Please do not use arm cortex-a15 for pandora2!!!!!!!!


EVERY SINGLE NUMBER SHOULD BE HIGHER ON THE P2..

MORE CORES! MORE RAM! MORE GHZ! MORE BUTTONS! MORE SCREEN! MORE MORE!!

SERIOUSLY LOOK AT HOW MANY MORES THIS OTHER THING HAS!!!
 
SNESFAN said:
So after my initial comment about using a big.Little i went looking into it a bit deeper. I have been looking at a mediatek quad core a7 (forget the actual chipset number) but its a 28nm process and it runs at 1ghz on the 4 cores. The benchmarks are actually quite impressive considering the low clock. Some marks that were shown were even able to edge out a galaxy s3. While it may not be competitive with the shield (that battle might not be worth fighting) it would have very adequate power and insane battery life (didnt see battery life specs) The S3 handles most emulators extremely fluidly and even drastic under the weight of android. It might be worth considering. They are even offering the a7 in an octacore variety later this year with either a Mali-400 or powervr 544 @300mhz
Click to expand...
I can only assume that whatever SoC you were looking at was a lot higher clocked than 1GHz if it was competitive with a Galaxy S3, or the benchmarks were strictly GPU related. Or something was wrong with the comparison (different browsers in Javacsript tests, for instance). Most mobile benchmarks suck anyway and should be taken with a ton of salt.

These A7-only SoCs are targeted for really low cost stuff and for something like Pandora that costs several hundred dollars you don't want to skimp a few dollars on a sub-par SoC. If MediaTek were somehow viable (all signs point to no) then it'd better to go with a higher end SoC from them.
 
Last edited by a moderator:
^ All I'm saying is that soc isn't going to bring anything new to the table for me.

Between my CC Pandora and Note 8.0 I'm covered.

P2 needs to emulate Dreamcast, PSP, fluidly and take a concerted stab at GameCube at least on par with the Shield or hopefully better.

I definitely think the Shield should be our benchmark for comparison.

What am I saying? - It's too soon for the P2.
 
@exophase- what I was looking at (found it agian) was the mt6589 and the benchmarks I found were antutu scores in the realm of 12-13k which were in line with the quad core a9 variant of the S3, this was maintained between most of the recent releases of the benchmark suite. But after looking at quadrant scores of 3k it is less than half that of the s3's 8k scores. So something is off for sure. As you mention current benchmarks are wildly deceptive and while trying to get a better image as to how the system performs as a complete system, it leaves lots of room for interpretation and could just be flat out wrong or it's possible to cheat the results. (recent questionable x86 results that imply some insider modifications) also creating low latency loops in I/O to artificially boost scores but doesn't help actual performance. I'm not sure if antutu or quadrant are the ones that are deceptive, but something is definitely fishy there.

So that specific chip might not be the best answer. Especially when you consider their stance on "open soure" is very limited if non-existant. Another manufacturer should probably be looked at. But the point I was making is that a multi-core only a7 might have some advantages in the long run that might be worth considering, not necessarily a specific chip or manufacturer. Being incredibly power and thermally efficient is what the pandora needs to set it apart from it's competition. Not necessarily being top dog, but power enough. If the performance of a7 only chipsets can get on par with what is needed to emulate nds, n64, psp and gamecube at a perfect rate but nothing more, then arbitrarily adding more processing power, unless it can touch the generation of consoles above that would be a waist. That's my rational. Agian, too early to start pointing out any one specific chip. The concept was the point of my post.

@ faeminx & essoair - My comment on the shield is because nvidia is wanting to do yearly generation remakes of the shield alongside it's tegra line just starting with tegra 4. The current shield is doing well based off what they're saying so it wouldn't surprise me at all if we see the next one a year - year and a half from now. So trying to keep up with that device and business plan for this small company with limited budget using 3rd party parts against a company that does all their stuff in house and huge budget is basically futile especially in the long run. Staying on top of the "most powerful gaming handheld" is at this point is going to be a very expensive losing battle. A different strategy is needed. What isn't a lost effort would be "the most open source gaming handheld with outrageous battery life and pc-like capabilities". That has proven to be successful in generating a good following and I feel that's why people love the pandora as much as they do. Just make sure it's power is enough to be relevant to the current needs, more kinda pushes it into another market it doesn't want to be in.

So going off that mindset

the number 1 priority should be finding the manufacturer with the most open source friendly business practices. Rockchip is a member of linaro and might be a good place to start looking instead. Samsung and nvidia have mainlined linux kernel sources as well (not for gpu) I would go as far as to say, screw the rest of the priorities though as this point is really all that matters and would be worth just saying forget performance as long as all the hardware source is open. Whatever you chose is going to be worked on and maintained for a long time and that's only possible if all the source is available or is possible to be reverse engineered. Closed source power machine becomes a lot less relevant years from now than a lower powered open source one.

Number 2 priority should be power efficiency, the cortex a7 architecture is built around being extremely power efficient so you will get some insane battery life if that's all you're using. It's downfall is obviously overall performance, but if that can hit the marks you need, then I strongly feel that's the best option for the goals of the pandora. If not a7 only chipsets because of performance, then the big.little a7/a15 is the next best thing. I just grossly underestimated how powerful those a7's were until I saw some numbers. I only saw it as an low power idle chip but it can actually do some crunching.

Number 3 priority should be availability is small numbers. Some of these Chinese manufactures are more likely to work with ED in the quantities he wants than companies like nvidia, samsung, or qualcomm. Going with these less than mainstream manufacturers is going to take you away from top of the line unlike the big names. If you do go with a big company, be prepared to lock yourself into one specfic chipset for the foreseeable future for shere quantity of parts you will need to order, moreso than the number of omaps that were purchased from TI. And that's only if their minimum order isn't so large it's too cost prohibitive to even consider. That's going to prevent any kind of rapid successive iterations to stay on top of what's out there now, which would be a dumb move to get into anyway. So that would mean there's no chance of a P3 even if in the far future, which might not be a bad thing as you will need exponentially more power to hit the emulation requirements of the consoles above a gamecube or psp. Thinking maybe 5+ years or 3 arm generations from now?

TLDR; Whatever the choice is, there's some guarantees. 1)it's going to be for the long run 2)it will quickly get outdated 3)it's going to be more expensive than it's competition. So get exactly what you need for the long lifespan of the device. Don't compromise on power/perf or versatility hitting a number that won't make a difference in a year from now. Think about what are be biggest selling points of what you got now is.
 
Last edited by a moderator:
^ I agree, though as has been mentioned... I would definitely consider a big.little soc over a pure a7 soc (which I feel brings nothing significantly new to the table that we can't pretty much achieve already).

You are right - battery life and openness have to be the two top priorities.

Though the trend may become: Shield version whatever for current gaming grunt, and Pandora version whatever for your all in one mobile computing solution (and personal collection of retro favorites).
 
SNESFAN said:
TLDR; Whatever the choice is, there's some guarantees. 1)it's going to be for the long run 2)it will quickly get outdated 3)it's going to be more expensive than it's competition. So get exactly what you need for the long lifespan of the device. Don't compromise on power/perf or versatility hitting a number that won't make a difference in a year from now. Think about what are be biggest selling points of what you got now is.
Click to expand...
Are there removable/socketable mobile ARM chipsets out there? Like PC Processors? Would be a great thing to continuously update the hardware that way. :) Make it a Mobile "Lego" Handheld! ^^
 
SNESFAN, don't use AnTuTu for comparison of anything. Ever. Please. Even if it were anywhere close to a meaningful or legitimate benchmark (it isn't) the aggregate score attempts to measure a whole bunch of stuff outside of CPU and even GPU performance.

In normal code 1GHz A7s won't come anywhere close to 1.4GHz A9s. I guarantee you that you'll notice the difference substantially in something like DraStic which you listed before. That chip would be a really terrible choice. Even MediaTek offers A9s, A15/A7s, and much higher clocked A7 SoCs. And an A7-only SoC would still be highly limited even at 2GHz, you'll be lucky if it performs like 1.5GHz A9s, and there's no way you'll ever get good Gamecube emulation on that.
 
Last edited by a moderator:
Ok, I can appreciate that, it was just a thought. If it has no possibility of doing any real grunt work then it shouldn't be considered. The only thing I don't particularly care for on the hybrid architectures is there is no way that I'm aware of to control when cores turn off and on. Like say you are fine with a little latency when browsing to save juice, but if the chipset sees a load at a certain level it will turn on higher energy consuming cores to handle the task. I've even read you can't even fool it with a governor at a system level. It would be cool if you could route specific tasks to run through certain cores like application profiles. Or have some sort of modern day "turbo" button like they once had on old PCs.
 
^ I'm not sure if it's legit yet, but apparently someone has released a mod for the Note 8.0 that does something similar. Set how many cores and at what Mhz per application.

And this isn't even a big.little soc... so I'm pretty sure that's it's doable.
 
Hopefully affinity works on these hybrid architectures as well, I don't know enough about it to say either way. From the things I've read the handover from an a7 core to an a15 core is handled completely transparent in hardware. But if this is something that can be manipulated then that would be awesome and would certainly have its benefits.

Dated post but at least it was addressed, don't know if it was ever sorted
http://lwn.net/Articles/481055/
 
Last edited by a moderator:
Hmmm...

One thing came to mind:

Did the Pandora ever benefit practically from choosing OMAP as a basis for the design from the fact that TI offers a "complete package", or "set" of chips?

I mean, sure, you can gobble up chips from various manufacturers, but was there any direct benefit from design, sourcing, etc. point of view for using specifically TI products?

Looking quickly from the hacker's guide, aren't there at least 3 other chips from TI there? Integration benefits or simply the best stuff that happened to be available were from TI at the time?

I got the impression from older threads that TI did help in some things... unfortunately including finding CC to work with (but lord only knows what exactly happened there and he ain't telling...).

For a small project like the Pandora things like longevity, I guess, matter quite a bit. Revisions to the MB can be done, sure, but it is nasty if you somehow end up needing them more often than convenient...

...though I'm not even sure myself what this had to do with anything :blink: .

Well... I'm not working in electronics, you can take this as a random clueless hobbyist thinking random clueless thoughts ;) .
 
You must log in or register to reply here.

Similar threads

_wb_
Overview of SoC options
16 17 18
Replies
348
Views
40K
urisma
U
Mr_Loon
What would the world and his dog think if P2 used an x86 SOC? (but not mini SD) ;)
9 10 11
Replies
200
Views
17K
ingoreis
ingoreis
D
I.mx51 Arm Cortex Soc
Replies
3
Views
4K
Exophase
E
Orz
My Thoughts on the Pyra
3 4 5
Replies
91
Views
11K
levi
levi
C
Another Linux on ARM user (not OpenPandora)
Replies
19
Views
4K
notaz
notaz
Back
Top