In a speech this morning at Intel Developer Forum in Beijing, senior vice president and long-time conference favorite Pat Gelsinger presented initial performance estimates for the latest prototypes of Penryn, the company's first high-k-plus-metal-gate technology processor at the 45nm lithography level. There, Gelsinger presented figures pointing to as much as 25% faster performance in 3D rendering operations for Penryn prototypes over today's Core 2 Duo, introduced last July.

Intel's tests tend to focus on processor performance rather than that of systems in which those processors are installed. So conceivably, with GPU speeds due to increase even between now and later in the year when the first Penryn CPUs are shipped, desktop system perceived speeds could actually be faster still.

Desktop systems with Penryn CPUs installed, Gelsinger went on, could enjoy 15% greater speed in imaging and 20% greater encoding speed, due in large measure to on-chip support within Penryn's new Super Shuffle Engine microcode (SSE4) for DivX. Gaming performance could be optimized by as much as 40% over Core 2 Duo.

What does this mean from a consumer's perspective? The Pentium 4 520, at 2.8 GHz, first made its way into consumer PCs in the spring of 2004. A Core 2 Quad QX6700 CPU has 88% better gaming performance, by reliable estimates, than that same P4 520. A 40% performance boost yet again, if these numbers are indeed borne out, would actually give Penryn 163% of the P4's performance in that category. That means a desktop PC with a single quad-core Penryn would be like using a "3P" 520 system.

Of course, with the P4 520 having been single-core, that performance increase should be just about right.

Gelsinger said the high-k-plus-metal-gate technology (the credit for whose discovery Intel declines to share with IBM) is enabling not only 20% increased switching speed among transistors, but is also enabling Intel to resume accelerating clock speeds - a practice Intel reluctantly began shunning as the power-saving era began in 2005.

Desktop Core 2 processors with Penryn architecture, he illustrated, could be clocked at greater than 3 GHz, while maintaining a 65 W thermal design point (TDP). 3 GHz versions of today's Core 2 Duos are rated at 80 W TDP.

Response to the news at IDF was particularly enthusiastic among notebook engineers, who are fully aware that Penryn will be the pinnacle component of Intel's forthcoming Centrino Pro platform, currently code-named "Santa Rosa." All the key parts of Santa Rosa appear to be coming together this week, including new systems-on-a-chip (SoC) for low-cost networking support. Intel confirmed its Next-gen Wireless-n 802.11n SoC will be supported. In recent years, the Centrino platform has been pivotal in determining WiFi standards; and Intel's leveraging Centrino Pro for its own flavor of 11n this time around could pose the most serious obstacle to date to rival network chip producer Qualcomm.

Centrino Pro, it was revealed, will also feature support for a radically new architectural concept code-named Geneseo. Here for the first time, Intel processors will support external processors, perhaps from third parties, who can lend integrated processing support directly to the system for vertical applications.

So a math-intensive server could be provided to an engineering department, for instance, or speech synthesis modules could be added to a system on the fly. Or perhaps a game console could be developed that's better suited for artificial intelligence.

Intel calls these external processors accelerators, borrowing a phrase from the '80s. Today, Intel fellow Ajay Bhatt projected the possibility that custom-made accelerators could be attached to PCs using the PCI Express bus, which would eliminate the problem of designers having to develop unique platforms to support individual accelerators. Potentially, customers could make those choices for themselves using interchangeable parts.

But in each of the above cases, accelerators would need to be supported by software. Here is where Intel's architecture may be leveraged to change the nature of software, perhaps restoring Intel's long ago advantage of being guaranteed compatible with the majority of the software base.

Intel's David Perlmutter is expected to reveal more about his company's next-generation MID platform in a speech now scheduled for Wednesday, Beijing time.