Let there be boards (2010 - 12 - 02)

[EvilDragon]

i worry about the time to wait to get this schedule, the time announced by the schedule, and finally the time that will exceed the schedule date to really have a complete production.

Well, it's pretty hard to get proper schedules shortly before christmas.


Christmas is the time before the world gets crazy.

I was wondering why you guys didn't decide to go with socketable RAM for this device... I think something like that would have been a great feature for the Pandora. I've been wondering for the past few days why you (OP) didn't decide to use socketable RAM. It's because you used OMAP? is there another sort of microprocessor you could have used to allow this feature? Had OP thought about it before deciding to go with the current design?

Michael knows that best, but looking at the board, I have no idea where we should've placed the RAM.. it's pretty much stuffed.

 

[mmielke]

This is good news


I am very much looking forward to using my Pandora with my University course on Games Programming


Was digging around my emails and I found this statement in them a little funny:

Yeah, you'll have statements like that. Most of the problems have been rectified, but ED is waiting on the forum member to visit the factory tomorrow. If the factory has told the truth, there will be pictures of thousands of boards (possibly video). When Circuit Co has stated that the boards would be completed by November 18th (October?) and gave us what I feel to be partial runaround, OP should have received some sort of discount. They do have a backup company that is being prepared in Taiwan at the moment. ED has mentioned that it should take less time, as most of the parts will be manufactured and assembled in one location. All in all, this is all great news. I have estimated that being 850-950 in the queue should give me a delivery date of April.


- Has he gone to Jared to buy the Pandora bracelet -

 

[Eridger]

Michael knows that best, but looking at the board, I have no idea where we should've placed the RAM.. it's pretty much stuffed.

That's why you'd add a daughterboard! I want my Pandora to be at least 8" thick!


Also, I haven't a clue who Michael is.

 

[Link]

That's why you'd add a daughterboard! I want my Pandora to be at least 8" thick!


Also, I haven't a clue who Michael is.

He's the mastermind behind the board and the innards

 

[Eridger]

He's the mastermind behind the board and the innards

Not someone I can easily get a hold of?

 

[Long]

Pandora coming!!! :lol:

 

[WizardStan]

Not someone I can easily get a hold of?

MWeston.


And no, there weren't any worthwhile processors they could have used, not without sacrificing several other features. I doubt we'd be getting 1Ghz clock speeds with the same power usage with anything else, for example. MWeston would know better, though, he's done far more research than I. Socketable RAM really isn't all it's cracked up to be, anyway. You may think more RAM is better (and generally speaking it is) but you eventually reach a limit of actual usable RAM, and I don't believe we're close to that limit yet, bearing in mind the principal of the project.

 

[ardje]

That's why you'd add a daughterboard! I want my Pandora to be at least 8" thick!

That's why they add a package-on-package.


With the ram on the board, four times as much I/O (if not more) have to be routed through the board, this means very expensive multi-layer boards.


It's impossible I guess.


Trust me, the current solution is *the* solution for these kind of devices.

 

[j0n]

MWeston.


And no, there weren't any worthwhile processors they could have used, not without sacrificing several other features. I doubt we'd be getting 1Ghz clock speeds with the same power usage with anything else, for example. MWeston would know better, though, he's done far more research than I. Socketable RAM really isn't all it's cracked up to be, anyway. You may think more RAM is better (and generally speaking it is) but you eventually reach a limit of actual usable RAM, and I don't believe we're close to that limit yet, bearing in mind the principal of the project.

Assuming a 32 bit processor and OS, the most a single process can use is 3GB. http://www.spack.org/wiki/LinuxRamLimits





Not an issue for most applications, but can become an issue for large research programs as I found out the hard way.


But yeah, its not going to be an issue unless the Pandora grows to having more than 3GB, which I doubt it will in the foreseeable future.

 

[slaeshjag]

The 3GB border is specific to x86. 32-bit processors usually have a 4GB address space, how much that is used up by other hardware devices depends on platform.

 

[j0n]

The 3GB border is specific to x86. 32-bit processors usually have a 4GB address space, how much that is used up by other hardware devices depends on platform.

Really? I would have assumed x86 was capable of addressing 4GB of address space, it being a 32 bit processor, why is the x86 different?


What I should have said is that I was assuming linux as the OS also, because I think that reserves 1GB for the kernel's usage by default, don't know if you can change this. From my experience, 32 bit linux doesn't like it if you exceed 2GB of memory usage, though I'm sure if you custom compile the kernel with crazy settings you could get nearer 4GB.

 

[marktuson]

Really? I would have assumed x86 was capable of addressing 4GB of address space, it being a 32 bit processor, why is the x86 different?


What I should have said is that I was assuming linux as the OS also, because I think that reserves 1GB for the kernel's usage by default, don't know if you can change this. From my experience, 32 bit linux doesn't like it if you exceed 2GB of memory usage, though I'm sure if you custom compile the kernel with crazy settings you could get nearer 4GB.

x86 is different because it's archaic. Its first incarnation was designed for glass terminals, and it hasn't really frown up apart from just getting bigger and faster. The 64-bit version is better, more powerful, and designed for actual computation.

 

[slaeshjag]

Really? I would have assumed x86 was capable of addressing 4GB of address space, it being a 32 bit processor, why is the x86 different?


What I should have said is that I was assuming linux as the OS also, because I think that reserves 1GB for the kernel's usage by default, don't know if you can change this. From my experience, 32 bit linux doesn't like it if you exceed 2GB of memory usage, though I'm sure if you custom compile the kernel with crazy settings you could get nearer 4GB.

It can, it just have a bunch of memory mapped devices in its memspace.


Also, there are hacks like PAE on x86 that allow you to use a memory bank like structure, so you can switch in memory outside of the 4GB memory space. This is something only the enterprise products from microsoft can, and most linux distros provide a kernel supporting it that is installable from the package manager.


I've not read up on what the kernel does with the 0xC000000~0xFFFFFFFF range, might be true that the kernel reserves that for its own purposes.

 

[marktuson]

It can, it just have a bunch of memory mapped devices in its memspace.


Also, there are hacks like PAE on x86 that allow you to use a memory bank like structure, so you can switch in memory outside of the 4GB memory space. This is something only the enterprise products from microsoft can, and most linux distros provide a kernel supporting it that is installable from the package manager.


I've not read up on what the kernel does with the 0xC000000~0xFFFFFFFF range, might be true that the kernel reserves that for its own purposes.

I'm the boss; I'm having this bit of core for myself.

 

[WizardStan]

Not an issue for most applications, but can become an issue for large research programs as I found out the hard way.

That's why I said "bearing in mind the principle of the project". If you're using the Pandora for huge research programs that require gigabytes of memory, you're probably doing something wrong


There's also swap space, if you really need more RAM: too slow to for large games that need fast memory, but taking an extra second to load your web page, or a few minutes more on an hour long calculation, wouldn't hurt too much.

 

[may88]

As I mentioned, they didn't get all needed PCBs for the nubs yet (the nubs also have PCBs in) so they couldn't finish more yet.


We are waiting for a schedule for the rest of the nubs.

Ah! That explains a lot. I miss read PCB to mean boards. Sorry, must learn my nub anatomy.


So actually the 90-90% bit makes sense now.

Depending on the width of your memory and how the system handles overflow, such a large number might even be interpreted as negative!

A number that size is likely to be zero.

Well, PC RAMs use different chips. We need PoP ones for the OMAP

Is this used for the NVRAM or is that different?

This is good news


I am very much looking forward to using my Pandora with my University course on Games Programming


Was digging around my emails and I found this statement in them a little funny:

Lost the original statement in this quote... But the OPT did not specify which Christmas.


Crazy statement/promise to be making. No lessons learnt. Sounds like one of Craig's.


Gett ED to go the PR is a good move although you kind of miss the wild stuff.

Pandora coming!!! :lol:

This is good news... for Pandora.

 

[WizardStan]

A number that size is likely to be zero.

Negative zero!

 

[j0n]

It can, it just have a bunch of memory mapped devices in its memspace.


Also, there are hacks like PAE on x86 that allow you to use a memory bank like structure, so you can switch in memory outside of the 4GB memory space. This is something only the enterprise products from microsoft can, and most linux distros provide a kernel supporting it that is installable from the package manager.


I've not read up on what the kernel does with the 0xC000000~0xFFFFFFFF range, might be true that the kernel reserves that for its own purposes.

Ah right, hadn't thought about memory mapped devices alright. PAE is only useful if you have lots of little processes though right? It doesn't extend the amount each process can use. Seems that processes can get around 2-3GB depending on OS and CPU in 32 bit.

 

[Thorium]

PAE is only useful if you have lots of little processes though right? It doesn't extend the amount each process can use. Seems that processes can get around 2-3GB depending on OS and CPU in 32 bit.

No a single process can actualy use more than 3GB with PAE. It can only address 3GB at a time but you can map a memory region from physical memory beyond the 4GB limit into your virtual address space. So you can map 500MB, process them, unmap them, map the next 500MB and so on. I dont know about Linux but Windows has a API for that.

 

[Seamus]

We seem do have drifted away from the burning topic when are we likely to see pandoras been shipped again.


I really wanted this for last xmas.


Now I know not even this Christmas.


A realistic delivery date please.


Ordered July 2009.