Raspberry Pi - Hardware Flaws and Fixups?

Now that people have actually started receiving the boards so the design in final, is the schematic close to being posted?  I looked around again in various likely places just now, but haven't spotted it yet.

actually, the design may not be quite final yet.  Liz said: "Happily, we’ve found it doesn’t need a shielded enclosure to reach Class B, although it will require a (very minimal) redesign." http://www.raspberrypi.org/archives/978, page 3, April 10, 2012 at 7:04 pm.

It hasn't been announced when such a change might be implemented, but probably better sooner rather than later, given the large production runs planned, perhaps along with a change to fix the SD card voltage issue.

Well, in principle you can use the FDX as a GPIO LED.  The difficulty is that you need to talk to the LAN9512 in USB-speak to configure and toggle it, and the driver may make this difficult.  Still, students would learn a lot about USB drivers trying to do it :-)

Was it here that someone asked me to measure the physical size of the 'pi? Mine is 55.9 x 84.9mm.

Yes me...  thx Roger.

Hans

I know it is only slightly different, and may not matter at all, but in a post on the Raspberrypi.org website one recipient (another case manufacturer actually) has said that: The board is 85.00 x 56.17

Hi David, yes I know   ModMyPi measured that....  That is why I asked Roger... I know he is very precise...

So probably the pcb board is 85.0 x 56.0  with 0.2 tolerance

All my cases are made 85.6 x 56.0, and I have 4 lying here waiting.... but I was not that lucky in getting a Pi...

So the pcb will fit,  even if it is .17 wider I can probably shave a little off the pcb board...

Hans

The schematic doesn't appear to show the memory chip-select lines,

but Gert has mentioned that the SoC has only a single chip-select line,

which prevents stacking more than one memory chip, as described here:

http://www.engadget.com/2011/03/25/samsung-starts-baking-30nm-4gb-lpddr2-chips-packaging-2gb-mobil/

Are the production volumes large enough now that Broadcom might consider

presumably minor changes such as adding an additional chip-select line?

The schematic does not show any of the connections to the Package on Package (POP) memory that actually sits on top of the BCM2835.  It never gets to the PCB so there is just a hidden attribute that calls up the part onto the parts list so it doesn't get forgotten!  I think I should make reference to that on the next schematic set.

More than one chip select would not be of any use unless a memory manufacturer decided to make a DRAM chip that passed through all the signals (and the additional chip select) to a package that sits above.

Theoretically it is possible, and there are examples of Package on Package on Package.  It's not uncommon for this to be done inside a single package (stacked die) but the interconnect is done usually with bond wires.

Unless we pass the 4-5 Million Pi's mark with no obvious tailing off of demand I think Broadcom would not be interested, I suspect the number is in reality much higher than this.  Making even a minor mod to silicon can cost millions of dollars.  One of the big costs used to be a mask set, but now I think direct imaging has taken this out of the loop. I'm sure there are people on the forum who could comment better than me - I just use the chips now. 

I used to make silicon but the advent of cheap (relatively) programmable logic (FPGAs)  has largely chopped away the mid volume market.  The stress of waiting for a PCB spin is nothing compared to 'first off' silicon - hats off to those guys, I'm to old to suffer that now!  You can fix a multitude of sins in an FPGA.

One option discussed before is to use a larger capacity POP memory chip but for the moment this is not in the roadmap.

Cheers

Pete

The engadget article I cited mentions getting 1GB by stacking 2 4Gb memory chips,

where otherwise stacking 4 2Gb chips would be required.  I'm not sure if that involves

multiple chip-select lines or not.  I'm also not certain what memory parts are actually

available since this article.

Looking at the article it looked like this is a stacked die ('chip') application inside a single package - should have made this clear in the previous post.  The BGA ball array is set up for Standard Mobile and not to sit on a POP architecture so I think we are out of luck here.

Pete

Even then...

It seems ARM code is pretty compact. I've compiled programs that seem to come out to 5 and 6k. That's nice.

Why do you NEED 1Gb of RAM?

A few weeks ago I was annoyed at my browser using over 1G of RAM. Then I calculated: It might cache a bitmap of that page for when I switch to that tab. I'd be annoyed if it would re-render it from the source if I switched tabs. If I Have 200 tabs open, a megapixel per page comes to 600 Mbytes.... A gig and a half isn't that bad for some overhead.

After loggin in a few times over the ethernet, I see the 'pi not touching the SD card to log me in. Everything needed has been cached. That's what it is supposed to be like.

Arduino-like board         raspberry pi.

16k flash.                     16G. (4G?)

2k? 4k? RAM                256M

serial 115k2                  USB 480Mbps

16x2 LCD on 8 pins       HDMI out / anallog video out.

No OS                          good OS.

To be able to run X I equipped my '486 machine with 8Mb instead of the standard 4Mb back in '91. A few years later, I had upgraded to 20 and then 32 Mb RAM and a new motherboard, I was swapping RAM modules around, and I thought my machine had crashed with the HDD led stuck on when the X weave showed up. Turns out the 32M was on the other motherboard I had forgotten to swap back: I was booting into X with 4Mb of memory. Things have bloated a bit. A kernel is now about 10x bigger than back then. A video card is no longer satisfied with "enough RAM to put the framebuffer in".

We've come a long way since then. The Raspberry pi is a nice machine with a whopping 256Mb of RAM. And because of the way it is built it can be sold cheap. And because of the way it's built, there is no system bus or memory slots. Tough luck.

Pete, While on the subject of RAM.... I've been running with a 208 / 48 split the last few days. How are my chances of running 240/16 ? I'm currently using a 656x416 framebuffer: No HDMI connected. Less than 300k pixels. i.e. a total of 1.2Mb for 32 bits per pixel (I can understand a GPU not supporting 24 BPP as it's a pain in the $%% (*))

(*) For the non-hardware guys, suppose you want to turn on the pixel at 101,101... Turns out that pixel is spread out over two 32-bit words. Really a mess to do.  Slows things down too. Best case, you have to do two writes to the framebuffer to turn on one pixel. Worst case you have to do two reads as well and delay those writes until you've gotten the results of those reads!

P.S. Look at: http://www.hynix.com/gl/products/mobile/mobile_info3.jsp

You can see there that they have up to 2Gbit of POP memory. About two years from now they will have 8Gbit and we can upgrade our 'pi to 1Gbyte of RAM.

Even then...

It seems ARM code is pretty compact. I've compiled programs that seem to come out to 5 and 6k. That's nice.

Why do you NEED 1Gb of RAM?

A few weeks ago I was annoyed at my browser using over 1G of RAM. Then I calculated: It might cache a bitmap of that page for when I switch to that tab. I'd be annoyed if it would re-render it from the source if I switched tabs. If I Have 200 tabs open, a megapixel per page comes to 600 Mbytes.... A gig and a half isn't that bad for some overhead.

After loggin in a few times over the ethernet, I see the 'pi not touching the SD card to log me in. Everything needed has been cached. That's what it is supposed to be like.

Arduino-like board         raspberry pi.

16k flash.                     16G. (4G?)

2k? 4k? RAM                256M

serial 115k2                  USB 480Mbps

16x2 LCD on 8 pins       HDMI out / anallog video out.

No OS                          good OS.

To be able to run X I equipped my '486 machine with 8Mb instead of the standard 4Mb back in '91. A few years later, I had upgraded to 20 and then 32 Mb RAM and a new motherboard, I was swapping RAM modules around, and I thought my machine had crashed with the HDD led stuck on when the X weave showed up. Turns out the 32M was on the other motherboard I had forgotten to swap back: I was booting into X with 4Mb of memory. Things have bloated a bit. A kernel is now about 10x bigger than back then. A video card is no longer satisfied with "enough RAM to put the framebuffer in".

We've come a long way since then. The Raspberry pi is a nice machine with a whopping 256Mb of RAM. And because of the way it is built it can be sold cheap. And because of the way it's built, there is no system bus or memory slots. Tough luck.

Pete, While on the subject of RAM.... I've been running with a 208 / 48 split the last few days. How are my chances of running 240/16 ? I'm currently using a 656x416 framebuffer: No HDMI connected. Less than 300k pixels. i.e. a total of 1.2Mb for 32 bits per pixel (I can understand a GPU not supporting 24 BPP as it's a pain in the $%% (*))

(*) For the non-hardware guys, suppose you want to turn on the pixel at 101,101... Turns out that pixel is spread out over two 32-bit words. Really a mess to do.  Slows things down too. Best case, you have to do two writes to the framebuffer to turn on one pixel. Worst case you have to do two reads as well and delay those writes until you've gotten the results of those reads!

P.S. Look at: http://www.hynix.com/gl/products/mobile/mobile_info3.jsp

You can see there that they have up to 2Gbit of POP memory. About two years from now they will have 8Gbit and we can upgrade our 'pi to 1Gbyte of RAM.

Roger,

   you ask "why do you NEED 1GB of RAM?", especially when ARM code is

relatively compact.

   First, the compactness of code has almost nothing to do with RAM usage.

Most RAM is used by the data an application is processing, not the code of

the application.

   There are many reasons why 1GB+ of RAM is needed, although certainly

everyone's milage will differ.

   Most Linux distros currently have stated memory requirements of around 1GB,

especially for non-command-line usage.

   Most PC's currently come with 4GB+ of RAM, and many applications are

taking advantage of that memory, including IDE's like Eclipse, word processors

such as LibreOffice, GUI's such as Gnome, multi-tabbed browsers, etc.

  Compiling large linux packages takes over 1GB, including webkit, gtk,

LibreOffice.  Some linux distros depend on packages being compiled

natively rather than cross compiled.

  RAM can be used for I/O buffers, and for ramdisk, to compensate for

slow disk access.  The RPi forum is hosted on a machine with 96GB

of ram for that purpose.

p.s.

  8Gb lpddr2 memory chips appear to be available today, although

not certain at what price, quantity, and form factor.

see http://sunrisenanotech.com/products06f.html

       http://www.dramexchange.com

You shouldn't host a popular website and forum on a raspberry pi.

The forum on raspberrypi.org is moving to phpbb instead of the wordpress thingy. Why? PHPBB is written more efficiently so that it will be able to handle increasing traffic.

Writing reasonably efficient code will allow you to do things with your 'pi that it was made for. You'll have to buy a bigger machine to do bigger things.

The fact that chips are listed does not mean that they are available. They often make nice pages advertizing new products cleverly hiding the fact that they are not available until 3Q2012. They hope you'll design with their component and only then find out that you can't buy them in quantity. That said, no 8Gb, 168 ball, 1.8V DDR2 is listed in the page you mentioned. (I don't know how likely it is that they would be compatible if it were listed). On the other site, I could not find any specs to see if it might be compatible.

right, you shouldn't host a popular website on a RPi, but you might want

to host a less popular website, with 1GB+ of the most frequently accessed

pages cached in RAM.

It looks like the 8Gb, 168 ball, DDR2 part listed is 1.2V rather than 1.8V.

If you have that kind of high hitrate, you shouldn't use a Pi for it..

The RPi is perfectly suitable for a server, costs about 64 - 80 MB in memory, remaining is for applications, caching...  more than enough to do all kind of things

I usually use the filesystem for storing webdata, no database overhead and very fast access, most of the time everything is cached automaticly by the filesystem.

I hope that in feature version they will stick to the 256 MB.....

That will make/force people to make/find more efficient solutions/programs.

Hans

Hans,

  It may seem to you that the RPi was designed with a ridiculously small RAM

in order to "make/force people to make/find more efficient solutions/programs".

But that isn't the case. 

It was designed the way it is because it was repurposed from the Roku design

as an educational computer rather than a media center, with no easy way to

enlarge the PoP memory.  As opportunities for enlarging the memory become

readily available, I'm sure the RPi will take advantage of them, even beyond 4GB.

If the RPi designers actually subscribed to the "hair shirt theory of programming

excellence", they would have reduced the cost by using 128MB instead of 256MB,

or maybe even left off the PoP RAM completely and used the 128KB L2 cache for

program memory.  That would still have been much more memory than the BBC micro.

Hynix 4Gb (=512MB) PoP RAM is now in mass production.

http://www.hynix.com/products/mobile/mobile.jsp?info.ramCategory=&info.ramKind=28&info.eol=NOT&posMap=MobileDDR2

previously this page showed the status as "customer samples".