Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on August 11, 2014 2:52 am wrote:
> Klimax (danklima.delete@this.gmail.com) on August 10, 2014 2:04 pm wrote:
> > Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on August 10, 2014 12:29 pm wrote:
> > > Klimax (danklima.delete@this.gmail.com) on August 10, 2014 11:01 am wrote:
> > > > juanrga (nospam.delete@this.juanrga.com) on August 9, 2014 5:26 am wrote:
> > > > > Klimax (danklima.delete@this.gmail.com) on August 9, 2014 2:51 am wrote:
> > > > > > Klimax (danklima.delete@this.gmail.com) on August 9, 2014 2:36 am wrote:
> > > > > > > iz (i.delete@this.z.x) on August 9, 2014 2:05 am wrote:
> > > > > > > > Klimax (danklima.delete@this.gmail.com) on August 9, 2014 1:31 am wrote:
> > > > > > > > > juanrga (nospam.delete@this.juanrga.com) on August 8, 2014 4:36 pm wrote:
> > > > > > > > > > Klimax (danklima.delete@this.gmail.com) on August 8, 2014 1:12 pm wrote:
> > > > > > > > > > > juanrga (nospam.delete@this.juanrga.com) on August 8, 2014 12:09 pm wrote:
> > > > > > > > > > > > Megol (golem960.delete@this.gmail.com) on August 8, 2014 11:23 am wrote:
> > > > > > > > > > > > > juanrga (nospam.delete@this.juanrga.com) on August 8, 2014 10:49 am wrote:
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > They compared older hardware. Migrating from SB-i7 to HW-i7 introduces little benefits
> > > > > > > > > > > > > > in performance (except when using new AVX2 extensions to x86) but in ARM each gen is
> > > > > > > > > > > > > > not a mere 5-10% faster than former gen but much more. Their choice favored x86.
> > > > > > > > > > > > >
> > > > > > > > > > > > > 5-10% is a huge difference given that it results from slight
> > > > > > > > > > > > > polishing. I have to say this point is nonsense.
> > > > > > > > > > > >
> > > > > > > > > > > > My point has been ignored.
> > > > > > > > > > > You'd first have to have point in the first place!
> > > > > > > > > > >
> > > > > > > > > > > You may not ever ignore base when comparing percentages. Which you just did.
> > > > > > > > > > > 5-10% when base is high is completely different then 50% from low base.
> > > > > > > > > > >
> > > > > > > > > > > 10% vs. 50% is in context of x86 versus ARM meaningless
> > > > > > > > > > > number, because ARM isn't even close to Sandy Bridge.
> > > > > > > > > > >
> > > > > > > > > > > Your point is nonsensical and is completely wrong.
> > > > > > > > > > >
> > > > > > > > > > > For some reason you have ignored evolution of Atom... because it demolishes entire thesis and your point.
> > > > > > > > > >
> > > > > > > > > > I notice that my point continue being ignored.
> > > > > > > > > >
> > > > > > > > > You have no point. There is nothing to be ignored.
> > > > > > > > >
> > > > > > > >
> > > > > > > > If you read what he actually said carefully, it's clear that his point is that comparing
> > > > > > > > old Intel hardware with old ARM hardware favours Intel, because ARM improved more than
> > > > > > > > Intel did in the same time frame, so their old hardware is relatively worse. His point
> > > > > > > > is that if they did the same comparison with new hardware ARM would do better.
> > > > > > > >
> > > > > > > > I think this point is only valid if the hardware was old when they started doing the tests, but if they
> > > > > > > > used the current stuff of the time and it took a while to
> > > > > > > > do all tests, write the paper and get it published,
> > > > > > > > then I think it's a bogus point because it's still a valid comparison for that time. Demanding that all
> > > > > > > > researches use brand new hardware for all their tests would favour Intel even more...
> > > > > > > >
> > > > > > > Ok. (To me he seemed to cross multiple approaches in discussion beside the paper)
> > > > > > >
> > > > > > > But I am pretty sure that paper in question did use best in
> > > > > > > that time. I'll dig out that old discussion we had here.
> > > > > > >
> > > > > >
> > > > > > Cortex A9 and Sandy Bridge appear to be of similar age.(2010 and 2011)Paper got released 2013.
> > > > > >
> > > > > > I don't think his point was valid anyway.(BayTrail)
> > > > > >
> > > > > > Original thread:
> > > > > > http://www.realworldtech.com/forum/?threadid=131745&curpostid=131745
> > > > >
> > > > > From same thread:
> > > > >
> > > > >

2. Why use ancient A8, and, worse, a Beagle board? For a 2013 paper you'd expect at least 2012 hardware.

> > > > >
> > > > >
> > > > Because paper didn't start in 2013? Have you missed what was used? Sandy Bridge. Not Ivy Bridge nor Haswell.
> > > >
> > > > Hell, they have Atom x450. Released in 2010.
> > > >
> > > > Your point never existed, because you ignored age of used
> > > > Intel's hardware and then you wonder why they have
> > > > "old" ARM. Because that paper was started in 2010! You cannot use unreleased stuff you don't have access to!
> > >
> > > I don't think it takes 3 years to write a paper like that. Cortex-A15 was widely available well before the
> > > paper was released, so it wouldn't have been much effort to include it. As others have already pointed out,
> > > the yearly gains in ARM performance are huge compared to x86. Unlike Atom, earlier ARM cores were limited
> > > by their memory system, and A15 changed that. So to be taken seriously you'd expect A15 to be included.
> > >
> > > If a similar paper was to be released again, would you expect it
> > > to include 64-bit ARMs and Silvermont or completely ignore them?
> > >
> > > Wilco
> > >
> >
> > Why it took years I have no idea. One delay might be time for peer review if done, another might be
> > assembling test suit and getting necessary runs. I can find unber of reasons why it took 3 years.
>
> Yes getting a test suite working is a lot of effort, but once you've
> got it then running it onto a new system is pretty easy.
>
> > That your ARM has huge gains is only because you had much lower bar to improvement.
> > Nothing more. Once ARM exhausts cheaper available stuff it will be similar to SB->Haswell.
> > And don't forget transition of Atom to Bay Trail. (Quite large jump)
>
> Atom had a huge jump in Q3 2013, but IIRC that was after the paper was released, so it's hard to argue
> it had to be in. And before then Atom performance had been completely flat for 5 years, so whether
> they used a 2008, 2010 or 2012 Atom would make little difference except perhaps for power.
Yes. I just wanted to point out that Intel produced same jump as ARM. (Because people love to use big jumps in ARM contrasted to SB->Haswell, yet miss that Bonnell->Silvermont jump)
Meant as beside paper under discussion. (Paper is just part of discussion)

Second I used Atom mostly as part of narrowing down year when they started. (the other is Sandy Bridge)

> > As for A15, do you have link when it was first properly available? Can't
> > find a thing, only announcement, which is not exactly useful.
>
> Check for the availability of the Chromebook with A15.
A15 announced in 2010. Design announced in 2011, shipping product in late 2012. Definitely late. (Otherwise they could use last "Old Atom" and Ivy Bridge, which would affect efficiency numbers.)

So A15 is out.

> > If released to day, no idea. Would depend on number of factors influencing when did actually
> > works start. SO it could easily miss both latest small cores from Intel, AMD and ARM.
>
> Well if it did miss the latest cores, what is the point exactly? To be relevant you've
> got to research stuff that is current, not something that is 3 years old.
>
> Wilco
>
For Intel side, it wouldn't matter much, because big cores have reached most efficient operation for given targets (10-95W regular, 90-150W Workstation/Servers)

For AMD for now it looks like it wouldn't be much different either.

ARM might be problematic, but if your objective is to look at ISA, then you need to cover latest version of ISA, so slightly older architectures are not that bit problem.

I guess one would have to ask authors what took most of time and if it could be avoided. I never did such comparison, so I got no idea on that front.