By: David Kanter (dkanter.delete@this.realworldtech.com), May 20, 2013 4:30 pm
Room: Moderated Discussions
Maynard Handley (name99.delete@this.name99.org) on May 19, 2013 6:27 pm wrote:
> David Kanter (dkanter.delete@this.realworldtech.com) on May 17, 2013 9:29 pm wrote:
>
> > Intel's comparisons vs. A15 were normalized to power and/or specific devices. That's
> > a very fair equalization point. Far more sensible than equal frequency.
> >
>
> I don't have a dog in this race, but this is, IMHO, NOT the
> most useful way to compare against CPUs for most users.
>
> What matters for most users is:
> (a) low power utilization for on-going background computation. This appears
> to be largely a solved problem by all vendors, and things like companion core
> of big.LITTLE are just minor refinements of what was already good.
>
> (b) peak performance that is as high as possible so the machine feels as snappy as possible.
> It doesn't matter if this peak uses a lot of power because it's not going to last long.
>
> The point is: three hours of on-going linear algebra, or even three hours of on-going CPU based video compression
> are NOT realistic use models for phones (and even tablets). So no-one should care much about things like
> how much energy it takes to run a computation and time intensive task. The focus should be what the CPU
> provides to make the device feel as snappy as possible, because snappiness is the primary place where performance
> matters, and it is how the device will be judged by users as "fast" or not.
That's not actually true. People watch movies on tablets, all the time. Some are feature length, some are just 30 minute TV shows, and some are 0.5-5 minute youtube clips. Those are all fair workloads.
I agree that Linpack is an odd benchmark.
And you really do care about power consumption.
> I don't know how to interpret the fact that Intel has not provided any benchmarks which show Atom as
> "snappier" than ARM.
What benchmark would that be? I'm not even sure how to define this problem, let alone test for it.
Obviously introducing a new type of benchmark suggests that you might be making
> special pleading for your device, but Intel have done this in the past, and doing so is the mark of
> a confident company, one that believes its tech is so superior that it can fight on the technical merits
> and take time to educate the public without having to play games. To me the fact that Intel is not pushing
> this angle suggests that they are NOT fundamentally superior on this most important metric.
What is the 'snappiness' metric?
Also I fundamentally disagree with you about energy. Power and energy always matter. A snappy experience that kills your battery isn't very interesting when you are away from a wall outlet.
David
> David Kanter (dkanter.delete@this.realworldtech.com) on May 17, 2013 9:29 pm wrote:
>
> > Intel's comparisons vs. A15 were normalized to power and/or specific devices. That's
> > a very fair equalization point. Far more sensible than equal frequency.
> >
>
> I don't have a dog in this race, but this is, IMHO, NOT the
> most useful way to compare against CPUs for most users.
>
> What matters for most users is:
> (a) low power utilization for on-going background computation. This appears
> to be largely a solved problem by all vendors, and things like companion core
> of big.LITTLE are just minor refinements of what was already good.
>
> (b) peak performance that is as high as possible so the machine feels as snappy as possible.
> It doesn't matter if this peak uses a lot of power because it's not going to last long.
>
> The point is: three hours of on-going linear algebra, or even three hours of on-going CPU based video compression
> are NOT realistic use models for phones (and even tablets). So no-one should care much about things like
> how much energy it takes to run a computation and time intensive task. The focus should be what the CPU
> provides to make the device feel as snappy as possible, because snappiness is the primary place where performance
> matters, and it is how the device will be judged by users as "fast" or not.
That's not actually true. People watch movies on tablets, all the time. Some are feature length, some are just 30 minute TV shows, and some are 0.5-5 minute youtube clips. Those are all fair workloads.
I agree that Linpack is an odd benchmark.
And you really do care about power consumption.
> I don't know how to interpret the fact that Intel has not provided any benchmarks which show Atom as
> "snappier" than ARM.
What benchmark would that be? I'm not even sure how to define this problem, let alone test for it.
Obviously introducing a new type of benchmark suggests that you might be making
> special pleading for your device, but Intel have done this in the past, and doing so is the mark of
> a confident company, one that believes its tech is so superior that it can fight on the technical merits
> and take time to educate the public without having to play games. To me the fact that Intel is not pushing
> this angle suggests that they are NOT fundamentally superior on this most important metric.
What is the 'snappiness' metric?
Also I fundamentally disagree with you about energy. Power and energy always matter. A snappy experience that kills your battery isn't very interesting when you are away from a wall outlet.
David