I was browsing the freescale web pages and stumbled upon the MPC8377E, which is an e300 based SoC that seems to offer a lot (most?) of those features that people missed in the 5200B based Efika (like Gigabit Ethernet, USB2, SATA 2.0, PCI-E, and a little (66%) faster).

In other words it is quite feature rich and would fit very well for many of those applications suggested by many of the developers in the Efika Developer Program.

So what are the downsides? Cost? RoHS?

Wow, could be the perfect NAS/DMS/GW machine with just a single chip.

_________________
CzP
http://czanik.blogs.balabit.com/

This is the development board (http://www.freescale.com/webapp/sps/sit ... DS&fsrch=1). It doesn't look "easy"...

None of the FreeScale eval boards are so nice and elegant, as the boards from Genesi. Just see what they did with 5200B :-)

http://www.freescale.com/files/32bit/do ... 200BPB.pdf

_________________
CzP
http://czanik.blogs.balabit.com/

We're looking at these (as always) but our current focus is on the MPC5200B/MPC5121E and MPC8610 since they need the most "love" right now. We have proven track record with the MPC5200B but the DIU (internal display) on these two chips means it can be a TRUE single board computer.

I would absolutely welcome an MPC8377E board if we had the time or resources to develop one; it is quite hard to sell these chips even with the "desktop like" features, to desktop markets and developers, since it implies having graphics cards available and far more support than any company is willing to put out there (XGI, AMD, nVidia, we know exactly where they all stand) unless you are pledging to sell 1,000,000 Pachinko machines a year.

We will see.

_________________
Matt Sealey

That's hardly surprising.

The purpose of an evaluation board as manufactured by the semiconductor company is completely different from someone like Genesi making a general-purpose computing platform such as the Efika.

On eval systems, it's common to hook some devices up in strange (or several, redundant) ways to show customers how it can be done (and verify that it is working). When it comes to board space, making it too crammed doesn't serve a purpose, it's more valuable to have room around the board for measurements, test points, debug connectors and board reworks in case of problems.

While for a systems company like Genesi, it's much more important to do a small nice system design without adding extra cost due to redundant hardware, and to package it nicely.

That's true, but we've always tried to market all our boards as more accessible reference designs; the Efika, for example, is just as feature-rich as the Lite5200B as long as you aren't doing anything with MOST (a hideous Automotive multimedia bus standard).

The Efika is far better as a development platform for something else - usually involving a new hardware design - simply because we make a big deal out of real Linux distribution support, and have exposed the features as best as can be done in a more "desktop oriented" fashion. After all, what good is hardware without software? The Efika can be used to develop that software (at a suitably high level) while real final target hardware isn't yet complete.
The idea with the Efika was to produce a board which could be sold for $99 and get people into the MPC5200B market - the same will be true of the MPC5121E and MPC8610 (although perhaps not at $99). The current range of development boards for these processors is upwards of $1500-3000.

The idea with the MPC5121E is we can use it to further our pervasive goal of that "cheap computer the size of a cigarette packet" - you remember the Eclipsis? :)

With external graphics and too much support required this isn't possible with the 5200B (but it's very close), but with integrated graphics and higher integration as on the MPC5121E it's really getting there. Of course we could do it now with i.MX or so but that would defeat the object; everyone does these boards and even smaller (see Gumstix for the best example). It wouldn't be unique or interesting on ARM.

Of course there are always better chips for the job, we have some concerns about the MPC5121E but are going to make a go of it anyway; it's as close as it could ever be to what we need for that highly embeddable, portable or even wearable system..

The first step is always a larger board though, ITX or something, then keep revising it down taking off the features people do not seem to use.

_________________
Matt Sealey

This whole discussion is digressing from the original topic, but still:
Yep, Efika is a good software development platform for that kind of parallel development, but it lacks a couple of crucial features to be based for hardware-near development, which will possibly require a "real" eval board instead. A JTAG header is the obvious first one. A closed-source commercial firmware is the second. It all depends on how low-level development is required for the application.

I'm not saying those choices weren't suitable for Genesi to make, they clearly were (I doubt many current Efika users miss JTAG or firmware sources).
This is where the mindset and market between the chip manufacturer and the system manufacturer differs.

I get the question all the time (why eval boards are so expensive). The answer is simple: They are there to be available for customers who are serious about buying _chips_ in volume, so they can evaluate them, and they are not meant for the end-user market. There's a significant amount of support involved per customer, boards are smaller volume and that really adds up in cost.

Also, and probably most significantly, the eval board price is rarely a hurdle for those looking to develop a new system while end users are much more price sensitive.
I remember the BlackDog systems that had a Xilinx with a 405 on them. Neat little things, but I don't think they ever went beyond the "toy" category. I'm extremely happy with my Nokia N810 at the moment, just the right combination of usability (with the added keyboard) and portability.
Exactly, and this is why the semiconductor eval boards are so often complex: so their customers can take a subset of that and base their design off of that. I'd expect that's exactly what Genesi did to make the Efika (base it on the reference design, added a few things and took off things no longer needed + add new firmware).

I disagree. Why do you need JTAG to develop application-level software? Why do you even need JTAG to develop Linux kernel software when there are plenty of kernel-level and application-level debuggers which operate over ethernet?
Right, and this stops x86 development somehow too, right? I think if you are developing hardware you need to consider licensing a firmware or spending a hell of a lot of time developing your own (or your own support for an existing Open Source firmware, at least).

It's actually far cheaper to license the Genesi Aura firmware than it is to pay an engineer to develop U-Boot support over the course of a month. It wouldn't be tenable to license it if it wasn't.
We have had literally 3 people express a genuine interest in JTAG. I mean, people who already own the hardware required (wigglers and host software are not cheap) and had a real need which could not be solved any other way (software or otherwise).
We have several examples where companies want to evaluate the MPC5200B for a solution and have jumped at the chance to buy a working Efika for $99 rather than then $3000 Lite5200B. Remember when you want to use a new chip (and if you are on the bleeding edge as with the MPC5121E or MPC8610) then you do not know if you are going to buy a huge quantity of chips. Buying 3 or 4 development boards for your team is a huge outlay to find out that this system does not meet your needs, compared to the $400 it would cost to buy the same number of Efika systems.

What if you are a startup and just don't have $10,000 of cash you can throw into "wondering if it's a good chip or not"?
The Efika design is "from scratch"; it is not directly or indirectly derived from the Lite5200B except by example and reference.

There are several things done wrong on the Lite5200B design for a start. Also, why start with something so complex?

Most of the peripherals on the MPC5200B are industry standard connectivity. USB connects via one of two standard PHY connections. Ethernet MII.. AC97.. PCI.. I2C for the system management PIC.. serial port.. a lot of the design where applicable was simply dropped in from the Pegasos (I'm sure you noticed the audio connectors are identical..)

You do not need a complex development board to implement this chip or any of Freescale's similar SoC designs on a suitable board. Sometimes it is better to go for a clean slate, than to take an existing design and rip out the things you do not need. We have built a substantial library of "parts" that can drop into new designs with very little effort..

_________________
Matt Sealey

Couple a MPC8377E with an ATI Radeon E2400 (@PCI-E x1 or x2) and you will have a very interesting system. Sure, it will not be as low cost/high margin as the MPC5121-e, but on the other hand it will have better specs, better performance and PCB sizes could still be quite small. And it will have cache coherency, which is a *big* plus to the overall equation, right? Maybe it would be worth it?
Isn't this a general obstacle that you will face in whatever endeavor you will ever undertake?

At some point Genesi considered using FPGA's:



Is this such a bad idea really? Replace the 5200B in the picture above with an MPC8377E (a lot of upgraded interfaces and buses) and you will have an Efika on steroids. Could be a building block for most of the LimePC devices, and what you lose in margin you gain in flexibility and control.

I mean if the big dragons are so tight about opening up their documentation of their GPU's, then what can you do really? And this must be a general problem in the smaller volume embedded industry, so there must be some kind of market there screaming for more open and reachable alternatives? Maybe this could even be a foundation of a new business? ;-)

Or are you "wanting what 'they' have" in the first picture and trying to deal with the *obvious downsides* of that one (it's not the first time Freescale fails your expectations) while hoping that Freescale will get it right *the next time* and manage to put coherency into the "chip 4u2"?

Do you have any odds for that one BTW?

Why do you want a ati video card? There wil be no linux drivers for powerpc supporting hardware acceleration in a proper way (at least not in short period of time).

As with the mpc5121 there may be a binary driver from PowerVR provided by genesi giving opengl es 1.x hardware acceleration.

But then again you may not be needing hardware acceleration. I for one would welcome it.

Also with the mpc5121 everything would be integrated on the mainboard not needing an extra graphics card.

That propably would save energy also. When the ati card on my efika-board goes in xwindows, the power useage goes way up.

You are right to start, but the second part is just false; we could get drivers if we wanted to. The problem is making the board cost too much; an ATI E2400 is not a cheap chip, it is certainly more costly than a ~400MHz SoC. There is also soaking up a whole peripheral bus on a chip which cannot share a bus; PCI Express is point to point, and AGP does not support multiple devices. It poses too many "makes the design suck" problems.

What we (I) don't want to see in the next boards is a solution where we force every customer to buy a specific Radeon model (as with Efika) since this is the only viable supported GPU. With the MPC5121E and MPC8610 there is the inbuilt DIU - it's a simple framebuffer graphics controller which is very very flexible. It's even got the same driver support in Linux (they are two slightly different workings of the same IP core).

We can AltiVec accelerate the DIU driver in X.org for the MPC8610 and use in-chip DMA to accelerate other functions which will work on the MPC5121E. It should be usable for video playback, along with AltiVec video codecs or the AXE DSP helping respectively. For extremely fast OpenGL support, I wouldn't bet your life on the idea unless you were running something highly embedded. I would love to see an X.org, DRI/DRM driver that works plus a libGL implementation which took full advantage of the PowerVR, but take a look; do you see it for the i.MX31 on Linux? We have an uphill struggle on our hands, as usual.
Even if not, the DIU will provide at the very least enough display support to get anything working. Any display on the chip is better than no display at all. The MPC5121E is a little.. wierd.. in this regard, though. We'll see which display we get when the board is finished, and we have bashed our heads on our desks for long enough to rattle out the graphics support! :D
Indeed. The MPC5121E is specced at less than 3W for the entire chip running with all units - it can be cut down using some new power management and by turning off certain unused units and clocks, too. The ATI chip is literally a rework of the mid-range desktop class chip from 2006. You would triple the power consumption of the board by only turning it on.. added to the triple cost, the driver support, there is already too much involved here.

Stefan, the FPGA was an idea at one point; it's actually not as good an idea as it was at the time when there may have been customers for it. It actually works better if you design a PCI card for the FPGA and put it there. Nobody who uses the Efika for FPGA development is using a Radeon, and the uses of the FPGA - video codec acceleration etc. - is superceded by the MPC5121E AXE unit anyway.

_________________
Matt Sealey

Have look at this: http://www.imgtec.com/powervr/insider/s ... asp#GLES1b (Freescale i.MX31)

(note: request OpenGLES 1.1 drivers from your platform supplier).

Anyway, that will be the embedded opengl-es way.

But that is not a bad thing. I think an opengl wrapper around opengl-es should just be possible. But there may be no need for that as also on normal opengl using vertex arrays is more efficient then calling glvertex3f .

Maybe even XGL could be layered upon opengl-es: http://en.wikipedia.org/wiki/XGL

We have all this for the MPC5121E. Like I said; we can get drivers if we want.
This too.
Yep and I had a chat with David Reveman too, about exactly that. It's somewhat possible.. but also somewhat not possible.

It's all possible but that driver kit is *NOT* mainline, distribution-oriented X.org, it's KDrive, which is an embedded subset, and a special OpenGL driver, and is missing a lot of glue and platform code even for the ARM driver (the customer is meant to fill that in themselves).

It is a considerable amount of work to even use that kit, none of which will benefit desktop users. Supporting desktop users would probably require contracting the original software developers (Alt Software) to write real X.org drivers and suchlike since we certainly do not have that talent in our pool.

_________________
Matt Sealey