MorphZone

ARM for the future?

Andreas_Wolf

# 523
Yokemate of Keyboards

Posts: 12074 from 2003/5/22

From: Germany

> AFAIK most ARM CPU's are bi-endian. At least I think that's the case with ARMv7/Cortex

After half a decade of assumptions, speculation and guesswork here on MorphZone, I think I found the definite answer on the question of ARM's endianness using these references:

http://translatedcode.wordpress.com/2012/04/02/this-end-up/
http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0203g/Chddgffb.html

The gist of it is that ARM knows one little-endian mode (on both data and code level, aka true little-endian) and two big-endian modes: BE32 and BE8. BE32 is big endianness on both data and code level (aka true big-endian), whereas BE8 is data-only big endianness.

BE32 is supported by:
- ARMv4
- ARMv5
- ARMv6

BE8 is supported by:
- ARMv6
- ARMv7
- ARMv8
»21.02.14 - 14:21

Zylesea

# 524
Yokemate of Keyboards

Posts: 2053 from 2003/6/4

Quote:

Andreas_Wolf schrieb:
> AFAIK most ARM CPU's are bi-endian. At least I think that's the case with ARMv7/Cortex

After half a decade of assumptions, speculation and guesswork here on MorphZone, I think I found the definite answer on the question of ARM's endianness using these references:

http://translatedcode.wordpress.com/2012/04/02/this-end-up/
http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0203g/Chddgffb.html

The gist of it is that ARM knows one little-endian mode (on both data and code level, aka true little-endian) and two big-endian modes: BE32 and BE8. BE32 is big endianness on both data and code level (aka true big-endian), whereas BE8 is data-only big endianness.

BE32 is supported by:
- ARMv4
- ARMv5
- ARMv6

BE8 is supported by:
- ARMv6
- ARMv7
- ARMv8

Andreas, you're the man! Good find. I almost gave up about that issue.
The beefier ARMs hence don't have the benefit of an easier to accomplish binary compability.

The Raspberry pi however is ARMv6 based....
--
http://via.bckrs.de

Whenever you're sad just remember the world is 4.543 billion years old and you somehow managed to exist at the same time as David Bowie.
...and Matthias , my friend - RIP
»21.02.14 - 22:37

Kronos

# 525
Moderator

Posts: 2236 from 2003/2/24

Quote:

Zylesea wrote:

The beefier ARMs hence don't have the benefit of an easier to accomplish binary compability.

Don't really see the problem here (or I failed to understand the linked text).

BE8 would still allow all system-structures to be 1:1 copies of 68k/PPC ones.

All data handled by an app internally can also be BE8 without problem.

The only thing we would need is an elf-loader making sure it load code-segments as LE and data-segments as BE (which I assume would be trivial).

Same for an 68k or PPC EMU, all data is BE and so is all legacy code (which is just data for the EMU), only the snippets of ARM-code it wants executed have to loaded a LE.
»21.02.14 - 23:13

Jim

# 526
Yokemate of Keyboards

Posts: 4977 from 2009/1/28

From: Delaware, USA

It is an additional complication, but less complicated than an X86 jump would be.
Still, couldn't we just stay with our current ISA?
"Never attribute to malice what can more readily explained by incompetence"
»22.02.14 - 00:50

ausPPC

# 527
Priest of the Order of the Butterfly

Posts: 543 from 2007/8/6

From: Pending...

Some other thoughts on ARM on a different project - http://reactos.org/node/779
PPC assembly ain't so bad... ;)
»01.03.14 - 04:02

Andreas_Wolf

# 528
Yokemate of Keyboards

Posts: 12074 from 2003/5/22

From: Germany

> http://reactos.org/node/779

From there:
"Apple, NVidia, Qualcomm, and Samsung are only some of the companies that have developed custom processors implement the ARM ISA. These aren't just designs that combine a generic ARM processor with a bunch of peripherals like graphics and memory, these are basically genuine custom CPU designs where groups experiment with pipelining, branch prediction, and even layout to maximize performance and minimize power usage."

For Apple and Qualcomm this is true, but chips with Nvidia's custom ARM core (Denver) have yet to emerge, and Samsung hasn't even announced any custom ARM core yet.
»01.03.14 - 13:49

minator

# 529
Order of the Butterfly

Posts: 365 from 2003/3/28

Looks like Freescale has some interesting plans:

http://www.techradar.com/news/computing-components/processors/freescale-sampling-64-bit-arm-processors-already-launching-later-this-year-1230508
»03.03.14 - 20:17

Andreas_Wolf

# 530
Yokemate of Keyboards

Posts: 12074 from 2003/5/22

From: Germany

> Looks like Freescale has some interesting plans:
> http://www.techradar.com/news/computing-components/processors/freescale-sampling-64-bit-arm-processors-already-launching-later-this-year-1230508

Yes, this is some nice additional info to what was already known about QorIQ LS chips based on Cortex-A50.

From the article:
"It is likely that Freescale will also bring 64-bit goodness to its i.MX application processor family"

This is not only likely but more or less a given for i.MX8.

From the article:
"Referring to MIPS [...] he also boldly stated that it "is on its way out", at least when it comes to Freescale."

This statement leaves me astounded. Was MIPS ever "in" when it comes to Freescale? Am I missing anything?
»03.03.14 - 22:52

minator

# 531
Order of the Butterfly

Posts: 365 from 2003/3/28

I don't think Freescale made MIPS chips but they are/were big in networking so would have been a big competitor.
»04.03.14 - 00:13

Andreas_Wolf

# 532
Yokemate of Keyboards

Posts: 12074 from 2003/5/22

From: Germany

> I don't think Freescale made MIPS chips but they are/were big in networking so
> would have been a big competitor.

Thanks, so it's more like he's meaning "the market(s) Freescale is in" when he says just "Freescale". That'd make sense at least.
»04.03.14 - 00:23

KimmoK

# 533
Acolyte of the Butterfly

Posts: 102 from 2003/5/19

From what I have learned, MIPS is not going away anytime soon.
Companies like Cavium seem to keep MIPS still on their high performance branch while developing also ARM core based chips.

It seems similar with Freescale. There exist higher performing Power core chips on the roadmap than the ARM ones.

But only future tells...
»04.03.14 - 14:19

Andreas_Wolf

# 534
Yokemate of Keyboards

Posts: 12074 from 2003/5/22

From: Germany

> Companies like Cavium seem to keep MIPS still on their high performance branch
> while developing also ARM core based chips.

...and even their own ARM cores:

https://morph.zone/modules/newbb_plus/viewtopic.php?forum=3&topic_id=7675&start=300
»04.03.14 - 20:21

minator

# 535
Order of the Butterfly

Posts: 365 from 2003/3/28

I'm surprised this hasn't appeared here yet:

apples-cyclone-microarchitecture-detailed

It's as suspected an absolute monster. It's a very wide OOO machine with 6 instruction decode per cycle.
Clock for clock it's as fast as anything Intel make. Pretty astonishing given that's Apple's first 64 bit core design.

I wanna see what they come up with next!
»05.04.14 - 17:39

Andreas_Wolf

# 536
Yokemate of Keyboards

Posts: 12074 from 2003/5/22

From: Germany

> I'm surprised this hasn't appeared here yet:
> http://www.anandtech.com/show/7910/apples-cyclone-microarchitecture-detailed

I read that some days ago. Unfortunately the article doesn't provide a DMIPS figure for the core. It would be nice to have one (also for the X-Gene) for my list :-)
»05.04.14 - 20:45

ausPPC

# 537
Priest of the Order of the Butterfly

Posts: 543 from 2007/8/6

From: Pending...

http://www.bunniestudios.com/ Open source laptop.
PPC assembly ain't so bad... ;)
»05.04.14 - 22:30

JuLieN

# 538
Caterpillar

Posts: 35 from 2008/4/16

From: France

On Talkchess forum we have a great test to benchmark ARM cpus : the Stockfish chess engine. It's perfect because :
- it uses all the cores you feed it with,
- it is available for Android and iOS
- chess programs love 64 bits, so you clearly see the difference between 32 and 64 bits CPUs.

We measure the average NPS Stockfish displays (NPS : Nodes Per Second : how many nodes in the chess tree Stockfish travels through pers second.

Here's our current charts (most powerful devices on top) :

Quote:

Ipad Air (2 64bit cores 1.4 ghz) 1.218.000 new highscore
Iphone 5s (2 64bit cores 1.3 ghz) 1.152.000 new
iPad Mini 2 (2 64bit cores 1.3 GHz) 1.128.000
Zopo ZP998 (8 cores) 1.100.000
Star Ulefone U9592 (8 cores 1.6 Ghz) 1.080.000
ASUS Transformer TF701t (4 cores ) 1.054.000 <- new
Samsung Note10.1 2014 (4cores exynos) 1.014.000 <- new
Samsung Galaxy Note2 (4 cores) 714.000
ASUS Transformer TF300T (4 cores) 661.000
Sony Xperia Z1 (4 cores) 602.000
Asus Infinity tf700 (4 cores) 590.000 new
Note 3 (4 cores) 581.000 new
Galaxy S3 (4 cores) 574.000
Cubot One (4 core MTK 6589T@ 4x1.5Ghz) 568.000 new
LG G2 556.000 new
Pomp c6 (4 cores MT6589T) 553.000
Pipo max M9 (4x cortex a9 1.6 ghz) 526.000
LG Optimus 4xHD (4 cores) 521.000
ZTE geek v975 (2+2HT intel atom) 511.000 new
iPad4 510.000
Nexus 10 (2 cores) 510.000 new
Google Nexus 5 (4 cores) 508.000 new
Amoi N828 (4 cores MTK6589 @ 4x1,2GHz) 493.000
iPhone 5 (2 cores) 476.800
HTC one X tegra3 (4 cores) 470.000
Motorola G (4 cores) 460.000 new
google nexus 7 tablet pc (4 cores) 457.000
SamsungGalaxy S4(1.9Ghz,4 cores,1GBhash)450.000 new
Asus Transformer Prime (4 threads) 440.435
HTC one 438.000 new
Huawei P6 (Hi-SiliconK3V2 4x1.5) 405.000 new
Google Nexus4 (4 cores) 400.000
Xiaomi Mi2 386.000
Sony Xperia Z (4 cores) 376.000
Oppo Find 5 368.000
Samsung Galaxy 3 (2 core) 333.567
Galaxy Note 329.000
Trekstor Surftab 305.000
iPad mini 277.500
iPad3 277.000
Ipad2 (2 cores) 274.000
Motorola Razr i (2 cores hyperthread) 271.000
Sony xperia S (2 cores) 259.000
Tablet Rockchip RK3066 2core @1.6Ghz 250.000
samsung galaxy S2 253.000
Motorola X 253.000 new
Motorola Razr (Slimkat Android 4.4.2) 250.000
iPhone 4S (Apple A5) 225.500
Lenovo A660 (2 cores) 222.000
Advent Vega Tegra2 220.500
Sony Tablet S 218.000
Motorola Razr xt910 (2 cores) 216.000
Motorola Razr HD 212.000
LG Optimus Speed P990 (2 cores) 200.000
HTC one S qualcomm s4 (2 cores) 200.000
Sony Xperia P (2 cores) 196.000
Samsung Galaxy SIII 1.5GHz (2 Cores) 189.867
Motorola Razr i (1 core) 186.000
Nuu Nu1 (qualcom dualcore 1.7 Ghz) 142.000
samsung epic 141.000
HTC Flyer, 1.5 Ghz 140.000
Advent Vega Tablet (2 cores) 137.126
Dell Streak 133.000
Samsung Vibrant 111.000
Samsung Galaxy Tab 108.500
LG Optimus 2x 102.000
HTC HD2 1 Ghz 1 core 94.000
HTC Desire S 90.000
HTC Desire 84.000
iPhone 4 80.000
ipod touch 4 78.000
samsung galaxy s 70.000
7-inch Barnes and Noble Color 62.500
Sony PRS-T1 ebook-reader 53.817
iPhone 3GS 51.000
HTC Wildfire S 31.000
Android phone Qualcomm 600MHz (1 core) 25.750
Ipod touch 2nd gen 16.500
Palm Pre oc. 1Ghz (Webos1.4.5) 16.000
ZT-180 10,2" Pad 15.345
Palm Pre oc. 800mhz (Webos1.4.5) 13.000
APAD Rockchip 600mhz (android1.5) 10.000
Palm Pre 500mhz (Webos1.4.5) 8.000

Apple's A7 is absolutely unrivaled, especially considering that its closest challengers have four or height cores and twice the clock speed. This is very impressive. No wonder people start to think apple will even switch Macs to the ARM architecture...

Those charts are updated frequently, as there's always someone to try Stockfish on a new device.

As the text formatting gets lost here, here's a screen capture of the original post :
ARMcharts.png

[ Edité par JuLieN 06.04.2014 - 01:51 ]
"A good bug is a dead bug" (Don Dailey)
»05.04.14 - 23:33

JuLieN

# 539
Caterpillar

Posts: 35 from 2008/4/16

From: France

By the way, if you want to contribute and give me the benchmarks of your non-listed devices here's how to proceed :

- iOS :
- Download Stockfish on the Appstore.
- chose "game -> new game -> analysis"
- wait for one minute so the figures stabilizes, and you've got the NPS measurement in kN/s. (Kilo Nodes per seconds).

- Android :
- Download Droidfish on the PlayStore.
- launch the game and go into its parameters : make sure "pondering" and "Display the computer's evaluation" are checked.
- in the "execution tasks" part of the settings, make sure it is set to the number of cores your device has.
- Now launch a new game as white and wait a minute so you get a good measurement.

On my iPad Mini 2 I get Around 1100 Kn/s (> 1.100.000 nodes /seconds), and on my Galaxy Note 3 I get around 480 Kns.

If you possess devices not listed above, please do this test and I'll update the charts (Thorsten and I are the main authors of these charts.)
"A good bug is a dead bug" (Don Dailey)
»05.04.14 - 23:43

JuLieN

# 540
Caterpillar

Posts: 35 from 2008/4/16

From: France

To compare with desktop CPU, my Macbook has a Core i7 running at 2.3 Ghz, with 4 cores (8 threads with hyper threading), and stockfish runs at around 5.3 M Nps. As it has 4 times more threads than the iPad Mini 2; you can see that both CPUs are about as powerful per core, despite the A7 runs at half the speed of the Core i7. This is VERY impressive !
"A good bug is a dead bug" (Don Dailey)
»05.04.14 - 23:47

takemehomegrandma

# 541
Yokemate of Keyboards

Posts: 2720 from 2003/2/24

Quote:

JuLieN wrote:
To compare with desktop CPU, my Macbook has a Core i7 running at 2.3 Ghz, with 4 cores (8 threads with hyper threading), and stockfish runs at around 5.3 M Nps. As it has 4 times more threads than the iPad Mini 2; you can see that both CPUs are about as powerful per core, despite the A7 runs at half the speed of the Core i7. This is VERY impressive !

This is *very* interesting, thanks everyone for sharing the info!

I'm sure more is to follow...
MorphOS is Amiga done right!
MorphOS NG will be AROS done right!
»06.04.14 - 07:37

takemehomegrandma

# 542
Yokemate of Keyboards

Posts: 2720 from 2003/2/24

Quote:

JuLieN wrote:
On Talkchess forum we have a great test to benchmark ARM cpus : the Stockfish chess engine. It's perfect because :
- it uses all the cores you feed it with,
- it is available for Android and iOS
- chess programs love 64 bits, so you clearly see the difference between 32 and 64 bits CPUs.

We measure the average NPS Stockfish displays (NPS : Nodes Per Second : how many nodes in the chess tree Stockfish travels through pers second.

Here's our current charts (most powerful devices on top) :

Quote:

Galaxy S3 (4 cores) 574.000

SamsungGalaxy S4(1.9Ghz,4 cores,1GBhash)450.000 new

So the Galaxy S3 is actually considerably faster than the Galaxy S4?
MorphOS is Amiga done right!
MorphOS NG will be AROS done right!
»06.04.14 - 07:48

ASiegel

# 543
Posts: 1370 from 2003/2/15

From: Central Europe

Quote:

JuLieN wrote:
As it has 4 times more threads than the iPad Mini 2; you can see that both CPUs are about as powerful per core, despite the A7 runs at half the speed of the Core i7. This is VERY impressive !

This assumes that the used software provides meaningful data for comparing both processors. This may be a fallacy. Unfortunately, many tools that are widely used to review the performance of desktop processors are not available on mobile platforms which makes it rather difficult to do comprehensive comparisons between mobile and desktop CPUs.

That being said, the performance of smartphones in general is most certainly impressive, especially if you consider that today's high-end phones are roughly ten times faster than their predecessors were a mere 5 years ago (iPhone 3GS vs iPhone 5S, for example).
»06.04.14 - 10:14

JuLieN

# 544
Caterpillar

Posts: 35 from 2008/4/16

From: France

Quote:

takemehomegrandma a écrit :
Quote:

JuLieN wrote:
On Talkchess forum we have a great test to benchmark ARM cpus : the Stockfish chess engine. It's perfect because :
- it uses all the cores you feed it with,
- it is available for Android and iOS
- chess programs love 64 bits, so you clearly see the difference between 32 and 64 bits CPUs.

We measure the average NPS Stockfish displays (NPS : Nodes Per Second : how many nodes in the chess tree Stockfish travels through pers second.

Here's our current charts (most powerful devices on top) :

Quote:

Galaxy S3 (4 cores) 574.000

SamsungGalaxy S4(1.9Ghz,4 cores,1GBhash)450.000 new

So the Galaxy S3 is actually considerably faster than the Galaxy S4?

Yes, we've been astonished as well. Sometimes, behind the advertising, there are regressions for real life applications...
"A good bug is a dead bug" (Don Dailey)
»06.04.14 - 11:55

JuLieN

# 545
Caterpillar

Posts: 35 from 2008/4/16

From: France

Quote:

ASiegel a écrit :
Quote:

JuLieN wrote:
As it has 4 times more threads than the iPad Mini 2; you can see that both CPUs are about as powerful per core, despite the A7 runs at half the speed of the Core i7. This is VERY impressive !

This assumes that the used software provides meaningful data for comparing both processors. This may be a fallacy. Unfortunately, many tools that are widely used to review the performance of desktop processors are not available on mobile platforms which makes it rather difficult to do comprehensive comparisons between mobile and desktop CPUs.

That being said, the performance of smartphones in general is most certainly impressive, especially if you consider that today's high-end phones are roughly ten times faster than their predecessors were a mere 5 years ago (iPhone 3GS vs iPhone 5S, for example).

Well this one tool, Stockfish, is available for both desktop and mobile platforms. And as speed is very important for chess engines, it is each time carefully optimized. For instance, the Core i7 version has a dedicated binary that makes use of the POPCNT instruction (very useful for bitboards chess engines).
"A good bug is a dead bug" (Don Dailey)
»06.04.14 - 12:02

Andreas_Wolf

# 546
Yokemate of Keyboards

Posts: 12074 from 2003/5/22

From: Germany

Update:

>> Looks like Freescale has some interesting plans:
>> http://www.techradar.com/news/computing-components/processors/freescale-sampling-64-bit-arm-processors-already-launching-later-this-year-1230508

> Yes, this is some nice additional info to what was already known
> about QorIQ LS chips based on Cortex-A50.

First members of the QorIQ LS2 series with Cortex-A57 core announced to be available in 2H 2014:

http://media.freescale.com/phoenix.zhtml?c=196520&p=irol-newsArticle&ID=1916610
»08.04.14 - 23:33

Jim

# 547
Yokemate of Keyboards

Posts: 4977 from 2009/1/28

From: Delaware, USA

Quote:

Andreas_Wolf wrote:
Update:

>> Looks like Freescale has some interesting plans:
>> http://www.techradar.com/news/computing-components/processors/freescale-sampling-64-bit-arm-processors-already-launching-later-this-year-1230508

> Yes, this is some nice additional info to what was already known
> about QorIQ LS chips based on Cortex-A50.

First members of the QorIQ LS2 series with Cortex-A57 core announced to be available in 2H 2014:

http://media.freescale.com/phoenix.zhtml?c=196520&p=irol-newsArticle&ID=1916610

Not quite as fast as I was hoping for and only A57 cores, but still promising.
RISC seems to be alive and well at Freescale.
Interesting seeing them make a point of mentioning the continuing development of PPCs.
BTW - What the heck does "already launching this year" mean? Already?
"Never attribute to malice what can more readily explained by incompetence"
»09.04.14 - 02:28