By Herman Lai, posted Oct 22, 2011 at 1:42 PM, 934 views,

Nvidia’s CEO Thinks GPU is More Powerful Than CPU

Nvidia’s dynamic leader Jen-Hsun Huang joined Walt Mossberg onstage at tech event AsiaD to talk about graphic chips and indicated GPU can do more than what we think, which makes GPU sounds like it has some mystical potentials. Rather than that, when he initially started his graphic company, he considered not to go toe-to-toe with Intel’s integrating chipsets. And this is how he had the spirit to make better GPU.

Nvidia’s CEO Jen-Hsun Huang — Interview with Walt Mossberg (Click here for video highlight)

In terms of the GPU, its Tegra 2 was the world’s first dual-core graphics processor that installed in many devices. Next big thing that the company will come up with is the first quad-core processor — Tegra 3, which is going to be named as Kal-El. Guess what? Kal-El is a name of a superman, and the entire products’ series will be based on comics like “Wayne” for Batman. At the moment, one of the most noticeable device that made use of GPU is what Lytro does with their camera, and we are waiting to see more surprise that those super-GPU bring to our life.

Toward the end of the conservation, Huang gave an advice to Microsoft — Go for tablets with Window-on-ARMs first, emphasis “these are not PCs”. Does that imply Microsoft did not treat their tablets like tablets? Moreover, he sggested it would be killer if Office can run on Window-on-ARMs tablets.

Walt: Who are your competitiors?

Huang: Qualcomm is a direct competitor. We started out as a graphics company, and evolved from serving gamers, designers, etc.

And we tried to grow initially by growing into the basic PC market by integrating chipsets. That didn’t work out for me, because “going toe-to-toe with Intel is not a good idea.” So now we’re trying increase the usefulness of GPUs in different products.

Walt: Now the GPU for many PCs is more powerful and complicated than CPU, right?

Huang: Yep. And they can do lots of stuff now: Computer graphics, image processing. Stuff like what Lytro is doing with their camera.

And the second part of the strategy was backing away from the chipset business, and investing in mobile computing. Back to competitors: Qualcomm is direct. Apple is an indirect competitor.

We partner with device makers, so in a lot of ways I’ve adopted their competition.

Walt: So what’s new?

Huang: When we first got into mobile, I started talking to the cellphone companies and telling them mobile devices would become a computer. And the iPhone proved that and woke people up. So our perspective is that these devices would be “rich computers — quite powerful computers” and that graphics would play an important role.

So we built the world’s first dual-core graphics processor, the Tegra 2, and that’s what we’re shipping today. And now we’re working on our first quad-core processor, Tegra 3, which is really going to be called Kal-El. Huang notes that Kal-El is Superman’s name, and that the rest of the product lineup is also based on comics: “Wayne” for Batman, etc.

Anyway: Getting performance is the easy part. The challenging part is delivering that performance with low power, and better and better efficiency. The low-power core with Kal-El is 20x “less hungry” for energy.

Second invention: Use the image-processing technique to look at the color content of an image. That allows us to save a ton of power that way.

It allows us to sort of “cheat” on colors and use similar colors, with different lighting — the human eye can’t tell.

Walt: Aren’t your competitors doing quad-cores as well, and trying to reduce power, too?

Huang: Sure. “Our challenge is to create something surprising every generation … we’ll have to come up with something magical or something surprising every generation.”

Walt: Is it a disadvantage for Apple to design its own chips?

Huang: Apple’s Bob Mansfield is really good at chip design. Also, we sell them chips for their PCs.

Anyway, there are a lot of good reasons to build your own chips — they can control the vertical stack, and they might have good insights on bringing stuff to market, and not sharing that with third-party competitors is a fine idea. But R&D costs will go up — I’ve already spent $2 billion on Tegra. So it may become more difficult for companies to build for lots of different segments.

“There might come a time when it becomes important to buy some or build some. We’ll see.”

Q&A

Q: You talked about building processors for phones or tablets. Do you think tablet-specific processors are really coming?

A: There’s not enough volume with 60 million tablets to build tablet-specific chips. But you can imagine building a processor where some versions are perfect for tablets, and other versions are perfect for high-end smartphones. That’s how you can make the economics work.

Source: AllThingsD