Question Phone processors faster than desktop ones?

Jacob 51

Upstanding
Dec 31, 2020
517
19
385
0
So now the mobiles have got 5 Nm lithography (Samsung), and the lowest we could get on a desktop is 7nm? (AMD) While intel is still stuck at 12 nm (correct me).

Is it easier to make phone processors or something?
 

Lutfij

Titan
Moderator
As technology advances you get higher pedigree tech crammed onto a smaller piece of substrate. Intel isn't stuck on 12nm...you need to read into their recent info about what they think is 7nm. As for your question at the end of the thread, phones don't do calculations/tasks like that seen on a desktop/PC, hence they are lower spec'd than what a desktop needs = less hardware to manage. You can't also slap an active fan on the back of a phone so they need to severely cut down to perform tasks but not burn themselves in the process.
 
Reactions: Howardohyea

Eximo

Titan
Ambassador
Hard to do a direct comparison between CISC and RISC processors as well. By their very nature it is 'simpler' to build a RISC processor (reduced instruction set vs complex instruction set). You basically have a fundamental design and then copy it as many times as your power and size window allow. (A bit simplified). You do the same with a CISC processor, but also have bespoke parts of the CPU capable of doing different and specific things. Each of these sections requires additional space and takes away from your bulk transistors.

So commonly the transistor count in a RISC chip is higher, and since power efficiency is the goal, they opt for the smallest process node possible, also extreme competition amongst phone hardware developers. PC space you still have efficiency concerns, but throughput is more important, so higher power budgets are allowed. Faster clock speeds is the end result. And those specific hardware designs allow some functions of the CPU to be very fast.

And of course the fundamental way you write software is different between the platforms.
 
So now the mobiles have got 5 Nm lithography (Samsung), and the lowest we could get on a desktop is 7nm? (AMD) While intel is still stuck at 12 nm (correct me).

Is it easier to make phone processors or something?
How small the transistor is has nothing to do with how well the processor performs. Here's a few counter examples to this:
  • ATi's Radeon 9700 (150nm) beating NVIDIA's GeForce FX 5800 (130nm)
  • AMD's first Athlon 64 (130nm) beating Intel's Pentium 4 (90nm)
  • NVIDIA's GeForce 10 series (16nm) going toe to toe if not better than AMD's Radeon Vega (14nm)
The day that phones are more powerful than PCs, then you will see them in data centers. Until then, phone/tablet=toy, PC=real computer. :D
If phones are more powerful than PCs, I'd say we're holding basically a nuclear device in our hands.

But seriously, trying to compare the two as-is is silly. Phones by design are power limited. However, I would argue that in terms of performance per watt, at least Apple's SoC makes a convincing case that the traditional PC processor is losing ground.
 
However, I would argue that in terms of performance per watt, at least Apple's SoC makes a convincing case that the traditional PC processor is losing ground.
The problem with any tightly integrated solution like a SoC is that there is basically no upgrade path, which is something that has helped PCs become the standard due to the ability to configure them as needed. With any embedded system like an SoC, it's a singular purpose which it does well (just like the SoC found in network switches), but beyond that it needs replacement/augmentation.
 
The problem with any tightly integrated solution like a SoC is that there is basically no upgrade path, which is something that has helped PCs become the standard due to the ability to configure them as needed. With any embedded system like an SoC, it's a singular purpose which it does well (just like the SoC found in network switches), but beyond that it needs replacement/augmentation.
What makes an SoC "singular purpose", given your example, is the software running on them. The SoC on a network switch can run anything else if you really want to hack into it. I mean heck, I think my router is running a minimal Linux build given it mentioned using busybox utilities.
 
What makes an SoC "singular purpose", given your example, is the software running on them. The SoC on a network switch can run anything else if you really want to hack into it. I mean heck, I think my router is running a minimal Linux build given it mentioned using busybox utilities.
Yes you can do this, but it's still quite limited to what you can do. For PCs, you're able to add ram if you need it, move from dual core to 8-core processors, tune for performance or thermals or anything in between. You simply are limited with an SoC compared to a PC. And that's how the SoC cost advantage comes into play. Make an SoC as flexible as a PC and the cost would be the same (at best).
 
Yes you can do this, but it's still quite limited to what you can do. For PCs, you're able to add ram if you need it, move from dual core to 8-core processors, tune for performance or thermals or anything in between. You simply are limited with an SoC compared to a PC. And that's how the SoC cost advantage comes into play. Make an SoC as flexible as a PC and the cost would be the same (at best).
I think the comparison here got lost in translation. I'm not comparing the flexibility or whatever of a system built around a SoC vs a PC. I'm comparing the performance of the CPU cores themselves in a vacuum.
 
I think the comparison here got lost in translation. I'm not comparing the flexibility or whatever of a system built around a SoC vs a PC. I'm comparing the performance of the CPU cores themselves in a vacuum.
The problem is that it's an apples versus oranges comparison--like a drag race between a turbine powered car and a gasoline one. Anything designed optimally for one function will be a master at that function, but that doesn't mean it will ever compete overall with something designed to be more flexible for multiple functions. It's why switches aren't x86 powered and are SoC/embedded.
 

wyliec2

Distinguished
Apr 4, 2014
139
11
18,715
11
I'm comparing the performance of the CPU cores themselves in a vacuum.
While this question on the surface might seem simple, it is fraught with ambiguity. The term PERFORMANCE must be defined in terms of something to be accomplished that can be quantified.

Even the simple term clock speed (GHz) is not comparable - what's the IPC, is it 32-bit, 64-bit, special purpose processors larger than 64-bit.

Are you comparing CPU's - iPhone CPUs have multiple cores of different types; there are CPUs with hundreds of cores available for purchase; there are massively parallel computing super computers.

Narrowing down to a single CORE - is it hyperthreaded? Is it commercially available? Do you include benchmarks of liquid nitrogen cooled CPUs which cannot run in a sustained manner? Again the questions of speed, IPC, word-size.

It's like asking what plane has the best PERFORMANCE:
In absolute speed?
In most thrust?
In longest range?
In best fuel efficiency?
In carrying the largest payload?
In shortest takeoff?
etc...
There's virtually NEVER an ABSOLUTE 'BEST'

Every 'BEST' is defined in terms of constraints - for mobile phones you have the most computing power to perform the tasks needed that will fit in X amount of space and draw no more than Y power and generate no more than Z heat....

IMHO the greater the ambiguity of the question, the more likely the answer will be 42....HGttG :)
 
Reactions: SamirD
While this question on the surface might seem simple, it is fraught with ambiguity. The term PERFORMANCE must be defined in terms of something to be accomplished that can be quantified.
Honestly, even if this was defined, people wouldn't accept the results anyway because people have a bias to justify spending a lot of money on something they really didn't need.
 

ASK THE COMMUNITY

TRENDING THREADS