Theoretical gaming/emulation performance level..?

[fusion_power]

Please stop going on about support and documentation, this problem with Pandora 1 not having enough documentation is in your mind, I don't know how many times I need to tell you this >_< No alternative solution is going to have open GPU specifications, none of them are going to come with OpenGL drivers, etc..

I Can't remember you've told me anything but to be honest I usualy can bareley remember what I've wrote yesterday, sorry. So yes, don't waste your time to explain, I will forget it tomorrow anyways. :lol: But I'm pretty sure I did not invent "this" I just repeat what was sayed so many times (not by me) since the Pandora was born.


So if everything HW/SW releated is great and fine and perfect for our beloved Programmers, then it's ok, I will not complain. I just will not pay any cent for Hardware that can't be used in it's full potential. For P2 I don't want to hear stuff again like "there is an DSP but we can't use it because of blahblah" or whatever.


Better use a more simple Chipset that can be easy coded for, instead of a (maybe) bloated 2GHz Monster where not even the Chip Manufacturers itself know what it can do and what not.


Thats basicly all of my concerns: paying alot for Expencive Hardware that is not worth the high price because nobody can pull out all the power...and we know the future P2 Hardware WILL be expencive.

 

[mcobit]

The dsp, that is in the Pandora is actually utilised in other products with this soc. The Problem with the Pandora was, that nobody knew how to get it to do something and didn't care to write specific stuff for it. Also the integration into the firmware was a problem.


It was not TIs fault, that it doesn't get used, but the problem that noone knew how to use it properly afaik.

 

[fusion_power]

Yes, that is actualy the point. There should be at least enough documentation for a propper usage of a DSP. I can't see where actualy the problem is, since many Devices use it without problems, so it can't be that hard imho. And if you don't know how to use it, then I would just kindly ask someone that does know about, this usualy works pretty well. As far as I know, the OMAP in the Pandora is pretty standard and used in many other devices, so I'm sure there are alot "professionals" that could help if here and there are problems.


I know that we now could use the DSP for example, at the cost of 32MB RAM but It was just an example what could be improved in the future, maybe. Remember, the HW is expencive and it would somehow feel be cheated if you only get half of the available power out of it.


Again, this is just my personal opinion and I don't want to complain or insult someone. I just want, that things like this (and other things) are kept in mind if there is a future Pandora 2 development some day. Better be prepared for everything that could go wrong than having bad surprises at the end.

 

[craigix]

Even a single Cortex-A15 core at 2GHz, on a 28nm SoC, will use much more power than a Cortex-A8 at 1GHz on a 45nm SoC (like DM3730). Never mind a second core. So unless Pandora 2 has a much higher capacity battery or the rest of the system uses a lot less power (doubt that a much higher resolution LCD and a bunch of faster RAM over multiple channels will facilitate this, not to mention a much stronger GPU) I don't see how the battery life at full utilization can be the same as the battery life at full utilization of Pandora 1. But this only matters for something that actually would put it at full utilization. For things that run full speed on both Pandora 2 could be capable of giving similar or better battery life.

We have designed around the battery maybe being up to 6200mah (the whole base of the lower part). We might not actually need that as really I feel the 10 hour point is the main goal.

 

[Binky]

^ I think the way battery life scale with use is important - how long will it last on standby? Playing music? etc - I'm sure this is largely a software problem, but even an avid gamer doesn't use their console to its maximum potential 24 -7 -52


This also relates to DSPs & the like - they may save a lot of electrical power if used correctly.


Would it be a good plan to find some vaguely similar open project and use similar hardware in the hope of sharing software (like the N900 and P1) ?


Or does the Linux - Android merger make this unimportant?

 

[Foxgod]

Would Tegra offer any advantage, considering it got a gpu on board that supports OpenGl and DX?


Right now i see many devs struggling with porting opengl instructions to GLES, wouldnt it bring benefit to gaming to simply have a SOC/gpu that supports OpenGL?


I know Tegra is not the fastest soc, but what is the point of the fastest soc if we are bottlenecked by poor 3d performance and OpenGL > GLES difficulties?

 

[fusion_power]

NVidia always have solid driver support so this could be an benefit. Not sure how much Power a Tegra (3) need but it could be more than OMAP Chips. Tegra uses Dual/Quad Core afaik, sounds pretty power consumpting.


On paper, everything sounds great (of course):


http://www.nvidia.com/object/tegra-3-processor.html

 

[SNESFAN]

Tegra 3 might be old news but think about this... partner with ouya? Have the same tegra chip set in both the Pandora and ouya?


The bridge is probally already burnt, but their library on the Pandora, all software Linux and android on both systems? A ouya in your pocket?


Dumb idea maybe... especially if you consider ouya a possible competitor... but joining the two would be awesome for the consumer IMO

 

[Exophase]

Yeah you know, for all those OUYA exclusives that will never get ported to mainline Android because of.. uh..


Tegra 3 in Pandora 2 would be ridiculous. I think a $99 console coming out spring 2013 has different hardware goals than a $699 mini-computer or whatever coming out in late 2013/early 2014.

 

[_wb_]

Tegra 4 maybe.


I don't really see the advantage of 4 cores over 2 cores (e.g. OMAP5) for something like the Pandora - especially if it's also gonna have 24 GPU cores that can be used for number crunching. OpenGL support is nice, but doesn't that add some kind of overhead? (why was GL|ES introduced in the first place?)


Does anyone have any clue on the availability and expected price of a Tegra 4 and an OMAP5 in low quantities? Are there power consumption specs/benchmarks available somewhere?

 

[Exophase]

No one knows much of anything about Tegra 4, much less a release date, or what the availability will be like to a customer like OPT. There's also no indication that OUYA will be using a Tegra 4 any time soon, if ever. Of course there aren't going to be benchmarks, not even of OMAP5, or at least not trustworthy ones.


OpenGL ES was introduced because it offers an easier spec to meet, which is important for low cost mobile SoCs that have special energy and area requirements. And it's also somewhat more modern/revised to remove some of the old cruft that no one used. I don't think it's really about being faster per se, where the hardware supports it. The key point is even if something like Tegra 4 supports it there still needs to be drivers, which can be a serious roadblock - why make full OpenGL drivers for Android when no one will want to support it because the other devices don't have it?


And "24 GPU cores"?? Geez, what are you reading? OMAP5 has two SGX544 cores which each have 4 USSE2 pipelines that you could I suppose call a core in some sort of sense. But there isn't even OpenCL drivers out for it and I wouldn't hold my breath for such a thing.

 

[Linux-SWAT]

A benchmark from maybe 4/5 years ago showed that PC games benefits from 3 cores, not more.


Of course this may have changed.

 

[_wb_]

This is what Wikipedia has to say about Tegra 4, I don't know how to decode some of that marketese:

Wayne

• Processor: Quad ARM Cortex-A15 MPCore + low power companion core
  
• Improved 24 (for the quad-core) and 32 to 64 (for the octa-core) GPU cores with support for DirectX 11+, OpenGL 4.X, and PhysX
  
• 28 nm HPL[33]
  
• About 10 times faster than Tegra 2
  
• To be released in Q1 2013
  

 

[Exophase]

Wikipedia, by itself, is not a source, it's what some random person thought is true (or just made up), and when it comes to nodes about hardware it's wrong all the time. And I shouldn't have to say this either, but it's not enough to have citations - you have to actually read the citations and confirm that they actually say what the source says they say. But you'll note that most of those claims don't even have citations.


As for 24 GPU cores, it looked like you were talking about OMAP5, not Tegra 3 or Tegra 4. When it comes to Tegra nVidia's usage of the word "core" is highly questionable.

 

[Linux-SWAT]

Does someone knows a good way to benchmark a phone cpu ?


I'd say some make -jn, but i'm not sure all phones are free enough to do it...

 

[Exophase]

Does someone knows a good way to benchmark a phone cpu ?


I'd say some make -jn, but i'm not sure all phones are free enough to do it...

The state of benchmarking on phone hardware is abysmal, the stuff reviews run is a joke


IMO - and bear in mind, I'm pretty damn biased, so take this with some salt (and if someone has a rebuttal I'd love to hear it) - emulators would be one option for performance testing. They're highly available on the platforms and fairly highly used, they have a fair amount of code coverage (doing whatever stuff the game does, doing branchy interpretation sort of stuff if not a recompiler, graphics rendering, etc), they tend to be pretty low level and written in low-ish level languages avoiding the massive domain-specific overhead of crap like Javascript, they're fairly easy to reproduce results in using savestates, and they often feature a fair amount of well optimized code so they're not always just a test of how fast poorly written code runs.


All they really need to make this work is a fast forward capability that runs it w/o throttling, unless the emulator can't run full speed to begin with. A frame rate counter is preferable for comparing performance but barring that you can just time how long it takes to get to some event after loading a savestate.


Of course this would just be one thing you can use to try to characterize performance but it could give you at least one solid set of data.. too bad no reviewers try something like this..

 

[Linux-SWAT]

Interesting.


So now, is there any benchmarking rom ? What system will be best ? N64, Psx, whatever ?

 

[Exophase]

I don't think a benchmark ROM is really necessary or even a good idea, just pick a bunch of games. As for platform, while N64 is good the emulators all use hardware for graphics so you should do some other stuff too. Even something like an NES emulator is good if there's a fast forward option.