need for devices that excel at creating content... and that demands a suitable nexus between processing power and cost, which isn't met by anything other than x86.

Maybe you are speaking of desktop user.

I would like just to point you the existance of some other architectures:

http://en.wikipedia.org/wiki/Digital_signal_processor

 

[ncalipari]

My math may be rusty, but I thought 20W would be 10X greater than 2W.

Yeah I know, if you dont practice math often you easily forgot it.

And you know this because? As soon as ARM starts adding 64-bit extensions and other instructions/cache in an attempt to increase performance, their power figures will increase.

64 bit is not more power hungry per se. If I have to address a very large ammount of memory (think about a vending machine of train tickets, with a local database) you will be more power efficient with a 64 bit CPU, no mater how good your PAE are written.

http://en.wikipedia.org/wiki/Physical_Address_Extension

Also, if you are going to start making statements like that, then I can make a statement like this: ARM will NEVER EVER be as powerful as x86.

Exactly. In single threaded performance ARM will never beat x86. Luckily the market is moving toward multi-threaded applications.

Let's rephrase my previous sentence: The theoretical lower bound for power efficiency for ARM is one order of magnitude smaller than the X86 bound.

And this is due to the cost of decode step, which is there just for legacy compatibility.

If we decide to drop legacy compatibility, then x86 is just a bad architecture.

 

[blckgrffn]

You realize that they are bringing this to the table by 2014 - likely 2015 - at the earliest?

And that Intel, not to mention AMD, will have spent gobs of money AND time increasing the performance and value of their platforms, not just their CPUs, in that amount of time?

Getting traction in the server market is not just about performance per watt. Intel is moving so far up the chain with Nehalem-EX that that by 2015 they will likely be seriously not only endangering Itanium into complete obsolescence but likely Power and Sparc as well (hot swap everything, hardware partitioning, etc.)

Finally, their (Intel) ability to scale their manufacturing processes so quickly means that they will be more power effecient than arm not solely due to their architecture (x86) but do to their ability to be consistently to be (at least) one manufacturing node ahead of their competition - typically with better technology around that manufacturing process to boot. What will change about that by 2015?

Do you know what you call an application server that has low IPC needs but can benefit from many threads?

A VM.

You can stack tens of them on a server now, hundreds by 2015. Running on big, nasty x86 CPUs.

ARM server hardware is going to have a huge uphill battle.