Google, Apple, or a smart startup could disrupt Intel, which increasingly looks awkwardly poised as the world's largest maker of microprocessors.
That's the view of Dave Ditzel, a veteran microprocessor designer. I talked to him about the microprocessor landscape after an interview for his upcoming paper at Hot Chips.
Ditzel led Sparc designs at the former Sun Microsystems before founding his own startup, Transmeta, that designed an x86-compatible chip. Most recently, he spent a little time at Intel on a microprocessor design that apparently got the axe from Brian Krzanich, Intel's new CEO. So he's been around the block and has something of an underdog's perspective.
Google may undermine Intel's x86 in servers with its work with IBM on the OpenPower Consortium, Ditzel said, and he makes a good case. The search company could probably save a lot of money and maybe even gain some performance/watt advantages if it could come up with a custom Power design for its data centers.
We know from seeing job reqs that Google has been hiring circuit designers and other kinds of chip engineers. Google's head of data center server technology leads the OpenPower group and has shown custom Power board designs.
I'm skeptical because I know Google tries everything and has the cash to do it. Nevertheless, Ditzel helped me see it also has the motive to make something real here that could deprive Intel of many hundreds of thousands of Xeon sales a year.
If Google gets something working, it might motivate Amazon, Facebook, and other big data center companies to follow. These folks represent 20% of the server business -- the hungry 20%.
Apple could put one of its next-generation A-series SoCs in a MacBook Air in the not-too-distant future, Ditzel said. Once it got the SoC up and running on its full Mac OS, it could easily spread use of the chip to other Mac notebooks and eventually desktops.
That's not a game changer for Intel, but it's another loss of x86 sockets and part of a scenario of death by a thousand cuts.
Finally, Ditzel said there's ample opportunity for a startup to do something really kick butt in microprocessors these days.
"Most of mainstream microprocessor design has slowed to a crawl. It's not happening at big companies, because they have lost the ability to design new architectures," he said, perhaps unconsciously thinking about his last project.
Multicore designs are running out of gas, given the lack of parallelism in most software. Nevertheless, "there are several really interesting opportunities for new microprocessors."
One path is in designing the first SoC from the ground up for 3D stacks, he said. No surprise there, given Ditzel is now working withThruChip Communications, a startup with an inductive coupling technique for chip-to-chip communications.
When asked about what will be hot at Hot Chips -- beyond his paper -- Ditzel agreed with analyst Kevin Krewell, who said in a Hot Chips preview that a paper about Nvidia's long-secret Project Denver will be hot.
"No one knows the details yet, but it looks like this may be first big disclosure," Ditzel said. "It's been a long time coming."
The project, believed to be Nvidia's first general-purpose CPU, was expected to be x86-based but now is reportedly a 64-bit ARM SoC. In this era when so many companies, including Intel's old rival AMD, are jumping on the ARM bandwagon for everything from mobile systems to servers, Denver may not be that disruptive in the long run.
There's no doubt Intel is on shaky ground, given its core PC market has slowed to a crawl, and the company still has not made significant progress in hot mobile markets where Qualcomm is king. For now, I'm keeping my eyes on Google, some still stealthy startup, or maybe those RISC-V folks at Berkeley for the next big disruption in microprocessors.
That's the view of Dave Ditzel, a veteran microprocessor designer. I talked to him about the microprocessor landscape after an interview for his upcoming paper at Hot Chips.
Ditzel led Sparc designs at the former Sun Microsystems before founding his own startup, Transmeta, that designed an x86-compatible chip. Most recently, he spent a little time at Intel on a microprocessor design that apparently got the axe from Brian Krzanich, Intel's new CEO. So he's been around the block and has something of an underdog's perspective.
Google may undermine Intel's x86 in servers with its work with IBM on the OpenPower Consortium, Ditzel said, and he makes a good case. The search company could probably save a lot of money and maybe even gain some performance/watt advantages if it could come up with a custom Power design for its data centers.
We know from seeing job reqs that Google has been hiring circuit designers and other kinds of chip engineers. Google's head of data center server technology leads the OpenPower group and has shown custom Power board designs.
I'm skeptical because I know Google tries everything and has the cash to do it. Nevertheless, Ditzel helped me see it also has the motive to make something real here that could deprive Intel of many hundreds of thousands of Xeon sales a year.
If Google gets something working, it might motivate Amazon, Facebook, and other big data center companies to follow. These folks represent 20% of the server business -- the hungry 20%.
Apple could put one of its next-generation A-series SoCs in a MacBook Air in the not-too-distant future, Ditzel said. Once it got the SoC up and running on its full Mac OS, it could easily spread use of the chip to other Mac notebooks and eventually desktops.
That's not a game changer for Intel, but it's another loss of x86 sockets and part of a scenario of death by a thousand cuts.
Finally, Ditzel said there's ample opportunity for a startup to do something really kick butt in microprocessors these days.
"Most of mainstream microprocessor design has slowed to a crawl. It's not happening at big companies, because they have lost the ability to design new architectures," he said, perhaps unconsciously thinking about his last project.Multicore designs are running out of gas, given the lack of parallelism in most software. Nevertheless, "there are several really interesting opportunities for new microprocessors."
One path is in designing the first SoC from the ground up for 3D stacks, he said. No surprise there, given Ditzel is now working withThruChip Communications, a startup with an inductive coupling technique for chip-to-chip communications.
When asked about what will be hot at Hot Chips -- beyond his paper -- Ditzel agreed with analyst Kevin Krewell, who said in a Hot Chips preview that a paper about Nvidia's long-secret Project Denver will be hot.
"No one knows the details yet, but it looks like this may be first big disclosure," Ditzel said. "It's been a long time coming."
The project, believed to be Nvidia's first general-purpose CPU, was expected to be x86-based but now is reportedly a 64-bit ARM SoC. In this era when so many companies, including Intel's old rival AMD, are jumping on the ARM bandwagon for everything from mobile systems to servers, Denver may not be that disruptive in the long run.
There's no doubt Intel is on shaky ground, given its core PC market has slowed to a crawl, and the company still has not made significant progress in hot mobile markets where Qualcomm is king. For now, I'm keeping my eyes on Google, some still stealthy startup, or maybe those RISC-V folks at Berkeley for the next big disruption in microprocessors.
0 comments:
Post a Comment