2004 Handhelds

[cthellis42]

Don't worry. At least you didn't do it to a Mars orbiter.

 

[V3]

PowerVR, earlier used in the Sega Dreamcast gaming console, uses little memory bandwidth or power, but is capable of delivering console-level performance in handheld devices, Harold said. "It is more feature-rich than the Dreamcast chip, and delivers a similar level of performance," he said.

So Dave, have you got any more info from Harold about the performance of MBX ?

When he said similar level of performance to Dreamcast, what is he referring to ? Clearly the fillrate is close to 5 times of Dreamcast.

 

[akira888]

But that's "effective fill rate" - PVR's way of giving the consumer a metric by which they can compare their products against IMRs made by competitors. The actual fill rate of a MBX is lower, and more comparable to the DC's CLX. An HRS at 80 MHz with two pixel pipes should have a fill rate of 160Mpixels/sec, about 60 percent better than CLX, still extremely impressive. What I don't get is how they integrated a full vertex shader with around 700000 transistors.

 

[Fafalada]

But that's "effective fill rate" - PVR's way of giving the consumer a metric by which they can compare their products against IMRs made by competitors.

That's how I thought of it as well, but the numbers and subsequent performance chart in that Arm pdf is pretty shady in that area - it's nigh impossible to determine what raw and what effective fillrate is supposed to be.

What I don't get is how they integrated a full vertex shader with around 700000 transistors.

Well the doc only says 'SIMD geometry processor' which could vary from a full general purpose APU like processor to a completely fixed T&L unit. Or am I missing something where they describe it in more detail?

 

[Dave Baumann]

So Dave, have you got any more info from Harold about the performance of MBX ?

When he said similar level of performance to Dreamcast, what is he referring to ? Clearly the fillrate is close to 5 times of Dreamcast.

No, I've not heard back from Harold, and I suspect he's in Las Vegas now. As to exectly what they are referring to I don't know because I don't know what level of ARM solution they are demoing.

But that's "effective fill rate" - PVR's way of giving the consumer a metric by which they can compare their products against IMRs made by competitors. The actual fill rate of a MBX is lower, and more comparable to the DC's CLX. An HRS at 80 MHz with two pixel pipes should have a fill rate of 160Mpixels/sec, about 60 percent better than CLX, still extremely impressive. What I don't get is how they integrated a full vertex shader with around 700000 transistors.

If they are listing effective fillrates they have previously always stated "effective fillrate with X times overdraw", nowhere does the ARM PDF state that, they only say that the effective performance will rise with increased complexity. My suspicion is that they are actual hardware rates, but what is defined as a pixel may be a little off.

 

[manux]

www.opengl.org has some quite interesting links in its main page... Seems that there will be quite a push for smart phones and 3d, which is nice...

http://ati.com/companyinfo/press/2004/4719.html
http://www.futuremark.com/pressroom/pressreleases/?010704
http://www.hybrid.fi/mobile.php

 

[akira888]

Well the doc only says 'SIMD geometry processor' which could vary from a full general purpose APU like processor to a completely fixed T&L unit. Or am I missing something where they describe it in more detail?

www.arm.com/pdfs/MBX 3D.pdf

On page 6 it describes the functionality of the VGP as being quite comparable with the DX8.0 vs1.1 spec. It also states that it requires ~160K gates (which I estimated to be ~700Ktrans) to implement. That seems to be much lower - by more than an order of magnitude actually - than the transistor cost estimates for VS units previously posted around here for the various PC graphics ASICs. My only (dumb) guess is that the FP ALUs aren't pipelined as aggressively since the speed requirement is much lower. Or the earlier estimates were off?

If they are listing effective fillrates they have previously always stated "effective fillrate with X times overdraw", nowhere does the ARM PDF state that, they only say that the effective performance will rise with increased complexity. My suspicion is that they are actual hardware rates, but what is defined as a pixel may be a little off.

I looked at the white paper again and I'm guessing - given my background is SW rather than HW everyone take this with a barrel of salt - that at 80 Mhz on 180nm litho with a dual pipe HR-S that can each process two "multisamples" per pixel gives a "AA pixel" rate of 320Mp/sec. Similiarly at 120 Mhz (130 nm litho) we arrive at 480Mp/sec, and thus the "320-480Mp/sec" number in page 9 of the PDF for HR-S. Or HR-S could have four full pipes, although that seems incorrect given what I've read.

 

[PC-Engine]

Looks like MBX has a very high fillrate compared to the new mobile 3D chips, however, I wonder how it does in polygon throughput.

http://www.commsdesign.com/showArticle.jhtml?articleID=17300183

Cellphone 3-D chips gear up for market

By Rick Merritt

EE Times

Jan 09, 2004

LAS VEGAS — Four chip designers and one software company rallied around the newly minted OpenGL ES applications programming interface at the Consumer Electronics Show here, pledging to deliver cores and games that bring 3-D graphics to cellphones starting this year.

ATI Technologies (Markham, Ontario) launched at CES the Imageon 2300, its first 3-D accelerator for cellphones. The chip processes up to 1 million vertices/s and 100 million pixels/s while dissipating just 75 mW peak power. It will sample by June for less than $10 in high volumes.

"People once told us 3-D would never make it in notebooks, and now that's a major business for us. We think the same will happen in cellphones," said Azzedine Boubguira, director of marketing for the handheld group at ATI.

ATI announced a design win for a 2-D accelerator in a Motorola phone last year and plans to launch another multimedia chip for cellphones at the 3GSM Congress in February, he added.

For its part, Takumi Corp. formed in Japan last October as part of a joint venture said it is working on a second generation of its cellphone 3-D accelerator core that could process up to 10 million triangles/s and 200 million pixels/s at a power consumption of about 1-3 mW/MHz, according to chief technology officer Hiroyuki Nitta. The company's first-generation GShark chip has been designed into phones from NEC and Sony. The GShark 2 core should be available by June.

The ATI and Takumi accelerators, along with the MiMagic 6 applications processor from NeoMagic and the MBX accelerator core from Imagination Technologies are all going through an OpenGL ES compliance testing process.

On the software side, HI Corp., a software accelerator maker from Japan expects to license and port dozens of games for OpenGL ES this year. The company opened a U.S. office this year to assist in developing global content for version 4.0 of its Mascot Capsule software acceleration engine that will be complaint with OpenGL ES.

"There's an effort going on to convince AAA gaming companies it's time to move into the mobile gaming space, mainly by licensing their titles," said Joshua Mogal, marketing director for the U.S. office of HI.

The Khronos group, an ad hoc association of about 50 companies that developed OpenGL ES, has begun work on a version 1.1 it expects to release at Siggraph in August. The update will include enhancements to the geometry and rasterization capabilities of the 1.0 version released last July, including new texture blending modes, said Khronos Chairman Neil Trevett.

 

[PC-Engine]

..

 

[PC-Engine]

..

 

[rabidrabbit]

hehe, did the N-gage really sell out last christmas in US at many EB stores? It was on the news here the other day, that the 'N-gage is really gaining support in US, and there is a massive advertaising campaing during some big football matches'

It said that the ngage is now 'the thing to have' in US, hehe

The newsflash did have another point of view too: A young african american kid was interviewed. He said something like this (while playing a PS2 FFX-2 game) "The ngage sucks! It really does! There's just too much in it! I don't want a phone, a gaming device and a MP3 player all in one, it's just too confusing! I want them all to be a separate device."

The whole newsflash... it was all a bit bizarre... hehe

 

[London Geezer]

He said something like this (while playing a PS2 FFX-2 game) "The ngage sucks! It really does! There's just too much in it! I don't want a phone, a gaming device and a MP3 player all in one, it's just too confusing! I want them all to be a separate device."

The whole newsflash... it was all a bit bizarre... hehe

"Yeah!!! I want to have 10 kilos of electrical equipment on me everytime i go out! And pockets the size of Arizona!!! YAY!!"

Really, unifications of those things (Cell phone, MP3, video, gaming etc) is what i've been waiting for since i came out of my mother's tummy.
This does not mean that Ngage is cool though.

One curious thing is that here in Britain you can get one for free with mobile phones providers contracts (of course, not on pay-as-you-go. don't know if it works like this everywhere else, doubt it)... So the price issue is non existant here. I got my Sony T610 for free with Orange.

 

[rabidrabbit]

yup, I bought my bf the SonyEricsson P900 smartphone for christmas.

It has ok gaming capabilities (V-Rally on it looks comparable to a PSOne game, but the controls make it unplayable), can play MP3 files from MemorySticks (up to 128MB) and even play MPEG4 clips (there are movies circulatin on the 'net on MPEG4 under 100Meg size, haven't tried those, but the quality of the pre-installed short clips at least is very good),
plus the phone and still/video cam.

It really is a nice piece of kit, and I can see the convenience and ease of use with all these being in one portable device.

 

[Simon F]

Good grief you are quick at getting these press announcements!

I'd almost swear that they arrive at B3D before the internal email system delivers them here!

What I don't get is how they integrated a full vertex shader with around 700000 transistors.

Well the doc only says 'SIMD geometry processor' which could vary from a full general purpose APU like processor to a completely fixed T&L unit. Or am I missing something where they describe it in more detail?

I think it's safe to say it's a DX8 compatible system.

 

[Dio]

DX8-class vertex shader units can be quite cheap when well designed. They can be fairly simple and don't need expensive items (like huge latency compensation). The clipping unit can be more expensive and/or much more complicated.

 

[marconelly!]

hehe, did the N-gage really sell out last christmas in US at many EB stores? It was on the news here the other day, that the 'N-gage is really gaining support in US, and there is a massive advertaising campaing during some big football matches'

It said that the ngage is now 'the thing to have' in US, hehe

That is absolutely untrue, though. nGage is about as good as dead in the good ol' US of A, and from the look of things it's not doing much better in Europe either (see the last CVG sales report).

 

[cthellis42]

Come now, 15k is great results, right? I mean, so long as they're making 500% profit and don't actually hope to succeed as a gaming device...