San Francisco (CA) – Day 1 of the Fall Intel Developer Forum 2007 has come to an end and we are digesting the news and information we have received so far. TG Daily’s Rick Hodgin sums up his impressions about the key products at the show, Penryn and Nehalem.
Pat Gelsinger took the stage early Tuesday morning to introduce the day’s kick-off keynote speaker, Intel’s CEO Paul Otellini. But before he handed off to Otellini he had something rather surprising to say: He referred to IDF 2007 as the “Industry Developer Forum” and not the “Intel Developer Forum“. This proclamation echoed a consistent theme repeated throughout the day: Intel is very interested in supporting their partnerships and alliances.
In truth, Intel came across today as a new company. They appeared to be something less than the cold, hard stand-alone company they have often been perceived. That fact was also echoed internally by more than one senior level executive at a late evening event held for the press to mingle with Intel executives.
In fact, visions of a company that’s not only reaching out externally, but also reaching out internally were conveyed. It seems the internal mindset of even a few years ago has been stripped away. Replaced with a more flexible, dynamic Intel, which at least one senior executive indicated is the reason why Intel has delivered on many products since those years with Netburst, the now famous micro-architecture of the Pentium 4 which nearly brought the company down to its knees.
Extreme to mainstream
The central theme for Otellini’s 1.5 hour message was wrapped up in this phrase: The extreme to the mainstream. The idea is taking what’s extreme today and making it mainstream tomorrow. He talked about this being true in all areas, not just mobile space, or not just server space. But rather it is the goal everywhere. He also cited history as an indicator of where we were, what we have now and what‘s likely to come as things progress. Imagine a typical Starbuck’s today. Countless notebooks and WiFi connectivity. Five years ago that would have been rare. And ten years ago, impossible. All, according to Otellini, thanks to Intel’s contribution in connectivity, something he called “Internet everywhere.”
While there were many key ingredients given for Intel’s success, the first processor discussed was Intel’s upcoming mobile chip, Penryn, which we will see on November 12. The updated Core architecture is key, as is the underlying 45nm process technology.
Otellini provided an overview of the history of the insulating layer which, in modern CPUs, is only five molecular layers of silicon dioxide (SiO2) thick. He explained that as far back as 15 years ago, Intel’s engineers saw this layer as problematic. The continued scaling of the insulating layer could not continue forever. And, we found out later in the day with Dr. Gordon Moore’s keynote, that five molecular layers is about the lowest you can go in practice. It’s a form of wall, and Intel was right up against it.
Penryn will come with a hafnium-based hi-k dielectric and a metal gate solution. This technology helps to reduce gate leakage by 10x over existing models, according to Intel. This 10x reduction is not in an overall leakage reduction, but only gate leakage. We were told future solutions will address aspects of the remaining leakage, though no details or timeframes were disclosed.
It’s interesting to note that Gordon Moore talked about hard limits being reached in 10-15 years. Up until now, he said, the walls which have been out there in front of the designers have always managed to fall down somehow. But now, they are real walls, ones which might truly stop further development until they are solved.
Penryn also comes with some fairly significant architectural changes, which vary somewhat from what we have understood Intel’s tick-tock strategy to be. “Ticks” are described to be process revisions while “tocks” are supposed to be new processor architectures. Intel showed several slides indicating the same thing – but Penryn will be introducing fairly significant architectural changes that are much more than a refresh.
These include Intel’s Wide Dynamic Execution engine with a new, untested Radix-16 divider (the first significant divider change in many years). Also, there are new components for faster OS responses and virtualization. There is the advanced smart cache of 12 MB, which is also 24-way set associative. Add to that Intel’s Smart Memory Access, new SSE4 instructions and a super-shuffle engine. And finally, there is a new deep power down mode (C6), which and allows logic units to off-load their state into other logic units for a truer power-down.
Clock per clock, Penryns are faster, Intel said. It is unclear how much of this comes from the 50% larger L2 cache and how much comes from new architectures. Otellini, however, did indicate Penryns deliver 20% greater performance due to the 45nm hi-k/metal gate solution.
The big surprise of the day? Two live demonstrations of Nehalem in a physical machine. In the morning, we had a simple Nehalem system setup running Windows XP and a few applications. Otellini told us that Nehalems had been taped out for about a month, and what we saw was three week old “A0” silicon. He then held up a wafer showing wide, native quad-core Nehalems that were in use. Intel representatives made sure that they weren’t using the AMD term “native quad-core” when describing Nehalem. Instead, they referred to it as “quad-core on a single die”.
Another demonstration focused on the hyper-threading capability of Nehalems: Each core can support two threads. Intel showed off aa 16-thread machine, running two threads on eight cores in a dual-socket, quad-core configuration. The Nehalems were running 3D demos showing their I/O prowess with the GPUs.
Nehalem will adopt nearly everything AMD has been touting with K8 and Barcelona. The architecture will see on-die memory controllers, “QuickPath” communication between sockets (formerly called “CSI” and comparable to HyperTransport), including a multi-ported crossbar for communicating in 4-way configurations. Nehalem will also come with a significantly larger L2 cache. It will be a full 12 MB, arranged in dual-banks of 6 MB per dual-core. While Nehalem is a single-die quad-core design, its cache is still broken out into two half-size segments.
Nehalem is a behemoth at 820 million transistors, yet its die is only 107 mm2 in size. That’s 36 mm2 smaller than Penryn, which is expected to carry also 820 million transistors in its quad-core version (and 410 million in dual-core models.)
During Otellini’s keynote, the audience was given a glimpse into what qualities he believes “Intel is exceptional at”. These included three main points. First was silicon process technology. Otellini talked about their hafnium solution, which was discovered in 2003 and later made ready for production as Gen1 in Penryn. He said no other competitor has developed a hafnium solution. He also said that Intel has found a way to break down every wall they’ve encountered today. Second was a related two-fold impact: Intel’s core microarchitecture and to an ever increasing degree, chipsets. This core microarchitecture claim does not refer specifically to the Core or Core 2 brand name, but rather the continued push of making better products. And lastly is Intel’s ability to scale and diffuse products into the marketplace. Otellini said Intel’s ability to create a product and then push it into the global markets with such volume is one key factor in their success.
Currently mobile CPUs account for nearly 40% of all desktop + mobile processors shipped by volume, according to Otellini. That number is expected to reach the crossover point in 2009 where at least 50% of those CPUs shipped will be mobile. That’s up from 20% shipped in prior to Centrino’s release, according to Intel. This also includes some ultra-mobile PC initiatives, of which Intel will release details on day 2.
Otellini said in addition to being 100% lead-free and 100% halogen-free by the end of 2008, Intel will also begin reducing the size of its form factor by up to 60% in future products. These reductions are not just for mobile products. Otellini said Intel is looking to address future markets which will begin using x86 based processors, as well as smaller form factor servers in the future. Some early Penryn and Nehalem blade designs were shown in one session which were about half the size of traditional blades.
The first day of IDF was an absolute bustle of activity. Of course, Intel is using IDF as an event to rally developers and convince the press that it has great technology in the works. And, just like in the previous 18 months, the company created an impressive display and from what we have seen on day one, Intel has a product line-up that AMD should be concerned of. Tick-tock appears to have been working flawlessly so far.