Click Here
home features news forums classifieds faqs links search
6071 members 
Amiga Q&A /  Free for All /  Emulation /  Gaming / (Latest Posts)
Login

Nickname

Password

Lost Password?

Don't have an account yet?
Register now!

Support Amigaworld.net
Your support is needed and is appreciated as Amigaworld.net is primarily dependent upon the support of its users.
Donate

Menu
Main sections
» Home
» Features
» News
» Forums
» Classifieds
» Links
» Downloads
Extras
» OS4 Zone
» IRC Network
» AmigaWorld Radio
» Newsfeed
» Top Members
» Amiga Dealers
Information
» About Us
» FAQs
» Advertise
» Polls
» Terms of Service
» Search

IRC Channel
Server: irc.amigaworld.net
Ports: 1024,5555, 6665-6669
SSL port: 6697
Channel: #Amigaworld
Channel Policy and Guidelines

Who's Online
20 crawler(s) on-line.
 76 guest(s) on-line.
 0 member(s) on-line.



You are an anonymous user.
Register Now!
 cncparts:  1 hr 14 mins ago
 saipaman4366:  2 hrs ago
 Beajar:  2 hrs 19 mins ago
 Rob:  2 hrs 22 mins ago
 DiscreetFX:  3 hrs 25 mins ago
 agami:  3 hrs 25 mins ago
 RobertB:  3 hrs 45 mins ago
 OlafS25:  4 hrs 35 mins ago
 Bruce72:  4 hrs 36 mins ago
 MEGA_RJ_MICAL:  5 hrs 24 mins ago

/  Forum Index
   /  Free For All
      /  inefficient x64 HW ... ;-)
Register To Post

PosterThread
KimmoK 
inefficient x64 HW ...
Posted on 22-May-2015 11:50:15
#1 ]
Elite Member
Joined: 14-Mar-2003
Posts: 5211
From: Ylikiiminki, Finland

We had some discussion about power consumptions.
While normal PC consume 60-110W when they are on, it can also be done in another way.

Cell phone with 2GB DDR2 RAM and 1,33Ghz baytrail, measured by feeding 4,2V directly to the device's battery contacts:
-playing music (screen off) 0,5W
-screen on, idle on win8.1pro desktop 1,4W
-720p video playback, full screen 1,6W (varies between 1,2w and 6W)
-1080p video playback, full screen 1,8W (varies between 1,2w and 6,5W)
-1080p causes CPU load of 3...7% (summary from four cores), cores run at 0,6Ghz,
so, almost all of the processing is being done by the integrated video decoder
-Max peak power usage I saw was 8,5W


(IMO: this has ok performance for basic/normal desktop needs and very good in enterntainment because of the video & 3D offloads ... I hope I get my hands on high performance variant of this one day ... and I must benchmark WinUAE on this on some day...)

back to work...

Last edited by KimmoK on 22-May-2015 at 11:56 AM.
Last edited by KimmoK on 22-May-2015 at 11:54 AM.
Last edited by KimmoK on 22-May-2015 at 11:50 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

 Status: Offline
Profile     Report this post  
cdimauro 
Re: inefficient x64 HW ...
Posted on 22-May-2015 19:26:49
#2 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3621
From: Germany

@KimmoK: sorry, I don't understand why you opened this thread.

Anyway, BayTrail is the previous Atom family. Now there's CherryTrail, which draws less power (it uses 14nm process), and AFAIK has H265 support.

However pay attention that performance aren't its target, so don't expect big numbers with WinUAE. Also something important thing is missing: the AVX SIMD extension (only SSE4 is available).

If you like performance and still want to have low power consumption, I suggest you to take a look at the Core M family.

 Status: Offline
Profile     Report this post  
KimmoK 
Re: inefficient x64 HW ...
Posted on 25-May-2015 11:48:34
#3 ]
Elite Member
Joined: 14-Mar-2003
Posts: 5211
From: Ylikiiminki, Finland

@cdimauro

I was joking about the HW inefficiency.
Intel is not inefficient in power usage any more.
I'm amazed how good BayTrail is. I think it is more than compareable to the rest (vs ARM & PPC) and most likely does more per core per Mhz than those others...
Oldies, but anyway: http://cdn2.wccftech.com/wp-content/uploads/2013/08/Atom-Z3000-Series-Bay-Trail-CPU-Performance.jpg

So far I have not any CherryTrail to play with (or any Apple ARM variants).
(btw. those intel efficient chips are not cheap, I get 2 ... 6 PPC chips for the price of single intel???
But it seems some variants of for example BayTrail chips are cheaper than what I have seen in BOM, because I can buy BayTrail tablet for less than the cost of certain BayTrail chip separately.)


++++
10 years ago it seemed impossible for intel computer to consume less than 10W when idle, now these BayTrail equipped devices do it easily. (all computers idle at same speed after all and computers are on idle most of the time)

++++++
CherryTrail...
http://techreport.com/news/27879/cherry-trail-debuts-as-the-atom-x5-and-x7-series
Atom x5-8300 PCIe 2.0 x1 (max 1080p OpenGL3.1, max 2GB RAM, USD150...249?)
Atom x5-8500 PCIe 2.0 x2 (max 4k OpenGL3.1, max 8GB RAM, USD250...349?)
Atom x7-8700 PCIe 2.0 x2 (max 4k OpenGL3.1, max 8GB RAM, USD350+?)
vs BayTrail "Intel says Cherry Trail offers "very similar" CPU performance"

I assume getting 3D drivers done for these is not simpler than for any other commercial core, where documents usually are not available for smallest niches. But it would be crusial as PCIe offering is so weak.

((I bet MorphOS team would use AMD CPU+GPU as they know radeon GFX already.))

There exist SATA capable versions as well I believe, just no time to dig those out for now.

++++++++++++++++++
Something else...

T4240 (EXISTING)
24 virtual cores
4 64bit memory buses
L3 cache
45W
(IIRC, LS3400P is planned to have 40vcores in future)

A72 (server) core (FUTURE)
20...40 cores (+some kind of half speed hyperthreading??)
4 64bit memory buses
L3 cache
30W??

ARM says A72 should perform like intel Core-M ??
And faster with heavily multithreaded load.
http://techreport.com/review/28189/inside-arm-cortex-a72-microarchitecture

(IIIRC, cavium and other huge multicore ARM chips are sampling, we should see some benchmarks...)

Last edited by KimmoK on 25-May-2015 at 01:13 PM.
Last edited by KimmoK on 25-May-2015 at 01:12 PM.
Last edited by KimmoK on 25-May-2015 at 01:11 PM.
Last edited by KimmoK on 25-May-2015 at 12:49 PM.
Last edited by KimmoK on 25-May-2015 at 12:12 PM.
Last edited by KimmoK on 25-May-2015 at 12:08 PM.
Last edited by KimmoK on 25-May-2015 at 11:56 AM.
Last edited by KimmoK on 25-May-2015 at 11:53 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

 Status: Offline
Profile     Report this post  
cdimauro 
Re: inefficient x64 HW ...
Posted on 25-May-2015 19:18:11
#4 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3621
From: Germany

@KimmoK

Quote:

KimmoK wrote:
@cdimauro

I was joking about the HW inefficiency.

ROFL. Owned.
Quote:
(btw. those intel efficient chips are not cheap, I get 2 ... 6 PPC chips for the price of single intel???

It depends by the SoC used. Some are very cheap. You can take a loot at the Minnowboards or the recently introduced Intel Compute Stick, which are boards/products and not only SoCs.
Quote:
But it seems some variants of for example BayTrail chips are cheaper than what I have seen in BOM, because I can buy BayTrail tablet for less than the cost of certain BayTrail chip separately.)

Exactly.
Quote:
I assume getting 3D drivers done for these is not simpler than for any other commercial core, where documents usually are not available for smallest niches. But it would be crusial as PCIe offering is so weak.

((I bet MorphOS team would use AMD CPU+GPU as they know radeon GFX already.))

Consider that Intel publishes all the documentation about its GPUs from loooong time ago, so they are well know and supported (they usually have the best support by the open source community).
Quote:
There exist SATA capable versions as well I believe, just no time to dig those out for now.

Yes. That's why you can find Atom computers also.
Quote:
++++++++++++++++++
Something else...

T4240 (EXISTING)
24 virtual cores
4 64bit memory buses
L3 cache
45W
(IIRC, LS3400P is planned to have 40vcores in future)

Price?
Quote:
A72 (server) core (FUTURE)
20...40 cores (+some kind of half speed hyperthreading??)
4 64bit memory buses
L3 cache
30W??

Prise?
Quote:
ARM says A72 should perform like intel Core-M ??
And faster with heavily multithreaded load.
http://techreport.com/review/28189/inside-arm-cortex-a72-microarchitecture

(IIIRC, cavium and other huge multicore ARM chips are sampling, we should see some benchmarks...)

They are preliminary benchmark, made using simulators, if I remember correctly. Let's wait for real-work benchmarks.

 Status: Offline
Profile     Report this post  
KimmoK 
Re: inefficient x64 HW ...
Posted on 28-May-2015 10:30:00
#5 ]
Elite Member
Joined: 14-Mar-2003
Posts: 5211
From: Ylikiiminki, Finland

Just spamming silly "MIPS" comparisson...

A1000 1MIPS
A1200 2MIPS
A1200+fastRAM 4MIPS
A4000/030 9MIPS
A1200/030/50 18MIPS
A4000/040 25MIPS
A4000/060 80MIPS
Babylon5 rendernet 1000MIPS (pilot episode rendered on networked A4000/040 HW)

PPC440 666Mhz 1300MIPS (3.3W)
PPC460ex 1Ghz 2000MIPS (5W)
e5500 1.4Ghz 1 core 4200MIPS (4 cores would consume 5,5W)
e6500 1.8Ghz 1 thread of 1 core 5900MIPS MAX 130 000 MIPS (24threads, 36W)

ARM:
RasberryPi2 1Ghz 1core 1200MIPS
SamsungExynos5250 2Ghz 1core 7000MIPS??

X86/x64:
Atom 2,4Ghz 1 core 1 thread 4800MIPS?
AMDFX8150 3,6Ghz 1core 14000MIPS, MAX 109 000 MIPS (8cores, 110W)
inteli7 3...3.5Ghz 1core 30000MIPS, MAX 298 000 MIPS (16 threads, turbo3.5Ghz, 140W)

Special:
PS3 10 200MIPS
xbox360 6400MIPS/core
PA6T 1,8Ghz 4000MIPS/core (

Last edited by KimmoK on 01-Jun-2015 at 07:25 AM.
Last edited by KimmoK on 29-May-2015 at 10:48 AM.
Last edited by KimmoK on 29-May-2015 at 10:36 AM.
Last edited by KimmoK on 29-May-2015 at 06:46 AM.
Last edited by KimmoK on 29-May-2015 at 06:42 AM.
Last edited by KimmoK on 29-May-2015 at 06:38 AM.
Last edited by KimmoK on 28-May-2015 at 10:38 AM.
Last edited by KimmoK on 28-May-2015 at 10:37 AM.
Last edited by KimmoK on 28-May-2015 at 10:33 AM.
Last edited by KimmoK on 28-May-2015 at 10:31 AM.
Last edited by KimmoK on 28-May-2015 at 10:30 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

 Status: Offline
Profile     Report this post  
olegil 
Re: inefficient x64 HW ...
Posted on 28-May-2015 14:53:58
#6 ]
Elite Member
Joined: 22-Aug-2003
Posts: 5895
From: Work

@KimmoK

298000 on 16 threads is 18k and change per thread. So 30k must be single-thread (SMT is less than twice as fast, after all) in turbo mode. It would seem about right. Not sure how it's issuing 9 instructions per clock, but it seems to manage it somehow. recognizing patterns of instructions and joining them helps, I guess.

Also, 1.8GHz * 3.3 instructions per clock is 5940 MIPS, not 5400. That's the T4240 in single-thread mode. So the i7 is three times as fast per clock, twice as fast clock. two thirds as many cores, three times as many watts.

And the i7 has rather pathetic SMT according to those numbers. 30000 in single thread, 18625 per thread SMT.

Last edited by olegil on 28-May-2015 at 04:38 PM.
Last edited by olegil on 28-May-2015 at 04:36 PM.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

 Status: Offline
Profile     Report this post  
cdimauro 
Re: inefficient x64 HW ...
Posted on 30-May-2015 6:49:45
#7 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3621
From: Germany

@KimmoK: MIPS numbers are totally useless, because:
- they change the meaning/value for processors of different families;
- they change the meaning/value even with processors of the same family.

An example of the first case. Processors like x86s are able to execute in one cycle complex instructions like this:
[code]ADD DWORD PTR [EBX+EDX*4+0xdeedf00d],0xc0dedbad[/code]
whereas RISC processors requires several instructions to emulate the same operation. So, CISCs like x86 (and our beloved 68K) are able to do more "useful work" per instruction.

Now an example of the the second case. Atom processors are able to execute 2 instructions per clock cycle, as we know. However Atoms before the BayTrail microarchitecture are in-order, where are from BayTrail they are out-of-order. The result is that the second ones are 2-3 time (if I remember correctly) better performer.

Conclusion: take REAL applications in different areas if you want to compare processors.

Some examples. Do you want to stress the "integer" performance? Try with a emulator. Do you want to stress floating points? Try with a ray tracer. And so on.

 Status: Offline
Profile     Report this post  
KimmoK 
Re: inefficient x64 HW ...
Posted on 1-Jun-2015 7:31:19
#8 ]
Elite Member
Joined: 14-Mar-2003
Posts: 5211
From: Ylikiiminki, Finland

@cdimauro

I often use MIPS value because:
-it gives rough outline of the performance ballpark
-often I have found they match pretty well to what I see in real life
-those values are videly available
-most of system operations are done via integer instruction
-The lightening speed of an OS needs integer performance (and bandwidth)


(IMO, if OS behaves slower on 1000MIPS HW than AmigaOS on 1MIPS HW, the OS is broken. (most are, it seems))

For heavy math tests blender is pretty good tool, but it's performance changes up to 50% between builds, so results are very hard to compare.

Just storing some ARM SBC info:
http://www.unhappyghost.com/2015/02/raspberry-pi-2-vs-odroid-c1-vs-humming-board-i2ex-vs-banana-pi-pro.html

Last edited by KimmoK on 15-Jun-2015 at 11:38 AM.
Last edited by KimmoK on 01-Jun-2015 at 07:37 AM.
Last edited by KimmoK on 01-Jun-2015 at 07:36 AM.
Last edited by KimmoK on 01-Jun-2015 at 07:35 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

 Status: Offline
Profile     Report this post  
cdimauro 
Re: inefficient x64 HW ...
Posted on 4-Jun-2015 7:32:24
#9 ]
Elite Member
Joined: 29-Oct-2012
Posts: 3621
From: Germany

@KimmoK: MIPS are misleading, like I stated. For example, can you explain me how an e5500 1.4Ghz 1 core can make 4200MIPS? Each core can decode and execute a maximum of 2 instructions per clock, so 1400 * 2 = 2800 MIPS, which well below the 4200 MIPS that you reported.

Yes, I know the o.s. and most applications usually use the so called "integer unit(s)", but it doesn't mean that you have to take the MIPS as a reference. As I also stated, MIPS generated by an in-order and out-of-order processor (even of the same family) can be the same (at same frequency and number of instructions decoded & executed), but practically the latter is MUCH faster than the former.

Regarding Blender, of course you have to use the same version, otherwise you cannot compare the results.

If you need some numbers which are accepted in the industry, you can use the SPEC int and SPE float, which are also widely available.

 Status: Offline
Profile     Report this post  

[ home ][ about us ][ privacy ] [ forums ][ classifieds ] [ links ][ news archive ] [ link to us ][ user account ]
Copyright (C) 2000 - 2019 Amigaworld.net.
Amigaworld.net was originally founded by David Doyle