bakaneko (nyan.delete@this.hyan.wan) on September 25, 2013 11:46 am wrote:
> anon (anon.delete@this.anon.com) on September 25, 2013 9:24 am wrote:
> > bakaneko (nyan.delete@this.hyan.wan) on September 25, 2013 9:00 am wrote:
> > > anon (anon.delete@this.anon.com) on September 25, 2013 7:43 am wrote:
> > > > bakaneko (nyan.delete@this.hyan.wan) on September 25, 2013 6:58 am wrote:
> > > >
> > > > > But how good is eDRAM? It increases
> > > > > complexity, but doesn't scale up or
> > > > > down.
> > > >
> > > > http://2.bp.blogspot.com/-AVJnvkbUQEU/UMd7j5NupdI/AAAAAAAAAUY/k31deuzDaMQ/s1600/Fig+5.png
> > >
> > > There is more than one way to scale.
> >
> > I just gave you a link that shows you it scales. Then there is further
> > reports (e.g., at IEDM) of roadmap to 3d stacking technologies.
>
> You gave me a link that shows that it scales
> with node shrinks.

Yes. It scales.

>
> You didn't give me a link that shows me
> how useful it will be if people try to adopt
> it into all the processes which come with
> real products.

With that link, I was rebutting your assertion that it does not scale. *Not* trying to support a claim that you just dreamed up then. Nice attempt at a strawman, though.

>
> Which is the kind of scaling that matters.

Sounds like marketing mumbo jumbo to me.

eDRAM scales (at least for several more generations), and that's the end of it.

Would it be suitable for phones as a replacement for main memory? This is what is being discussed here.

>
> Just because it is feasible doesn't make it
> useful. Same for die stacking.
>
> > > How
> > > well does eDRAM scale between SRAM and
> > > normal DRAM? Would you replace all your
> > > SRAM with eDRAM? All your DRAM with eDRAM?
> >
> > This is what is being discussed.
>
> But it is only one application of eDRAM.

By amazing coincidence, that is the one application that we are talking about.

>
> > > IOW, if it can't replace either, it will
> > > be yet another solution which needs to
> > > be carried around.
> >
> > You mean to say, another tool in the bag of microarchitects? Then that's great.
>
> Just because it is feasible doesn't make it
> useful.

What is this semantics game you're trying to play?

> And useless tools just mean wasted
> time.
>
> > And, FYI, it has already been widely used for a long time in such capacity.
>
> Where? This is the first time I heard of it. I
> may have missed some special products, but I doubt
> everyone switched to eDRAM while I wasn't looking.

Pretty trivial to find them.

>
> > > > > It's just yet another
> > > > > indirection.
> > > >
> > > > What does this mean?
> > >
> > > IBM and Intel use it in separate dies
> > > between normal DRAM and their internal
> > > caches.
> >
> > That is not what is being discussed.
>
> It sounds important, why not discuss it?

Because that was not the topic of this thread.

>
> > > IIRC IBM also uses it inside the dies
> > > which carry the cores as caches. So,
> > > I'm off here.
> > >
> > > > > And it may be replaced
> > > > > with a different technology in a few
> > > > > years.
> > > >
> > > > Same can be said for everything.
> > >
> > > Not in context. For how many decades was DRAM
> > > (whatever version) as a separate package sold?
> >
> > For smartphones? Either zero or one, depending on when you start counting.
> >
> > But the good thing is that they are not wedded to long term solutions
> > and form factors. If something is cheaper, it will get used.
>
> Long term solutions are part of the cost.

What do you mean, and what is your point?