Kal-El will succeed Nvidia's Tegra 2 applications processor later this year, bringing two additional cores and an extra helping of performance. We already knew a little bit about the chip, but Nvidia has just released two whitepapers containing a wealth of fresh information—including news that Kal-El will feature a fifth CPU core designed to handle menial tasks while the four regular cores are asleep:
The first whitepaper outlines the patented fifth core. This extra core – which we call the “companion core” – runs at a lower frequency and operates at exceptionally low power.
During less power-hungry tasks like web reading, music playback and video playback, Kal-El completely powers down its four performance-tuned cores and instead uses its fifth companion core. For higher performance tasks, Kal-El disables its companion core and turns on its four performance cores, one at a time, as the work load increases.
The Variable SMP architecture is also completely OS transparent, which means that operating systems and applications don’t need to be redesigned to take advantage of the fifth core.
The Variable SMP whitepaper (PDF) explains that the companion core will be used both when the host device is in active standby mode and when running applications "do not require significant CPU processing power." Apps without high CPU power requirements include audio and video playback, either online and offline, which will be "largely processed by hardware-based encoders and decoders."
According to the whitepaper, the companion core is based on the same ARM Cortex-A9 architecture as the regular cores, but it's fabbed on a low-power process and optimized for operation up to 500MHz only. The regular cores, meanwhile, are fabbed using a "General/Fast" process and can climb into the gigahertz range. The whole CPU will be manufactured by TSMC at the 40-nm node, just like Tegra 2. The mention of different process technologies suggests Kal-El's companion core might be a separate chip, but the diagram in the whitepaper suggests otherwise. We've asked Nvidia for clarification.
In any case, judging by the chart below, the low-power companion core will enable Kal-El to be quite a bit more power-efficient than Tegra 2—even though Tegra 2 has two fewer cores:
Sadly, the whitepaper doesn't say if the companion core has a pink heart etched on it.
The first Kal-El devices could ship sooner rather than later. Whispers from the rumor mill have hinted that Asus could unleash a new tablet—perhaps a second-gen Eee Pad Transformer—based on the new Nvidia silicon some time in October. Nvidia CEO Jen-Hsun Huang boasted in May of this year about the number of Kal-El design wins: "It's got to be at least 10. We have five major phone companies and we have five major PC (manufacturers)," he stated.
Update 9/21 8:59 AM: Nvidia responded to our query about Kal-El manufacturing with the following statement:
Project Kal-El is a single chip. We have two processes (manufacturing techniques) for making chips: one optimized for power, one for performance. Kal-El uses a mixture of both. The 5th companion core uses a process that is low leakage at low frequencies. The quad cores use a process that exhibits very low dynamic power at high frequencies.In other words, the companion core resides on the same piece of silicon as the four regular cores, but it's fabbed using a different process. Interesting.
|1. Ryszard - $503||2. punkUser - $502||3. the - $306|
|4. SomeOtherGeek - $300||5. Ryu Connor - $250||6. doubtful500 - $200|
|7. Anonymous Gerbil - $150||8. danny e. - $125||9. SecretSquirrel - $125|
|10. Techonomics - $100|
|The TR Podcast is live, so come ask us stuff!||0|
|AMD shows off DirectX 12 performance with new 3DMark benchmark||30|
|Intel and Micron sampling 3D NAND based on floating gates||16|
|Report: Microsoft to build an Intel-powered, non-Pro Surface tablet||47|
|Toshiba's 3D flash spreads 16GB over 48 layers||6|
|Cougar's 300M gaming mouse looks awfully familiar||14|
|Wednesday Night Shortbread||69|
|EVGA Hybrid liquid-cools the GeForce GTX 980||20|
|Rowhammer attack exploits shrinking process size in DRAM||40|