Thursday, December 17, 2009

Nvidia Tegra 2 for Dual Core Netbooks/Smarbooks in 2010

UPDATE: in the original post I wrote that the Nvidia Tegra 1 uses an ARM Cortex A8 core, this is wrong (thank you Brian!), an ARM11 core is used.
UPDATE2: Qualcomm Snapdragon uses the "Scorpion CPU", a Qualcomm derivate of the ARM Cortex A8. (thank you alxlr8!)

While Nvidia Ion is "only" providing graphics acceleration to netbooks and nettops, the new Nvidia Tegra 2 is supposed to be the first Nvidia SoC (System on a Chip) implementing both the core CPU and GPU functionality for netbooks/smartbooks.
According to Xbitlabs, Nvidia is going to make a major announcement during next CES (Jan 2010): the first netbooks/smartbooks with Nvidia Tegra 2 are going to be launched already in 2010.
Very few details are available: it will be a dual core processor with 40nm technology.
As you may remember the Nvidia Tegra 1 (used in Microsoft Zune HD player) implements a single core ARM Cortex A8 ARM11 processor. Now the easiest step for Nvidia would be to just add a second ARM11 core to the SoC, on the other hand the ARM11 is a rather "old" processor, especially compared to the more efficient Cortex A8 and A9.

The new ARM Cortex processors are used in other ARM SoC targeting netbooks/smartbooks, for example the Qualcom Snapdragon (actually a Qualcomm derivate called Scorpion CPU).
For this reason I guess that Tegra 2 will also use the dual core Cortex A9 MPcore, the max CPU clock should be at most 2GHz. Of course this is a maximum clock, the clock used in Tegra 2 the can be much lower.
2GHz seems to be the maximum clock possible at 40nm for these two processors: in recent press announcements Synopsis presented a 2GHz ARM Cortex A8 core and ARM a 2GHz Cortex A9 core, both using a 40nm geometry.

Of course Nvidia Tegra 2 based netbooks/smartbooks can only run Linux variant: a distribution or at most Android/Chrome OS

8 comments:

Brian said...

Tegra 1 is not Cortex-A8, it is ARM11.

djsm said...

Thank you Brian!
Yes I definetly made a mistake: the Nvidia Tegra 1 uses an ARM11 core, not an ARM Cortex A8

alxlr8 said...

Hi.

Just a quick comment. There is no MPCore variant of the Cortex-A8. Cortex-A8 is a single core processor. The only other MPCore product is indeed the Cortex-A9 (aside from the older ARM11 MPCore).

alxlr8 said...

Another quick comment (sorry).

Qualcomm's Snapdragon chipset doesn't use any ARM Cortex products. Instead, it uses Qualcomm's Scorpion CPU, which is of their own design under their ARM Architecture License.

djsm said...

@alxlr8

thank you for the very useful comments!

About MPCore: Thank you, I didn't know that ARM didn't develop a Cortex A8 MPCore.
To be precise I'm pretty sure an MPCore version exists also for the Cortex A5, beside ARM11 and Cortex A9.

About Snapdragon: You're again right. Technically the Scorption CPU isn't a Cortex A8. On the other hand Qualcomm got an ARMv7 architecture licence and I read somewhere that they took much from the Cortex A8 design since when they got the license the Cortex A8 was the only ARMv7 available. Of course I may be wrong.

I'll post corrections as soon as I get a few minutes!

alxlr8 said...

@djsm@

That's OK. It is sometimes difficult to find reliable information about ARM products in the "blogosphere", since ARM has always been relatively unknown.

As you say, though, Cortex-A5 is indeed scalable from 1 to 4 CPUs, and is an "MPCore" product.

Scorpion does indeed have ARMv7 architectural features that are also found on Cortex-A8, but they are not taken directly from Cortex-A8. In obeying the same architecture, they will have similar capabilities (although not all parts of the architecture are compulsory). Qualcomm have designed everything inside Scorpion from scratch.

djsm said...

@alxlr8

Thank you for clarifying!

Now you make me curious: "Qualcomm have designed everything inside Scorpion from scratch"
I read this information in a few blogs and I ask myself what's really meant.
I couldn't imagine that Qualcomm didn't reuse any logic module, on the other hand it looks like you got insider information that I don't have ;-)

I can't understand why to reinvent the wheel? AFAIK with an architecture license you get the soft macro and you can modify whatever you like in order to achieve better performance, lower power consuption or just to add new features.

alxlr8 said...

There is no special information here - its all on Qualcomm's web site, and ARMs. An architecture license enables a company to design their own micro-architecture (a soft macro, or a custom hardened implementation etc.), and maintain compatibility with ARMs architecture. [Note: If all micro-architectures obey a given architecture, they are all compatible.]

There is no logic re-used by Qualcomm from any ARM macrocells. The soft macros that ARM designs are licensed to companies that don't want to (or can't afford to) design their own processor. ARM designs it once, and sells it many times.

Qualcomm have a big team of designers, and obviously have their own reasons for designing their own processor (such as being able to own their whole IP chain). It's a very nice processor. Their reasons for "re-inventing the wheel", as it were, might be to target higher frequencies than an off-the-shelf ARM core, or to implement higher performance memory/data processing than the more generic ARM designs.