By: anon (anon.delete@this.anon.com), June 30, 2013 10:41 am
Room: Moderated Discussions
EduardoS (no.delete@this.spam.com) on June 30, 2013 5:24 am wrote:
> anon (anon.delete@this.anon.com) on June 30, 2013 2:53 am wrote:
> > It is important to win benchmarks that are vaguely relevant.
> > [...] How important is it to win irrelevant benchmarks.
>
> Some times users base their decision on wich product to buy by looking at benchmarks,
> sometimes those benchmarks aren't really relevant to their uses.
This is your only justification for your assertion that multiple producers in some highly competitive markets are spending effort on useless product changes?
>
> > What do you mean, easier/cheaper? If it "improves" the chip for things
> > that don't matter, then it is not actually an improvement, is it?
>
> I don't think "doesn't matter" is the right expression, let's say that, doubling a particular metric will
> improve end user performance by 1%, that particular metric isn't very important, right? Let's also say that,
> to double that particular metric costs you 1% of the cost of the chip, worth? Likely, game-changer? No.
How easy do you think it is to double floating point performance? You can't on one hand say "that costs a lot" to an alternative that does not fit your narrative, and on the other hand say "doubling floating point performance is easy and cheap".
If it was so easy and cheap, and so important for useless benchmarks, then why would not Intel have added huge floating point performance to Silvermont? Answer that.
>
> > You could easily and cheaply "improve" a chip for an artificial test simply by adding a
> > vast amount of slow L4 cache to it, for example. Does not require any logic changes.
>
> But costs a lot, if it were free or very cheap manufactors would already have done that.
It costs depending on how much you put on, and how you measure the costs exactly.
Implementing floating point performance that you find in A15, for example, is not "free or very cheap".
>
> > Your logic simply doesn't hold up. If they just wanted to get flops, the CPU would look like a GPU.
>
> That would costs a lot as well.
>
No it doesn't. GPU can do more flops/watt than a CPU, and more flops/area. Just put a little A7 core in one corner to run the OS, and dedicate the rest to a GPGPU array.
> anon (anon.delete@this.anon.com) on June 30, 2013 2:53 am wrote:
> > It is important to win benchmarks that are vaguely relevant.
> > [...] How important is it to win irrelevant benchmarks.
>
> Some times users base their decision on wich product to buy by looking at benchmarks,
> sometimes those benchmarks aren't really relevant to their uses.
This is your only justification for your assertion that multiple producers in some highly competitive markets are spending effort on useless product changes?
>
> > What do you mean, easier/cheaper? If it "improves" the chip for things
> > that don't matter, then it is not actually an improvement, is it?
>
> I don't think "doesn't matter" is the right expression, let's say that, doubling a particular metric will
> improve end user performance by 1%, that particular metric isn't very important, right? Let's also say that,
> to double that particular metric costs you 1% of the cost of the chip, worth? Likely, game-changer? No.
How easy do you think it is to double floating point performance? You can't on one hand say "that costs a lot" to an alternative that does not fit your narrative, and on the other hand say "doubling floating point performance is easy and cheap".
If it was so easy and cheap, and so important for useless benchmarks, then why would not Intel have added huge floating point performance to Silvermont? Answer that.
>
> > You could easily and cheaply "improve" a chip for an artificial test simply by adding a
> > vast amount of slow L4 cache to it, for example. Does not require any logic changes.
>
> But costs a lot, if it were free or very cheap manufactors would already have done that.
It costs depending on how much you put on, and how you measure the costs exactly.
Implementing floating point performance that you find in A15, for example, is not "free or very cheap".
>
> > Your logic simply doesn't hold up. If they just wanted to get flops, the CPU would look like a GPU.
>
> That would costs a lot as well.
>
No it doesn't. GPU can do more flops/watt than a CPU, and more flops/area. Just put a little A7 core in one corner to run the OS, and dedicate the rest to a GPGPU array.
