[GTALUG] the meaning of Core i[3579] (was: NUC NUC NUC)

[Lennart Sorensen]

On Wed, May 22, 2019 at 12:24:33PM -0400, D. Hugh Redelmeier via talk wrote:
> TL;DR: Intel uses brands.  We care about technology.  Sometimes brands
> map cleanly onto technology, but sometimes they don't.  You cannot
> trust that mapping.
> 
> Here's my inference on the matter (not based on careful research):
> 
> - these names are brands, not technology
> 
> - we (techies) want them to represent technologies.  For example, we
>   kind of know what "Core i5" means
> 
> - when times are good, and changes are smooth, intel tries to align
>   its brands with technologies
> 
> - when Intel is desperate, they muddy the waters of their brands
> 
> Example trajectory:
> 
> - Pentium meant one thing.
> 
> - Then it got extended with the II and III branding AND technologies.
>   Still related, and the branding evolved in a coherent way

Actually the II and !!! were extensions of the Pentium Pro which was
completely different from the Pentium.

Why did intel marketing decide on "Pentium !!!"?

> - Then the Pentium 4 came along and it was something completely
>   different

Seems that netburst architecture tainted the pentium brand enough that
they needed to replace it and hence Core came along.  First Core chips
were essentially rebranded Pentium M chips as far as I remember.  Core 2
were the first new ones.

> - Pentium M was also technically different, but that was signified by
>   a sufficiently different name (in my perception)

The Pentium M is as far as I have understood it a power optimized extension of the Pentium !!!

The Pentium Pro, II, !!! and M were all P6 architecture.  The Pentium
was P5.

> - "Pentium" lay dormant for a bit, and then got attached to some
>   Atom-architecture and low-end (crippled) Core chips.  Talk about
>   confusing!  It became a pricing / value designator and not a
>   technology.

Seems Pentium these days has replaced the old Celeron brand.

> Intel is in trouble.  It's latest fab processes (10 nanometer) are not
> coming on stream as quickly as they had planned and promised.
> Historically they have been significantly ahead of other fabs.  Now
> they are behind and squirming visibly.  This has thrown their whole
> roadmap off.
> 
> The i9 seems to be a way of showing forward motion without needing the
> new fab processes.  A way to appear to "refresh" their line without
> stressing their manufacturing capabilities.

Seems like the i9 are essentially Xeon server chips tweaked for high
end desktop use.

> As far as I recall, they do have one 10nm product that they claim is
> shipping but it's not very good and so probably nobody is buying it.
> It is probably so expensive to produce (low yields) that they don't
> want anyone to buy it.  It's a mobile chip without a GPU (probably due
> to yield issues) that isn't notably better than the normal 14nm
> versions that are shipping.
> 
> Since 10nm isn't productive, they are shipping as much 14nm product as
> they can make.  But it's not enough for the market.  They didn't build
> up 14nm capacity because they thought they would be using 10nm fabs.
> 
> Intel also has AMD as a competitor and threat for CPUs, something they
> could ignore for a long time.  The upcoming "Rome" AMD server chip
> looks to be particularly scary.  AMD is selectively using TSMC's 7nm
> technology (apparently comparable to Intel's 10nm -- don't get fooled
> by the number).
> 
> In the eighth generation of Core, they moved some extra cores and HT
> down to lower-numbered models, probably to ward off the mild AMD
> threats.  Perhaps to stimulate consumer excitement.  In my mind, this
> happens to make the i5 a "sweet spot".  And make the 8th gen fairly
> attractive (up until now, I've been content sitting at the 4th gen).
> But maybe the next one will better deal with the speculation problems.
> 
> Don't count Intel out.  Their server chips were inferior to AMD's
> Opterons for a long period but the server market didn't significantly
> switch to AMD.  Then Intel technology caught up and surpassed AMD.

Some did though.  Intel did eventually design something new and QPI
was better than hypertransport when they eventually released it, but
for many years they were behind.  Intel also seems to have better FPUs
in general than AMD, although the next generation AMD chips coming out
this year should narrow the gap quite a bit.

> ================
> 
> The term "microarchitecure" is used to describe how a chip is
> organized.  The Core "microarchitecture" started with the Core 2 Duo
> branding.  The "Core" branding before that had a different
> microarchitecture (32-bit).  So confusing.
> 
> Intel's current Core-brand processors (and XEON-brand too) are still
> based on an evolved Core microarchitecture.

That P6 based design just keeps going.  Unlike that netburst that followed
it.  I even remember when netburst was first talked about in magazines
thinking "This contradicts a lot of what I have learned about good cpu
architecture at university.  How can they expect this to ever work well?".
Turns out it didn't work well for exactly the expected reasons.

> AMD's Zen microarchitecture is newish, and it is what has put them
> back into the game.
> 
> When I talk about "Atom", I'm refering to microarchitecture.  The
> branding of those products is way too hard to follow.

The different generations of atom have also been quite different.
Some were in-order designs, later they had out-of-order execution.
Some had SMT, some did not.  Just another brand for low power/embedded
it seems.

-- 
Len Sorensen