Thread: General Exynos or Snapdragon ?
View Single Post
Old March 15th, 2013, 03:19 PM   #14 (permalink)
EarlyMon
The PearlyMon
 
EarlyMon's Avatar
 
Join Date: Jun 2010
Location: New Mexico, USA
Posts: 46,620
 
Device(s): M8, LTEvo, 3vo, and Shift - Evo retired
Carrier: Sprint

Thanks: 42,834
Thanked 57,365 Times in 23,085 Posts
Default

Quote:
Originally Posted by davoid View Post
but... then is it only miniaturization that makes the newer cpu designs more efficient, and not improved memory controllers etc? I ask because it's heard said that newer processors can do more work for the same clock speed. But if that is the case, why not use newer processors at lower clock speeds rather than the older processors? Oh dear, I started off well...
No no - great questions!

Smaller structure geometry in silicon does indeed equate exactly to higher performance and lower power consumption.

And - we know a lot of what to expect by examining the manufacturing processes.

Let's use some numbers to sort it all out.

Both the S4 Pro and the Snapdragon 600, for example, are made with the same 28 nanometer manufacturing process.

The 600 is 15% faster because of its design, not the transistor size. Let's see the possible outcomes to know what that means, and assuming the same number of transistors (close enough for this discussion's purposes) -

I could do the same job on the 600 as the S4 Pro but run the clock 15% slower, and I'd get a power savings with no performance difference.

Or, I could run my 600 and S4 Pro at the same rates, but get 15% more processing power out of the 600 without using any extra battery.

See how that works?

Ok, let's use that, turn it around and see how that explains the Samsung chip -

Both the A7 and A15 quads are made on the same silicon, probably no bigger than the result of a 32 nanometer manufacturing process.

The A15 will have minimum and maximum possible clock speeds - ditto for the A7, but they will both be lower than for the A15.

For the A7, I don't care a lot about design efficiency - that affects performance. I don't care about performance there - I care about power efficiency. If I want performance, I'll go to the A15 quad. If I want power saving, I go with fewer transistors, running slower than my A15, all made the same size per transistor.

If I have two chips and run one at half the speed of the other, I'll save a lot of power. If I cut the number of transistors in half, I'll double my power savings (not saying that those are the exact numbers between the A7/A15, just laying it out to clarify).

And that's what the A7 is all about for power efficiency.

This approach actually appeared last year in the Tegra 3. That used an A9 quad plus a single slow, small core for the power savings.

Like the new Sammy, that too was a big/little design.

Some complained that the transition on the Tegra 3 from the single tiny core was kind of great when jumping to the A9 or vice versa.

And that would have gotten worse with an A15 quad.

So, Samsung evidently decided to combat that with a quad of smaller cores. I don't know, but I could see that it might scale up from 1 to all 4 cores on the A7 and then switch over to the A15 - just as one design example.

I'll stop there - does this help any??
EarlyMon is offline  
Last edited by EarlyMon; March 15th, 2013 at 03:47 PM.
Reply With Quote
The Following User Says Thank You to EarlyMon For This Useful Post:
davoid (March 15th, 2013)