Brett (ggtgp.delete@this.yahoo.com) on November 9, 2014 1:51 am wrote:
> David Kanter (dkanter.delete@this.realworldtech.com) on November 9, 2014 12:08 am wrote:
> > Brett (ggtgp.delete@this.yahoo.com) on November 8, 2014 4:10 pm wrote:
> > > David Kanter (dkanter.delete@this.realworldtech.com) on November 1, 2014 3:26 pm wrote:
> > > > Brett (ggtgp.delete@this.yahoo.com) on November 1, 2014 12:25 pm wrote:
> > > > > David Kanter (dkanter.delete@this.realworldtech.com) on October 31, 2014 10:26 pm wrote:
> > > > > > > The bulldozer core was over optimized for server use, the server market is where Intel
> > > > > > > makes it's money. Money that is used to punish competitors in the other markets.
> > > > > >
> > > > > > Intel makes profits in servers, but their stronghold is really in high volume (e.g.,
> > > > > > NB+DT). Without those, the server market would be much more difficult for Intel.
> > > > >
> > > > > Today, Intel was more than happy to make little money in the volume desktop market
> > > > > when AMD had competitive products, Intel made up the difference in mobile and
> > > > > server. AMD only has the design dollars to attack one market at a time.
> > > > >
> > > > > > >AMD has
> > > > > > > fewer problems making chips for mid range mobile and mid range desktop.
> > > > > >
> > > > > > The problem is that AMD has been confined to a value play.
> > > > >
> > > > > Even when AMD had faster chips than Intel, Intel's market power
> > > > > confined AMD to the value plays, the consumer market.
> > > >
> > > > AMD used to have around 50% of the 4S server market. They used to have ~25% of the
> > > > 2S server market. That's not a value play, and their server ASPs were quite high.
> > >
> > > AMD was competing against Intel's antique shared front side bus systems. Four sockets with
> > > four on chip memory controllers against four chips with one off chip memory controller.
> > > And still AMD was mostly only able to get the University/academic big computing clusters. You tell
> > > me, what part of business server sales did AMD get, 10%? Yes, AMD got a few google type sales, but
> > > at the fortune 1000 AMD was largely absent, and locked out by threats from Intel against Dell and
> > > HP. Until SuperMicro(?) started making serious inroads, but by then the tide was turning.
> >
> > Brett, AMD had >50% of the MP server market. HPC is and was predominantly DP. Their ASPs were quite
> > high. For 2006, AMD's ASPs were about 30% higher than Intel for servers (~$400 vs. ~$300).
>
> Fine, AMD got to >50% of the MP server market. I was obsessing over how hard it was
> to create a channel that would even sell AMD servers. It took years before SuperMicro
> stole enough share that Dell and HP had no choice but to sell AMD servers.

It took roughly 1.5-2 years. Opteron was introduced in late 2003, and they started taking significant share in 2005.

> Years that Intel could continue being lazy and unresponsive to the demands of the customers.
> A company that tried to force the mobile market to the P4 fireball. Plus Itanic.

> A company that puts all it's efforts into just one design, (plus token Atom) when it
> has ten times the engineers of AMD and could easily afford say three different designs
> for different markets. Instead of just cache size slicing the one and only design.

I think you are talking about CPU cores here, but could you clarify?

Today server chips use the same CPU core, but different fabric, cache, I/O, memory controllers, etc.

Going forward the fabric topology will be even more different (e.g., single ring for client, mesh for servers).

There's also a lot of variety in CPU cores at Intel:

1. Quark
2. Silvermont/Atom
3. KNL core
4. Haswell/big core

I think Intel is pretty serious about Atom/silvermont and will invest heavily there.

> > Please get your facts straight and stop spouting incorrect data.
>
> "At its peak during the second quarter of 2006, AMD held a 26.2 percent [all server types] market share."

Yes, that's right. Their market share was even higher in MP servers, although MP servers are a small portion of the server market.

> The multi socket market also peaked at around that percent.

I'm not sure I understood your sentence correctly. Do you mean to say that at its peak, the MP server market was only about 25% of the server market?

The multi-socket server share for AMD was >50%, based on data from Intel/IDC/Gartner.

> Lets shift over to the single socket server market, what was AMD's peak single socket server >market share?

I don't have 1P data from 2005-2007 (when they hit their peak). I do have AMD's server ASP, which was around $400-$450 compared to $300ish for Intel.

My real point here is that Intel's poor strategic choices (e.g., P4 core, Itanium, FSB, 32-bit) all compounded with AMD's good choices (e.g., 64-bit, integrated memory controller, K8 core) to create an opportunity for AMD to break into the server market and across the whole price spectrum (not just value; they hit the MP market and had really high ASPs). So it can be done.

Intel simply did not have the power to keep AMD's superior products out of the server space. However, once Intel started executing well (e.g., Woodcrest onwards), the tables turned and AMD lost share. When AMD mis-executed, they basically exited the market entirely.

What I take away is the following lesson:

-If Intel executes poorly and a competitor executes well, the server market will shift and Intel will suffer.

Looking forward, I don't see Intel really stumbling the way they were around 2000-2005. They are tightly focused on servers and making smart moves. It's not obvious to me whether competitors can really make inroads across the whole market when Intel is executing well.

I think the biggest hope for a competitor is to target a specific niche where they can out-execute Intel, since a big perf/power advantage will yield results.

David