Exophase (exophase.delete@this.gmail.com) on February 7, 2015 3:35 pm wrote:
> Alberto (git.delete@this.git.it) on February 7, 2015 1:44 pm wrote:
> > I meant the topic of the slides, the performance at the same
> > power. I could be wrong but the "IPC" is not even named.
> > The problem is that they not say the "workload", yes because there are thousands of workloads and many of
> > them do not strees the cpu at all, allowing it to run at
> > lower clock speeds (with the right power management)
> > and at lower voltages. We all know like FInFet main strongness
> > is at the lower voltages available (the gains
> > are amazing !!!!) and they are ummm.....not so exciting but good enough as the voltage goes up.
> >
> > So i am pretty certain that their claims are about an average handset usage without pretending
> > much form the SOC. This is good!! being the more common situation in pratics.
> > Still as the things become serious, IMO all this gold becames
> > silver, forget this gold with the SOC under full load !!!.
> >
> > I am pretty in line with David, most of the advantage comes from the dramatic process shift.
> >
>
> Why would you mention Peter's 10-50% number, then go on to talk about the slides which we all
> already know are poor marketing? That makes it clear you read his post, yet you keep talking
> about it as if you have no idea which is incredibly misleading. And you can't keep saying that
> most of the advantage comes from process when it has now been revealed that there's a significant,
> sometimes huge IPC gain AND decrease in power consumption while on the SAME PROCESS.

The reason is, i can show you some syntetic bench (category so loved by ARM community) in which Haswell is around 60% faster at the same clock vs. the previous generation in few workloads :). Say 10-50% means nothing without additional informations, the bar is predominantly to the left or to the right ?? and the bench suite utilized to evaluate the estimation is industry standard or ......shitty.
So better stay on the official slides ignoring the IPC for now.

>
> > Sorry i have forgot your nickname :). This is a new thread,
> > i not want to criticize further the topic of many others.
> > Ha! Samsung plans to leave ARM IP in both Cpu and Gpu :).
> >
>
> I have no idea what you're saying. You were making the same old claims about Krait and
> I did nothing but question them, but I guess you don't want to talk about it now.

No, obviously, this is the wrong thread......anyway ARM said many times to OEMs to stay careful with clock speed in A15 and A57, but never was heard....
Sure the power consumption/clock speed ramp of Krait is less steep, and this is an advantage shared with Intel.