User avatar
Jim Manley
Posts: 1600
Joined: Thu Feb 23, 2012 8:41 pm
Location: SillyCon Valley, California, and Powell, Wyoming, USA, plus The Universe
Contact: Website

Re: Is a new Raspberry Pi model coming?

Sat Nov 28, 2015 9:53 pm

It's pretty clear no one has gotten my drift on what my unpublished speculation would be, and that's actually making this more fun. I will say this, don't think about why the BCM-2835 was apparently (and it may have just been a package change, mind you) slightly downsized, and I'm assuming to fit the new DRAM device. Think about what a certain kind of substantial change in die will allow in development of future, much higher-performance-for-same-cost Pi SoCs (plural, over time). First, they adapt the package to fit a new family of DRAM (including BGA connection points) but don't change the internals of the SoC, other than to get the gozintas and gozoutas aligned for the new package.

Next, they do some photolithographic sleight-of-hand to take advantage of the new die capabilities to adapt the BCM-2835/6 with no change in functionality, but it reduces cost, increases performance, and reduces heat. Some time later, they reengineer the components to add capabilities while maintaining compatibility with the older generation Pii. Eventually (likely years from now, depending on continued exponential growth in Pii demand), they finish the transition with either an entirely updated SoC (multiple ARM Cortex cores from a later generation, a new GPU, the whole enchilada), or an entirely new SoC family that provides compatibility via any number of options, such as having both BCM-2835/6 cores _and_ new circuitry for advanced capabilities. There's only one way to accommodate all of that, and that's the 64-bit, I mean, million-dollar question ... but this is all pure blue-sky, uninformed by Reality, wild-a$$-guessing on my part, for now.

I hope the Foundation can comment on the tech needed to create the Zero so that we don't have to sit out here and wonder ... not that I'm not enjoying playing the game, mind you, nor does it seem anyone else minds! :lol:
The best things in life aren't things ... but, a Pi comes pretty darned close! :D
"Education is not the filling of a pail, but the lighting of a fire." -- W.B. Yeats
In theory, theory & practice are the same - in practice, they aren't!!!

User avatar
Jim Manley
Posts: 1600
Joined: Thu Feb 23, 2012 8:41 pm
Location: SillyCon Valley, California, and Powell, Wyoming, USA, plus The Universe
Contact: Website

Re: Is a new Raspberry Pi model coming?

Sat Nov 28, 2015 10:12 pm

alexeames wrote:As I wrote on my blog, Eben told me last week that not only is there a CPU overclock, but also L2 cache is running faster too (and something about a core running faster, that sadly I forgot the exact details if any that he gave me on that). I very much doubt that there was any monkeying with the silicon on such a now old chip.

Also, I measured the power consumption. The max current draw I managed was 140 mA at 5.09V. http://raspi.tv/2015/raspberry-pi-zero- ... asurements
This was pretty much the same as the model A+
Hey Alex, can you provide a link to the blog post with Eben's comment? He discussed the Zero with you before the launch?

BTW, is the max current draw when running the GPU flat out doing the most complex 3-D transformations possible, enough to fill all of the pipelines? The new DRAM means more objects can be handled by the GPU, and that means more power draw. I think everyone forgets that 99% of the die area in the SoC is the GPU and it's not just generating video if it's being used for its intended Eben-given purpose of slamming 40,000,000 shaded polygons per second into our eyeballs. Yes, I think it's time we start talking about elevating Him to God status, much to the chagrin of Liz and Mooncake, the only cat capable of rolling its eyes :roll:
The best things in life aren't things ... but, a Pi comes pretty darned close! :D
"Education is not the filling of a pail, but the lighting of a fire." -- W.B. Yeats
In theory, theory & practice are the same - in practice, they aren't!!!

User avatar
bstrobl
Posts: 97
Joined: Wed Jun 04, 2014 8:31 pm
Location: Germany

Re: Is a new Raspberry Pi model coming?

Sat Nov 28, 2015 11:35 pm

Jim Manley wrote:It's pretty clear no one has gotten my drift on what my unpublished speculation would be, and that's actually making this more fun. I will say this, don't think about why the BCM-2835 was apparently (and it may have just been a package change, mind you) slightly downsized, and I'm assuming to fit the new DRAM device. Think about what a certain kind of substantial change in die will allow in development of future, much higher-performance-for-same-cost Pi SoCs (plural, over time). First, they adapt the package to fit a new family of DRAM (including BGA connection points) but don't change the internals of the SoC, other than to get the gozintas and gozoutas aligned for the new package.

Next, they do some photolithographic sleight-of-hand to take advantage of the new die capabilities to adapt the BCM-2835/6 with no change in functionality, but it reduces cost, increases performance, and reduces heat. Some time later, they reengineer the components to add capabilities while maintaining compatibility with the older generation Pii. Eventually (likely years from now, depending on continued exponential growth in Pii demand), they finish the transition with either an entirely updated SoC (multiple ARM Cortex cores from a later generation, a new GPU, the whole enchilada), or an entirely new SoC family that provides compatibility via any number of options, such as having both BCM-2835/6 cores _and_ new circuitry for advanced capabilities. There's only one way to accommodate all of that, and that's the 64-bit, I mean, million-dollar question ... but this is all pure blue-sky, uninformed by Reality, wild-a$$-guessing on my part, for now.

I hope the Foundation can comment on the tech needed to create the Zero so that we don't have to sit out here and wonder ... not that I'm not enjoying playing the game, mind you, nor does it seem anyone else minds! :lol:
From what I can tell its probably just a better binned 2835 made to run with a standard clock speed of up to 1Ghz (lower at idle). Hence you see no difference in idle consumption vs the a+ which you would see if were using a better manufacturing process.

You also have to remember that the die is being used in different packages (the 2835 is a bga version of the 2708 die). The die usually ends up being smaller by quite a bit compared to the package, the size of which depends more on the bga grid size and amount of connections needed to the mainboard.

Also, while tiny tweaks at a process node are possible and may have been used to improve the operating margins of the 2835 thats being used in the zero, an improved process node is pretty much out of the question. The costs for shrinking the design, validating it and producing the mask set would pretty much make the pi zero impossible to produce at the price point.

So, as gordon has already stated in a comment in the reveal post, these 2835s have been "bled dry" and dont have much if any overclocking headroom.

Hopefully one of the mods (jamesh ;) ) will correct me if i am wrong.

fruitoftheloom
Posts: 20937
Joined: Tue Mar 25, 2014 12:40 pm
Location: Delightful Dorset

Re: Is a new Raspberry Pi model coming?

Sat Nov 28, 2015 11:37 pm

In case anyone missed it:
Regards the 1Ghz speed, that is actually the maximum the ARM11 was designed for: https://en.wikipedia.org/wiki/ARM11
Retired disgracefully.....

User avatar
bstrobl
Posts: 97
Joined: Wed Jun 04, 2014 8:31 pm
Location: Germany

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 12:26 am

fruitoftheloom wrote:In case anyone missed it:
Regards the 1Ghz speed, that is actually the maximum the ARM11 was designed for: https://en.wikipedia.org/wiki/ARM11
Not quite true, arm11 was designed to surpass 1 Ghz on a 100nm or better process. Nothing stops a chip architecture from going faster other than the constrains of the hardware, and the speed of the electrons on the longest path. In this case the 2835 probably can't feed enough power to the arm core for above 1Ghz operation.

emma1997
Posts: 286
Joined: Sun Nov 08, 2015 7:00 pm
Location: New England (not the old one)

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 12:42 am

alexeames wrote: Is that still true post-Jessie? I was under the impression it was not on by default any more? (We'll need to check that though)
I can verify the latest Raspbian outputs 3v serial at 115kbps. I needed to do this to get the IP address. However I ended up using Minibian which is a truly minimum version of the latest Jessie. Unlike Dietpig and the new "lite" Jessie which IMO are not lite at all.

User avatar
Jim Manley
Posts: 1600
Joined: Thu Feb 23, 2012 8:41 pm
Location: SillyCon Valley, California, and Powell, Wyoming, USA, plus The Universe
Contact: Website

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 2:23 am

bstrobl wrote:From what I can tell its probably just a better binned 2835 made to run with a standard clock speed of up to 1Ghz (lower at idle). Hence you see no difference in idle consumption vs the a+ which you would see if were using a better manufacturing process.
When you say "better binned", what exactly do you mean (sorry, I speak the "stuperior" version of English :lol: ). Do you mean better actually made, or the top of the standard yield (the expensive "vintage", or perhaps "bintage", in this case? :))? They wouldn't be able to toss the 80% (or whatever fraction) of the stuff that didn't perform below 1 GHz and get to this price point - they'd need to be well above 99.99% yield in certified devices, according to the industry stats to which I have access.
bstrobl wrote:You also have to remember that the die is being used in different packages (the 2835 is a bga version of the 2708 die). The die usually ends up being smaller by quite a bit compared to the package, the size of which depends more on the bga grid size and amount of connections needed to the mainboard.
The Zero's SoC is _smaller_ to fit a _larger_ number of BGA points for the _larger_ amount of DRAM addressable, and I'll bet you anything that array can handle a lot more DRAM - I'd guess at least 4 ~ 8 GB. Now, sure, the array could be more dense than on the original 2835 given how long in the tooth that puppy is now, and we'll just have to see.
bstrobl wrote:Also, while tiny tweaks at a process node are possible and may have been used to improve the operating margins of the 2835 thats being used in the zero, an improved process node is pretty much out of the question. The costs for shrinking the design, validating it and producing the mask set would pretty much make the pi zero impossible to produce at the price point.
Ahhh, finally, someone has taken the bait without quite saying so using the five-letter word I was thinking of, but you said "process", not me ("Note entering People's Exhibit #2837 into the record, if it pleases The Court, Your Hono(u)r." :)). Remember that there's nothing saying that you have to use the limit of the photolithography for a given process level - you can build gargantuan discrete power transistors in an 11 nm process fab - it just wouldn't be very cost-effective. Likewise, you can take the masks for say, oh, I don't know, a 40 nm process and use them in, oh, I don't know, say, a 28 nm process and get exactly the same die, albeit probably at a (much?) better yield (in total units produced as well as higher average and peak performance).

That's because there is a continuing increase in optimization knowledge and equipment capabilities as we slide down the slippery process slope, in photolithography, doping chemistry, bake times and temps, cool-down rates, etc., etc., etc. So, one outcome is that you could wind up with a larger number of total certified devices (and hence lower per-unit cost, possibly much lower since you're using larger trace widths), with a, oh, say, 40% increase in performance, and possibly a _reduction_ in heat generation, improved heat dissipation (through more advanced materials and edge design), and even a slight _reduction_ in voltage required, again just due to improved materials and other process advances (this is pure guesswork based on memories from long ago in a fab far, far away ;)).
bstrobl wrote:So, as gordon has already stated in a comment in the reveal post, these 2835s have been "bled dry" and dont have much if any overclocking headroom.
Hopefully one of the mods (jamesh ;) ) will correct me if i am wrong.
Oh, Gordon, Gordon, Gordon, you'd think he has a PhD and decades of experience in this stuff the way people keep bowing, courtseying, and bootlicking in his presence ... what's that? He does? Ohhhh, wellll, that's very different! Never mind! :lol: Remember that he also wears funny hats and clothes that show off his bony physique as he careens around the countryside literally at breakneck speeds, and what kind of level of intellect and judgment does that portray, hmmmm? :o

BTW, there will _always_ be the odd device that overclocks quite nicely much higher than its siblings, the so-called Golden Child, and somewhere Out There is a Zero that can approach the speed and agility of an actual Mitsubishi Zero!!! ;) There are just a lot fewer of them when the yield is nearly uniform (virtually all devices pass certification), just as there will be fewer DOA devices, too, increasing total usable yield.

Taking more/full advantage of the more advanced process are steps that may not yet have been done in production, but somewhere in a test run in a fab right now there could be the next step in the transition baking away, actually using a more advanced process ... and the Foundation ain't saying, and neither am I! 8-) . It has to happen sooner or later, and as we climb the hockey stick on the Pi production curve, that day gets closer and closer ... I can hardly wait ... which may be nearly as long as when anyone can order any quantity of Zeroes and pay $5.00 each for them, delivered :roll:.
The best things in life aren't things ... but, a Pi comes pretty darned close! :D
"Education is not the filling of a pail, but the lighting of a fire." -- W.B. Yeats
In theory, theory & practice are the same - in practice, they aren't!!!

User avatar
rpdom
Posts: 15448
Joined: Sun May 06, 2012 5:17 am
Location: Chelmsford, Essex, UK

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 9:49 am

Jim Manley wrote:The Zero's SoC is _smaller_ to fit a _larger_ number of BGA points for the _larger_ amount of DRAM addressable, and I'll bet you anything that array can handle a lot more DRAM - I'd guess at least 4 ~ 8 GB.
This is obviously incorrect. The BCM2835 can only support a maximum of 1GB of RAM anyway. However, 1GB PoP RAM modules of with the right configuration haven't been available and probably still aren't.
There was no need to alter anything for the Pi Zero. The BCM chip already supported 512MB RAM as has been used in the B and B+ for some time now. No changes required.

The higher clock rate may be down to some new tweaks in settings in the "firmware" (it has been hinted at) and the improved power regulation circuits over the original B (people have commented that the B+ seems more stable on overclock than the B while using the same chip).

Are you certain the package has shrunk? I don't have one to hand to check, but the 2835 is a bit smaller than the 2836. Perhaps you are comparing it to the 2B SoC? I can't see that it would have been worth even a minor redesign/repackage of the 2835 for the Zero.

fruitoftheloom
Posts: 20937
Joined: Tue Mar 25, 2014 12:40 pm
Location: Delightful Dorset

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 10:01 am

rpdom wrote: This is obviously incorrect. The BCM2835 can only support a maximum of 1GB of RAM anyway. However, 1GB PoP RAM modules of with the right configuration haven't been available and probably still aren't.
There was no need to alter anything for the Pi Zero. The BCM chip already supported 512MB RAM as has been used in the B and B+ for some time now. No changes required.

The higher clock rate may be down to some new tweaks in settings in the "firmware" (it has been hinted at) and the improved power regulation circuits over the original B (people have commented that the B+ seems more stable on overclock than the B while using the same chip).

Are you certain the package has shrunk? I don't have one to hand to check, but the 2835 is a bit smaller than the 2836. Perhaps you are comparing it to the 2B SoC? I can't see that it would have been worth even a minor redesign/repackage of the 2835 for the Zero.
+1 totally agree with above from my research and feedback on this forum over the last 4 years


As I stated the reference design of the ARM11 had a maximum 1Ghz speed and the RPF have now managed that, considering that ROKU STB released 2011 had only 600Mhz it is a fantastic achievement to be able to almost double the Mhz....
Retired disgracefully.....

User avatar
bstrobl
Posts: 97
Joined: Wed Jun 04, 2014 8:31 pm
Location: Germany

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 10:40 am

Jim Manley wrote: When you say "better binned", what exactly do you mean (sorry, I speak the "stuperior" version of English :lol: ). Do you mean better actually made, or the top of the standard yield (the expensive "vintage", or perhaps "bintage", in this case? :))? They wouldn't be able to toss the 80% (or whatever fraction) of the stuff that didn't perform below 1 GHz and get to this price point - they'd need to be well above 99.99% yield in certified devices, according to the industry stats to which I have access.
https://en.wikipedia.org/wiki/Product_binning
They can still put the dies that are not 100% stable at 1 GHz in other A+, B, B+ and CM Pis which make up a significant amount of units.
Jim Manley wrote: The Zero's SoC is _smaller_ to fit a _larger_ number of BGA points for the _larger_ amount of DRAM addressable, and I'll bet you anything that array can handle a lot more DRAM - I'd guess at least 4 ~ 8 GB. Now, sure, the array could be more dense than on the original 2835 given how long in the tooth that puppy is now, and we'll just have to see.
That does not make sense. Usually you use a larger BGA to have more connection points for RAM etc. so that you can mount and solder them reliably. The Zero does not have the camera and display connectors, so the BGA can be made smaller here (if, as others have questioned, the package is smaller, which it may not be).
Jim Manley wrote: Ahhh, finally, someone has taken the bait without quite saying so using the five-letter word I was thinking of, but you said "process", not me ("Note entering People's Exhibit #2837 into the record, if it pleases The Court, Your Hono(u)r." :)). Remember that there's nothing saying that you have to use the limit of the photolithography for a given process level - you can build gargantuan discrete power transistors in an 11 nm process fab - it just wouldn't be very cost-effective. Likewise, you can take the masks for say, oh, I don't know, a 40 nm process and use them in, oh, I don't know, say, a 28 nm process and get exactly the same die, albeit probably at a (much?) better yield (in total units produced as well as higher average and peak performance).
You can't use the same Masks for a different node as most lithographic processes have become very different and complicated at different node levels.

https://en.wikipedia.org/wiki/Optical_p ... correction

This isn't the 1970s anymore where huge margins allowed masks to be reused and produced at a much cheaper price.
Jim Manley wrote: That's because there is a continuing increase in optimization knowledge and equipment capabilities as we slide down the slippery process slope, in photolithography, doping chemistry, bake times and temps, cool-down rates, etc., etc., etc. So, one outcome is that you could wind up with a larger number of total certified devices (and hence lower per-unit cost, possibly much lower since you're using larger trace widths), with a, oh, say, 40% increase in performance, and possibly a _reduction_ in heat generation, improved heat dissipation (through more advanced materials and edge design), and even a slight _reduction_ in voltage required, again just due to improved materials and other process advances (this is pure guesswork based on memories from long ago in a fab far, far away ;)).
Each Wafer has to go through 500+ steps in manufacturing. While the process that uses the masks in the steppers is important, others could just as easily be tweaked to improve the perfection of each die. One Intel guy said that Intel had a problem where the amount of argon in the air (around 0.93%) can differ in tiny amounts between fabs and mess up the yield.
Jim Manley wrote: Oh, Gordon, Gordon, Gordon, you'd think he has a PhD and decades of experience in this stuff the way people keep bowing, courtseying, and bootlicking in his presence ... what's that? He does? Ohhhh, wellll, that's very different! Never mind! :lol: Remember that he also wears funny hats and clothes that show off his bony physique as he careens around the countryside literally at breakneck speeds, and what kind of level of intellect and judgment does that portray, hmmmm? :o
Seriously?
Jim Manley wrote: BTW, there will _always_ be the odd device that overclocks quite nicely much higher than its siblings, the so-called Golden Child, and somewhere Out There is a Zero that can approach the speed and agility of an actual Mitsubishi Zero!!! ;) There are just a lot fewer of them when the yield is nearly uniform (virtually all devices pass certification), just as there will be fewer DOA devices, too, increasing total usable yield.

Taking more/full advantage of the more advanced process are steps that may not yet have been done in production, but somewhere in a test run in a fab right now there could be the next step in the transition baking away, actually using a more advanced process ... and the Foundation ain't saying, and neither am I! 8-) . It has to happen sooner or later, and as we climb the hockey stick on the Pi production curve, that day gets closer and closer ... I can hardly wait ... which may be nearly as long as when anyone can order any quantity of Zeroes and pay $5.00 each for them, delivered :roll:.
You will probably have to wait till we get the 2837 or some such. While Fabs do tweak their manufacturing processes to improve yields and perfection, designing a new chip (such as the 2836 which did not even change from the 2835's low-power 40nm process) costs Millions (around $2 million for the 2836).

rpdom wrote: This is obviously incorrect. The BCM2835 can only support a maximum of 1GB of RAM anyway. However, 1GB PoP RAM modules of with the right configuration haven't been available and probably still aren't.
There was no need to alter anything for the Pi Zero. The BCM chip already supported 512MB RAM as has been used in the B and B+ for some time now. No changes required.

The higher clock rate may be down to some new tweaks in settings in the "firmware" (it has been hinted at) and the improved power regulation circuits over the original B (people have commented that the B+ seems more stable on overclock than the B while using the same chip).

Are you certain the package has shrunk? I don't have one to hand to check, but the 2835 is a bit smaller than the 2836. Perhaps you are comparing it to the 2B SoC? I can't see that it would have been worth even a minor redesign/repackage of the 2835 for the Zero.
I believe the 2835 also has a bug in it which was fixed in the 2836, so 512 MB really is the max it can support even if there are 1GB PoP modules available.

User avatar
alexeames
Forum Moderator
Forum Moderator
Posts: 2869
Joined: Sat Mar 03, 2012 11:57 am
Location: UK
Contact: Website

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 2:02 pm

I just put the calipers on my oldest Pi's BCM. 11.96x11.97mm

Then an A+ 12.02 x 12.01mm

And then a Pi Zero 11.97 x 11.99mm

So I put it to the jury that...

The package has not been downsized (unless I have grossly misunderstood what Jim's been talking about - which is of course entirely possible). :lol:
Alex Eames RasPi.TV, RasP.iO

fruitoftheloom
Posts: 20937
Joined: Tue Mar 25, 2014 12:40 pm
Location: Delightful Dorset

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 2:10 pm

alexeames wrote:I just put the calipers on my oldest Pi's BCM. 11.96x11.97mm

Then an A+ 12.02 x 12.01mm

And then a Pi Zero 11.97 x 11.99mm

So I put it to the jury that...

The package has not been downsized (unless I have grossly misunderstood what Jim's been talking about - which is of course entirely possible). :lol:
The BCM2835 is an established SoC so yes JM is mis-taken :shock:
Retired disgracefully.....

User avatar
liz
Raspberry Pi Foundation Employee & Forum Moderator
Raspberry Pi Foundation Employee & Forum Moderator
Posts: 5202
Joined: Thu Jul 28, 2011 7:22 pm
Contact: Website

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 2:57 pm

I can put this one to bed: the BCM2835 hasn't changed between Pis.
Director of Communications, Raspberry Pi

fruitoftheloom
Posts: 20937
Joined: Tue Mar 25, 2014 12:40 pm
Location: Delightful Dorset

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 3:04 pm

liz wrote:I can put this one to bed: the BCM2835 hasn't changed between Pis.
:D :D :D :D :D :D :D :D :D :D :D :D :D
Retired disgracefully.....

User avatar
bstrobl
Posts: 97
Joined: Wed Jun 04, 2014 8:31 pm
Location: Germany

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 3:17 pm

liz wrote:I can put this one to bed: the BCM2835 hasn't changed between Pis.
Can you tell us what has allowed you to raise the baseline speed to 1GHz? Or are you keeping this quiet to announce a future speed upgrade for the B+ and A+ :lol: .

Has anyone who owns a Zero also checked the config file to see if the overclocking options have been used/can be set to something else or is the clock speed baked into the firmware.

Also, just calculated that the H.264 licensing fees probably make up nearly 5% of the Zero's final cost :?

mic_s
Posts: 92
Joined: Sun Oct 26, 2014 4:15 pm

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 4:04 pm

Has anyone who owns a Zero also checked the config file to see if the overclocking options have been used/can be set to something else or is the clock speed baked into the firmware.
MagPi Issue 40, Page 21 :
We compared the boot time of the Raspberry Pi
Zero to the Raspberry Pi 2. How does it fare?

Raspberry Pi 2 :
Boot to command line : 14.5 seconds

Raspberry Pi zero :
Boot to command line : 27.0 seconds

Raspberry Pi zero TURBO OVERCLOCK:
Boot to command line : 23.0 seconds
It's your turn to calculate the exact clock they used in turbo mode.
Try it, no garantee … and … please … let your zero alive.

michael

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 23965
Joined: Sat Jul 30, 2011 7:41 pm

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 4:46 pm

bstrobl wrote:
liz wrote:I can put this one to bed: the BCM2835 hasn't changed between Pis.
Can you tell us what has allowed you to raise the baseline speed to 1GHz? Or are you keeping this quiet to announce a future speed upgrade for the B+ and A+ :lol: .
The knowledge that almost every single 2835 Pi could be overclocked to at least 1Ghz, meant that it was safe to raise base clock speed and voltage. Voltage increase means lower lifespan, but not enough to make any difference (still in decades). It's really just a firmware tweak.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
“I think it’s wrong that only one company makes the game Monopoly.” – Steven Wright

User avatar
bstrobl
Posts: 97
Joined: Wed Jun 04, 2014 8:31 pm
Location: Germany

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 5:09 pm

mic_s wrote: It's your turn to calculate the exact clock they used in turbo mode.
Try it, no garantee … and … please … let your zero alive.

michael
Thanks for that, turbo should be around 1.2 GHz then. Might get a Zero to sand off the cpu and see if I can produce a die shot of the chip. The low price justifies a couple of sacrifices :twisted:
jamesh wrote: The knowledge that almost every single 2835 Pi could be overclocked to at least 1Ghz, meant that it was safe to raise base clock speed and voltage. Voltage increase means lower lifespan, but not enough to make any difference (still in decades). It's really just a firmware tweak.
Will this be propagated to the other models or is it exclusive to the Zero for now?

fruitoftheloom
Posts: 20937
Joined: Tue Mar 25, 2014 12:40 pm
Location: Delightful Dorset

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 5:11 pm

jamesh wrote:
bstrobl wrote:
liz wrote:I can put this one to bed: the BCM2835 hasn't changed between Pis.
Can you tell us what has allowed you to raise the baseline speed to 1GHz? Or are you keeping this quiet to announce a future speed upgrade for the B+ and A+ :lol: .
The knowledge that almost every single 2835 Pi could be overclocked to at least 1Ghz, meant that it was safe to raise base clock speed and voltage. Voltage increase means lower lifespan, but not enough to make any difference (still in decades). It's really just a firmware tweak.
What are the default settings for the RPi0 ??

http://elinux.org/RPiconfig#Tested_values
Retired disgracefully.....

dom
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 5349
Joined: Wed Aug 17, 2011 7:41 pm
Location: Cambridge

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 5:47 pm

fruitoftheloom wrote: What are the default settings for the RPi0 ??

Code: Select all

arm_freq=1000
gpu_freq=300
core_freq=400
over_voltage=6
If you have a Pi1 then feel free to add those settings to config.txt They are very likely (although not guaranteed) to work.

fruitoftheloom
Posts: 20937
Joined: Tue Mar 25, 2014 12:40 pm
Location: Delightful Dorset

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 5:56 pm

dom wrote:
fruitoftheloom wrote: What are the default settings for the RPi0 ??

Code: Select all

arm_freq=1000
gpu_freq=300
core_freq=400
over_voltage=6
If you have a Pi1 then feel free to add those settings to config.txt They are very likely (although not guaranteed) to work.
Thank you.....
Retired disgracefully.....

User avatar
ric96
Posts: 1253
Joined: Sun Mar 17, 2013 6:03 am
Location: NOIDA, India
Contact: Website

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 6:31 pm

alexeames wrote:I just put the calipers on my oldest Pi's BCM. 11.96x11.97mm

Then an A+ 12.02 x 12.01mm

And then a Pi Zero 11.97 x 11.99mm

So I put it to the jury that...

The package has not been downsized (unless I have grossly misunderstood what Jim's been talking about - which is of course entirely possible). :lol:
If you havn't removed the POP RAM then this is meaningless... i do have an actual dead piB with the pop ram removed
https://www.youtube.com/watch?v=0oi53Y6r2SI

I don't have a caliper but just using a scale bcm2835 soc on a b rev2 measured:
9mm x 9mm
My apologies for shameless YouTube Plugs...
youtube.com/sahajsarup
twitter @sahajsarup
skype srics1996
e-mail: sahajsarup@gmail.com
Blog: http://www.geektillithertz.com/wordpress
Web: http://www.geektillithertz.com

fruitoftheloom
Posts: 20937
Joined: Tue Mar 25, 2014 12:40 pm
Location: Delightful Dorset

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 6:39 pm

ric96 wrote:
alexeames wrote:I just put the calipers on my oldest Pi's BCM. 11.96x11.97mm

Then an A+ 12.02 x 12.01mm

And then a Pi Zero 11.97 x 11.99mm

So I put it to the jury that...

The package has not been downsized (unless I have grossly misunderstood what Jim's been talking about - which is of course entirely possible). :lol:
If you havn't removed the POP RAM then this is meaningless... i do have an actual dead piB with the pop ram removed
https://www.youtube.com/watch?v=0oi53Y6r2SI

I don't have a caliper but just using a scale bcm2835 soc on a b rev2 measured:
9mm x 9mm
viewtopic.php?f=63&t=126849&start=200#p852235
Retired disgracefully.....

User avatar
ric96
Posts: 1253
Joined: Sun Mar 17, 2013 6:03 am
Location: NOIDA, India
Contact: Website

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 6:43 pm

fruitoftheloom wrote:
ric96 wrote:
alexeames wrote:I just put the calipers on my oldest Pi's BCM. 11.96x11.97mm

Then an A+ 12.02 x 12.01mm

And then a Pi Zero 11.97 x 11.99mm

So I put it to the jury that...

The package has not been downsized (unless I have grossly misunderstood what Jim's been talking about - which is of course entirely possible). :lol:
If you havn't removed the POP RAM then this is meaningless... i do have an actual dead piB with the pop ram removed
https://www.youtube.com/watch?v=0oi53Y6r2SI

I don't have a caliper but just using a scale bcm2835 soc on a b rev2 measured:
9mm x 9mm
viewtopic.php?f=63&t=126849&start=200#p852235
i know, i did read that.
It's more of like "Just saying". #NoHardFeelings
....oops not twitter :lol:
My apologies for shameless YouTube Plugs...
youtube.com/sahajsarup
twitter @sahajsarup
skype srics1996
e-mail: sahajsarup@gmail.com
Blog: http://www.geektillithertz.com/wordpress
Web: http://www.geektillithertz.com

User avatar
Jim Manley
Posts: 1600
Joined: Thu Feb 23, 2012 8:41 pm
Location: SillyCon Valley, California, and Powell, Wyoming, USA, plus The Universe
Contact: Website

Re: Is a new Raspberry Pi model coming?

Sun Nov 29, 2015 6:43 pm

bstrobl wrote:https://en.wikipedia.org/wiki/Product_binning
They can still put the dies that are not 100% stable at 1 GHz in other A+, B, B+ and CM Pis which make up a significant amount of units.
There's no such thing as "better binning" - the process is just binning (aka sorting in some production environments) and then you can _select_ the "better" parts.
Jim Manley wrote: The Zero's SoC is _smaller_ to fit a _larger_ number of BGA points for the _larger_ amount of DRAM addressable, and I'll bet you anything that array can handle a lot more DRAM - I'd guess at least 4 ~ 8 GB. Now, sure, the array could be more dense than on the original 2835 given how long in the tooth that puppy is now, and we'll just have to see.
bstrobl wrote:That does not make sense. Usually you use a larger BGA to have more connection points for RAM etc. so that you can mount and solder them reliably. The Zero does not have the camera and display connectors, so the BGA can be made smaller here (if, as others have questioned, the package is smaller, which it may not be).
No, you're talking about the bottom of the SoC for CSI and DSI connections - I'm talking about the top of the SoC where the DRAM POP is mounted.
Jim Manley wrote: Ahhh, finally, someone has taken the bait without quite saying so using the five-letter word I was thinking of, but you said "process", not me ("Note entering People's Exhibit #2837 into the record, if it pleases The Court, Your Hono(u)r." :)). Remember that there's nothing saying that you have to use the limit of the photolithography for a given process level - you can build gargantuan discrete power transistors in an 11 nm process fab - it just wouldn't be very cost-effective. Likewise, you can take the masks for say, oh, I don't know, a 40 nm process and use them in, oh, I don't know, say, a 28 nm process and get exactly the same die, albeit probably at a (much?) better yield (in total units produced as well as higher average and peak performance).
bstrobl wrote:You can't use the same Masks for a different node as most lithographic processes have become very different and complicated at different node levels.
https://en.wikipedia.org/wiki/Optical_p ... correction
This isn't the 1970s anymore where huge margins allowed masks to be reused and produced at a much cheaper price.
I should have said "mask files/libraries" since no one actually manipulates/modifies actual mask models any more (e.g., the cutting and gluing of ruby mask Mylar in the step just before photoreduction to produce the final size masks used in production in Ye Olde Dayes). You can certainly use files/libraries for larger node processes in smaller node processes, and normally you wouldn't, but this isn't a normal situation. I'm talking about how you would transition incrementally to initially reduce risk while increasing volume and possibly performance. OPC only matters when going down in actual trace width, not when using a larger trace width than a given node level supports, IIRC.
Jim Manley wrote: That's because there is a continuing increase in optimization knowledge and equipment capabilities as we slide down the slippery process slope, in photolithography, doping chemistry, bake times and temps, cool-down rates, etc., etc., etc. So, one outcome is that you could wind up with a larger number of total certified devices (and hence lower per-unit cost, possibly much lower since you're using larger trace widths), with a, oh, say, 40% increase in performance, and possibly a _reduction_ in heat generation, improved heat dissipation (through more advanced materials and edge design), and even a slight _reduction_ in voltage required, again just due to improved materials and other process advances (this is pure guesswork based on memories from long ago in a fab far, far away ;)).
bstrobl wrote:Each Wafer has to go through 500+ steps in manufacturing. While the process that uses the masks in the steppers is important, others could just as easily be tweaked to improve the perfection of each die. One Intel guy said that Intel had a problem where the amount of argon in the air (around 0.93%) can differ in tiny amounts between fabs and mess up the yield.
OK, it doesn't sound like you've worked in a fab or done process design, because there is no such thing as "just as easily be tweaked" when it comes to changing process steps. An Intel engineer didn't properly document just one step (gas/plasma exposure time and/or temperature) in one of the early Pentium products and it took them weeks of trial-and-error in the R&D fab to recover it. CEO Andy Grove was down in the fab every day looking over the techs' and engineers' shoulders monitoring and commenting on prospective changes because they were losing tens of millions of dollars a day in potential revenue during the delay. Your example about the argon concentration demonstrates that in spades.
Jim Manley wrote: Oh, Gordon, Gordon, Gordon, you'd think he has a PhD and decades of experience in this stuff the way people keep bowing, courtseying, and bootlicking in his presence ... what's that? He does? Ohhhh, wellll, that's very different! Never mind! :lol: Remember that he also wears funny hats and clothes that show off his bony physique as he careens around the countryside literally at breakneck speeds, and what kind of level of intellect and judgment does that portray, hmmmm? :o
bstrobl wrote:Seriously?
NO! That's what the giant emoticons are about! Gordon and I are fellow cyclists and look much alike, except that I'm way taller and much better looking ... because I have 20/10 eyesight ... and only when I'm in something other than Lycra Spandex, else women's and children's gazes must be averted lest they become horrified at my visage. You critics really need to take a chill pill and stop looking at the world through your darkened glasses as if everything said by others is meant in the dour tones you use to describe the world. Sheesh!
Jim Manley wrote:Taking more/full advantage of the more advanced process are steps that may not yet have been done in production, but somewhere in a test run in a fab right now there could be the next step in the transition baking away, actually using a more advanced process ... and the Foundation ain't saying, and neither am I! 8-) . It has to happen sooner or later, and as we climb the hockey stick on the Pi production curve, that day gets closer and closer ... I can hardly wait ... which may be nearly as long as when anyone can order any quantity of Zeroes and pay $5.00 each for them, delivered :roll:.
bstrobl wrote:You will probably have to wait till we get the 2837 or some such. While Fabs do tweak their manufacturing processes to improve yields and perfection, designing a new chip (such as the 2836 which did not even change from the 2835's low-power 40nm process) costs Millions (around $2 million for the 2836).
Well, DUH! However, we're now at a level of volume where that can be supported, especially if the Zero can increase the market to tens of millions of systems fairly quickly. In educational institutions, the shift from the $20 Model A to the $5 Zero is going to mean a "YUUUUGE" (to quote a current Presidential candidate ;)) change in the number of systems that will be able to be acquired for students. It will be to the point where, not only will every student who wants one be able to have one regardless of income status, but they'll be able to take them home to do whatever their little hearts please with them. THAT'S when education really happens, not when listening to some old sod/soddess blathering on in a lecture, or stepping through some boring lab concocted by an educational theoretician. Their younger siblings will see that and it will rub off on them, and even their parents can learn ... if they spend the time with the kids which, sadly, is really the number one reason for declining academic performance.

Then, there are the industrial controller and embedded products markets that have found the Pi to be an outstanding platform for their purposes at a cost-busting price point that's only going to accelerate demand for these puppies, especially as they get smaller and more capable, as well as cheaper. This was a totally unforeseen market when the Foundation started all of this, but a very fortuitous one that is helping drive the exponential growth of Pi sales. I wouldn't be surprised at all if needs of that market influenced the decision to go with production of the Zero, as it's deep in the middle of their sweet spot.
rpdom wrote:This is obviously incorrect. The BCM2835 can only support a maximum of 1GB of RAM anyway. However, 1GB PoP RAM modules of with the right configuration haven't been available and probably still aren't. There was no need to alter anything for the Pi Zero. The BCM chip already supported 512MB RAM as has been used in the B and B+ for some time now. No changes required.

The higher clock rate may be down to some new tweaks in settings in the "firmware" (it has been hinted at) and the improved power regulation circuits over the original B (people have commented that the B+ seems more stable on overclock than the B while using the same chip).
When I was talking about increased DRAM addressing capability, I was referring to future devices, not the 2835/6, and since I can't access any info on the Elpida/Micron DRAM device, I don't know whether it's the youngest sibling of a family of devices with the same, or compatible, BGA layout, which is typical for a new device like this. A Foundation member specifically said that they had to rejigger the BGA layout on the Zero's SoC to accommodate this, which is what started me thinking down this path.
bstrobl wrote:I believe the 2835 also has a bug in it which was fixed in the 2836, so 512 MB really is the max it can support even if there are 1GB PoP modules available.
Again, I'm talking about future SoCs, not the 2835/6, and I'm pretty sure that the Foundation/Broadcom/Element 14 wouldn't have made the change to the BGA just for the Zero - what if everyone had just turned their nose up and said, "No thanks, we want something better-performing than the Model 2B, not with fewer overall features than the Model A+."? We won't know what the Pi roadmap is until each step is revealed, but knowing Eben and the rest of the team, I'm very confident that there are strategic steps being made that are going to have even more profound effects than the astounding price and increased performance and DRAM over the Model A+ embodied in the Zero.

The Zero is just probably the first step among a number of equally-if-not-more-amazing steps that are going to unfold over the coming years, and even if I'm wrong about the specifics (and I couldn't have included more caveats than blue-sky, wild-a$$-guessing, etc.), it's clear (at least to me, but what do I know?) that the market expansion the Zero is making possible has already started a seismic shift at what can be done at the lowest levels of computing.
rpdom wrote:Are you certain the package has shrunk? I don't have one to hand to check, but the 2835 is a bit smaller than the 2836. Perhaps you are comparing it to the 2B SoC? I can't see that it would have been worth even a minor redesign/repackage of the 2835 for the Zero.
Well, you know what, that's exactly what I did, and I wasn't aware that there was an increase in size in the 2836 over the 2835, nor would I have expected one. Except that now that I think about it, that's how they got the 2836 connected to the 1 GB DRAM device on the underside of the 2B - they had to add addressing lines to get to 1 GB - DOH!!! OK, where's the forehead-palm-slap emoticon? :oops: I guess the "embarrassed" one will have to do, for now.

OK, as I have done before, I will admit that I screwed that one up. However, my line of reasoning still stands further down the line in that the Foundation clearly has to maintain compatibility with the 2835 for a very long time, but at the same time, they have to move down to a more capable node/pitch size in order to expand the feature set, improve yield, increase performance, further reduce costs long-term, etc. For the moment, that's more expensive, but that probably won't be true much longer for 28 nm. I don't think anyone here doubts that a 64-bit Pi is somewhere off in the future, no doubt years away, but not that many considering what's happened just in the last year with the 2B and now the Zero.

If sales continue to increase exponentially, the Foundation, and Broadcom let's not forget, will have increased latitude to pursue even more mind-bending stunts of prestidigitation. It should also be noted that all of this sort of effort can be written off for tax purposes, not only because it's R&D, but because it's benefiting a charitable educational cause. All I'm saying is "Watch this space" and think about what could be in the offing eventually, if not this week. Once again, all caveats apply as this is all blue-sky, WAG-and-a-half, what-if, just-brain-farting-out-loud activity on a cold and rainy day (hmmm, maybe I need to lay off the cold medication a bit! ... :lol: ).
The best things in life aren't things ... but, a Pi comes pretty darned close! :D
"Education is not the filling of a pail, but the lighting of a fire." -- W.B. Yeats
In theory, theory & practice are the same - in practice, they aren't!!!

Return to “Off topic discussion”