News From Gamepark Re: Gpkids/mini Development

[Epicenter]

Allow me to quote an e-mail I recently recieved from GAMEPARK:

Among the 3 systems, GPKiDS will be the first to be released. Whether GPKiDS will be released under GAMEPARK’s brand or under other firms’ brand has not yet decided. Our rep said at E3, we planned to launch GPKiDS three months ago however; we have been encountered several problems with GPKiDS and thought we should fix the problems right before the launch. This is why it is taking longer than we had first anticipated. The release date for GPKiDS cannot be said right now. We have not yet fixed the problems and we are unsure how long it would take to get things right. In addition, we don’t want to make the same mistakes we have made at E3 and other occasions. However, in my opinion and mine only, GPKiDS will be launched no later than the end of this year.

Now for XGP and XGPmini, it has been over a year since we thought out and designed the two systems. Because it has been over a year, using what we said we were going to use, in terms of specification, will put our systems between no where. They will no longer have an edge (technology and price wise) over other portable systems. So we are planning to make some changes so that our systems have up-to-date technologies and still have price competitiveness over other handhelds. So far we have finished with sculpturing, if I were to say in figure of speech. All we need to do is adding more details to our systems so that there won’t be major faults with the systems. Right now we are concentrating on finalizing GPKiDS and dirstribution of GPKiDS. Once we have found strong ground for GPKiDS, we will move on to XGPmini. If I may ask a quick question to you, as a developer what would you think of a system which uses Samsung S3C 2410 chipset and separate 3D graphic accelerator? Would you still prefer a system which uses S3C 2460? Lastly, the release of GPKiDS in October is totally incorrect information.

I'm first of all glad to see GP is taking the required time to iron out issues before putting out their systems, I can forgive the delay since they intend to compensate for it with increased hardware specifications. This is an interesting question posed, however. Let's take a look at the features of the SoCs in question.

S3C2460 Main Features:
- 200/266 MHz ARM926EJ-S Microprocessor w/ 16KB Instruction/Data caches, Jazelle Java acceleration
- VFP9 ARM-designed Vector/Floating Point Unit
- 100 MHz TEAK 16-bit DSP
- MobileDDR/MobileSDRAM Controller
- Onboard PowerVR MBX RS 2D/3D Graphics Accelerator (1.0M Polys/sec+)
- Hardware MPEG4 Decoding/Encoding support
- Hardware JPEG Decoding/Encoding support

S3C2410 Main Features:
- 200/266 MHz ARM920T Microprocessor w/ 16KB Instruction/Data caches, no Java acceleration
- No VFPU, No DSP
- SDRAM/MobileSDRAM Supported, MobileDDR not supported
- No onboard 2D/3D graphics acceleration
- Hardware MPEG4 Decoding/Encoding support
- No JPEG Decoding/Encoding support

Also in the fray is the S3C2413, which adds MobileDDR support to the S3C2410's features, but does not support MPEG4 decoding in hardware.

A quick breakdown of features..
- The VFPU (AFP9) will accelerate any non-integer math, particularly useful in 3D or 2D applications featuring complex object movement. It will also act like a secondary processor when Integer and Floating-Point code is used equitably, providing a large speed improvement without special code to operate it, making it painless to use, unlike say, the ARM940T in the GP2X.
- MPEG4 decoding support will speed up operations involving playback of video. Without it, video playback will rely on the CPU and might not be quite as snappy. It is my belief 320x240 video playback will work fine with no hardware acceleration, but playing video at other resolutions would be ill-advised. JPEG decoding support accelerates the loading of JPEG images, useful for web browsing and textures in 3D games.
- MobileDDR memory support will considerably accelerate memory access, especially useful in situations where multiple components might compete for memory. MobileDDR memory may also improve battery life. MobileSDRAM uses less power than SDRAM, but is not as fast as MobileDDR.

So, this raises some interesting issues. Stay with the S3C2460 which may be more expensive? Or settle for a simpler chip, and add an accelerator? To keep floating point capability, a coprocessor must be added. I have not found separate chips that are based on the VFP9 technology by ARM, but I suspect that they do in fact exist. Samsung makes a separate MPEG4 decoding unit called SA3A480.

An issue is also raised about if the XGP/XGP Mini should use different 2D/3D accelerators. For the record, my recommendation to Gamepark was the S3C2460 ideally, and if that was too expensive, a cheaper option might be the S3C24A0, coupled with a separate VFP9-based coprocessor, Samsung SA3A480 for MPEG4, and a weaker 2D/3D accelerator than the XGP, but still supporting the OpenGL-ES standard API.

Just thought I'd share my news, and see if anyone else has their own views on the matter they'd like to share.

 

[sehs33]

Thanks for sharing the news Epicenter, several topics and polls about what do we/GamePark think to be the optimal solution are around now (of these many were started by you ) so I will try giving my opinion about this matter briefly:
I believe they are doing it the wrong way, they shuldn't be trying to find the optimal hardware and then price it, they should decide a price quote and start picking components that will keep them in that price's range, they have three handhelds, each for a segment, and all what will stop a lower segment from buying the higher segment's handheld is price, so they should set their prices, and see what hardware they will get for those prices.
Last thing, they should also take in consideration other handheld prices.

 

[Vimacs]

I realy think that a v/fpu is essential for the next generation of handheld if you dont want to drasticaly up the mhz and still get a good performance.

I am quite sure that a 200mhz chip with v/fpu can outperform a 250-300mhz chip in many cases and still use less power.

The S3C2460 seems like a very good choice to me,
while the S3C2410 would be pretty senseless in a new handheld as it would probably be even slower than the mmsp2 and good video playback would be hard.

 

[Aninhumer]

I think the XGP should use the S3C2460, I don't think losing a VFPU and Graphics accellerator would be worth the saving. (Although I would like to hear price estimates)

The mini wouldn't need something as powerful as the S3C2460, but I still think the S3C2410 is lacking.
And if the S3C2460 isn't too expensive, and there isn't a "middle ground", it would still be a good choice.


Be aware, I'm not exactly a 1337 hardware expert, I'm just basing my opinion on what I know, but I think I know enough...

 

[yaustar]

I would also vote in for the S3C2460 or something similar. As long as it has a FPU and a half decent graphics accelerator, I be happy.

 

[shinneri]

So the release of the other two relies on the success of the GPKiDS? That's troubling...

 

[Epicenter]

So the release of the other two relies on the success of the GPKiDS? That's troubling...

From how I've heard the Korean games market described, it sounds like they might sell a decent number of these things if the games are alright. From what I've heard, hardcore gaming isn't all that popular in Korea. ED even described to me that only 1,000 of approximately 26,000 GP2X sales were in Korea, and those users only want to use the machine for PMP functions. GamePark Holdings barely understood that anyone played games on the GP2X until a few months ago and could barely understand the concept of what we DO with the unit if not use it for multimedia. In such an environment where cell phone games seem to make up the majority of mainstream gaming, the GPKiDS may not be such a disaster after all, though I can only see it having limited appeal to the sort of users and developers who frequent GP32X.

I believe they are doing it the wrong way, they shuldn't be trying to find the optimal hardware and then price it, they should decide a price quote and start picking components that will keep them in that price's range, they have three handhelds, each for a segment, and all what will stop a lower segment from buying the higher segment's handheld is price, so they should set their prices, and see what hardware they will get for those prices.

That sounds a lot like what they are doing. They set a desired price point of $150 for the XGPmini, and $300 for XGP. Those would be final purchase-time consumer prices. I'll cite the example of GP stating they wish to replace the DMB television reciever functionality with GPS or another feature in the units not sold to nations that support the DMB broadcast standard, to make the $300 price point justified.

I'd like to entertain a hypothetical discussion topic-- let's say the S3C2460 is too expensive to use in the XGP Mini, and another chipset must be used instead. Given Samsung's SoC lineup, and known coprocessors, what would be the most ideal combination? These are Samsung's offered SoC's:

S3C2410 - ARM920T, 200/266MHz, MMU, NAND Flash Boot Loader, ROM/SRAM/SDRAM Control, STN/TFT LCD Control, Touch Panel Control, MMC/SD Card, USB, 10bit ADC
S3C2412 - ARM 926EJ, 200/266MHz S3C2410's feature + mSDRAM
S3C2413 - ARM 926EJ, 266MHz S3C2410's feature + mSDRAM, mDDR, OneNAND + Camera Interface
S3C2440 - ARM920T, 300/400MHz, 2410's Feature + Camera Interface
S3C2443 - ARM920T, 400/533 MHz, 2440 Feature + USB2.0, CF•ATA I/F, HS MMC, SPI
S3C2460 - ARM 926EJ, 24A0's Feature + Teak DSP + 3D(1M) + USB OTG
S3C24A0 - ARM 926EJ, 200/266MHz, 2410's Feature+MPEG4 + Camera Interface
SC32442 - ARM920T, S3C2442 (S3C2440's Feature) + MCP (Multi Chip Package)

(From http://www.samsung.com/Products/Semiconduc...mily_cd=MSC0102, where you can find more details about the individual chips.) Not mentioned in the description of the S3C2460 is its integrated V/FPU coprocessor, and some memory controller features.

My ideal combination would be the S3C24A0, a VFP9-based coprocessor for vector/floating point functionality, Samsung SA3A480 coprocessor for MPEG-4 decoding, and a separate 2D/3D accelerator, perhaps less powerful than the PowerVR MBX Lite, but powerful enough for high-end 2D Operations and acceptable 3D performance. I'd shoot for something with no less than 300,000 polygons/sec rendering capability, and of course, full OpenGL-ES compatibility.

 

[aapje89]

sounds cool

I would vote for the S3C2460, but not that I know anything about it...

 

[Aninhumer]

I think those descriptions are useless, they're describing "packages", only some of which would be sensible in the XGP.

I would have thought a complete SoC would be cheaper, and better designed, than using seperate components. But I trust Epicenter's judgement if he thinks there's a better way.

 

[Epicenter]

Aninhumer: They aren't packages, like kits-- for example S3C2460 is one chip containing an ARM926EJ-S processor, 2D/3D accelerator, VFP9, MPEG-4 decoder... so on and so forth. Coprocessors are attached externally as an option.

 

[icurafu]

If they use the Samsung S3C 2410, then I wonder what the seperate 3d accelerator will be?

Integrated is generally cheaper and more energy efficient. With 266 ARM9s, I don't think they need more power then what they already got with the PowerVRs.

 

[Alex.]

GamePark Holdings barely understood that anyone played games on the GP2X until a few months ago and could barely understand the concept of what we DO with the unit if not use it for multimedia.

Then how come the Korean GP2X site, more than a year ago, boasted emulation, interpreters, and homebrew game programming as the GP2X's main features?

- Alex

 

[JaqMs]

GamePark Holdings barely understood that anyone played games on the GP2X until a few months ago and could barely understand the concept of what we DO with the unit if not use it for multimedia.

Then how come the Korean GP2X site, more than a year ago, boasted emulation, interpreters, and homebrew game programming as the GP2X's main features?

- Alex

Yeah I remember that. They said things like PSX, GBA, and Dosbox were in the works. And for original games, they mentioned things like "Naruto 3D" "Wipeout 3D clone" and "Gunbound clone". Oh, and that was in September of LAST YEAR. IIRC, the first GP2X's came out around December/January.

 

[Shikaku]

I'd say allow Java acceleration, for the XGP at least. That way I won't have to learn C/C++ and still make good games =p If you all want the XGP to replace the GP2X and be better, I say crush the GP2X's specs flat completely. FPU, 640x480 screen, 266 MHz ARM with good overclock, Java acceleration, and 4/6 AAs either as required or OPTIONAL. Hell, tell GP that they should also sell a AA battery pack holder or an li-ion battery seperately that will fit the XGP, meaning you don't have to buy the li-ion battery, but instead can opt for the AA holder pack. This can be only for the international sales obviously. Give us the option I say....

Don't think I am against my own unit, I just have the mindset that the XGP won't be out until 2008 at the least from the posts by Epicenter and JaqMs (if they are true).

 

[Alex.]

My friend, software is what makes or breaks a system. The GP2X has plenty, and will grow exponentially. Don't believe me? Just wait and see what happens after the community contest is over and people will start releasing their games more freely, more often. The GP2X will also likely be cheaper than ther XGP (hell, it may even decrease in price).

- Alex

 

[shinneri]

GamePark Holdings barely understood that anyone played games on the GP2X until a few months ago and could barely understand the concept of what we DO with the unit if not use it for multimedia.

Then how come the Korean GP2X site, more than a year ago, boasted emulation, interpreters, and homebrew game programming as the GP2X's main features?

- Alex

Yeah I remember that. They said things like PSX, GBA, and Dosbox were in the works. And for original games, they mentioned things like "Naruto 3D" "Wipeout 3D clone" and "Gunbound clone". Oh, and that was in September of LAST YEAR. IIRC, the first GP2X's came out around December/January.

GP2X came out in November, but you're right. They said:

1. Game SDK with GCC 4.x
2. Genesis/Megadrive/MegaCD
3. Sound core, Doom porting, SNES
4. Neo-Geo
5. Quake engine porting, 3D work
6. NES, SMS, MSX, GBC, GB
7. Others: Utility development for ARM
8. 3D graphic library, 3D engine and examples
9. Naruto 3D porting, New 3D racing game(like wipeout), Gunbound clone
10. MAME, PSX, Dos Box(PC emulator with MS-DOS)
11. Build SDL/Lib C cross compiling toolchain, VBA32(GBA emulater)

So clearly they knew the GP2X was going to be used to a great extent for gaming.

 

[JaqMs]

But why'd they have to lie...

 

[shinneri]

But why'd they have to lie...

It was probably homebrew. They can't force a homebrew developer to complete a project.

 

[Aninhumer]

Aninhumer: They aren't packages, like kits-- for example S3C2460 is one chip containing an ARM926EJ-S processor, 2D/3D accelerator, VFP9, MPEG-4 decoder... so on and so forth. Coprocessors are attached externally as an option.

I got confused by the packages numbers, and what seemed to be a disturbing lack of features...

I just saw that it says "+2410's Features", it makes sense now.

EDIT: It would be much easier to compare if we had price estimates.

 

[Epicenter]

Then how come the Korean GP2X site, more than a year ago, boasted emulation, interpreters, and homebrew game programming as the GP2X's main features?

My knowlege that GPH is largely clueless about how much gaming actually OCCURS on GP2X handhelds is right from EvilDragon, if you do not believe me. With his clearance, I'll give you a chatlog of that discussion. They were probably given suggestions on how best to market it to the GP32 crowd before release from people like Craig/ED though. Clearly a lot of them were wishful thinking.

My friend, software is what makes or breaks a system. The GP2X has plenty, and will grow exponentially. Don't believe me? Just wait and see what happens after the community contest is over and people will start releasing their games more freely, more often. The GP2X will also likely be cheaper than ther XGP (hell, it may even decrease in price).

All GP2X software should run fine on XGP machines (with the exception of the GPkids, for fairly obvious CPU speed, RAM, resolution and OS reasons). They use compatible processors, compatible resolutions, will have sufficient RAM and graphical capability, and will apparently be using the same OS. Software won't be a problem. I do not feel we need to repeat this exact same argument from General GP2X Discussion where you extolled the GP2X's praises ad nauseum, in much DaveC-esque fashion, at me at such a rate it must have made breathing difficult.