45 nm Vs. 65 nm
I already talked about some of the details. Wolfdale is significantly smaller than Conroe, with a die size of 107 mm² compared to its predecessor's 143 mm² device size. This means that a 45 nm 6 MB Wolfdale processor requires only 74% the silicon size of a 65 nm 4 MB Conroe. Although Intel has rated all three versions (2.66 to 3.16 GHz) at 65 W, we found that the actual power requirement in fact stays well below the maximum of this so-called thermal envelope. According to Intel, the transition from 65 nm to 45 nm has been easy to achieve, as two thirds of the tools and technical equipment can still be used.
While the die size decreased, transistor size increased by 40% from 291 million to 410 million. Most of the additional transistors can clearly be attributed to the inflated L2 cache capacity. According to Intel, these Hafnium-based High-k transistors switch 20% faster, while requiring 30% less power when switching. Besides some changes inside the cores including a fast Radix 16 divider, improved virtualization support and a 128 bit shuffle unit, Intel also added a new set of instructions, summarized under the term SSE4.
SSE stands for Streaming SIMD Extensions, where the SIMD represents the term "single instruction, multiple data." There are 47 new processor instructions, all of which are designed to accelerate the creation and processing of digital content such as images, video or audio files. While SSE4 must be supported by applications in order to offer a performance benefit, the other modifications together with the increased L2 caches have a direct impact on performance in almost all of our benchmarks.
And there is another feature Intel introduces with Penryn: the Deep Power Down state. This is another C-states extension, which a processor can access to save power. While the L2 cache typically is flush to save power, the Deep Power Down state allows for power otherwise channeled to the L2 cache and to individual cores to be completely interrupted when not needed. Once execution of a core is resumed, the system will restore the architecture's last state.
Unfortunately, the Deep Power Down feature will only be available in the mobile Core 2 Penryn processors and not in the Wolfdale cores for the desktop. This approach might make limited sense from a product positioning standpoint, as Intel wants to sell mobile processors (at higher prices) for low-power applications and notebooks. But it certainly does not bode well for Intel's contribution to the industry's efforts to reduce power consumption and waste. However, since transitioning to deeper C-states does require energy by itself, Deep Power Down might not always make sense in a desktop environment, where more applications and services are active than on the desktop.
Lastly, Intel also already offers its Dynamic Acceleration Technology (DAT) in the 65 nm and 45 nm mobile processors. While this isn't used for E8000 series desktop models, DAT allows the processor to run one core at increased clock speeds as soon as the other core is in C3 or in a deeper state of sleep because it is idle. This is beneficial for single-threaded applications that will run best at high clock speeds.
There is one little change in the Core 2 processors' idle clock speed: While the multiplier for Enhanced SpeedStep still is x6 (think of it as the highest gear Intel used to reduce the "engine's" RPM in an effort to save gas), the idle clock speed is no longer 1600 MHz as with the FSB1066 processors, but it is 2000 MHz (333 MHz x6). While this has also been the case with the Core 2 Duo E6x50 models, we thought mentioning it doesn't hurt. Remember that the core is more efficient anyway.