AMD Llano: Mobile discrete GPU killer?

[IntelUser2000]

I don't know if anyone thought of this.

On the desktop, the graphics cards aren't so bound by TDP and the current generation HD5870 parts already perform way beyond what Llano is rumored to feature for the GPU core.

Northern Islands could probably create another 30-40% gap with some TDP increase.

Llano's GPU core:
~480 SPs
25.6GB/s bandwidth
"Evergreen" core

~Radeon HD5570 performance

Now look at mobile: http://www.anandtech.com/show/3758/avadirect-clevo-w860cu-5870-vs-285m/3

The Mobility Radeon 5870 features nearly identical specs to the desktop Radeon 5770. It's ~2.5x faster than the 5570.

Could "Northern Islands" derivative close the gap? Not much, as they are already power bound.

If Wikipedia's specs are correct, the performance of Llano's GPU core will be at least Mobility Radeon HD5650.

Now let's look at hypothetical 2011. Unless the mobile version significantly sacrifices clock/shaders to save power, this is what will happen.

Laptop 1: $2000, weighs 10 pounds, 1.5 inch thick, 15.6 inch screen, less than 1 hour battery, burns your lap, and your chances for fertility.

Laptop 2: $800, 6 pound weight, 4-6 hour battery life, running nowhere near "hot", sleek 1 inch

The first laptop might perform 3x faster, but is much worse in everything else. Before, that difference would have netted you 5-15x difference in performance.

Essentially, it'll obsolete ALL mobile discrete GPUs except the very high end. Low end, mid-end, mid-high end, all gone. The marketshare of integrated graphics in mobile goes from 90% to 98%.

It's very likely AMD has thought of that, soooo..

1. They'll take that cannibalization for 2-3 quarters until 28nm discrete GPUs are ready
2. Cannibalization is inevitable, might as well get used to it now
3. Related to #2, carry some of the GPU pricing on the CPUs that feature the GPUs
4. Some poor marketing and technical decisions.
5. Related to #4, the "hype" is really all hype and nothing much

 

[Scali]

Well, isn't it Intel that has cannibalized discrete GPU sales for ages now?
Intel is actually the largest GPU supplier in the world, and a lot of that is because of their low-cost, power-efficient IGPs.
So perhaps Llano will give AMD a piece of that action.

 

[Dribble]

It's a low end gpu with a cpu. It can never be more then that. It has very limited power requirements and doesn't have it's own main memory - sharing the very low bandwidth main memory bus with the cpu.

It's competing with intel's gpu+cpu offering, and for that I expect the gpu will be a bit better but the cpu will be worse. It won't change the world - it'll just give amd what intel already has, and hence help them compete a bit better in the budget notebook market.

 

[Phynaz]

I've heard various numbers tossed around about number of shaders and performance. I do not believe AMD has released any information on this, so I don't believe any of them.

 

[Lonyo]

Killer? Not so much.
Helper? Maybe yes please.

Optimus type technology plus Crossfire in a laptop. The power consumption of integrated when real graphics isn't needed, with the performance of a higher powered GPU when using Llano + discrete GPU in Crossfire (assuming a relatively moderate discrete GPU like mobile HD5570 type level).

 

[Rezist]

It sounds like Llano should be a nice step up from the HD3200, which will be very nice, if it's a good step up with similar battery life that would be great.

 

[GodisanAtheist]

Didn't any of you guys see the AMD presentation with Llano/Fusion running AvP in DX11?

Its really very impressive.

 

[nonameo]

Most people don't game on their computers, so I wouldn't imagine that this would matter much. HD3200 level graphics is enough for most. llano is just overkill.

 

[cbn]

I don't know if anyone thought of this.

On the desktop, the graphics cards aren't so bound by TDP and the current generation HD5870 parts already perform way beyond what Llano is rumored to feature for the GPU core.

Northern Islands could probably create another 30-40% gap with some TDP increase.

Llano's GPU core:
~480 SPs
25.6GB/s bandwidth
"Evergreen" core

~Radeon HD5570 performance

Now look at mobile: http://www.anandtech.com/show/3758/avadirect-clevo-w860cu-5870-vs-285m/3

The Mobility Radeon 5870 features nearly identical specs to the desktop Radeon 5770. It's ~2.5x faster than the 5570.

Could "Northern Islands" derivative close the gap? Not much, as they are already power bound.

If Wikipedia's specs are correct, the performance of Llano's GPU core will be at least Mobility Radeon HD5650.

Now let's look at hypothetical 2011. Unless the mobile version significantly sacrifices clock/shaders to save power, this is what will happen.

Laptop 1: $2000, weighs 10 pounds, 1.5 inch thick, 15.6 inch screen, less than 1 hour battery, burns your lap, and your chances for fertility.

Laptop 2: $800, 6 pound weight, 4-6 hour battery life, running nowhere near "hot", sleek 1 inch

The first laptop might perform 3x faster, but is much worse in everything else. Before, that difference would have netted you 5-15x difference in performance.

Essentially, it'll obsolete ALL mobile discrete GPUs except the very high end. Low end, mid-end, mid-high end, all gone. The marketshare of integrated graphics in mobile goes from 90% to 98%.

It's very likely AMD has thought of that, soooo..

1. They'll take that cannibalization for 2-3 quarters until 28nm discrete GPUs are ready
2. Cannibalization is inevitable, might as well get used to it now
3. Related to #2, carry some of the GPU pricing on the CPUs that feature the GPUs
4. Some poor marketing and technical decisions.
5. Related to #4, the "hype" is really all hype and nothing much

I remember reading that desktop Llano would have 5570 performance, but isn't a weaker fusion chip being used for laptops?

 

[IntelUser2000]

Didn't any of you guys see the AMD presentation with Llano/Fusion running AvP in DX11?

Its really very impressive.

That's not even Llano, but Ontario, which is based on the low power Bobcat.

 

[cbn]

That's not even Llano, but Ontario, which is based on the low power Bobcat.

What about the "Sabine" platform Llano? How much power compared to desktop Llano?

 

[OCGuy]

No....just....no.

Not yet.

 

[GaiaHunter]

Most people don't game on their computers, so I wouldn't imagine that this would matter much. HD3200 level graphics is enough for most. llano is just overkill.

And how many of those people don't game on PCs because their PCs aren't up for the task?

 

[Idontcare]

And how many of those people don't game on PCs because their PCs aren't up for the task?

That is a fair question but the answer for my case is that I don't know a single person in real-life that (a) wants to game, but (b) doesn't because their graphics in desktop/laptop isn't performing at level xyz.

Every one I know, and that is a lot of non-techie people, either prioritizes their cash flow so they have the hardware to play games at performance level xyz or they just turn down the eye-candy (this is what I do, gaming on my DELL laptop with Intel graphics) and resolution until the gameplay is adequately smooth.

 

[MarcVenice]

That is a fair question but the answer for my case is that I don't know a single person in real-life that (a) wants to game, but (b) doesn't because their graphics in desktop/laptop isn't performing at level xyz.

Every one I know, and that is a lot of non-techie people, either prioritizes their cash flow so they have the hardware to play games at performance level xyz or they just turn down the eye-candy (this is what I do, gaming on my DELL laptop with Intel graphics) and resolution until the gameplay is adequately smooth.

That's the thing, they prioritize gaming. I know many people who buy a laptop with a decent cpu, a screen with a certain size/resolution, and a decent harddisk, because that's what they absolutely need. I can think of two people right now who play only older games, or who have tried installing games on their laptop, but their laptops have to slow graphics to make it worthwhile.

Llano and ontario might change that, because they can still buy 'cheap' laptops, but ones that actually do have decent graphics that enables them to play certain games.

 

[Dribble]

Your average non techy who likes to game will buy a console - that gives them a consistently good experience. You really need quite a lot of technical knowledge if you want a good gaming experience on a pc.

 

[Idontcare]

Your average non techy who likes to game will buy a console - that gives them a consistently good experience. You really need quite a lot of technical knowledge if you want a good gaming experience on a pc.

Or in the case of one of my brothers you just need loads of ego, arrogance, and ignorance and you can have a rather crappy gaming experience on a DIY pc but insist it delivers teh uber pwnage with reckless abandon

 

[GaiaHunter]

That is a fair question but the answer for my case is that I don't know a single person in real-life that (a) wants to game, but (b) doesn't because their graphics in desktop/laptop isn't performing at level xyz.

Every one I know, and that is a lot of non-techie people, either prioritizes their cash flow so they have the hardware to play games at performance level xyz or they just turn down the eye-candy (this is what I do, gaming on my DELL laptop with Intel graphics) and resolution until the gameplay is adequately smooth.

And that is a fair answer.

But there are also the case of people that bought a PC/laptop with no interest in gaming and then later wish they could and then have all sorts of problems, or those that are interested in a game and try it but it is so slow/so ugly they decide consoles are the way to play.

It is like mobile phones and their built in cams/music players.

If you really want to take pictures you need a real camera for it. For music some will be passable, especially if you don't listen to it all the time.

Now if APUs can bring "real camera" level instead of "mobile camera" level that would be quite positive, IMO.

Sure, for the ultimate experience you still need "professional level" cameras, but that is another story.

But similar to consoles quality (actually means more quality due to resolution differences) to laptops that don't weight a ton, cost a pair of your organs and dry up in 30 minutes... I'm in for one.

 

[cbn]

I can think of two people right now who play only older games, or who have tried installing games on their laptop, but their laptops have to slow graphics to make it worthwhile.

Llano and ontario might change that, because they can still buy 'cheap' laptops, but ones that actually do have decent graphics that enables them to play certain games.

I agree. For a lot of people I'm sure these cheaper APU laptops will be sufficient for their needs.

 

[cbn]

http://www.anandtech.com/show/3762/...ly-2010-reviewed-shaking-the-cpugpu-balance/2

From the looks of it, Apple is trying to broaden its install base of OpenCL compliant machines. In preparation for what I'm not really sure, but something is coming.

Isn't Apple interested in Fusion? I wonder what kind of programs they are planning for Open CL at the consumer level? If something strong pans out could this end up also giving a good boost to Linux based machines also using Fusion?

Intel/Microsoft vs a fusion backed Apple/Linux install base?

 

[Scali]

Microsoft is interested in Fusion just as much as Apple.
Don't forget, they introduced DirectCompute in DX11, and using GPU-acceleration is one of the new features of Windows 7. They already have support for GPU-accelerated transcoding: http://www.bit-tech.net/news/bits/2009/06/02/microsoft-reveals-native-win-7-video-transc/1

 

[cbn]

What about PC gaming?

OPEN GL vs MS D3D?

Will we see more competition between developers on Steam?

High cost (Apple)/Low cost (Linux) installation base vs moderate cost MS install base? (vs console gaming market)

 

[Scali]

I don't think much will change really. If anything, Steam on Mac has demonstrated just how mighty MS and D3D are.
As far as I know, Valve has no plans of replacing their D3D code with OpenGL on Windows. So that pretty much implies that they prefer extra development effort on Windows, in exchange for the best possible performance and image quality.

On the other hand, Valve may also show that you don't necessarily need the best possible performance and image quality to have nice and successful games on a variety of platforms... but it's too early to say at this point.

It's a bit of a chicken-and-egg problem... Developers may choose to go for cross-platform code, so they can cover Apple, linux and Windows in one go, with OGL.
But that means they lose their competitive edge in terms of performance and image quality... so what do the gamers go for then? Will they stick to Windows and D3D games? Or will they accept slightly less sophisticated games and start using Apple/linux platforms more?
With the current marketshare of Apple/linux, it's not that interesting for developers to take the risk. I'm not sure why Valve made this decision, to be honest.

For my own private projects I haven't quite decided what to do either. I have been working with D3D for many years, and my current codebase supports DX9/10/11. However, I recently started an open source project, for which I made an OpenGL sample program. That sample got a bit out of hand, and ended up being a pretty complete rendering backend. So I could move to OpenGL at this point.
I still prefer D3D as an API... but my OpenGL code runs on linux, FreeBSD, Solaris and OS X aswell, not just Windows.

 

[cbn]

With the current marketshare of Apple/linux, it's not that interesting for developers to take the risk. I'm not sure why Valve made this decision, to be honest.

I don't know much about Apple, but it seems people like them for the user friendly features.

If they could take this to the next level with graphics (possibly as a useful tool) then so much the better.

Is there anything about OPEN GL that lends itself better to different tasks (compared to DX11)?

 

[Scali]

I don't know much about Apple, but it seems people like them for the user friendly features.

If they could take this to the next level with graphics (possibly as a useful tool) then so much the better.

Sure, but even on Windows, OpenGL is already problematic. Since so few games use OpenGL, driver developers don't spend a lot of time on optimizing and debugging the OpenGL drivers.
On the Mac the problem is even larger, so both Apple and AMD/nVidia have their work cut out, if they want to make the Mac competitive with Windows in terms of gaming.
I don't think that's where their priorities lie.

Is there anything about OPEN GL that lends itself better to different tasks (compared to DX11)?

Not really. Ever since OpenGL 2.0, they've pretty much tried to be more like D3D with every update.