NVIDIA's Project Kal-El

[Mopetar]

Yeah, I think we've come to a point where improvements in web browser code are providing bigger advances than more powerful hardware. It seems like every time a new version of some browser comes out it's claiming to be even faster than the competitors that just beat it a few months back.

GPU acceleration is probably the next best way to improve performance for systems that don't already have it, but more CPU power beyond a certain point isn't going to do anything expect for the most complex and large websites.

 

[Ajay]

Most reviews of new dual core phones have said that there isn't a whole lot of use for the extra core right now. The most useful thing they've found you can do is get better performance while running several things in the background, which most phone users don't like to do because it erodes battery life.

It's unlikely that developers are building apps to take advantage of parallel code. I haven't taken a look at any of the various SDKs, so I don't know what kind of support even exists for devs. One would think that basic multi-threading is available (to separate out the UI, IO, etc.). I really should have a look at the Android SDK.

 

[Mopetar]

It's unlikely that developers are building apps to take advantage of parallel code. I haven't taken a look at any of the various SDKs, so I don't know what kind of support even exists for devs. One would think that basic multi-threading is available (to separate out the UI, IO, etc.). I really should have a look at the Android SDK.

One of the problems is that if you write code to use a specific number of threads it can have adverse effects on performance based on the actual hardware. If you write code to use four threads, but have one CPU core, it may not be as efficient if the code were written to use fewer threads. Alternatively, if you write to use only one thread, you're wasting any extra cores that may exist.

Does Android (or iOS for that matter) implement anything similar to GCD in OS X where developers can designate blocks of code or entire functions to be performed in parallel? I remember hearing about it some time ago as one of the big new features in their OS release, but haven't heard a lot about it since them.

Edit: Wikipedia says that GCD is included in iOS 4. However, considering now iOS devices have multiple cores yet, I don't know how much developers have been using it.

 

[IntelUser2000]

One of the problems is that if you write code to use a specific number of threads it can have adverse effects on performance based on the actual hardware. If you write code to use four threads, but have one CPU core, it may not be as efficient if the code were written to use fewer threads. Alternatively, if you write to use only one thread, you're wasting any extra cores that may exist.

You can program for both. Base it on CPUID or something.

 

[P4man]

If im not mistaken, Tegra 2 Dual core ARM A9 CPU is at 150 to 200mW at full load so i will guess that a quad core will be double at 300-400mW.

No way. Those arent full load numbers, those are roughly the numbers while playing HD video, but thats done on a dedicated h264 decode unit, and 150mW would be roughly around the power consumption while browsing the web. But they are not TDP or "full load" figures.

ARM states 0.5W for the power optimised cortex A9, but note that is just the cpu cores, it likely doesnt even include caches, much less IMC, GPU, IO, video decode units and what not. In reality I suspect a power virus could drive a Tegra 2 to 2W, with 1W being a more typical loaded power consumption, and a fairly normal number for mobile phone SoCs.

As for Tegra 3, unless nVidia pulled off a miracle, or included some thermal or power throtteling, it will have a considerable higher peak power, even though real world power might be similar or even lower while running typical tasks. I also wonder if it wont be too hot for a phone, and probably dual cores will be used there mostly, but for a tablet it should be perfectly okay. A TDP of 3W or more should be okay for a tablet.

 

[P4man]

It looks like Coremark does scale perfectly with multiple cores. (assuming each Tegra 3 cpu core is the same as a Tegra 2 cpu core)

Yes, it scores perfectly with cores and perfectly with frequency. Coremark runs completely in the L1 cache, it really only tests the core. It doesnt touch L2 cache, let alone memory or anything else. Its a useful benchmark to compare.. well, cores really, particularly for the embedded market, but dont put too much weight on it for end user application performance.

So according to the results (found in the video) each Cortex A9 cpu core gets roughly half the coremarks of a 2 Ghz Core2duo T7200 cpu core.

But if you look carefully, you will notice an outdated compiler was used for the C2D. Doing the same benchmark with the same compiler (GCC 4.4.x), the C2D outperforms the Tegra 3 by a significant margin. Even more so if you enable 64 bit. Benchmarketing at its best.

Still, even when using honest results, a tegra 3 (or any 1 GHz quad core A9) would outperform a ~1.4 GHz Core2Duo on that benchmark, which isnt too shabby, considering thats still the cpu powering a macbook air.

edit: someone else already linked those results.

 

[Dark Shroud]

According to this article from April 9, 2010 Web kit 2 will only use two cores.

Does anyone know if Google (or other companies) have plans to have browsers scale on cpu cores beyond this?

IE9 makes use of multi-core support. If you have the spare cores MS's script engine, Chakra, will designate one core for it's own use. While IE9 will use the remaining cores/threads as well as the GPU for the rest of the processing. IE9 also hardware accelerates everything in the webpage not just the Java Script.

MS just released the IE9 RC-1 for download, if you have Vista or Win7 download and give it a try. The performance is very impressive.

 

[Dark Shroud]

You guys seem to be thinking phones and tablets only.

ARM will be Windows-8 compatible and easily fit into a low-mid range solution for full fledged PCs. Nvidia has a boatload of chipset experience under their belts.

Tegra 4 on an Nforce style motherboard coming in at a lower total system price for a similar intel offering will be an automatic win.

Atom and Tegra will collide in notebooks. the problem here is Intel will always have the manufacturing advantage due to the smaller process.

If intel gets serious about competing here, they can enter the market within a year and dominate.

Nvidia has already said they're not making anymore chipsets for mobos. Intel has also quietly licensed AMD GPU tech when AMD & Intel updated their cross licensing agreements. As soon as Intel's GPU cores can run DX11 MS's APIs (DXVA, Direct2Draw) will be able to use those cores for hardware acceleration of software. Combine that with the power of AMD's APUs where the vector (GPU) cores are already able to run hardware acceleration with very low power use as well as work with AMD add-on cards with Crossfire. If AMD adopts eDRAM into the mix ARM SOCs will feel pain.

I just don't see ARM laptops/notebooks taking off. It won't matter that they should theoretically be able to match the same price of a Intel netbook. They won't be able to run the x86 software that you expect a laptop to run nor will they match the power of Intel & AMD systems that will only cost a bit more.

I think we can all agree that Atom will soon be getting a solid die shrink and probably a quad core option. I also think tablets are a "it's the new big thing" fad that will fade off into a decent niche market. MS wants them running full versions of Windows hence the ARM option. But once again those will be limited in the software they can run. Although putting one of those tablets into a charging dock and running a wireless keyboard & mouse will be a nice feature set.

I'm sorry but I just don't see ARM & specifically Nvidia taking the various markets by storm.

 

[Ajay]

This looks very promising (although higher clocks would be better ATM):

Perhaps the biggest change in Honeycomb is that the Dalvik VM environment is now multi-core aware. This means the OS can now leverage the processing power and potential power savings of multi-core SoCs to speed things up and offer enhanced battery life. In fact, even single threaded apps can see speed improvements in Honeycomb. While running in a dual-core environment, the OS automatically lets the bytecode run on one core and the Dalvik garbage collector run on the other. Apps specifically written to support symmetric multiprocessing will see even larger performance gains. With companies like ARM (Cortex A15) and Qualcomm (Snapdragon “Krait”) already announcing their next-gen dual/quad-core SoC architectures, it will be exciting to see what kind of mobile performance we have in store for the future!

 

[drizek]

Yeah, I think we've come to a point where improvements in web browser code are providing bigger advances than more powerful hardware. It seems like every time a new version of some browser comes out it's claiming to be even faster than the competitors that just beat it a few months back.

Part of it is due to the software taking advantage of faster hardware. I feel like Chrome uses more ram than any other browser, and I think it threads better than other browsers. Old computers with 256MB of ram or even 512MB might have a bit of trouble running it compared to firefox. I have my fair share of tabs open, but right now Chrome 10 is using 733MB of ram. I've actually had experiences where Chrome by itself would start swapping on my system with 4GB ram and nothing else running(granted, this was after running continuously for a week or so).

 

[Mopetar]

You can program for both. Base it on CPUID or something.

Something like that sounds as though it would take a non-trivial amount of development time. It also makes bug-fixes a pain if you have duplicated code that needs to be fixed.

If we're going to have a four-core Tegra 3 this fall, there will be one core tablets, two core tablets, and four core tablets, although the one core tablets are likely to be in the minority. That's three different hardware configurations to optimize for, not counting the possibilities of different GPU optimizations.

Unless developers want to take the time to tune performance by hand, you end up with lowest common denominator results. At least Google is getting parts of the OS to run on a separate core which should help, but they should really implement something like GCD to make it easier for the developers to get the best performance. Apple released the code for it under the Apache license so Google could take it and port for Android.

 

[drizek]

one core tablets = all iPads ever made (except for the iPad 2s in Steve's secret warehouse).

 

[Mopetar]

Part of it is due to the software taking advantage of faster hardware. I feel like Chrome uses more ram than any other browser, and I think it threads better than other browsers. Old computers with 256MB of ram or even 512MB might have a bit of trouble running it compared to firefox. I have my fair share of tabs open, but right now Chrome 10 is using 733MB of ram. I've actually had experiences where Chrome by itself would start swapping on my system with 4GB ram and nothing else running(granted, this was after running continuously for a week or so).

I'm guessing that's because Chrome uses separate threads for almost everything in addition to having its own built in version of the Flash player.

However, the software is definitely getting faster independent of the hardware it's being run on. However, as the software gets older, more and more features start to creep in that require faster hardware. Probably the best example of this is iTunes which is a bloated sea cow compared to its former self.

 

[Mopetar]

one core tablets = all iPads ever made (except for the iPad 2s in Steve's secret warehouse).

Was mostly referring to Android tablets. I'm assuming that the iPad has the ability to run GCD style functions and code blocks since it has iOS 4. Of course if developers don't take advantage of it it's not going to be very useful, but at least it's there.

It's also unlikely that we'll see four core iPads anytime soon (Unless the next version is going that route which seems unlikely.) as Apple seems to stick to a yearly release cycle for their iOS devices. New Android hardware comes out as fast as the manufacturers can get the newest bleeding edge parts so we'll probably see a few as soon as the Tegra 3 is ready to ship.

 

[drizek]

iPad 2 is coming out literally any minute now and it will almost certainly run tegra 2 or something similar.

 

[Mopetar]

iPad 2 is coming out literally any minute now and it will almost certainly run tegra 2 or something similar.

Probably some chip of Apple's own design that's similar to the Tegra 2. Guessing dual Cortex-A9 cores and at least a SGX540. However, I don't expect it to have four cores which is what I was speculating about.

Apple has fewer hardware profiles, which makes it easier from a developer perspective, but they have a fairly long product cycle so from a consumer perspective there're going to be four-core Android tablets long before an iPad with that many is available, especially if Nvidia has the Tegra 3 ready by the fall.

 

[drizek]

The iPad was the first 45nm/1GHz Cortex A8 on the market. Apple will hang out on the bleeding edge when it makes sense to do so.

I don't think quad core is worth it until it gets shrunk.

Who knows though, maybe the iPad 2 will surprise us all and have a quad core. There is no technical reason why they can't pull it off.

All I know is that my iPad 1 is going on craigslist ASAP.

 

[DeathReborn]

Nvidia has already said they're not making anymore chipsets for mobos. Intel has also quietly licensed AMD GPU tech when AMD & Intel updated their cross licensing agreements. As soon as Intel's GPU cores can run DX11 MS's APIs (DXVA, Direct2Draw) will be able to use those cores for hardware acceleration of software. Combine that with the power of AMD's APUs where the vector (GPU) cores are already able to run hardware acceleration with very low power use as well as work with AMD add-on cards with Crossfire. If AMD adopts eDRAM into the mix ARM SOCs will feel pain.

I just don't see ARM laptops/notebooks taking off. It won't matter that they should theoretically be able to match the same price of a Intel netbook. They won't be able to run the x86 software that you expect a laptop to run nor will they match the power of Intel & AMD systems that will only cost a bit more.

I think we can all agree that Atom will soon be getting a solid die shrink and probably a quad core option. I also think tablets are a "it's the new big thing" fad that will fade off into a decent niche market. MS wants them running full versions of Windows hence the ARM option. But once again those will be limited in the software they can run. Although putting one of those tablets into a charging dock and running a wireless keyboard & mouse will be a nice feature set.

I'm sorry but I just don't see ARM & specifically Nvidia taking the various markets by storm.

WhileI agre that Tegra won't be taking the PC/Notebook market by storm, it could take the Console market as it would enable quick & easy ports from consoles to phones/tablets & vice versa. x86 has been tried in consoles before and got replaced by PowerPC based chips, now with ARM becoming more prevalent in Gaming devices it could just replace powerPC in the next 2 generations of consoles.

 

[P4man]

I just don't see ARM laptops/notebooks taking off. It won't matter that they should theoretically be able to match the same price of a Intel netbook. They won't be able to run the x86 software that you expect a laptop to run nor will they match the power of Intel & AMD systems that will only cost a bit more.

You are assuming everyone wants to run x86 software. Im not convinced, especially for lowend notebooks that are used mainly for web/mail and the like. I suspect android and chrome os will not only become a viable alternative, but a more desirable alternative for many people who want something simple, thats fast, without having to be a part-time sysadmin to keep their machines running and virus free. I dont think its a coincidence Apple is doing so well in the notebook and tablet market; its not just marketing, both have a non windows OS that simply does the job and is easy to use.

If quite a lot of people are willing to pay a premium for that, imagine them getting it for a substantial discount, as those new OSs are free, require substantially less RAM, harddisk space, battery power and even CPU power. The combination of those could make ARM/Chrome/android notebooks substantially cheaper (and/or more profitable for OEMs) as well as more suitable for a non trivial portion of the market.

I think we can all agree that Atom will soon be getting a solid die shrink and probably a quad core option. I also think tablets are a "it's the new big thing" fad that will fade off into a decent niche market.

I somewhat agree. But I believe the reason for the current tablet hype isnt so much the formfactor; that isnt even new, its been around for over a decade. What appeals about it is battery life and ease of use; better OS's than windows for the job. Some day soon people will realize that an ARMdroid netbook will offer those same advantages without sacrificing a proper keyboard and convenient clamshell.

 

[Dribble]

Android has really been developed from the ground up to have a touch interface, it would probably be rubbish with a mouse and keyboard.

If you want ARM + netbook then you'll need windows 8, which I suspect will do the mixed mouse/keyboard/touch thing very well. It'll also provide the most important x86 apps (i.e. MS office). On the downside it's still bound to have all that history bloating and slowing it down so will run like a dog on any ARM cpu.

That said I don't see the netbook surviving outside of a niche market - it's tasks mostly better served by a tablet (lightweight browsing, watching stuff, flash games, etc) or a proper laptop (content creation needing cpu grunt and a big screen).

 

[P4man]

Android has really been developed from the ground up to have a touch interface, it would probably be rubbish with a mouse and keyboard.

thats only the UI; not exactly the hardest thing to change. Besides; I mentioned Chrome OS for a reason, as it will be Google's OS for anything with keyboard and mouse. Initially at least as ChromeOS and android are going to merge:
http://www.silicon.com/technology/s...-converge-says-googles-eric-schmidt-39746988/

BTW; keyboard isnt an issue anyway. Surely nothing could be easier than disabling the onscreen keyboard. Touchscreen vs mouse is perhaps a bit more work, but hardly a huge problem. Lastly, who said you cant have a touchscreen on a netbook:
http://www.engadget.com/2011/02/17/nec-busts-out-an-android-netbook-the-lifetouch-note/

If you want ARM + netbook then you'll need windows 8,

I hope not! Besides Android and ChromeOS, there is also this thing called iOS or OS-X. Then there is Ubuntu. With some luck there might even still be Meego. Frankly, if ARMs future success in this market depended on the windows port, Id strongly bet against it. I believe the opposite is closer to the truth, Windows hope of survival in these markets will depend on their ARM port.

 

[Mopetar]

Android has really been developed from the ground up to have a touch interface, it would probably be rubbish with a mouse and keyboard.

If you want ARM + netbook then you'll need windows 8, which I suspect will do the mixed mouse/keyboard/touch thing very well. It'll also provide the most important x86 apps (i.e. MS office). On the downside it's still bound to have all that history bloating and slowing it down so will run like a dog on any ARM cpu.

That said I don't see the netbook surviving outside of a niche market - it's tasks mostly better served by a tablet (lightweight browsing, watching stuff, flash games, etc) or a proper laptop (content creation needing cpu grunt and a big screen).

If you don't need existing x86 applications you could easily just use Linux with an ARM netbook. OpenOffice can substitute for MS Office for the most part and the browsing experience isn't going to be any different.

Microsoft may be using their move to ARM to cut out a lot of the bloat from Windows. There's a lot of legacy support in Windows and switching to the ARM architecture could allow them to drop almost all of it. Personally, I think they should just cut out all of the legacy stuff and include a feature that allows users to run it inside of a VM of an older version of Windows if it doesn't work in the new version of Windows.

ARM netbooks should be cheaper than the current x86-based versions. I don't know how well they can compete with tablets price-wise since tablets aren't really much different from a component perspective. The other problem is that not many companies are going to want to start a race to the bottom and erode profits. I think that most will be content to hang out around the iPad's price range and compete on additional features.

 

[Mopetar]

thats only the UI; not exactly the hardest thing to change. Besides; I mentioned Chrome OS for a reason, as it will be Google's OS for anything with keyboard and mouse. Initially at least as ChromeOS and android are going to merge:
http://www.silicon.com/technology/s...-converge-says-googles-eric-schmidt-39746988/

I think Chrome OS is more or less dead and isn't going to see more development other than having its best features stripped and added to Android.

Chrome OS made a lot of sense when netbooks were the new hotness and no one thought that tablets were viable. After all, Microsoft had been pimping them for nearly a decade and they never took off. Then the iPad came along and changed things and everyone and their grandmother was racing to get a tablet out the door. Manufacturers were much more excited about selling tablets for around $600 or more than Chrome OS netbooks for about half as much.

Google was still pushing Chrome OS, but the manufacturers were already trying to shoehorn Android onto tablets whether it was a perfect fit or not. Google realized they were going to do it wether the user experience was abysmal or not so they've really rushed to get their tablet version of the OS ready. If the manufacturers kept using non-tablet optimized versions of the Android OS it would probably just result in a tarnished brand.

Focusing on tablets keeps the hardware manufacturers happy because the margins are better and it also allows Google to leverage their Android Marketplace app store to bring content to the devices quickly, keeping the software developers and consumers happy. Chrome OS doesn't have much of a future and will most likely just be harvested for parts so that Android can run on a netbook.

 

[Dribble]

If you don't need existing x86 applications you could easily just use Linux with an ARM netbook. OpenOffice can substitute for MS Office for the most part and the browsing experience isn't going to be any different.

Define *easily*. No one but a hardware geek can use Linux cause it only ever kind-of works - you have to know how to debug and hack it to make it work properly. Equally the ui isn't up to windows standards, well it might be if you are a linux geek and know how to do all the mirad of things to make it work nicely but no *normal* person is going to do that.

 

[bryanW1995]

I'm mostly worried about AMD.

They are too small of a company to miss the boat on something like this.

AMD is probably smart to sit this one out. They can't compete with intel in cpu, nvidia in gpu, and nvidia/intel/samsung/ti/etc etc etc in mobile with their limited resources. Even intel with their nearly limitless R&D budget isn't going after all of those markets simultaneously. Bobcat and differnt versions of it are AMD's play for now, we'll see how things work out. Maybe they get bought out. Maybe they go bust. But, if they do manage to scale bobcat well, and if BD is at least somewhat competitive, and they continue their gpu overall competitiveness, they at least have a roadmap that could keep them independent and, dare I say, profitable, in the not-too-distant future.