Maximum overclock varies per individual Pi


12 posts
by kraades » Sun Mar 31, 2013 8:34 am
I have seen quite a lot of Pi's and some (most) of them will gladly run on a maximum overclock while others cannot stand any overclock at all (any overclock will almost immediately cause sd card corruption).

Because the hardware seems to vary, It makes it very hard to give advise to fellow Pi owners!

Is this an known issue?

BTW
Tested with the same power supply, same sd card (restored from backup), same everything. Only a different Pi (mostly from the same batches).
Posts: 6
Joined: Wed Jul 18, 2012 8:03 am
by joan » Sun Mar 31, 2013 8:36 am
Yes.

Overclocking/overvolting is the luck of the draw.
User avatar
Posts: 6337
Joined: Thu Jul 05, 2012 5:09 pm
Location: UK
by jamesh » Sun Mar 31, 2013 8:43 am
The advice to owners is simply that the chip is rated (and sold) for 700Mhz, but you can try overclocking if you want. It is, as Joan said, the luck of the draw. Most chips can be pushed, occasionally they cannot.
Volunteer at the Raspberry Pi Foundation, helper at Picademy September and October 2014.
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 12130
Joined: Sat Jul 30, 2011 7:41 pm
by malakai » Sun Mar 31, 2013 8:48 am
Similar posts have been around since September http://www.raspberrypi.org/phpBB3/viewtopic.php?f=63&t=17847 not sure if I would define it as a known issue. As each board will perform differently than the one before. Any computer like device should have variances upon performance / stability to within a degree as the overall process of manufacturing memory, cpu's, and the like have variances that cannot be controlled. I think you will even find varying corruption and or performance among your best 2 performing boards. Such as the time at which they corrupt or exhibit some issue should vary as well.
http://www.raspians.com - always looking for content feel free to ask to have it posted. Or sign up and message me to become a contributor to the site. Raspians is not affiliated with the Raspberry Pi Foundation. (RPi's + You = Raspians)
User avatar
Posts: 1383
Joined: Sat Sep 15, 2012 10:35 am
by 4thdwarflord » Fri Apr 05, 2013 6:58 pm
Why do you think it is rated at 700Mhz and not higher? It's because of this reason.
Posts: 142
Joined: Mon Dec 03, 2012 4:26 pm
Location: Deep in the mines of Moria
by LemmeFatale » Sat Apr 06, 2013 4:37 pm
It's well worth noting that this isn't something that's Raspberry Pi-specific - it's the case with all CPUs. There are just natural variations in manufacture, as with anything else. :P

EDIT: And malakai already explained that much better than I did. I missed that - sorry!
Classic - Raspberry Pi Model B (512MB) with Motorola Atrix Lapdock
Lemcon-One - Raspberry Pi Model B (256MB) PiMAME TV-Box
User avatar
Posts: 254
Joined: Fri Feb 01, 2013 8:47 pm
Location: UK
by Jim Manley » Sat Apr 06, 2013 6:39 pm
This is what we educators call a teachable moment. Others have pointed out that it's the luck of the draw as to how far a given Pi can be overclocked, but you may be interested as to why that's the case. First, you should look at the word "overclock" - as the first two syllables suggest, overclocking means exceeding the rated speed of a device, which is 700 MHz, and no more, for the Pi. Any additional performance you may obtain is a statistical anomaly and this is true of any mass-manufactured device, whether it's based on a semiconductor process or anything else.

You may be surprised to find out that when any semiconductor device is produced that performs at a given level of performance (speed, in this case), it's one of a number (hundreds, possibly thousands) of other devices that may have higher, lower, or equal performance that can be in direct proximity to the device in question, usually on a semiconductor wafer. That's a very thin, circular disk up to 300 mm (~12inches) in diameter originally made of 99.9999999% pure silicon or other material with similar properties, such as germanium (first used for semiconductors due to its naturally-occurring properties) or gallium arsenide, used for extremely high-speed devices with high radiation immunity (also particularly fragile and expensive, only appropriate for highest-end military and space applications).

So, a single wafer, which is manufactured exactly like many thousands of others, actually has devices over the full range of possible levels of performance from worst to best. There aren't separate manufacturing lines for CPUs with lower, medium, and higher clock speeds in otherwise-identical models - they all come from the same manufacturing process. It's only through testing of each device that its performance level actually becomes known. Very few will reach the highest speed (around 1 GHz for the Broadcom BCM2835 used in the Pi) and a small number will only reach the rated 700 MHz, but probably half will perform at a speed of at least 850 MHz.

The statistical distribution of the performance of individual devices depends on many factors, including dozens to hundreds of discrete steps in the fabrication of each device, depending on its complexity. Any slight variance in temperature, time, material purity and concentration, etc., at the location of a device during a specific manufacturing step on a wafer can disproportionately affect the performance of that device. There is no way to predict what the performance of any particular device will be - it's strictly a numbers game and only testing can ascertain what attributes such as speed actually are.

Your mileage (per hour, or instruction cycles per second in this case) not only may vary, it will!
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
Posts: 1358
Joined: Thu Feb 23, 2012 8:41 pm
Location: SillyCon Valley, California, USA
by W. H. Heydt » Sat Apr 06, 2013 9:56 pm
Jim gave a good explanation of it all. There are a couple of other points worth making...

The manufacturing process is a game of numbers. You want to get the maximum number of "good" chips from each wafer. "Good" meaning (a) they work, (b) they work at the specified performance that you sell them on, and (c) you're not making piles of chips that perform so well that you could sell them as being better than you claim.

If you violate rule (a), you wind up throwing away a lot of chips, wasting time (mostly) and resources (some).

If you violate (b)...well it depends on how you sell your chips. If you sell a "family" of chips with different speed ratings, you test each chip to find out how fast it will run reliably and mark the package for that speed and adjust the price accordingly. If you have too many chips that run really fast, you mark some of them for a lower speed and sell them for what you can get for them (Intel is rather famous for this).

If you violate (c), you might want to consider increasing the price and selling the chips as being faster than you originally claimed.

Broadcom is selling the BCM2835 as a 700MHz chip. That means that every chip that goes out the door has to be tested to run *reliably* at 700MHz. Since they are not (at least to the Pi manufacturers) selling BCM2835s rated faster than 700MHz, they will not claim that you can count on getting them to run faster. Since they want to get as many good 700MHz BCM2835 chips off each wafer as possible, it's good practice to do the design and manufacturing actually targeting a bit higher speed, say (picking an arbitrary number out of the air), 800MHz. That way, the *relative* "duds" that actually work will probably still be good at 700MHz, and can be sold, but they're still not selling lots and lots of 1GHz chips marked for 700MHz.
Posts: 1783
Joined: Fri Mar 09, 2012 7:36 pm
Location: Vallejo, CA (US)
by LemmeFatale » Sat Apr 06, 2013 10:02 pm
Jim Manley wrote:First, you should look at the word "overclock" - as the first two syllables suggest, overclocking means exceeding the rated speed of a device, which is 700 MHz, and no more, for the Pi. Any additional performance you may obtain is a statistical anomaly and this is true of any mass-manufactured device, whether it's based on a semiconductor process or anything else.

You'd be amazed (or, alternatively, probably won't be amazed :P) at how many people forget this.

Long ago, I recall once seeing a person suggest that they would keep requesting RMAs on a particular product (which is of course unfair on the vendor, because it costs them, and it's just wrong to do this with perfectly-working hardware, as this person claimed they would) until they got a unit that would overclock to the particular speed that they wanted it to! (I can only assume that they were of the impression that they had some sort of right to a higher-than-rated speed.)
Classic - Raspberry Pi Model B (512MB) with Motorola Atrix Lapdock
Lemcon-One - Raspberry Pi Model B (256MB) PiMAME TV-Box
User avatar
Posts: 254
Joined: Fri Feb 01, 2013 8:47 pm
Location: UK
by liz » Sat Apr 06, 2013 10:12 pm
We have a framed wafer of 2835s in the office - I'll try to remember to take a picture of it next week so that "teachable moment" Jim mentions can carry on a bit longer!
--
Head of Comms, Raspberry Pi Foundation
User avatar
Raspberry Pi Foundation Employee & Forum Moderator
Raspberry Pi Foundation Employee & Forum Moderator
Posts: 4141
Joined: Thu Jul 28, 2011 7:22 pm
by Jim Manley » Sat Apr 06, 2013 11:34 pm
liz wrote:We have a framed wafer of 2835s in the office - I'll try to remember to take a picture of it next week so that "teachable moment" Jim mentions can carry on a bit longer!

Well, that does it ... now I'm just going to have to visit sooner, rather than later, to see it in person unless someone can be convinced to sneak another wafer out of the fab to become at least a temporary-loan artifact for a certain Computer History Museum at 1401 N. Shoreline Blvd., Mountain View, CA 94043 ;)

James and Gert each posted about bringing a BCM2835 wafer to a Jam - is it the same as the one in the office, or are those so plentiful in Cambridge that the streets are now paved with them? :D
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
Posts: 1358
Joined: Thu Feb 23, 2012 8:41 pm
Location: SillyCon Valley, California, USA
by jamesh » Sun Apr 07, 2013 7:45 am
Jim Manley wrote:
liz wrote:We have a framed wafer of 2835s in the office - I'll try to remember to take a picture of it next week so that "teachable moment" Jim mentions can carry on a bit longer!

Well, that does it ... now I'm just going to have to visit sooner, rather than later, to see it in person unless someone can be convinced to sneak another wafer out of the fab to become at least a temporary-loan artifact for a certain Computer History Museum at 1401 N. Shoreline Blvd., Mountain View, CA 94043 ;)

James and Gert each posted about bringing a BCM2835 wafer to a Jam - is it the same as the one in the office, or are those so plentiful in Cambridge that the streets are now paved with them? :D


I did take a wafer to the first Jam, but it wasn't a 2835 one - in fact, not sure what it was - old I think! I did have some unpackaged chips as well (i.e. cut out from the wafer but not in the black resin package we know and love), but again, they were not 2835's!
Volunteer at the Raspberry Pi Foundation, helper at Picademy September and October 2014.
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 12130
Joined: Sat Jul 30, 2011 7:41 pm