Raspberry Pi 2 unveiled with faster processor and more memory

[Exophase]

As for floating point passing, I don't do any in my code but the compiler (FPC) does for things like maths and trig etc. There's nothing I can do about that. As a result, the Pi clocked lower than my Pandora runs PandaBAS much, much faster - sometimes up to 50% faster depending on what the user's doing in the BASIC code.

This is probably as much or more due to VFP being a lot faster on ARM11 than it is on Cortex-A8 due to the latter having an unpipelined implementation. This is especially true if it uses 64-bit types.

Complex functions like trig, while still losing a significant number of cycles in the softfp passing, will spend comparatively more time on the function itself. And if the compiler is calling functions for simple operations like floating point addition or multiplication then it's doing a very, very bad job.

 

[Green7]

RS Components have conned me (seemingly not an unusual occurrence).


I placed my order which they said would be next day delivery as they had units in stock only to then be sent an email saying they had no stock and would inform me when they got some.


Same thing happened to some people on the Raspi forums.

 

[pmprog]

That's common, companies seem to have this need to pretend they have stock when they really don't. One of my work projects required 12 16x2 LCD displays and a Raspberry Pi. The site we ordered from said they had 500+ LCDs, but 9 came and they said the rest were on back order as they had none left. Somehow I don't think there were 500 more units ordered in that day

 

[Exophase]

Lordus recently got both an RPi 2 and an Odroid C1. He did some benchmark tests with DraStic, also comparing it with Pandora. The test was for 300 frames of Mario's head in SM64:
 


1 thread:                          C1 / Rpi2    / Pandora
Full run time:                20.96ms / 29.85ms / 22.53ms
Video 2D run time:             2.03ms /  2.68ms /  2.93ms
Video 3D run time:            13.73ms / 18.74ms / 11.53ms
Video Geometry run time:       1.55ms /  2.70ms /  2.60ms
Screen Update run time:        0.00ms /  0.00ms /  0.24ms
Audio run time:                0.43ms /  0.76ms /  0.69ms


Threading in the 3D renderer was turned off, to get an idea of single threaded performance. The 2D, on the other hand, was still rendered using two threads. Everything was at native clock speed, meaning 1.5GHz for C1, 900MHz for RPi 2, and 1GHz for Pandora. Note that the "screen update" time is 0 for the non-Pandora platforms because on other Linux targets SDL is used to render the screen and it can be very slow and erratic. So it's just disabled altogether for the benchmark. This means actual performance will be worse.

Using 3 threads for 3D increases its performance by nearly 3x for the C1, but only somewhat over 2x for the RPi 2. I only have single threaded 2D results for the C1, where it's about 1.5x slower.

What this shows is that Cortex-A7 is actually much worse at NEON than Pandora (the 3D rendering is very NEON heavy), and the A5 is about the same. At least for this test, which mostly uses integer NEON code. Normalized for clock speed, both are about 1.8x slower than the Pandora at 3D rendering. This leads me to believe that A7 also lacks 128-bit integer support. But, threading the rendering across 3 cores more than makes up for it (in my tests I've found on pretty much all platforms that performance doesn't improve or gets worse after three threads, not totally sure why)

If you subtract everything else from the full run time, you get kind of an idea of how long the CPU, event management, DMA, and so on parts of the emulation takes. This is 3.22ms for the C1, 4.97ms for the RPi 2, and 4.54ms for the Pandora. If you normalize for clock speed, you get 4.83, 4.473 and 4.54 - numbers that are all surprisingly close to each other. Not really sure what to make of this yet. May be better to get some data on some more CPU emulation heavy scenes.

 

[Grench]

Would it be possible for those benchmarks to be run on a Pyra dev board too?

 

[Exophase]

Would it be possible for those benchmarks to be run on a Pyra dev board too?

Neither of us have one, so no, unless notaz wants to. They'd probably be pretty similar to results on my Exynos 5250 Chromebook, although I don't really want to dig that out now either.

The motivation for this was to give some data point for how C1 and RPi 2 compare, because other than CPU type they're very similar. After some more tests I'm pretty confident that the C1 will be faster almost all of the time, contrary to what that blog post suggested.

 

[Exophase]

Okay, Lordus was nice enough to do some tests with his Odroid XU configured to two threads only and Cortex-A7s disabled to best simulate Pyra. It runs at 1.66GHz, while AFAIK Pyra will be able to clock up to 1.7GHz:

http://pastebin.com/KzU7ZBpb

Scene from Golden Sun was at its slowest point in the intro, where it bogs down a lot. Alice in Wonderland is I think the most demanding game I've seen, for some reason (although there might be some others that are pathological exceptions).

Note that real performance is lower on all but Pandora because the screen will actually be rendered. But, Pyra should have very low overhead for this if I can access the display controller the same way I do on Pandora, probably in the noise.

Takeaways:

- C1 is much faster than RPi 2 on every test

- RPi 2 still outperforms Pandora with threading enabled in 2/3 of the tests. But only because of the threading on 2D + 3D, meaning that Pandora will have an easier time reaching real time speed w/frameskip

- XU and by proxy Pyra should have no problem with almost any DS games, although I doubt anyone thought it would.

- For this particular application (possibly NOT representative of others, so take with salt) the Pyra-like XU configuration is consistently ~2.7x faster in single thread performance, which is also exactly what I remember seeing on the Chromebook. Using the second core to speed up rendering raises the improvement to 3.1-3.9x. I think you'll see most other emulators won't get such a big improvement from the second core, though.

 

[TrashyMG]

while AFAIK Pyra will be able to clock up to 1.7GHz:

While right now I can't clock higher than 1.5Ghz, I've seen several claims that it may be able to clock up to 2Ghz... however I seen the 1.7Ghz claim as well...  I suppose we won't know until overclocking is unlocked. 

http://www.ti.com/devnet/docs/catalog/endequipmentproductfolder.tsp?actionPerformed=productFolder&productId=18100

 

[Ian J]

Which of the above devices Odroid C1 or RPi2 would users recommend as a media player and for emulation of the following:

Amiga

ScummVM

Commodore 64

PC Engine

MAME or similar

Playstation 1

I was looking at getting a 5v 2a output mains adaptor would this be enough to power either device and a Western Digital My Passport drive?

I don't really fancy getting a powered USB hub just to connect a hard drive.

 

[Exophase]

Well, according to Lordus plus other posts, C1 has serious problems with video performance. To the point where if your output is 1080p forget about even getting more than 40Hz. If it's 720p it's possible, though. Since it usually has a core to soak the overhead this wouldn't normally affect emulator performance outside of effectively skipping frames.

And he linked me to this:

http://forum.odroid.com/viewtopic.php?f=116&t=7852#p61066

He also says RPi is smoother because of working vsync.. but it has its own problems with timing resolution.

 

[Ian J]

balls

Excellent link though, thank you for the info.

So it looks like a RPi2 then.

 

[TDT]

Just a warning to those considering buying one...they're kinda backordered right now.

I tried my hardest to buy 2 of them as soon as they were announced, and was successful within 24 hours of launch.  Unfortunately, I already hit backorder status.

So yeah, I don't know when mine is coming, but I'm looking forward to it.  Should replace my current raspberry pi home server.

 

[gadgetoid]

This makes it even more obvious to me that RPi 2 is a conscious response to Odroid C1. 

You're missing a piece of the puzzle there. The chip on the RPi 2 is custom silicon, they didn't just cough it up overnight, it was in development long before the Odroid C1 existed.

Pi isn't under any threat from Odroid, or indeed anyone at the moment. And I don't think they're likely to be for a long time. They're surely approaching 5 million units sold now, and that's a hard ecosystem to usurp!

I'm actually interested in testing some of our products on the Odroid C1 and building up a decent compatibility list- it's good to have higher level hobbyists on board who can come up with good projects and ideas we can backport for Pi users.

 

[Exophase]

You're missing a piece of the puzzle there. The chip on the RPi 2 is custom silicon, they didn't just cough it up overnight, it was in development long before the Odroid C1 existed.

Yes, that is the public story, RPi Foundation says that the SoC was made only for them. I'm personally skeptical. With as secretive as Broadcom is I doubt we'll ever know.

You talk about RPi having been under no threat. Maybe if all they cared about was being number one that would be the case, but losing any significant number of sales matters, even to a non-profit foundation. And I'm positive that if Hardkernel invested much in volume C1 would have stolen a big chunk of RPi 1's sales.

I do think an updated RPi would have happened sooner or later regardless, but to me the timing is conspicuous. Coming out with RPi 2 quite shortly after C1 is released, changing only the SoC and RAM which were the two big weaknesses compared to it, and making (questionable) comments on blogs that are clearly directed towards convincing people RPi 2 is better all look suspect to me. I'm not going to say the whole thing was reactionary, but I do think it at least played some role.

 

[gadgetoid]

Yes, that is the public story, RPi Foundation says that the SoC was made only for them. I'm personally skeptical. With as secretive as Broadcom is I doubt we'll ever know.

If Raspberry Pi Trading did suddenly react to the Odroid C1 then it was an awfully convenient coincidence that Broadcom just so happened to have the right chip sitting in the corner of their facility just waiting for its moment to shine!

The Pi 2 was certainly in the works before the C1, but there are also clues that the C1 caused it to be rushed out a little earlier than it should have been. Looking at the direction the software was taking, and taking into account the number of B+ kits sold over Christmas, I would have expected the Pi 2 to launch in the middle of this year.

For example, It would have been nice to have features like Device Tree ( a massive change to the Pi community, since it breaks guides and how-tos the whole internet over ) to be rolled out, tested and fully embraced before the Pi 2 landed and foisted them upon thousands of unsuspecting users.

Either way, it's nice to have a faster Pi no matter how it came about. If Odroid want to keep throwing down the gauntlet, then I'm more than happy with that!

 

[Exophase]

If Raspberry Pi Trading did suddenly react to the Odroid C1 then it was an awfully convenient coincidence that Broadcom just so happened to have the right chip sitting in the corner of their facility just waiting for its moment to shine!

Not as coincidental as you think. Check out this quad Cortex-A7 + VideoCore IV SoC that Broadcom announced over a year and a half ago:

http://www.broadcom.com/press/release.php?id=s771153

The one in the RPi 2 could literally be a BCM23550 with the baseband parts fused off.

Broadcom has since announced that they're out of the baseband market (http://www.forbes.com/sites/greatspeculations/2014/06/04/rising-competition-forces-broadcom-to-exit-the-baseband-market/) which probably includes such SoCs. So they could have a large stock without any customers, and a good opportunity for RPi to negotiate a new SoC for cheap. RPi also (at least in theory) could act as an eval board for Broadcom, so maybe they're hoping someone will want to mass produce a design prototyped with it, using the same SoC (and therefore perhaps they are saving better bins for such an outcome)

The Pi 2 was certainly in the works before the C1, but there are also clues that the C1 caused it to be rushed out a little earlier than it should have been.

Exactly, knowing more about it I think that's the case, that RPi 2 was in planning stages but C1 caused it to come out to market faster.

That said, I think it's entirely possible that RPi Foundation knew about C1 well before we did, or at least saw the writing on the wall with the direction that competitors have been taking.

 

[elwing]

Yes, that is the public story, RPi Foundation says that the SoC was made only for them. I'm personally skeptical. With as secretive as Broadcom is I doubt we'll ever know.

 
we'll never know, but I can imagine that being true, since this upgraded chip still got the same old graphic processor, that was opensourced for the RPi... I doubt someone else would keep that old graphic chip for another reason...

 

[Exophase]

we'll never know, but I can imagine that being true, since this upgraded chip still got the same old graphic processor, that was opensourced for the RPi... I doubt someone else would keep that old graphic chip for another reason...

See the post above. The CPU + GPU configuration is something Broadcom already used in another SoC, so given that it's hardly a stretch to think that they'd have such a thing made for more than just RPi.

RPi Foundation's claim that they selected Videocore IV because it's the most open option is, in my opinion, a farce. They didn't select anything, they have little choice but to use what comes in the SoC Broadcom supplies them, which uses their own GPU technology. For example, their tablet-oriented SoCs (http://www.broadcom.com/products/Applications-and-Multimedia-Processors/Tablet-Application-Processors) all use Videocore IV, so they're not just kicking it around for RPi's sake. I'm not aware that they've licensed outside IP for graphics in anything else. Making a GPU is a big investment, you tend to want to get as much return on it as you can.

This is a lot like back in the days of something like GP2X, where the hardware manufacturers would say they're using Linux because it's all open source and stuff. That may have been true, but really they used it because it was free for them to do so and a good fit. RPi is basically tied to the hip with Broadcom so they'll use what Broadcom can give them, that's pretty much all there is to it, any other explanation is marketing or making excuses.

 

[mcobit]

The RPi2 also still has the slower RAM and no Ethernetcontroller directly in the soc but bodged on via usb what makes it slow abd restricts usb bandwidth if you use both.


The C1 has hardwareproblems like non functioning CEC because of wrong voltage levels.


It has a more beefy GPU but closed drivers.


I like that the RPI used the same Videocore graphics to maintain good documentation and free drivers but they could have upped the corecount there.


Edit: I ordered a PI for me as I want to update my retropie and kodi setup. N64 and psx emulation seem to very good.

 

[Ian J]

The RPi2 also still has the slower RAM and no Ethernetcontroller directly in the soc but bodged on via usb what makes it slow abd restricts usb bandwidth if you use both.

How bad is this?  If I was watching a movie via USB hard-drive what sort of speed could I expect via wired ethernet?