can you write an 'OS' that will fry a CPU?

[IBdaMac]

Couldn't get a solid answer in CPU/Overclockin so I figured I'd try here...

please see http://forums.anandtech.com/messageview.cfm?catid=28&threadid=1245265

Thank you for any help

 

[Matthias99]

Bullshit.

AMD itself provides *maximum* thermal dissipation parameters (as well as "typical" ones), which represent the absolute highest amount of heat you can generate with the processor. Usually there's a specific loop of internal commands that result in this amount of heat being generated (and most often this is a sequence of commands that does absolutely nothing useful, thus the need for a custom OS or program running in realtime mode to force it to happen). Even if your "friend" managed to create and run such a program on your processor, nothing bad would happen. What exactly do you think the heatsink is for? It's designed to remove enough heat that even at the absolute maximum thermal output, the processor will still remain within its operating limits. And even if you did start to massively overheat it (because of, say, bad airflow in your case, or a busted heatsink fan), it would take long enough that the motherboard's protection circuitry would shut it down long before it fried itself.

Basically, who do you trust more: AMD (a multibillion dollar company that employs hundreds of very smart engineers, and DESIGNED THE FRIGGIN CHIP) or some guy you know?

 

[Sunner]

They're probably just messing with you.

Or they're just stupid, you should know that better than me

 

[buleyb]

After some conversing here, we all call bullsh*t. Unless you are running with a poor cooling solution (like improperly installed HSF), a damaged or poor sample CPU, or an FPGA, it just won't happen.

I say you call their bluff and let them give it a shot. Its a stupid thing to test though, if you make any progress and manage to kill a CPU, you need a new test box

 

[Rainsford]

I agree with those people that call BS on that. On the simplest level, you would have to run a series of instructions somehow faster than the CPU designers designed the CPU to run them. Since that is 100% impossible without altering the hardware, I think those guys who say they can are full or it.

 

[DrPizza]

I'm far from a hardware expert, but IMO, without changing the voltage, you have a maximum amount of power available to the CPU. I think that your heatsink should be able to handle it.

 

[mattsaccount]

If AMD/Intel were stupid enough to allow their chips to be ruined by software, why do you suppose no virus EVER (that's a long time) has been written to take advantage of it?

 

[AFB]

Originally posted by: mattsaccount
If AMD/Intel were stupid enough to allow their chips to be ruined by software, why do you suppose no virus EVER (that's a long time) has been written to take advantage of it?

Well said!

 

[compslckr]

there are anti cmos viruses where it says the file is a windows 98 boot disk or some crap like that and it ups the voltages to the cmos like it would when new data is being written to the cmos, but insted of upping it a little it boosts it high enough to smoke.

yes this is real, i have seen my applied electronics teacher do this to an old 133mhz system, pretty cool actually

as far as killing a cpu, there must be a way to do it since you can do software overclocking of processors, but i have not seen anysoftware that allows you to make voltage changes, only FSB

 

[Rainsford]

Originally posted by: compslckr
there are anti cmos viruses where it says the file is a windows 98 boot disk or some crap like that and it ups the voltages to the cmos like it would when new data is being written to the cmos, but insted of upping it a little it boosts it high enough to smoke.

yes this is real, i have seen my applied electronics teacher do this to an old 133mhz system, pretty cool actually

as far as killing a cpu, there must be a way to do it since you can do software overclocking of processors, but i have not seen anysoftware that allows you to make voltage changes, only FSB

I would be EXTREMELY surprised if the CMOS voltage can be modified by software enough to break something. What would the utility be of including hardware instructions to do that?

And as far as overclocking with software, I think that was the stupidest decision ever from an engineering standpoint. It is a well known hardware mantra that hardware can be trusted, software can't. I personally would much rather have to move dip switches than have the capability to allow some random windows app to modify my hardware operating parameters.

 

[Peter]

Upping the voltage to the CMOS when writing to it? That's news to me, and I'm a BIOS engineer. It's all bull, folks.

But of course, when the system hardware contains overclocker's features that allow you to break stuff on purpose, some mean software written specifically for this particular mainboard model can do so too, without you wanting it.

 

[uart]

Looking at the original thread it is clear that these guys are talking purely about heating through normal software, without messing with voltage or FSB or doing anything overtly malicious.

It is true that the CPU power can vary substantially depending on the code being executed. Matthias99 is correct in saying that the particular sequences of instructions that totally maximize CPU power will not usually occur in practical code, it generally will be a "meaningless" mix of instructions. However it should be noted that good software optimization (good scheduling etc) does generally increase CPU power. That is, the more efficiently that a piece of software is programmed in terms of making maximum utilization of a CPU's available execution units, then the higher the CPU power will be.

The normal requirement that a program implement a particular algorithm or generally do some pre-defined task puts very considerable constraints on exactly what instructions can be used and in what order they can be issued (obviously hehe). Someone programming purely to produce heat however does not have these constraints, so it is possible to write some specific "heat em up" software which can give quite a significant temperature rise.

I doubt that your friends could do much better than some of the excellent CPU thermal stress test programs that are readily available.

I suggest you give burnk7 a try. It is part of the cpuburn suite which is only a tiny 21k download available here. http://users.ev1.net/~redelm/

BTW, Don?t run any other programs, let burnk7 have all available cpu time and watchout, it can give some impressive temperature rises.

PS. Just download the ZIP file at the line : MS-Windows [95|NT] : .README file. ZIP file ver 1.4 (21 kB)

 

[borealiss]

i think this is possible, albeit extremely improbable, as it would require NDA information on the spec of whoever manufactured the cpu, and would require specific debug hardware to be present for thermal testing. most cpus for any platform have a specific set of machine specific registers (MSR's), both public, private, and extremely private. let's take a p4 for instance. hypothetically, say there exists msr's that disable thermal throttling. there are also might be msr's that you can write to to change the internal default voltage. again, depending on how the manufacturer implements voltage identification, this could be possible. if the bios on mainboard used something similar to the CPUID string stored in the cpu, then usually one can write to an msr to overwrite these, which in turn would cause the motherboard to read a different default voltage. usually an msr is given by a specific address, so you would do something like:

mov ecx, msr_address ;select the msr address and load into ecx
db 0fh, 30h ;write to msr

so if one were to know of the exact msr's to write to, know how the motherboard would interface with the cpu to extract voltage information, and know the settings to change in the cpu, you could overvolt the cpu. again, this is highly improbably, as most bios' have their own internal lookup table for voltages set depending on what the mainboard reads from the capabilities registers for a specific architecture. so an intimate knowledge of the firmware of a platform and the cpu would be required, on top of the fact that this hardware must be present in the cpu to begin with. given that most hardware implementations include some type of debug features, especially ones that control the thermal characteristics of a high power design, this would be possible. if the actual instruction to disable thermal features on a design were implemented through jtag instructions and required the use of specific hardware to issue these instructions, you're on your own. heck, since we're assuming someone performing this is all knowledgeable, they may be able to rewrite the firmware for a motherboard to turn off the cpu fan if it can be controlled by some sort of onboard monitoring IC. that might fry it.

 

[Matthias99]

Originally posted by: Peter
Upping the voltage to the CMOS when writing to it? That's news to me, and I'm a BIOS engineer. It's all bull, folks.

Don't EEPROMs use a high overvoltage to rewrite the floating gates (at least in one direction)? What would happen if you held it in the "write" state (with +12V or +20V or whatever driving into the gate) for an extended period of time? Wouldn't you eventually damage it? I thought this was an issue with the Xilinx FPGA chips we used in my EE classes, but maybe it's not with BIOS CMOS chips.

 

[CTho9305]

Originally posted by: Matthias99

Originally posted by: Peter
Upping the voltage to the CMOS when writing to it? That's news to me, and I'm a BIOS engineer. It's all bull, folks.

Don't EEPROMs use a high overvoltage to rewrite the floating gates (at least in one direction)? What would happen if you held it in the "write" state (with +12V or +20V or whatever driving into the gate) for an extended period of time? Wouldn't you eventually damage it? I thought this was an issue with the Xilinx FPGA chips we used in my EE classes, but maybe it's not with BIOS CMOS chips.

I also thought 12V was used for programming vs 5V for reading.

 

[Peter]

That was ages ago. It's been a while that FlashROM chips have been single-voltage, 5V and now more and more 3.3V.

And even the old ones had the higher programming voltage supplied to a separate "Programming Voltage" pin. The normal supply voltage is kept unchanged during the process, no harm done even if you leave Vpp on for ages.

 

[Abzstrak]

Originally posted by: uart
Looking at the original thread it is clear that these guys are talking purely about heating through normal software, without messing with voltage or FSB or doing anything overtly malicious.

It is true that the CPU power can vary substantially depending on the code being executed. Matthias99 is correct in saying that the particular sequences of instructions that totally maximize CPU power will not usually occur in practical code, it generally will be a "meaningless" mix of instructions. However it should be noted that good software optimization (good scheduling etc) does generally increase CPU power. That is, the more efficiently that a piece of software is programmed in terms of making maximum utilization of a CPU's available execution units, then the higher the CPU power will be.

The normal requirement that a program implement a particular algorithm or generally do some pre-defined task puts very considerable constraints on exactly what instructions can be used and in what order they can be issued (obviously hehe). Someone programming purely to produce heat however does not have these constraints, so it is possible to write some specific "heat em up" software which can give quite a significant temperature rise.

I doubt that your friends could do much better than some of the excellent CPU thermal stress test programs that are readily available.

I suggest you give burnk7 a try. It is part of the cpuburn suite which is only a tiny 21k download available here. http://users.ev1.net/~redelm/

BTW, Don?t run any other programs, let burnk7 have all available cpu time and watchout, it can give some impressive temperature rises.

PS. Just download the ZIP file at the line : MS-Windows [95|NT] : .README file. ZIP file ver 1.4 (21 kB)

I'm impressed, after 25 minutes it finally got my cpu to 45C, I've never seen it above 43C after running prime95 for 24 hours... (Barton core 2500+ @ 2500MHz @ 1.825V -- 200MHz x 12.5 -- air cooled)

 

[DerwenArtos12]

I call BS. The only way this would be possible is a very very complicated combination of things. Fist the program would have to be huge if you wanted it to work on more than one mainboard. Second the "program" would have to detect what motherboard it is and write information to the cmos(which is only psooible on overcloking friendly boards) that would up the voltage into the 2.5 area which is not available on most motherboards, which means it would have to add it, whcich would require basically writing a new bios which as Peter can tell you is EXTREMELY dificult and would take about a year per board unless you had the manufacturer documentation.

 

[glugglug]

You could always give the CPU a bad microcode update.

 

[borealiss]

microcode patches are usually loaded into a small onboard sram in the cpu, i.e. they're not permanent, and need to usually be loaded by the bios each time. i don't know of any cpus out there that support reconfigurability like a fpga, so once a cpu powers off, the microcode update usually is gone too until the next bootup.

 

[Peter]

Right. The microcode format is completely undocumented and wildly unknown. Pair that with the fact that the CPU rejects inconsistent microcode uploads, and you'll see that this is pretty much impossible. It's not like it hasn't been attempted.

And even if you'd succeed, you'd merely crash the machine, not break it. Big difference.

 

[DrPizza]

From a purely physics point of view, the "record" reached at Tom's hardware site sometime in the recent past was 80-something watts dissipated by the processor. Again, not being much of a hardware person, I'm treading in unfamiliar territory, but isn't the maximum power available to the motherboard just a fraction of the peak power output of a PSU? (maybe 1/2?) Thus, with a 400 Watt power supply, probably less than 200 watts are available to the motherboard. With a processor running at 100 watts and a watercooling system, the intake water would increase by at most a degree or two as it flowed through the heatsink. Double that for 200 watts. So, the only real issue would have to be the thermal conductivity from the processor itself to the water.

 

[Hector13]

Originally posted by: Rainsford
I agree with those people that call BS on that. On the simplest level, you would have to run a series of instructions somehow faster than the CPU designers designed the CPU to run them. Since that is 100% impossible without altering the hardware, I think those guys who say they can are full or it.

Didn't this used to be (or perhaps still is) an issue with video cards/monitors??

I remember when having to set up x-windows several years ago (probably 3-4), people would warn about the dangers of mis-setting your monitor's refresh rates. They would warn of the possibilities of burning out your own monitor and video card... don't know if this was (is) true or not.

 

[CTho9305]

Originally posted by: Hector13

Originally posted by: Rainsford
I agree with those people that call BS on that. On the simplest level, you would have to run a series of instructions somehow faster than the CPU designers designed the CPU to run them. Since that is 100% impossible without altering the hardware, I think those guys who say they can are full or it.

Didn't this used to be (or perhaps still is) an issue with video cards/monitors??

I remember when having to set up x-windows several years ago (probably 3-4), people would warn about the dangers of mis-setting your monitor's refresh rates. They would warn of the possibilities of burning out your own monitor and video card... don't know if this was (is) true or not.

I don't think you can damage the video card, but old monitors would let you run them at scan rates that were too high, and there's something bad about that (I don't know why, but I'd speculate that some parts draw more current at higher frequencies or something, which could cause other parts to burn up). Modern monitors won't do that - they turn off or display a message saying the refresh is too high.

 

[buleyb]

Originally posted by: DrPizza
rom a purely physics point of view, the "record" reached at Tom's hardware site sometime in the recent past was 80-something watts dissipated by the processor. Again, not being much of a hardware person, I'm treading in unfamiliar territory, but isn't the maximum power available to the motherboard just a fraction of the peak power output of a PSU? (maybe 1/2?) Thus, with a 400 Watt power supply, probably less than 200 watts are available to the motherboard.

Perhaps I'm just not understanding you here...but a 400watt power supply is rated to supply 400-ish Watts to the internal components, but if you actually needed all 400watts for the system, you'd be pulling more than 400 watts from the wall (480-520watts, just a guess though), because of the inefficiencies in switching power supplies...

anyone else to confirm this?


Oh, and the topic....I still claim bullsh*t, and have your magical friends try it or come here and discuss why they thing its possible.