Mediatek MOAR COARS!

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

ShintaiDK

Lifer
Apr 22, 2012
20,378
146
106
That's why increasing core count and MT performance is the only realistic way to improve performance in any meaningful way.

You forgot the missing software and all the countless issues and limitations making multithreaded software.
 

Fjodor2001

Diamond Member
Feb 6, 2010
4,224
589
126
You forgot the missing software and all the countless issues and limitations making multithreaded software.

Depends on the workload. 8 cores vs 4 will give 100% higher performance on some workloads, 0 % in others, and anything in between is possible.

Even if 8 cores on average only provides 50% higher performance, that corresponds to the aggregate improvements of ~6 CPU generations each providing ~7% CPU increase. (1.07^6=1.5)
 

scannall

Golden Member
Jan 1, 2012
1,960
1,678
136
Depends on the workload. 8 cores vs 4 will give 100% higher performance on some workloads, 0 % in others, and anything in between is possible.

Even if 8 cores on average only provides 50% higher performance, that corresponds to the aggregate improvements of ~6 CPU generations each providing ~7% CPU increase. (1.07^6=1.5)

Back to the point of this thread. If you are running that many programs on your phone, you're doing it wrong. Seriously. You aren't going to be ripping DVD's with your Galaxy or your Oppo any time soon.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
146
106
Depends on the workload. 8 cores vs 4 will give 100% higher performance on some workloads, 0 % in others, and anything in between is possible.

Even if 8 cores on average only provides 50% higher performance, that corresponds to the aggregate improvements of ~6 CPU generations each providing ~7% CPU increase. (1.07^6=1.5)

What with all those where performance regresses due to your 8 cores.

http://anandtech.com/bench/product/1317?vs=1260

If it wasnt for the 5960X(140W) 20MB cache, how do you think its 3.0/3.5Ghz cores would do in average workloads against a 4/4.4Ghz 4790K(88W)?
 
Last edited:

Fjodor2001

Diamond Member
Feb 6, 2010
4,224
589
126
If you are running that many programs on your phone, you're doing it wrong.

Number of programs being executed concurrently != Number of cores in use.

One single program can use all available cores at once.
 
Last edited:

monstercameron

Diamond Member
Feb 12, 2013
3,818
1
0
Back to the point of this thread. If you are running that many programs on your phone, you're doing it wrong. Seriously. You aren't going to be ripping DVD's with your Galaxy or your Oppo any time soon.


Maybe not but you could run the cores in a lower power state and do all the background tasks simultaneously.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
146
106
Maybe not but you could run the cores in a lower power state and do all the background tasks simultaneously.

That fact you even need the cores turned on is the life killer for the battery.

There is a reason why Apple got the best smartphones out there. And no synthetic multithreaded benches.

Apple isnt under the lowest bar determiantor in its products. So Apple went the sane way like desktop and went for a fast dualcore. This you can feel with Apple products. Its a far superiour user experience. Ofcourse some comes from IOS, but most come from the sane design approach.

The only reason we even see these 8 and 10 cores with big.Little and god knows what to call this bastard version is because ARM failed to make a core covering all the performance/watt ranges. This is also why you rarely, if ever see all the cores on in normal life. But rather a context switch between the cores whos power profile matches the current usage.
 
Last edited:

scannall

Golden Member
Jan 1, 2012
1,960
1,678
136
Maybe not but you could run the cores in a lower power state and do all the background tasks simultaneously.

Background tasks have been around far longer than multiple cores. For desktops, everyone runs AMD down for having lots of cores, but lousy IPC. But for mobile people are running down great IPC, in favor of more cores?

For a tech site this seems pretty confusing to me.....
 

poofyhairguy

Lifer
Nov 20, 2005
14,612
318
126
There is a reason why Apple got the best smartphones out there.

The phones with 1GB of ram running 64Bit apps are the best phones out there? No way.

SoCs are more than CPUs, and the iPhone 6+ has less effective RAM than the 2012 iPhone 5 when you consider 64bit apps. Even leaving all the OS stuff to the side, Apple CPUs don't overcome the fact that you CANT multitask thanks to RAM limits. Its like a race car engine on a dune buggy.

If you want to say that the iPad Air 2 is the best tablet or even mobile device on the earth I won't argue. It takes that CPU, adds a core, and gives it the RAM it needs. But the iPhones aren't in its class.
 

poofyhairguy

Lifer
Nov 20, 2005
14,612
318
126
Back to the point of this thread. If you are running that many programs on your phone, you're doing it wrong.

That is a very ethnocentric point of view. For a lot of the world a smartphone is their first COMPUTER. We forget that in our lives when many have laptops, desktops, tablets and smartphones. In our deck the smartphone is at the bottom, not because of power (my smartphone is more powerful than the desktop that got me through college) but because we have form factor options.

Do you remember your first computer? Would you have liked someone implying you shouldn't do something with it because they thought it was a task too good for your device even if it could maybe technically do it?
 

Abwx

Lifer
Apr 2, 2011
11,885
4,873
136
In what metric?

In a fairy tale metric..

Also note that raising clock speeds might not be as feasible as it it on the desktop, higher clocks means higher voltage more heat and more power.

Raising IPC will yield the same result, that is power increasing as a square of perf delta.

Take a single core that consume 10W, wether you use IPC or frequency, or a mix of both, you wont have more than 3x the perfs by scaling this chip to 100W, now put 10 of thoses CPUs in a single chip and you ll have 10x the throughput for 10x the power comsumption, that s what killed the DT market, that is, to direct the market where it suits Intel s profitability but certainly not the enthusiasts and consumers needs.
 

scannall

Golden Member
Jan 1, 2012
1,960
1,678
136
That is a very ethnocentric point of view. For a lot of the world a smartphone is their first COMPUTER. We forget that in our lives when many have laptops, desktops, tablets and smartphones. In our deck the smartphone is at the bottom, not because of power (my smartphone is more powerful than the desktop that got me through college) but because we have form factor options.

Do you remember your first computer? Would you have liked someone implying you shouldn't do something with it because they thought it was a task too good for your device even if it could maybe technically do it?


My 8 bit, 1Mhz 6502? Yeah, that was a while ago. It's not 'ethnocentric' to point out poor design choices that were made creating that CPU. Poor choices endure regardless of borders or incomes.
 

Fjodor2001

Diamond Member
Feb 6, 2010
4,224
589
126
That fact you even need the cores turned on is the life killer for the battery.

There is a reason why Apple got the best smartphones out there. And no synthetic multithreaded benches.

They do? And who determines that? By what metrics? And how come their market share is so small compared to the Android phones?
 

Fjodor2001

Diamond Member
Feb 6, 2010
4,224
589
126
The only reason we even see these 8 and 10 cores with big.Little and god knows what to call this bastard version is because ARM failed to make a core covering all the performance/watt ranges.
No, it's because it's impossible to design a CPU core so that it performs ideally across the complete performance spectrum. You can always do better with two separate core designes, each optimized for it's particular performance range.
This is also why you rarely, if ever see all the cores on in normal life. But rather a context switch between the cores whos power profile matches the current usage.

Yes, that's the whole point of big.LITTLE! In low performance use cases, use only the low power cores with better perf/watt. In higher performance use cases, use the high performance cores with higher performance but also lower perf/watt.

And in case max performance is needed, use all cores.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
146
106
No, it's because it's impossible to design a CPU core so that it performs ideally across the complete performance spectrum. You can always do better with two separate core designes, each optimized for it's particular performance range.


Yes, that's the whole point of big.LITTLE! In low performance use cases, use only the low power cores with better perf/watt. In higher performance use cases, use the high performance cores with higher performance but also lower perf/watt.

And in case max performance is needed, use all cores.

You do know that all the ARM companies doing custom cores are not going big.LITTLE? Take a wild guess on why.

big.LITTLE is nothing but a hotfix bandage because someone couldnt afford the R&D and have to pay the penalty in increased manufactoring and support cost.
 
Last edited:

ShintaiDK

Lifer
Apr 22, 2012
20,378
146
106
They do? And who determines that? By what metrics? And how come their market share is so small compared to the Android phones?

Mainly because of the price range. Ask yourself, how many Galaxy S6 does Samsung ship for example?
 

imported_ats

Senior member
Mar 21, 2008
422
64
86
Depends on the workload. 8 cores vs 4 will give 100% higher performance on some workloads, 0 % in others, and anything in between is possible.

Even if 8 cores on average only provides 50% higher performance, that corresponds to the aggregate improvements of ~6 CPU generations each providing ~7% CPU increase. (1.07^6=1.5)

The problem is that in lots and lots of workloads, 8 cores is going to give WORSE performance than 2 cores.
 

imported_ats

Senior member
Mar 21, 2008
422
64
86
No, it's because it's impossible to design a CPU core so that it performs ideally across the complete performance spectrum. You can always do better with two separate core designes, each optimized for it's particular performance range.


Yes, that's the whole point of big.LITTLE! In low performance use cases, use only the low power cores with better perf/watt. In higher performance use cases, use the high performance cores with higher performance but also lower perf/watt.

And in case max performance is needed, use all cores.

You left how some major issues there, like overheads, scheduling, et al. Software isn't really there that makes bl work, and likely never will be there. So far, for actual delivered performance, you are better off just turning off the big part of bl.
 

Fjodor2001

Diamond Member
Feb 6, 2010
4,224
589
126
You left how some major issues there, like overheads, scheduling, et al. Software isn't really there that makes bl work, and likely never will be there. So far, for actual delivered performance, you are better off just turning off the big part of bl.

Care to be more specific about those issues, and do you have any data indicating to what extent that actually impacts performance?

Also, if it would be such a problem and bad solution, then how come ARM, Samsung and Qualcomm all are using b.L?
 

monstercameron

Diamond Member
Feb 12, 2013
3,818
1
0
The problem is that in lots and lots of workloads, 8 cores is going to give WORSE performance than 2 cores.


You have to qualify that statement, are you saying that a higher ipc 8 core will lose to a lower ipc dual? I intentionally twisted that to prove a point.

Also don't let legacy prevent you from moving forward, going forward a lot of algorithms might have to be redesigned to scale with core count.

Also note that ST perf has stumbled on a roadblock due to clock speeds hitting a wall, ipc and core count bare really the only way to achieve more throughput.
 

Fjodor2001

Diamond Member
Feb 6, 2010
4,224
589
126
You do know that all the ARM companies doing custom cores are not going big.LITTLE? Take a wild guess on why.
All the major ARM mobile phone players (ARM, Qualcomm and Samsung) are using b.L.
big.LITTLE is nothing but a hotfix bandage because someone couldnt afford the R&D and have to pay the penalty in increased manufactoring and support cost.

No they use it because it's a better solution to be able to cover the complete performance spectrum while maintaining optimal perf/watt throughout the complete range.

Qualcomm and Samsung does not have enough R&D money? You must be joking.
 

ShintaiDK

Lifer
Apr 22, 2012
20,378
146
106
All the major ARM mobile phone players (ARM, Qualcomm and Samsung) are using b.L.


No they use it because it's a better solution to be able to cover the complete performance spectrum while maintaining optimal perf/watt throughout the complete range.

Qualcomm and Samsung does not have enough R&D money? You must be joking.

Qualcomm only use big.LITTLE until they got their own custom cores ready. Apple doesnt use big.LITTLE at all. So even Qualcomm tells you that big.LITTLE is subpair. (Not to mention a disaster software support wise).

Samsung actually dont have that much R&D. Thats why they use vanilla designs. Just like the rest that cant afford custom cores.

bulletin20150224Fig01.png


Without the foundry Samsung may not even be in top10.
 
Last edited:

ShintaiDK

Lifer
Apr 22, 2012
20,378
146
106
You have to qualify that statement, are you saying that a higher ipc 8 core will lose to a lower ipc dual? I intentionally twisted that to prove a point.

But you forgot the point when you twisted your question into something without relevance to justify more cores.

Plenty of examples on 8 cores being slower than 4. I even supplied it in this thread. For some reason the notion is that you can just add cores without penalty. But there are plenty of penalties, software and hardware.