Designing The Gp2x's Successor - Hardware Selection Poll

[nubie]

Not that it probably matters anymore, but I doubt you could really purchase a chip with such a robust featureset for $20. Maybe in quantities of a million? :rolleyes: In the quantity the GP2X or its successor would be made I wouldn't expect to be able to pull that off with an acceptable price. At any rate it'd require a good bit of engineering to get a discrete ARM processor and a custom multimedia IC working together nicely; money I don't believe GPH has to throw around. Hence their affinity for SoCs..

Sorry, that is what I heard, I agree that it is likely more difficult to get ahold of than the cheapo SoC that we use, but until someone offers hard facts I am sticking to my guns (please prove me wrong, at least point me in the right direction to find the facts).

Engineering? Maybe, but since it is nVidia they probably have a reference design to rip-off (or use) just the way they "designed" the GP2X on the Magiceyes devkit.
Edit: Here is from the FAQ on mobilemedia page. (http://www.nvidia.com/object/mobilemedia_faqs.html:

Q: What processors do you support?
A: We have reference designs on show at Computex on both Freescale and Intel platforms, however we also have existing designs with TI as well so our solution remains relatively agnostic and we can support many other solutions as well depending on customer requirements.

Isn't Freescale an ARM style unit? And they claim support for other solutions.

Of course we all want the fastest machine around, but don't forget we have to balance performance and battery life here, people. Sticking a 400Mhz ARM9 and a GoForce3D chip in there is going to suck a lot of juice compared with a SoC that has it all in one chip at a slightly slower clock rate (Option 1).

Indeed, especially when that SoC is rated for a maximum power draw of just 200mA at 1.8V. I believe that'd come in under the GP2X's MMSP, even. More advanced fabrication process than 130nm is probably to thank.

I am officially confused, are you for or against the goforce? FYI it is already in a cellphone you or someone you know owns. Second here are some specs on the 5500. Please provide me some links or some facts, I want to learn more about your side.
http://www.nvidia.com/page/handheld.html
http://www.nvidia.com/object/IO_15896.html

NVIDIA NPOWER POWER MANAGEMENT
• Fully-static CMOS technology
• Low-power 90nm process
• Individual module enables
• Automatic shut-off of unused pipeline stages
PACKAGING & VOLTAGE
• Available with 2MB stacked, 8MB stacked, or
external memory interface
• JTAG boundary scan & BIST
• 0.95 to 1.32V core, 1.71V to 3.30V I/O

2nd Generation NVIDIA nPower Technology
Minimizes the drain on your battery, so you get more hours of music, videos and games and still have plenty of talk time.
TV-Out
Show your photos and videos on a bigger screen so everyone can share the memories.

The TV-Out is just interesting, I wonder if that means it has the TV encoder on-chip, knowing nVidia it's possible, in which case that cuts out another useless power-sucking chip (damn Conexant isn't made for handhelds)

 

[icurafu]

I would be interested in some debate about using the AMD Geode GX 533@1.1W Processor, or whatever else is cheap in a couple of years. I understand the Geode is moving to Athlon based architecture in the near future. Would it be a positive thing, to move from ARM9 to x86? The technical specifications have some interesting points.

http://www.amd.com/us-en/assets/content_ty...odegx533_pb.pdf

Geode LX is another alternative. (Might be too big)

http://www.amd.com/us-en/assets/content_ty..._geodelx_pb.pdf


You get a FPU, 3DNow! and MMX extensions. Looks like a fast 2D accelerator only.

In two years, we would expect something similar to the Geode NX with embedded video.

Shrinking and embedding current Geode NX.

http://www.amd.com/us-en/assets/content_ty...nxfamily_pb.pdf

Hopefully, we’ll see something from ATI in there by then, considering the merger.


John technology from VIA is also interesting, and soon to come. Intergrated VIA CPUs intergrating unichrome (Savage2000 2D with Savage4 3D) on insanely low watts.

Mostly, this is just a question. Is it time to connsider x86?

 

[nubie]

Mostly, this is just a question. Is it time to connsider x86?

I think the answer in the previous thread was a resounding no!!

The Geode NX is the thing that really gets me going, it is basically a low-voltage Athlon, it is even Socket A/462 drop-in compatible. Now with the ATI merger will we see ultra-low power Athlon 64 based cores at 0.8 volts (and less) and integrated xpress200 solutions on the die?? We can only hope.

As far as an open-source low-cost handheld is concerned the people writing the programs much prefer the ARM architecture to the x86. And the hardware people in the know are correct that ARM has them beat for power consumption; forget the ballpark, x86 isn't even in the same league.

I mention the goforce 5500 because it is a fully functional and useful device, its sheet reads like any good media-center bound SoC, except it is without a processor. And the technology is already proven in phones that are out there in large numbers (the goforce4800 in the slivr and razr mobile phones).

The 90nm process and the low-power modes, coupled with the TV-Out make it a no-brainer when adding an accelerator. 2d, mpeg4, 3d and more are just icing on the cake.

The question is whether there are single ARM cores interfaceable with it that run at high enough clocks (300-400mhz) to make it worthwhile.

Screw the whole idea if it is too complicated, just do a SoC with a ARM core sporting a FPU and a decent clock and I guess we could all be happy. I just wanted decent 2D acceleration and the option of 3D.

 

[Aninhumer]

I think if you had a good clockspeed on the main processor, an FPU and 2D acceleration, good 3D wouldn't be that difficult (for the hardware). Obviously it wouldn't be as fast as if it did have the 3D, but if it would be overly complicated/expensive/powerneedy it's probably better to not have one.

And we really can't decide what hardware we want in there for at least a few years, things change so fast. They might even make a major breakthrough next year, and GPH would be plodding along with their feeble 400Mhz proccessor.

 

[scanti]

What would be a cool over the top feature, would be a "HDR" LCD display.

Instead of having a few super bright white LEDs behind the LCD you have an array of say 80 x 60 LEDs (for a 320 x 240 display). Each LED can have 256 levels of brightness, from off to fully on. You then have a normal RGB frame buffer that drives the LCD and an 8 bit brightness frame buffer that drives the LEDs.

You can then use the LCD for the colour information and the LEDs for the lower resolution brightness information. If you don't want to bother with the additional HDR features you just set all the LEDs to full brightness.

Of course using 4,800 LEDs (I didn't realise it would be that many) instead of the usual 3 or 4 (?) would put up the price and power consumption (although each LED wouldn't have to be as bright) but I did say it was an over the top feature.

Unfeasible (maybe less so in the future) but it would be cool.

 

[pkostrze]

Know what I'd like? It's really simple... a small watch battery CMOS to store date/time. That's it. The rest of the stuff would be cool, god knows I'd like a touchscreen but it looks like, with a little hacking, the GP2X could be retro-fitted with one. If the date/time could be stored then Qtopia would work as a PIM and the GP2X would be perfect as it is.

Let's live in the now; we were damn lucky GPH split off and made the GP2X in the first place - we could all have been shafted with the vaporware XGP.

 

[Gadget]

If they release one with WiFi, I will buy 2 =D I wish there was a simple WiFi solution for the GP2X, that doesn't involve a BOB ='(

 

[xafier]

I think option 1 is by far the best of the two options... even the power of an FPU would be a great advantage alone for any 3D work, but dedicated 2D/3D power would improve things dramatically... I'm all for anything with an FPU... I really wish we'd had an FPU instead of the 940, it would have got used a lot more in non-commercial stuff

 

[Epicenter]

I contacted GamePark about using the "Option 1" SoC (Samsung S3C2640) if GPH wasn't interested; they informed me they have been working on a platform for that exact chip for an upcoming product line. Whether it's the XGP or something in development for later release I am still waiting to hear. That's some excellent news; a machine using this chip would be truly incredible to develop for.

 

[Gadget]

I contacted GamePark about using the "Option 1" SoC (Samsung S3C2640) if GPH wasn't interested; they informed me they have been working on a platform for that exact chip for an upcoming product line. Whether it's the XGP or something in development for later release I am still waiting to hear. That's some excellent news; a machine using this chip would be truly incredible to develop for.

Excellent =D Tell them to build an embedded WiFi adapter as well, or at least provide a plug in WiFi upgrade module, or even just build a USB host in there...

 

[Epicenter]

I did; I suggested a low-cost WiFi chipset like the (was it .. Zyxel?) one that worked so readily with the GP2X's linux implementation attached via USB Host capability to the S3C2640 with adjustable transmit power (for battery concerns) and an internal antenna or antenna jack (e.g. RP-SMA, not something that is flimsy and easily broken like the MC jack from an Orinoco PCMCIA adapter.)

Among my other suggestions were a modular control system with a jack to plug control units into; e.g. a D-Pad, Click-stick or perhaps an analog device if they wanted to do a bit more engineering; 2 onboard USB ports (mini or regular size depending on form factor) .. a distinct TV/Audio output port as not to need to use a combined plug like the GP2X, a hardware volume potentiometer, and 2 SD card slots (the S3C2640's capable of that.)

GP was very receptive and positive, and the representative I spoke to responded quickly and with very good English. GP doesn't seem like they deserve the scorn they often recieve here, and it looks like they're putting together some pretty solid products. I wasn't wild about the GP32, but it seems they've got the right idea.

 

[Dzz]

GP doesn't seem like they deserve the scorn they often recieve here, and it looks like they're putting together some pretty solid products. I wasn't wild about the GP32, but it seems they've got the right idea.

I think a lot of the "scorn" is not directed at their rumored hardware choices. Tomorrow is the one year anniversary of the first XGP announcement and it doesn't seem particularly closer to actually being launched than it ever did; I think that's the source of most of the criticism.

 

[Epicenter]

Well, I can't blame them for taking their time. The GP2X came onto the market without any testing, that's pretty much the root cause for the first run's massive faults. And I will say GP is at least doing a better job of communication and PR, even if they can't always meet the deadlines they set for themselves. (neither does GPH. Look at the new firmware announcement.. mid-July? And now it's mid-August and no sign of it?)

 

[Dzz]

Well, I can't blame them for taking their time. The GP2X came onto the market without any testing, that's pretty much the root cause for the first run's massive faults. And I will say GP is at least doing a better job of communication and PR, even if they can't always meet the deadlines they set for themselves. (neither does GPH. Look at the new firmware announcement.. mid-July? And now it's mid-August and no sign of it?)

Yup, GPH gets bashed a lot too, but for different reasons than GP

 

[DaveC]

I did; I suggested a low-cost WiFi chipset like the (was it .. Zyxel?) one that worked so readily with the GP2X's linux implementation attached via USB Host capability to the S3C2640 with adjustable transmit power (for battery concerns) and an internal antenna or antenna jack (e.g. RP-SMA, not something that is flimsy and easily broken like the MC jack from an Orinoco PCMCIA adapter.)

Among my other suggestions were a modular control system with a jack to plug control units into; e.g. a D-Pad, Click-stick or perhaps an analog device if they wanted to do a bit more engineering; 2 onboard USB ports (mini or regular size depending on form factor) .. a distinct TV/Audio output port as not to need to use a combined plug like the GP2X, a hardware volume potentiometer, and 2 SD card slots (the S3C2640's capable of that.)

GP was very receptive and positive, and the representative I spoke to responded quickly and with very good English. GP doesn't seem like they deserve the scorn they often recieve here, and it looks like they're putting together some pretty solid products. I wasn't wild about the GP32, but it seems they've got the right idea.

And how is this 266 MHz only device going to be much better for emulation than the GP2X is now? The GP2X can OC even higher many times right now. Yes the GPU even 3D will help a little bit but will it really make all that much of a difference with stuff like SNES or GBA? The PSP with its 3D GPU still struggles at times to emulate SNES full speed at 333 MHz. So either the code is inefficient or the 3D chip doesn't give all that much of a boost.

I know you are looking at it from a mostly 3D homebrew angle rather than emulation but a HUGE portion of the users want emulation as a main use as is is a big niche. Many of us that want 3D will just go buy a DS or PSP for that purpose. Honestly there are so many 3D devices including home consoles and PCs that yet another doesn't excite me much. So wouldn't it be a better use of that cost to put in a faster CPU and lots of RAM, (NeoGeo emulation), and have 2 SD slots one with SDIO for wireless or other expansion that the user wants?

 

[Epicenter]

And how is this 266 MHz only device going to be much better for emulation than the GP2X is now? The GP2X can OC even higher many times right now. Yes the GPU even 3D will help a little bit but will it really make all that much of a difference with stuff like SNES or GBA? The PSP with its 3D GPU still struggles at times to emulate SNES full speed at 333 MHz. So either the code is inefficient or the 3D chip doesn't give all that much of a boost.

You're not really following and I doubt you've looked at the S3C2460 specifications; you clearly didn't even read my whole post thoroughly. If you had; you'd know that:
- The CPU is natively 266 MHz, not a 200 MHz part. The ARM926EJ-S architecture spans up to 400 MHz right now, so it's nowhere near the clock ceiling-- overclockability to 333 MHz+ on most units is probable.
- There is an AFP9 Floating Point/Vector Processing unit parallel to the CPU. On its own, it is about as powerful in MFLOPS as the GP2X's ARM920T is in MIPS. This can be used to offload stress from the main CPU and enable the emulation of other ICs in the SNES or add-on helper ICs added by cartridges (e.g. Starfox and Mario World 2's 'SuperFX' chips.)
- There is a programmable 16-bit DSP supplementing the two.
- The 2D/3D accelerator is fully OpenGL-ES compliant and can be utilized for 2D functionality such as transparency handling, blending, potentially scaling/rotation with the right implementation. This would give SNES emulation a significant boost. I have no reason to believe the SNES GPU is being utilized to accelerate graphics emulation to its fullest potential, if at all, at this time.
- Other benefits include that instead of avoiding the use of floating point math, integer and FP math could be used in relatively equal amounts per frame allowing equitable loading of the Integer processing capability of the ARM9 and the FP processing capability of the AFP9, providing a potential performance benefit approaching 100% in many applications without the hassle of writing multiprocessor code.

I know you are looking at it from a mostly 3D homebrew angle rather than emulation but a HUGE portion of the users want emulation as a main use as is is a big niche. Many of us that want 3D will just go buy a DS or PSP for that purpose. Honestly there are so many 3D devices including home consoles and PCs that yet another doesn't excite me much. So wouldn't it be a better use of that cost to put in a faster CPU and lots of RAM, (NeoGeo emulation), and have 2 SD slots one with SDIO for wireless or other expansion that the user wants?

You 'know' incorrectly; I am interested predominantly in 2D capabilities. 3D is a side benefit, and although it would provide enormous performance gains to games like Quake 1/2 and PSX emulation (I spoke to Zodttd who stated his/U-nai's intentions to utilize a GPU in successive GPH hardware through the OpenGL API) .. it's not my primary concern by any means.

The CPU is significantly faster. Most NeoGeo games run at full speed with the ARM920T at 266 MHz; and at 300 MHz I am unable to find one game that doesn't run at full speed except during scaling operations which are simply slow because of not-fully-optimized code (which could also be accelerated by a GPU, by offloading the tasks from the CPU entirely.)

Also, I did suggest 2 SD card slots. SDIO capability is supported natively on the S3C2460 as well. And 96MB+ of RAM .. Mobile DDR to be specific.

 

[NoidZ]

Option 1: I wanna play Tony Hawk on a handheld. O, and I wanna play Dreamcast games on a handheld too

 

[Epicenter]

You'll never be able to emulate the Dreamcast on that hardware. Or any handheld for a long time. The CPU is much, much too powerful. The rough figures for the 200 MHz SH-4 CPU are 360 MIPS and 1.4 GFLOPS. That overshoots the CPU and AF9's capabilities big-time, especially the floating point performance-- ~300 Million floating point operations per second on the AFP9, compared to *1.4 BILLION* per second on the Dreamcast's SH-4.

That's peak performance too, it's not including the enormous overhead of emulation. Perhaps an ARM11 at near 1 GHz could pull it off. Perhaps. But probably not. Also its graphical capabilities are far in excess of what the 3D accelerator can do; 3-7 Million triangles/sec. versus 1 Million on the S3C2460-- and it probably doesn't support all the same effects.

 

[nubie]

And how is this 266 MHz only device going to be much better for emulation than the GP2X is now? The GP2X can OC even higher many times right now. Yes the GPU even 3D will help a little bit but will it really make all that much of a difference with stuff like SNES or GBA? The PSP with its 3D GPU still struggles at times to emulate SNES full speed at 333 MHz. So either the code is inefficient or the 3D chip doesn't give all that much of a boost.

I know you are looking at it from a mostly 3D homebrew angle rather than emulation but a HUGE portion of the users want emulation as a main use as is is a big niche. Many of us that want 3D will just go buy a DS or PSP for that purpose. Honestly there are so many 3D devices including home consoles and PCs that yet another doesn't excite me much. So wouldn't it be a better use of that cost to put in a faster CPU and lots of RAM, (NeoGeo emulation), and have 2 SD slots one with SDIO for wireless or other expansion that the user wants?

DaveC is wrong, that is too funny. (PS, so, so wrong, very very very very wrong). EDIT:my MK2 wouldn't clock over 240mhz, so this chip IS a huge upgrade in mhz for the stock clock. The additional computational units are very needed, and useful. The difference between floating and fixed math is BIG.

The PSP is NOT OPEN, you can't compare the struggles of people on a closed system to those of a scene full of people on a GPL console .

All of Epicenter's points were valid, in fact I was unaware that option 1 had 2d/3d acceleration by PowerVR, if I could I would likely choose that in the poll. Sorry. I also would like to say that now I have been studying the PowerVR chip in the Dreamcast that I see what you are saying. (does DaveC recall how powerful the Dreamcast was at 2D, and how well it emulates stuff?? This is the same technology for video processing. He should type a formal aplogy.)

I wish you had just pointed me to some tech documents, the nVidia chips look to be brute-forcing stuff like programmable pixel shaders and giga-bandwidth. Whereas the PowerVR architecture is perfect, and so simple and elegant it should do nicely.

Option 1 it is then, any cost on this vs the current SoC numbers??

Option 1: I wanna play Tony Hawk on a handheld. O, and I wanna play Dreamcast games on a handheld too

PSX YES!! Dreamcast NOOO!! (I just spent the last week reading up on the dreamcast, I am so excited to be getting a used bundle on eBay. I have wanted to try one for so long, and I find that I don't game so much on the PC or a handheld, plus it is so much cheaper to buy a Dreamcast and DC games, bleemcast is under $20 btw.)
I thought I would try it out, it turns out that it is better than PS2, I was unaware of the PS2 being crappier, I knew it was crappy, but the DS is so much better . It can do PSX @640x480 and 60fps . WITH Fullscreen AA (check the PS2 for aliasing, I can't even see the next corner in Need for speed, just a damn mass of speckles and crap).

 

[scanti]

I just had a thought. Would it be possible to design the handheld so you can hot swap the batteries. In other words remove the batteries while the console is still on, have the processor go into a low power standby mode - powered by a lithium coin battery or maybe a chunky capacitor (you'll only need a few minutes worth of power) and then restore the system when the new batteries are insterted.

Handy for those marathon gaming sessions.