portets
Posts: 186
Joined: Sat Oct 29, 2011 6:24 am

Re: Overclocking

Tue Jul 31, 2012 3:38 am

@Dom -After looking at my benchmark, there wasn't much difference at all. With some tests I found that ssv-membench is actually cpu bottlenecked on the pi. arm_freq at 980 made the ram benchmarks improve dramatically up to sdram_freq=600. After that, I get no improvement. I haven't heard anyone else having the same issue I'm having where ram clock can't crash the system.

Do I have a hardware fault to do with ram clock?

dom
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 5410
Joined: Wed Aug 17, 2011 7:41 pm
Location: Cambridge

Re: Overclocking

Tue Jul 31, 2012 9:00 am

portets wrote: Do I have a hardware fault to do with ram clock?
No, I think the sdram freq tops out at 600MHz (all overclock values have a max frequency limit before the PLLs go out of spec, normally something else fails before you reach that). I'll have a look at the code and see if that restriction can be removed.

adamdbz
Posts: 15
Joined: Thu Jul 19, 2012 8:40 am

Re: Overclocking

Tue Jul 31, 2012 4:01 pm

Hi
I can easily overclock too 900mhz in raspbian
But in xbmc it crashes
Any idea why?

-Adam Ericson

modin
Posts: 1
Joined: Tue Jul 31, 2012 4:15 pm

Re: Overclocking

Tue Jul 31, 2012 4:20 pm

I somehow can´t seem to get the "time echo "scale=2000;4*a" | bc -l" command to work after i reinstalled Raspbian on my flash card. Any tips for how i can get this burn in test up and running again?

User avatar
AndrewS
Posts: 3625
Joined: Sun Apr 22, 2012 4:50 pm
Location: Cambridge, UK
Contact: Website

Re: Overclocking

Tue Jul 31, 2012 4:43 pm

adamdbz wrote:I can easily overclock too 900mhz in raspbian
But in xbmc it crashes
Any idea why?
Possibly because xbmc makes heaver use of the GPU? Or maybe those two distros are using different versions of the firmware/kernel?

User avatar
Lob0426
Posts: 2198
Joined: Fri Aug 05, 2011 4:30 pm
Location: Susanville CA.
Contact: Website

Re: Overclocking

Tue Jul 31, 2012 5:47 pm

Most likely the GPU. I can overclock to 850 in everything except Quake3 and Openarena, then it crashes. Did you try to raise your gpu_freq also? that will crash mine at anything over 320 (and others are saying the same) in the games. Works fine for everything else. This is without any overvolting. I am going to do some overvolting when I get the other two RasPii tomorrow. Did not want to risk the only RasPi I had, even though @dom says it is relatively safe, I do not have that good of luck sometimes. :oops:
512MB version 2.0 as WordPress Server
Motorola Lapdock with Pi2B
Modded Rev 1.0 with pin headers at USB

http://rich1.dyndns.tv/
(RS)Allied ships old stock to reward its Customers for long wait!

FX4
Posts: 64
Joined: Fri Jul 27, 2012 2:39 pm

Re: Overclocking

Tue Jul 31, 2012 5:48 pm

adamdbz wrote:Hi
I can easily overclock too 900mhz in raspbian
But in xbmc it crashes
Any idea why?

-Adam Ericson
Then you are not successfully overclocking to 900MHz. You are just getting lucky until you hit the right instruction set or process the wrong data. You need to back both off. Either it is stable at x frequency or it is not. It's not a kind of sort of thing. Mine is 100% stable at 933Mhz, I back it off to 925MHz for a little head room.

shalo
Posts: 74
Joined: Tue May 08, 2012 7:25 pm

Re: Overclocking

Tue Jul 31, 2012 8:11 pm

To add to these recent replies, I recently received my second pi which means I can actually play a little since the first became a workhorse pretty quickly.

I've been using q3 as a test and I want to find the non-overvolted limit first.

I was able to run benchmarks and such at 920mhz ARM, however not truly stable. In fact I retreated all the way down where I saw varying successes. I've now see 900mhz last over an hour. 890 and 880, I've seen 2hrs with no crash only to fail in later retests. I thought I had found the limit at 875mhz ARM where Q3 ran in loops successfully for 12hrs. However, I've now seen a crash at 875 after 32mins of a retest.

I can see now that I have tainted the limited gpu testing I had done which were all done at 890mhz ARM when I had originally assumed 2hrs of stability was enough. However I am going to overvolt before experimenting with the gpu further since even 700arm/280gpu was immediately unstable in q3 and 700/200gpu/300core unstable also.

Things I've noticed with q3 (on raspbian). You nearly always need to issue at least one /snd_restart or you get crackly audio (over analogue anyway) and this seems to affect the fps result. To benchmark you want to go with /timedemo 1 and then /demo four.

To have it loop demos you can /exec demoloop.cfg if you put the following commands in a file with that name and put it in /.q3a/baseq3

timedemo 1
set loop1 "demo four;wait;set nextdemo vstr loop2"
set loop2 "demo four;wait;set nextdemo vstr loop1"
vstr loop1

FX4
Posts: 64
Joined: Fri Jul 27, 2012 2:39 pm

Re: Overclocking

Tue Jul 31, 2012 10:11 pm

If you are going to burn in I would suggest a heat synch. That is a lot of stress on the chip. I know it's a popular way to burn in overclocked PCs but these little chips get hot under load with no way to dissipate the heat you will slowly degrade the chip. You need to find a program that will monitor the chips operating temp and see where it stabilizes vs the manufacturer spec. You may have to use a temp probe. I don't know if these ARM chips have built in temp monitoring.

User avatar
Lob0426
Posts: 2198
Joined: Fri Aug 05, 2011 4:30 pm
Location: Susanville CA.
Contact: Website

Re: Overclocking

Tue Jul 31, 2012 11:02 pm

FX4 wrote:If you are going to burn in I would suggest a heat synch. That is a lot of stress on the chip. I know it's a popular way to burn in overclocked PCs but these little chips get hot under load with no way to dissipate the heat you will slowly degrade the chip. You need to find a program that will monitor the chips operating temp and see where it stabilizes vs the manufacturer spec. You may have to use a temp probe. I don't know if these ARM chips have built in temp monitoring.
There are several threads that discuss heat sinks. It is the considered opinion, of knowledgeable members of the design team, that heat sinks are not really nesessary.

My opinion is that if you intend to run them in a hot environment, such as a car installation or a closed case, or other environment other than at room temperature, you should think about a heat sink or two. There has been much discussion about the topic. These Cambridge people that live at 69F (20.5C) all year round just do not agree with me :lol: It is 101F (38.3C) outside right now. Do not want to know how hot it is in the car right now. :D Guess that is why it is called sunny California! 8-)

@dom has been running his overvolted and overclocked for at least half a year so far and no heat sinks.

But I will keep trying to bring him and @liz over to the darkside with heatsinks in multiple colors. The "bluey" ones someone posted almost brought her over. :twisted:
512MB version 2.0 as WordPress Server
Motorola Lapdock with Pi2B
Modded Rev 1.0 with pin headers at USB

http://rich1.dyndns.tv/
(RS)Allied ships old stock to reward its Customers for long wait!

User avatar
AndrewS
Posts: 3625
Joined: Sun Apr 22, 2012 4:50 pm
Location: Cambridge, UK
Contact: Website

Re: Overclocking

Tue Jul 31, 2012 11:16 pm

Lob0426 wrote:But I will keep trying to bring him and @liz over to the darkside with heatsinks in multiple colors. The "bluey" ones someone posted almost brought her over. :twisted:
In another thread someone suggested diamond heatsinks ;)

FX4
Posts: 64
Joined: Fri Jul 27, 2012 2:39 pm

Re: Overclocking

Wed Aug 01, 2012 1:35 am

Lob0426 wrote:
FX4 wrote:If you are going to burn in I would suggest a heat synch. That is a lot of stress on the chip. I know it's a popular way to burn in overclocked PCs but these little chips get hot under load with no way to dissipate the heat you will slowly degrade the chip. You need to find a program that will monitor the chips operating temp and see where it stabilizes vs the manufacturer spec. You may have to use a temp probe. I don't know if these ARM chips have built in temp monitoring.
There are several threads that discuss heat sinks. It is the considered opinion, of knowledgeable members of the design team, that heat sinks are not really nesessary.

My opinion is that if you intend to run them in a hot environment, such as a car installation or a closed case, or other environment other than at room temperature, you should think about a heat sink or two. There has been much discussion about the topic. These Cambridge people that live at 69F (20.5C) all year round just do not agree with me :lol: It is 101F (38.3C) outside right now. Do not want to know how hot it is in the car right now. :D Guess that is why it is called sunny California! 8-)

@dom has been running his overvolted and overclocked for at least half a year so far and no heat sinks.

But I will keep trying to bring him and @liz over to the darkside with heatsinks in multiple colors. The "bluey" ones someone posted almost brought her over. :twisted:
Well you are certainly free to do as you choose. I'll just leave this as I used to be a manufacturing engineer in the microprocessor industry and you have my professional opinion.

FX4
Posts: 64
Joined: Fri Jul 27, 2012 2:39 pm

Re: Overclocking

Wed Aug 01, 2012 3:06 am

I just want to add a couple of things here because I am concerned others are going to willy nilly overclock and overvolt their RPi in an attempt to get maximum performance. Then turn around and perform severe burn in testing. Both increasing clock speed and increasing voltage will cause PN junction temperatures to rise, this can in certain cases cause thermal runaway and infant mortality of the processor. It can also cause what is known as a race condition. This will mostly just result in messed up calculations but I have seen race conditions cause shorts in an asic leading to failure. If you are playing around by upping voltage and clock speed you are moving the processor out of its design parameters. Most chips have some headroom in the rated specification that is safe to play, but honestly in this case there are probably just a handful of engineers at Broadcom that really know where this chip will meltdown and fail. The easiest way to keep things thermally under control is to just add a heat synch properly bonded to the processor. If I were pushing the limits I would also want to monitor core die temperature. Short of that a temp probe on the hottest spot on the chip package. Of course on the other hand it is only $35.00, play away.

Flojer0
Posts: 31
Joined: Sun Jun 10, 2012 8:10 am

Re: Overclocking

Wed Aug 01, 2012 4:46 am

FX4 wrote:I just want to add a couple of things here because I am concerned others are going to willy nilly overclock and overvolt their RPi in an attempt to get maximum performance. Then turn around and perform severe burn in testing. Both increasing clock speed and increasing voltage will cause PN junction temperatures to rise, this can in certain cases cause thermal runaway and infant mortality of the processor. It can also cause what is known as a race condition. This will mostly just result in messed up calculations but I have seen race conditions cause shorts in an asic leading to failure. If you are playing around by upping voltage and clock speed you are moving the processor out of its design parameters. Most chips have some headroom in the rated specification that is safe to play, but honestly in this case there are probably just a handful of engineers at Broadcom that really know where this chip will meltdown and fail. The easiest way to keep things thermally under control is to just add a heat sink properly bonded to the processor. If I were pushing the limits I would also want to monitor core die temperature. Short of that a temp probe on the hottest spot on the chip package. Of course on the other hand it is only $35.00, play away.
First, you spelled heat sink wrong. Second, seeing as the CPU is hiding under its own ram I believe a temp probe would be useless. And third, Eben himself has said that due to the low powered nature of the SoC overvoltage is a MUCH larger threat to the life of the chip than heat. The whole system doesn't even consume 4W for pete's sake.

I almost want to order a second pi and overclock/overvolt the balls off of it just to see if I can kill it in a reasonable amount of time. :)

User avatar
Lob0426
Posts: 2198
Joined: Fri Aug 05, 2011 4:30 pm
Location: Susanville CA.
Contact: Website

Re: Overclocking

Wed Aug 01, 2012 5:50 am

FX4 wrote:I but honestly in this case there are probably just a handful of engineers at Broadcom that really know where this chip will meltdown and fail.
Some of those "engineers" started the Raspberry Pi foundation and some more were part of the design team for the Raspberry PI. Heat Sinks are "optional" equipment! But I have them on mine. There is a post in a thread claiming that heat sinks helped him up his overclock speed. A lot more testing would be needed to gain any data as whether this really was the case. There are people running at 8 overvolt, with no heat sinks at all, and they do not report any heat related problems. There really is no real data as to the absolute life of a Raspberry Pi. The device that has been using it (the BCM2835)longest is the Roku2. They do not have any heat sinks on the Roku2. It was introduced around July 2011. So there really is no track record to speak of yet.
512MB version 2.0 as WordPress Server
Motorola Lapdock with Pi2B
Modded Rev 1.0 with pin headers at USB

http://rich1.dyndns.tv/
(RS)Allied ships old stock to reward its Customers for long wait!

JRT
Posts: 28
Joined: Tue Jul 17, 2012 2:44 pm

Re: Overclocking

Wed Aug 01, 2012 9:33 am

I'm going to try overclocking tonight and would appreciate a little advice. This thread is a bit long to find what I'm looking for easily!

So, to overclock I just use "arm_freq" to set the cpu speed to ~800mhz (as a starting point)?

This is where it gets confusing and I need help - what's the relationship between that setting and all the others (if any). I'm aware (vaguely) that there's some relationship between all the settings, but I'm not sure what it is... Is it recommended to set "gpu_freq", "sdram_freq" and "core_freq", etc to anything other than the defaults? What effect will this have?

Sorry for all the questions! Thanks.

portets
Posts: 186
Joined: Sat Oct 29, 2011 6:24 am

Re: Overclocking

Wed Aug 01, 2012 10:10 am

JRT wrote:I'm going to try overclocking tonight and would appreciate a little advice. This thread is a bit long to find what I'm looking for easily!

So, to overclock I just use "arm_freq" to set the cpu speed to ~800mhz (as a starting point)?
Yeah, arm_freq=800 is a good starting point. Remember over-volting voids the warranty, but overclocking doesn't.
The default sdram_clock=400, try 450 or so and work up after playing with arm_clock. The only clocks that have a relationship are core_freq and gpu_freq(gpu_freq sets a few others, but only worry about gpu_freq). core_freq is 250, gpu_freq is also 250. core_freq should always be equal to or twice gpu_freq. Or if 2x core_freq is a number divisible by 3, you can set that for gpu_freq.

For example: core_freq=420
2x420 = 840
840/3 = 280
gpu_freq=280

robwriter
Posts: 114
Joined: Sat Feb 25, 2012 6:26 pm

Re: Overclocking

Wed Aug 01, 2012 10:17 am

Surely it's a fairly simple question as to whether the overclocking limit of the Pi is heat limited or limited by the chip design. If its not heat limited then heat sinks are a waste of time.

JRT
Posts: 28
Joined: Tue Jul 17, 2012 2:44 pm

Re: Overclocking

Wed Aug 01, 2012 10:43 am

portets wrote:
JRT wrote:I'm going to try overclocking tonight and would appreciate a little advice. This thread is a bit long to find what I'm looking for easily!

So, to overclock I just use "arm_freq" to set the cpu speed to ~800mhz (as a starting point)?
Yeah, arm_freq=800 is a good starting point. Remember over-volting voids the warranty, but overclocking doesn't.
The default sdram_clock=400, try 450 or so and work up after playing with arm_clock. The only clocks that have a relationship are core_freq and gpu_freq(gpu_freq sets a few others, but only worry about gpu_freq). core_freq is 250, gpu_freq is also 250. core_freq should always be equal to or twice gpu_freq. Or if 2x core_freq is a number divisible by 3, you can set that for gpu_freq.

For example: core_freq=420
2x420 = 840
840/3 = 280
gpu_freq=280
Perfect - thanks a lot! I won't be doing many graphics-intensive things (at first) so I'll probably stick to just arm_freq for now. It seems to be the place you gain most performance...

shalo
Posts: 74
Joined: Tue May 08, 2012 7:25 pm

Re: Overclocking

Wed Aug 01, 2012 12:20 pm

I work on the premise that the raspberry pi is fine at default settings, called (A). It should never have an issue from there no matter how much load you give it. It only makes sense that I should ensure any overclock is equally as stable. I'll call this limit (B).

Once I am extremely confident it is stable, I can increase the voltage. If that first overclock limit (B) was due to heat, there ought to be zero benefit from adding voltage since presumably heat generation is a product of voltage squared, so additional voltage should actually make it less stable and crash faster.

If I plan to use an over_voltage of 6 and find that limit, I will label it (C). It makes sense to see if heat is an issue at this point. I could add a heat sink, or try over_voltage of 8 for a while.

Given the thorough nature of my testing, I should have no trouble seeing if (B) or (C) has changed over time. If I add a heatsink it will be interesting to see the difference, if any, between this new (D) and (C).

I am not convinced that heat is an issue and would like to see for myself. If you search for posts by one of the people who is not a member of the foundation (and so is a little more loose-lipped) but is an engineer at Broadcom, you will find lots of valuable information.

Of interest is that these chips don't work in a "conventional" manner. In their testing they are not restricted to an over-voltage equal to the 8 setting and presumably this is how they could kill it with over-voltage but that heat was never the problem even at extreme voltage levels that are unavailable to us.

Elsewhere you might see a fairly blatant suggestion that the BCM2835 is estimated to last 40 years and that an (unspecified but in the context of the 6 setting) over-voltage setting might reduce that by around half.

FX4
Posts: 64
Joined: Fri Jul 27, 2012 2:39 pm

Re: Overclocking

Wed Aug 01, 2012 2:49 pm

First, you spelled heat sink wrong. Second, seeing as the CPU is hiding under its own ram I believe a temp probe would be useless. And third, Eben himself has said that due to the low powered nature of the SoC overvoltage is a MUCH larger threat to the life of the chip than heat. The whole system doesn't even consume 4W for pete's sake.

I almost want to order a second pi and overclock/overvolt the balls off of it just to see if I can kill it in a reasonable amount of time. :)



You are correct, I did misspell heat sink. I never spell well when I am tired.

I'm not going to disagree with Eben, he has put a lot of research and time into the design. Further I haven't worked as an EE in fifteen years. I don't think short term overclocking is going to kill the chip, what I do expect to see is the chips that are severely overclocked to die young deaths for a lot of reasons I am not going to discuss here. That being said, processes have changed a fair amount since I left the industry and chips do seem to be more robust. I also pulled several fairly current research papers from Intel and oddly the state of California last night that shows temperature rise and burn in testing as major causes asic infant mortality. So my concerns are still valid. I do expect to see burn in testing kill a few of these boards.

What raised the flag in my eyes was, that he was overclocked discussing over voltage and seeing chip performance degradation after running a burn in test. My best guess is his torture testing is causing chip degradation. Specifically I think he is causing some metal migration in the chip so performance isn't what it was new. PN junction temperature rise during burn in testing is a known primary cause of premature metal migration.

I think one thing a lot of people assume is that every chip is the same. They are not. Die on the outside of the wafer have larger geometries so they are slower. Die towards the center are faster (although this is not always a good thing). Wafer to wafer the process can vary sometimes a lot depending on the fab and line in the fab.

And one final thing, the processor being under it's own ram I am sure was accounted for during normal operating parameters +/- x.

I think the thing that is kind of funny here in your defense of pushing the boundaries here is that you make the assumption that because the design is good within its design parameters means it's good no matter what you do to it. This assumption is almost never true. If Eben feels applying too much voltage (overvolting) and maxing out the clock frequency, then running a burn in test until the chip stops running is safe then I can only assume he did a ton of expensive destructive testing. I'm betting not given the nature of this project. I'm betting any overclock testing done was just upping the clock frequency until they started seeing calculation errors. His observations were that all chips pretty safely make 800MHz. 800MHz makes sense because it is almost a 15% above the manufactures production specification. It gives the manufacturer headroom for fab variance and packaging problems.

shalo
Posts: 74
Joined: Tue May 08, 2012 7:25 pm

Re: Overclocking

Wed Aug 01, 2012 3:23 pm

FX4 wrote:I'm betting any overclock testing done was just upping the clock frequency until they started seeing calculation errors.
I wouldn't make that bet if I were you.

If you mean me, I was not referring to either Eben or indeed dom. I was very specific and here are some direct quotes from one of the guys who gets paid to mess about with these things.

"As real engineers, yes, we have overclocked the ARM. But we can't tell you about it as this is an open website and one of our marketing managers might read it. They will immediately sell it with that spec. ;) Yes, we have overclocked the SDRAM but we raised the operating voltage beyond the safety limit. So if you want an ultra fast top-of-the-range Raspberry you can have, for a short while until it is "baked"."

"I know that is what a lot of big processors do but electronics does not always work that way. Think of it as a reverse-diode. No matter how far you cool it, it will blow up at a certain voltage.
Same with frequency. I can turn the frequency up for a circuit and at a certain point it will stop working. Cooling it down might help a bit but not that you can 'double the speed'. Energy efficient devices like the BCM2835 are the same. You can get 20-80% more out of them by over-voltage, then another 5-10% by cooling, but eventually the over-voltage blows them up, not the overheating."

"I played with SDRAM over voltageing and over clocking. Some manufacturers are faster then others. Unfortunately the better and faster ones are also more expensive so we don't use those on the product. So, yes I have managed to run the SDRAM much faster then 400MHz, but also reduced the lifetime a lot."

"Strictly speaking there are two options: Over-clocking and over-voltage. Overclocking does not void your warranty but over-voltage does! The speed which Dom gets, is by doing both because you can run a lot fast at a higher voltage but the life span of your device will reduce. I know the next question: how much? There is no really good answer to that question. The estimate is that it reduces it by half. Disclaimer: the following is just a ball-park of an approximate of a guess of an estimate: Your normal BCM2835 lifespan is estimated to be about 40 years. No: you can not hold me to that if you manage to blow it up in two years or in two hours! As soon as you over voltage your warranty is void even if it blows up ten milliseconds later. (I am luckily not a member of the Foundation as they would never be able to say things like that)"

Presumably you can agree that the raspberry pi is designed to work at stock settings. Presumably you agree that this is not the absolute limit. In an overclocking thread you are going to be dealing with people interested in finding that limit.

If you can explain how to actually find the limit of a specific board without finding the limit of a specific board, that would really be worth reading. In the meantime, I'm more inclined to believe the numerous not-so-subtle hints dropped by the people that are working on these on a daily basis.

FX4
Posts: 64
Joined: Fri Jul 27, 2012 2:39 pm

Re: Overclocking

Wed Aug 01, 2012 3:39 pm

Short answer, because I used to do this. Random sample of boards, crank up the clock, crank up the the voltage. Wait for them to fail. Document the failure points and then look for trends. It is a lot more complicated than this but a high level it is how destructive testing performed.

robwriter
Posts: 114
Joined: Sat Feb 25, 2012 6:26 pm

Re: Overclocking

Wed Aug 01, 2012 3:52 pm

All this holds true if you can overvolt as much as you like, but we have a pretty tight range to play with, and I'd like to think that range was chosen for a reason.

shalo
Posts: 74
Joined: Tue May 08, 2012 7:25 pm

Re: Overclocking

Wed Aug 01, 2012 4:14 pm

FX4 wrote:Short answer, because I used to do this. Random sample of boards, crank up the clock, crank up the the voltage. Wait for them to fail. Document the failure points and then look for trends. It is a lot more complicated than this but a high level it is how destructive testing performed.
But you cannot stress test a raspberry pi without stress testing a raspberry pi. You're in an overclocking thread so it is a given that people here and interested in overclocking and you are saying not to stress test the raspberry pi.

You are likely to cause yourself way more problems running an unstable pi than ensuring it is stable. When it crashes say once every couple of weeks, you will have a nightmare trying to narrow down exactly what the cause was and of course, the crash might just be inconvenient.

The people who designed and work with these very chips on a daily basis are telling you that heat is basically not an issue, if they can't then I obviously can't convince you otherwise :)

I'm genuinely not sure I understand what point you are trying to convey other than you worked in a related field 15 years ago back when the first 3D gpus were only just becoming mainstream. I don't think overclocking is for you and that's fine too.

Return to “Advanced users”