% fortune -ae paul murphy

Bias and Kantor's Intel CSI analysis

Early last week a guy named David Kantor at Real World Technologies released a careful analyis of Intel's planned common systems interconnect (CSI) and its associated integrated memory controller (IMC) technology. One of the earliest comments on this came from zdnet's own George Ou who provided a quick review along with Kantor's own executive summary in a blog filed early Tuesday morning under the title: Details of Intel CSI "QuickPath" released.

One of the people who contributed to the discussion on this, signing himself as Uber Dweeb, had this to say about the column:

...and here's the spin

"This report explains how CSI works and why it will likely achieve lower latency and higher bandwidth than AMD?s HyperTransport."

I read the report this morning. There is much speculation which David Kanter is very upfront about. Nowhere did he make this assertion. His point was that both AMD and Intel are making changes which will keep both very competitive. It is unknown who will have the advantage. David Kanter is a prime example of how an author writes an article without an obvious prejudice. Any bias he may have doesn't overshadow his reason or logic. This is a great article about "Real World IT".

If you don't want to read the entire piece by Kanter, please at least skip to the conclusion page. http://www.realworldtech.com/page.cfm?ArticleID=RWT082807020032&p=13

"These changes will put Intel on a more equal footing with AMD, which has had a leg up in system architecture with their integrated memory controller and HyperTransport."

How George can completely skip over the multiple statements regarding AMD's advantages with HyperTransport to turn this little summary into a piece about how 'Intel's CSI is an AMD killer', is yet just another example of his prejudice. This is not "Real World IT" and happens so often it brings a reader to question a possible motive.

Unfortunately for Dweeb's credibility the sentence he takes such exception to is quoted from Kantor's executive summary, not George's interpretative text - and the headline he attributes to George: "Intel's CSI is an AMD killer" is neither on nor in the piece.

It's possible, therefore, to look at that sentence: "This report explains how CSI works and why it will likely achieve lower latency and higher bandwidth than AMD?s HyperTransport" as stating a reasonable conclusion developed by the original author on the basis of careful analysis.

It's equally possible, however, to view the coming to of that conclusion in the paper as the expression of Kantor's biases - and that's certainly what I thought when a friend's email brought the paper to my attention. Here's most of the response I sent him after skimming the report that morning:

ummm...

1 - was this written in 2003? If so it seems to have been reasonably forward looking... In particular, the author claims that CSI will erase Intel's performance deficit relative to AMD - but assumes throughout the article that AMD offers a stationary target with manufacturing but no design improvement since about 2004.

2 - the article is so x86 centric that it's annoying. Sentences like "Hardware prefetching is nothing new to modern CPUs, it has been around since the 130nm Pentium III", reflect an intentional ignorance combining arrogance with mis-information.

3 - I don't see any serious consideration of the integration issues affecting multi-core architectures, no knowledge that the 4 way SMP limit reflects MS compilers more than HW costs, and no acknowledgement that Intel's heading back to the gigahertz race in response to problems with Microsoft's software.

4 - the comments on Itanium are (unintentionally) tragi-comic: "The combination of doubling the core count, massively increasing bandwidth and reducing latency should prove compelling for Itanium customers and will likely cause a wave of upgrades and migrations similar to the one triggered by the release of Montecito in 2006." - that's the wave in which HP lost market share to IBM's P5 and AMD's opteron.

However, the real extent of the author's professional blindness to non Intel architectures other than Dec's Alpha line is most ironically highlighted by his view of the Intel Pentium Pro. Right at the beginning he notes, quite correctly, that the Pentium Pro introduced the technologies underlying today's Xeon servers and was a genuine Intel design success. Here's the second paragraph from that introduction:

Intel's front-side bus has a long history that dates back to 1995 with the release of the Pentium Pro (P6). The P6 was the first processor to offer cheap and effective multiprocessing support; up to four CPUs could be connected to a single shared bus with very little additional effort for an OEM. The importance of cheap and effective cannot be underestimated. Before the P6, multiprocessor systems used special chipsets and usually a proprietary variant of UNIX; consequently they were quite expensive. Initially, Intel?s P6 could not always match the performance of these high end systems from the likes of IBM, DEC or Sun, but the price was so much lower that the performance gap became a secondary consideration. The workstation and low-end server markets embraced the P6 precisely because the front-side bus enabled inexpensive multiprocessors.

All of that's true - with the significant note that all of the companies, like NCR and Sequent, then making high end intel based systems are now gone - but all of it's wrong too. What actually happened was that the P6 abandonment of the older, 16bit, Windows code most people were using at the time allowed AMD to steal much of Intel's market by issuing a series of chips combining backwards compatibility to the 16bit universe with the ability to run more modern 32 bit code.

In that context it's important to note that AMD's introduction of 64bit x86 machines was a Sun motivated replay of Intel's P6 debacle aimed at killing off Itanium - with hypertransport both a necessary consequence of the higher data flows required to support x64 and not that far off design ideas first embedded in Sun's scaleable processor architecture.

What seems to happen in this paper is that some parts of history, and therefore their continuations into the future, are simply lost. Consider what isn't there in these further re-statements of a view in which Intel is DEC's successor in high end CPU design and nobody else counts:

When the P6 front side bus was first released, it caused a substantial shift in the computer industry by supporting up to four processors without any chipset modifications. As a result, Intel based systems using Linux or Windows penetrated and dominated the workstation and entry level server market, largely because the existing architectures were priced vastly higher. (page 8)

and again in his Conclusions:

The success of the Pentium Pro and its lineage captured the multi-billion dollar RISC workstation and low-end server market, but that success also created inertia around the bus interface. Politics within the company and with existing partners, OEMs and customers conspired to keep Intel content with the status quo. Unfortunately for Intel, AMD was not content to play second fiddle forever. The Opteron took a portion of the server market, largely by virtue of its superior system architecture and Intel's simultaneous weakness with the Pentium 4 (Prescott) microarchitecture.

What isn't there is first that it took four years for Microsoft to get enough 32bit software into the market to vitiate AMD's ability to gain market share from 16bit compatibility, and thus let Intel get Xeon out the door in any volume - and what else isn't there is the entire dual core replay in which AMD forced Intel to dust off a rejected 2002/3 set of design ideas, forced Intel to adopt x64, forced Intel to adopt basic protections like execute bits, and is now forcing Intel to contemplate preserving only Itanium's instruction set through risc core microcode in a minor variation on the same machine all of their x86 products rely on.

Kantor's paper is about the specifics of Intel's new CSI/IMC technologies, not about Intel's competitive position, but the extent to which this reveals Intel's attempts to catch up with where AMD was several years ago suggests another question pro-Intel commentators don't seem to be asking: if Intel is so great, where does their ten year history of adapting to AMD design ideas come from?


Paul Murphy wrote and published The Unix Guide to Defenestration. Murphy is a 25-year veteran of the I.T. consulting industry, specializing in Unix and Unix-related management issues.