Nokia N93 - first Nokia with HW accelerator

[jkemp]

If all N-Gage 2 phones used MBX, what would be the point of not exposing vendor specific extensions? Surely Nokia plans on using different GPU vendors and using OpenGL ES 1.0 for N-Gage 2 as it is the lowest common denominator?

Let´s just put it this way: why would we deliberately limit ourselves in using just one hw vendor, giving them all the control and basically freedom to dictate the core price and terms. Since we have this great standard for abstracting hardware, why not utilizing that.
Minimizing fragmentatiion, maximising hw options..

 

[Entropy]

NOKIA will do what is in it's best interest and that's completely irrelevant where each piece of IP comes from. Large companies like NOKIA or in extension Texas Instruments aren't the average consumer that can be swayed by any sort of funky marketing or PR stunts.

You can't be serious.
There is a lot of money spent marketing to corporate entities. It pays for itself. And at the end of the day corporate decisions are made by people, flesh and blood, meeting - talking - deciding. The whole human interaction thing is still a part of the decision process with the intricacies of corporate interaction and policy layered on top. Hell, not only do corporations feed happy marketing bullshit to their industry customers, they even do it to their owners!

Bottom line, it really is a complete jungle. Granted the norm is not consumption for selfgratification as in consumer space, but...

I'm not making any statements in this particular case. But I have to say that the slivers of in the trenches perspective on the development of these markets that jkemp and others offer is one of the few reasons I still visit B3D at all.

 

[Lazy8s]

The level of vertex functionality offered strictly by the OpenGL ES 1.1 spec is probably respectable, but a neat feature like the curved surface acceleration offered as an extension by a processor like MBX+VGP would be missed, in addition to the shader programmability obviously.

In ATi's deal with Nokia, ATi even admitted that they don't expect to get all of Nokia's graphics business; they're only realisitically hoping for a significant portion.

Regarding the N93's relatively deep graphics pipeline: while avoiding stalls isn't harder, it's assumedly more important due to the larger penalty. Is the higher throughput gained from the tradeoff what developers, those who have experience on multiple companies' GPUs, are finding to be more notable?

 

[Ailuros]

You can't be serious.
There is a lot of money spent marketing to corporate entities. It pays for itself. And at the end of the day corporate decisions are made by people, flesh and blood, meeting - talking - deciding. The whole human interaction thing is still a part of the decision process with the intricacies of corporate interaction and policy layered on top. Hell, not only do corporations feed happy marketing bullshit to their industry customers, they even do it to their owners!

Bottom line, it really is a complete jungle. Granted the norm is not consumption for selfgratification as in consumer space, but...

Your last sentence comes closer to what I actually meant; there is a slight difference between the average consumer and the large company representative, both being customers in the end.

You're not trying to tell me that professionals (and if they aren't tough luck for the company they're representing) aren't times harder to trick then the average Joe are you?

 

[Xmas]

The level of vertex functionality offered strictly by the OpenGL ES 1.1 spec is probably respectable, but a neat feature like the curved surface acceleration offered as an extension by a processor like MBX+VGP would be missed, in addition to the shader programmability obviously.

Sorry, there is no such extension.

Regarding the N93's relatively deep graphics pipeline: while avoiding stalls isn't harder, it's assumedly more important due to the larger penalty. Is the higher throughput gained from the tradeoff what developers, those who have experience on multiple companies' GPUs, are finding to be more notable?

The only stall situation I can think of you might be referring to is trying to read pixels from the framebuffer, but that is expensive on any platform with 3D hardware (though admittedly more expensive on a TBDR where there are two frames in flight). And it's rarely ever needed, at least for games.

 

[Lazy8s]

Can't stalling in general be more costly on a TBDR and brought on by other unadvisable practices like trying to update a texture or mixing 2D and 3D within a frame?

All of MBX's curved surface functionality is exposed with basic OpenGL ES 1.1?

 

[Simon F]

All of MBX's curved surface functionality is exposed with basic OpenGL ES 1.1?

The curved surface package is simply a library built on top of MGL, but it could be ported to OpenGL ES (with extensions) if there is demand.

 

[Xmas]

Can't stalling in general be more costly on a TBDR and brought on by other unadvisable practices like trying to update a texture or mixing 2D and 3D within a frame?

In the absence of pixmap support (which is indeed rare), mixing CPU 2D and GPU 3D requires reading the pixels from the render target, doing 2D, then writing the pixels back to the framebuffer (if you want it to appear on screen). That's expensive on any GPU platform.

 

[Lazy8s]

Why does the accuracy of the N93's trilinear filtering score low in the JBenchmark Pro test? The texture size doesn't seem to be a problem. If the trilinear filtering had passed like the PNX4008 Sony Ericssons' did, the N93 would dominate the Pro rankings by a lot more than it already is.

The bilinear test, this time, thankfully bumped up the texture size to a more reasonable 8x8.

Also, what's holding back the score on the Chess test? Shouldn't an ARM11 have some kind of advantage over ARM9s?

 

[Xmas]

Why does the accuracy of the N93's trilinear filtering score low in the JBenchmark Pro test? The texture size doesn't seem to be a problem. If the trilinear filtering had passed like the PNX4008 Sony Ericssons' did, the N93 would dominate the Pro rankings by a lot more than it already is.

A simple software bug. I guess it might get fixed with a firmware update.

 

[roninja]

rumours N95 will be announced shortly, wonder if it will keep with the same CPU as the N93....;-)

 

[hidefguy]

rumours N95 will be announced shortly, wonder if it will keep with the same CPU as the N93....;-)

With a 5 megapixel cam and the same "DVD-like recording" as found in the N93, I'm guesssing that this is powered with at least OMAP 2. The available pictures make it look like it's touch-screen. With GPS built in, *if* this does have OMAP 2 (presumably not OMAP 3 - now that would be amazing - hopefully not a 2430 but the 2420 instead) this handset will blow other devices off the picture. It's too bad they opted for a QVGA display with this device, but presumably that's built-in redundancy to encourage users to buy a VGA display upgraded device later.

[CORRECTION: the N95 seems to have the same display resolution as the N80, almost HVGA]

 

[Lazy8s]

OMAP3s don't launch until mid-2007.

While OMAP2420 gives the strongest 3D performance of the OMAP2s, the less expensive 2430 and on have an improved IVA imaging and video core, so they might become the preferred choice of phone makers, especially those of camera/camcorder phones, once they launch.

 

[Lazy8s]

An OpenGL ES, "native" version of the JBenchmark HD game test increased the test's speed by about 4x and 5x on the Nokia N93 and Sony Ericsson M600, respectively, compared to the Java version. That raised its framerate on the N93, the highest rated phone for that benchmark in either API, to almost 78-fps.

 

[roninja]

The best mobile phone in the world

Timo Poropudas

Nordic Wireless Watch - November 6, 2006 at 17:40 GMT

http://www.nordicwirelesswatch.com/wireless/story.html?story_id=5052


The Nokia N93 is the world’s best mobile phone platform. This was the finding of Futuremark’s graphics benchmarking tests, the first results of a new cooperative program between Futuremark and MobileMonday. The Nokia N93 scored off the chart when Futuremark tested ten mobile phones available on the Finnish market.

Petri Talala, the head of the mobile division of benchmarking specialist Futuremark, thinks he might get even better test results in Japan where he hopes to be testing soon.

Futuremark’s benchmarking focuses on the ability of mobile phones to show graphic information. They show how fast and smoothly a phone can show a video, and how well animations and games run on the screen.

Nokia N93 tops the graphics comparison.

 

[jkemp]

GLBenchmark scores seem to appear online..

N93 wipes the floor with the rest

 

[Lazy8s]

Like JBenchmark, GLBenchmark is comprehensive for a public tool.

The trilinear filtering tests correctly this time under these OpenGL ES benchmarks.

The tests are quite conservative for hardware accelerated devices, though, as the N93 maxes out the scale in every graphics measure.

 

[kishonti]

Like JBenchmark, GLBenchmark is comprehensive for a public tool.
The tests are quite conservative for hardware accelerated devices, though, as the N93 maxes out the scale in every graphics measure.

N93 gets maximum "performance stars" as it has the highest results in every performance subtests. These stars show the relative performance compared to the actual highest score inside the database.
(These stars help the general public to interpret numerical results.)

 

[Lazy8s]

Thanks for the clarification.

The actual N93 has ended up being even more powerful than the old OMAP2420 development systems, as Nokia had suggested it might, as measured by 3DMarkMobile ES 1.1.

http://futuremark.com/bdp/certified/

A stock N93 runs the Proxycon graphics demo at over 30 fps.

 

[Lazy8s]

I think I noticed the N93 in the Transformers movie, given a little cameo with some impressive exposure.