NVIDIA Enters the Mobile CPU Market

Vigile writes "NVIDIA just announced the new Tegra line, a complete system architecture on one chip. Built around a licensed x86 ARM 11 CPU, this tiny chip (smaller than a US dime) includes a processor, memory controller, southbridge, and 3D and video processors. The SoC design is meant to give iPhone-type devices a more impressive visual experiences while maintaining idle power consumption under 100 mW. While not a direct competitor to Intel's Atom or VIA's Nano processors, the NVIDIA Tegra will no doubt push the envelope in handhelds and cement NVIDIA's place in the world of computing going forward."

Not x86

[zsazsa (141679)]

The article summary is wrong or has a typo or something. This is not on some weird hybrid x86/ARM platform; it's just ARM.

Re:

[by Anonymous Coward]

Typo? Ha, fat chance! When was the last time a /. summary had a typo in it?

Okay, sometimes they get it right by accident.

[Futurepower(R) (558542)]

When was the last time a Slashdot summary didn't have something wrong with it?

The title of the referenced article is "Its not X86, but who cares?" There's no X86 in an ARM processor. It's a licensed design.

It was too much for the Slashdot editor to read even the first 3 words of the summary?

To much work to read the title of the article?

[Futurepower(R) (558542)]

"... even the first 3 words of the summary?" should be ... even the first 3 words of the article?

Re:Not x86

[OrangeTide (124937)]

Screw that, I will only buy an ARM if it's an x86 ARM. I need to run Vista, a computer is not a computer without Vista!!

Re:Not x86

[jd (1658)]

If it was an x86, it would be mostly 'armless.

Re:

[lysse (516445)]

It's just a flesh wound!

Cat got your tongue? (something important seems to

[DrunkenTerror (561616)]

Ars had a good article about it [arstechnica.com]

Stupid lameness filter.

Not x86

[larry bagina (561269)]

oh yeah, kdawson. Go figure. From the article title: "Its not X86, but who cares?"

Dupe

[Hektor_Troy (262592)]

We had the exact same thing yesterday

http://hardware.slashdot.org/article.pl?sid=08/06/02/1441214 [slashdot.org]

Re:Dupe

[mrbluze (1034940)]

We had the exact same thing yesterday

And you'll be getting it again tomorrow if you don't eat yer veggies!

Re:Dupe

[Dahamma (304068)]

Not quite the exact same thing... this one is LESS accurate ;)

WinCE...

[fuzzyfuzzyfungus (1223518)]

I am really hoping that they plan on expanding the lineup of supported operating systems. Obviously Windows is never going to happen, this being an ARM core; but Linux would be nice. In particular, Linux is pretty much the only way that this chip will have any shot of crawling out of the smartphone/PMP ghetto and making its way into general purpose small computers(I'd love to see a laptop built around something of this sort).

Windows CE just isn't a very pleasant OS period, and its flaws really start to show once you get outside of the smartphone market, where at least it has some experience, or the thin-client market, where abject suck doesn't matter too much. Whether or not you like Linux or Windows better, you'd be hard pressed to argue that Windows CE is better for anything resembling a real computer. Unfortunately, given that this is Nvidia, and the chip looks tuned to "support premium content" I'm not going to be holding my breath. It's a pity, really. This setup looks rather cooler than Atom, and capable of some really fun stuff, but I'm not sure how good the odds are of it ever making its way into a mininotebook or small desktop form factor.

Re:

[dreamchaser (49529)]

It doesn't even look like it's designed for mini notebooks or tiny desktops. This is aimed at the smartphone/handheld market.

So what's the difference??

[EmbeddedJanitor (597831)]

We have a multi-hundred MHz ARM in a cell phone/PDA these days, with a display controller and a bunch of peripherals built in to the same chip. That's enough to do an EEEPC - perhaps using q bit more RAM than is standard in a phone.

If you were really keen you could stuff a few extra 1x2inch ARM cards in the box and have a Beowulf cluster in a sub-laptop box.

The sub-notebook is nothing new. I have an old Psion7 (http://newth.net/psion7/index.html) that must be 6 or seven years old now. It was a bit slow, but only had a 100MHz StrongARM CPU. A re-jig with a modern 600MHz+ ARM would fly!

Like on the Handheld PC

[tepples (727027)]

Obviously Windows is never going to happen, this being an ARM core [...] you'd be hard pressed to argue that Windows CE is better for anything resembling a real computer.

Some versions of Windows Mobile were designed to run on a form factor called Handheld PC [wikipedia.org], which was in essence a Pocket PC with a bigger screen and a keyboard. Can you demonstrate a problem with building a subnotebook around such a platform, especially now that the XO and Eee have stimulated the low-end subnotebook market?

(I'd love to see a laptop built around something of this sort)

This northern summer, the Pandora PDA [openpandora.org] comes out. If you like handheld computers, you might want to check it out.

Re:Like on the Handheld PC

[fuzzyfuzzyfungus (1223518)]

I've been looking forward to the Pandora. I'm not much of a gamer anymore; but it looks to be shaping up into a very capable little box. I'm quite tempted to pick one up when the time comes.

As for WinCE, I know that it can be made to act something like a real desktop/notebook OS, I've used and deployed loads of HP "semithin" clients(mostly citrix or RDP; but limited builds of IE, WMP, etc are available locally) running WinCE of various versions, and I've used a few PDAs running it. It just doesn't measure up as a desktop OS. It looks just enough like Windows to screw with your expectations; but doesn't run Windows programs, use Windows drivers, or even behave all that much like Windows in terms of shell look and feel. Given their sterling performance with "Vista capable/ready" I'm sure that MS marketing would have no trouble explaining to Cletus and Maybell User why their bargain bin software won't run on that Windows...

Essentially, WinCE suffers from the majority of the shortcomings commonly ascribed to running a non-Windows OS on the desktop, without possessing any significant upsides.

Re:

[bhtooefr (649901)]

There was a version of the Psion NetBook which came with WinMob... (and it's about the size of an Eee, I'd say.)

Please not WinCE

[EmbeddedJanitor (597831)]

Linux and ARM are great partners. There are probably more ARM Linux systems than x86 Linux systems (if you count all those ARM Linux cell phones as systems).

The only valid reason to design in x86 these days is to run Windows. ARM is lower wattage and cheaper. Once you look at whole systems costs (battery etc) ARM comes out streets ahead. Most OSS can be readily redeployed on ARM. There is even an ARM Ubuntu.

WinCE is a very limited architecture and has no support for SMP etc. It is basically a toy version of Windows.

I don't think so.

[EmbeddedJanitor (597831)]

I've been a WinCE OS-level developer for quite a few years (since 1.0) and currently have the low level dev tools for WinCE6 on a PC.

No SMP in sight, not even in the emulator. If you know differently, I'd like to know.

Re:

[pimpimpim (811140)]

Seriously, can someone who is in this business explain me why you would need SMP for a mobile device? If you are going to deploy small CPUs in parallel, is the performance loss in communication not so huge that it just makes more sense to switch to a more powerful CPU instead?

Re:I don't think so.

[jrumney (197329)]

Most mobile devices run several ARM cores. Usually they are arranged as peripherals to the main CPU rather than working together as an SMP system. Typically the baseband (the bit that handles commincation with the cellular network) will have its own CPU in all but the lowest end devices, and increasingly graphics/video/audio processing is offloaded to a specialised chip. Nvidia are pulling this all into a single package, just as Marvell have with the PXA320, and TI have with several of their OMAP processors. In a typical configuration, it will appear as a single core to WinCE, so SMP support will not be needed, as the other cores are being used for specialised processing rather than as general CPUs. The PSP does something similar with its dual core MIPS processor.

Re:WinCE...

[rbanffy (584143)]

With its power envelope and die size, they could put 8-way multi-processing and some vectorized FPUs before they hit either Atom's die size or power consumption ;-)

Seriously, with Via making a complete reference design freely available, what could stop someone from making a compatible motherboard that could fit in the place intended for a VIA-based motherboard?

The Eee shows that any processor that can drive a web browser and an e-mail program can be the core of a successful sub-notebook. In NVidia's shoes, I would invest a decent amount of money to make sure Gnash runs fine on it and is as compatible with Flash as compatibility can be.

I would love to see a small sub-notebook with a fast multi-core, multi-threaded, ARM-based CPU, an XO-derived widescreen display, an iPod-class HDD and a battery that could power it all for 24 hours of continued abuse.

And I bet it could be done for peanuts, which means a huge profit margin beyond the wildest dreams of the commodity PC-compatible market.

Of course, I assume someone from Microsoft mentioned casually to someone at NVidia, perhaps during golf, that they could consider dropping some NVidia support on an upcoming Vista service pack if those Linux-running small CPUs start making inroads in the undead XP camp.

Re:

[Antique Geekmeister (740220)]

> And I bet it could be done for peanuts, which means a huge profit margin beyond the wildest dreams of the commodity PC-compatible market.

Not the huge profit you might think, because competitors would also be making it, and selling it, for peanuts. Unless you were able to lock it up with patents, which involves its own costs in getting them passed and in keeping the lawyers available for the inevitable violations of such an attractive market.

Vista and NVidia makes me shudder anyway: the high-end graphic

Re:

[QuantumRiff (120817)]

Obviously Windows is never going to happen

Why not? Windows currently runs on X86 and Itanium architectures. NT4 also ran on MIPS, and Alpha, if I remember correctly. If a processor gets popular enough, you better believe that MS will support it to generate sales. (BTW, Windows Mobile runs on ARM cores.)

Re:WinCE...

[fuzzyfuzzyfungus (1223518)]

The main problem, even if MS were willing to put the resources into a port, is that Window's value is severely reduced without its massive library of third party stuff. Since the overwhelming majority of the Windows ecosystem is proprietary in nature, and much of it made by outfits other than MS, it would be an absolute miracle if MS were able to get more than a tiny bit of it ported over.

As for the various NT ports, they aren't an encouraging picture. NT MIPS and Alpha are dead, XP for Itanium is dead, and there is no Vista for Itanium. Only Windows Server and a few of its variants are still alive. And look at the slow pace of the x86/x86_64 transition.

I suspect that limited platform support is a decent business decision for MS, they aren't stupid, and they probably know better than we do how much fun it would be to attempt to bludgeon every last software vendor for Windows into shipping multiarchitecture support; but I seriously doubt that anything short of the probable extinction of the x86 architecture would motiveate a wholesale move on MS' part.(Maybe a stopgap of some sort, x86 virtualization built into the kernel or similar, or a move of existing MS tech to a new environment .NET on WinCE or Singularity or something; but classic NT and friends not so much).

Re:

[Bill_the_Engineer (772575)]

Obviously Windows is never going to happen, this being an ARM core; but Linux would be nice.

Linux runs on the ARM. I understand that even Debian has an ARM port...

Re:

[fuzzyfuzzyfungus (1223518)]

Linux supports ARM without trouble; but NVIDIA has, thus far, not mentioned supporting Linux. I suspect that a roomful of decent hackers will take a case of jolt cola and ten minutes to boot a Linux kernel on the first device to hit the market. Actually supporting the device without NVIDIA's cooperation could be a good deal trickier. The GPU, integrated peripherals, and video decode engines would probably need some serious reverse engineering. And odds are that not too many OEMs would be willing to venture

Re:WinCE...

[GreyWolf3000 (468618)]

To support a CPU is not enough. ARM cpu cores will typically connect to an AMBA bus (like hypertransport), but these SoCs will usually have the entire bus internal. You'll have a whole set of peripherals which need to be programmed all over again with little or no code reuse from existing projects. You need to understand how these peripherals interact with the boot rom and CPU in order just to load a bootloader onto it. If you have perfect documentation, you'll still probably need at least a decent oscilloscope or logic analyzer to get a heartbeat out of it. Talk to any firmware or digital design engineer about 'board bring up' on an unproven cpu platform, and you'll likely hear quite a few nasty anecdotes.

All this can be done, but it would save everyone a lot of time if nvidia supports Linux with a real board support package.

Re:

[Antique Geekmeister (740220)]

You mean like this? http://www.arm.linux.org.uk/ [linux.org.uk]

Don't underestimate ARM.

[serviscope_minor (664417)]

Modern ARMs are very good. Much, much better than the XScales. They can come with floating point units which are many times faster than software emulation and do not drain much power. Even under load they draw much less than the atom. Also, it looks like the modern cores come with OoO, something which is missing from the atom.

Secondly, the ASUS EEE has shown that it may not br necessary to be x86 compatible any more. It is compatible, but it has sold many copies running Linux. That means that x86 compatibility is not required, since Linux runs well on many processor types.

OoO ??

[by Anonymous Coward]

"Modern cores come with OoO"... is that some weird new smiley or have they etched a copy of openoffice into the microcode prom?

Re:OoO ??

[klingens (147173)]

Out of Order execution

Re:

[aproposofwhat (1019098)]

screw that - mine just page faulted :-(

Re:

[serviscope_minor (664417)]

or have they etched a copy of openoffice into the microcode prom?

I doubt it. After all the die is less than one square meter.

Re:

[FrankDrebin (238464)]

Umm, last I looked XScale was an ARM (core).

Re:

[ady1 (873490)]

>>Modern ARMs are very good. Much, much better than the XScales

Xscale IS ARM. Look it up.

Re:

[RiotingPacifist (1228016)]

Given that I spent my day battling a Linux x86_64 I doubt that non-x86 architectures are widely used by end users. Just because Linux runs on non x86 in no way means people are willing to support/use/accept it on anything else.

Being open source does mean that the code can be ported, doesn't guarantee it will compile & be stable on all architectures. Even in the kernel non-x86 is generally 1-2 kernel releases behind before new features are properly supported and there's a lot more to an OS than the kerne

Re:Don't underestimate ARM.

[serviscope_minor (664417)]

The point is that the EEE has sold many, many units running Linux. This demonstrates that people are prepared to accept Linux and therefore x86 is unnecessary since Linux does not require it.

Linux being OK implies that x86 is unnecessary.

Performance

[by Anonymous Coward]

What is the performance like compared to Blue Steel and Magnum?

The mother of all UARTS

[serviscope_minor (664417)]

From TFA, on the second page, the UART is the same size as the NAND flash! for it to be that big, they must be fabbing a full-sized 8250 on its original process.

Re:

[by Anonymous Coward]

In order to drive a serial cable directly, a UART may need to withstand +/-25V and deliver .5A or more. It makes no sense to get that out of an 11-layer process.

Re:

[CuriHP (741480)]

I don't think that's the case here. It looks like those labels are really just a rough block diagram superimposed on a photo of the die. For example, most of what is below the label "Image Processor" looks like it is probably the cache for the ARM11. (Hard to be 100% sure since I mostly do chips with ARM7s and ARM9s.)

Re:

[iabervon (1971)]

In order to make the chips easier to design, they decided to standardize the size and shape of their functional components to one of two squares. Sure, the UART wastes some space, but it avoids the need to come up with a funny-shaped peripheral to go next to it. It's like why the AT91SAM7X has the same size MMIO space for the ethernet and for the reset controller.

Actually, the real reason is that, if the UART were smaller, it wouldn't cover up enough of the secret design in the next layer down, which you ca

Re:

[CuriHP (741480)]

While this may be done for large CPUs from Intel, AMD, etc., it's not standard practice for these kind of smaller SOCs. It's not worth the effort to lay out small peripherals like UARTs as separate blocks and tile them together. The CAD tools that are available can do it automatically and yield a much faster design when they're allowed to smudge the boundaries of blocks when it makes sense to do so.

I can tell you for certain that AT91SAM7X are not done with the method you describe.

Idle power

[Kanasta (70274)]

"impressive visual experiences while maintaining idle power consumption under 100 mW"

Qualcomm's 7200 already maintains idle power consumption under about 50mW on HTC devices even with screen drawing power. Of course, the claimed "impressive visual experiences" on this chip can't be verified since Qualcomm/HTC neglected to provide any drivers and their chips run slower than 4yr old hardware. Turns out to be a licensing issue, not a technical issue.

So depending how nVIDIA prices their chip, we may once agai

Re:

[DragonTHC (208439)]

highly doubtful.

these devices will carry a "Powered by nVidia" sticker and I seriously doubt that nVidia would allow manufacturers to use these chips and disable the features.

We know that nVidia prefers a unified driver structure. I doubt they'd make special drivers for certain devices.

Death by interactive TV

[heroine (1220)]

Hopefully NVidia won't sink in this mine field. Just look what happened to ATI & Intel's ventures into interactive TV. Broadcom is the immobile giant in interactive TV. U better know what you're doing if you're going to go up against The Broad. Now let's port OCAP to Tega or Tegra, or whatever it is.

Awesome!

[crhylove (205956)]

Now can I have one in my phone, one in a EEE type notebook, one in a regular full size notebook, one in a desktop, one in my car, and have them all communicate via wifi N and turn into a cluster if they are close by?

Thanks Nvidia!

Re:

[by Anonymous Coward]

Yeah, "corpie" isn't obnoxious at all.