AMD And THG update

Mhrmnhrm writes "In the interests of responsible journalism, the gang over at Tom's Hardware has developed this article in the interests of setting the record straight about their original AMD burn-out video, and the new release (possibly from AMD) of this past week. It would seem that BOTH videos are correct, and that the question of whether or not somebody is hiding something depends entirely upon your own point of view."

Comments

test with fan failure, not heatsink falloff

[doughnuthole]

It is much more likely for a fan to die on the heatsink, than for the heatsink to fall off on its own. Someone should do a test to see if this also causes chip failure or if the sensors on the motherboard can shut down the computer before anything is damaged. The results would mean a lot more to actual users.

But in the end

[baptiste]

the important thing is that publicity of Tom's article got AMD off their butts to try and resolve the problem. Yes, this won't help folks with existing motherboards and I doubt we'll see the new Maxim chips on motherboards for a little while (since many aren't even reading the thermal diode yet) I think its great that Tom's HW pushed AMD in the right direction. So we can debate about who did what, conspiracy, etc but its pointless. Obviously THW blew up Athlons in less than a second. Obviously someone (possibly AMD, who knows0 had one running without a heatsink. Personally I think the 2nd video is fake in some way. I'm an AMD/Athlon fanatic - love them. But I can't believe they managed to play Quake on a CPU without a heatsink unless the mobo they used was doctored in some way to throttle speed based on the thermal diode reading.

But in the end - its really not an issue. Yes, existing Athlon owners are at a SLIGHT risk of failure if their heatsinks fall off (I'd love to see REAL stats on how often THAT happens) But in the end, its still cheaper to replace your Athlon once than to go with an equivalent Pentium 4. So lets be glad AMD listened to the folks at Tom's Hardware - realized they were getting a black eye, and did something about it. Hopefully in a few months we can buy mobos with the Maxim chip safty valve or some tryp of clock throttler. Then the Pentium freaks will have to argue over real stuff like benchmarks and performance instead of making snide comments about Athlons burning you house down.

Re:But in the end

[Anonymous Coward]

But I can't believe they managed to play Quake on a CPU without a heatsink unless the mobo they used was doctored in some way to throttle speed based on the thermal diode reading.

Actually, with the proper chipset, the thermal diode can be used to throttle the speed. Go look at the AMD 766 register programming manual, section 4.6.1.4 and register C3A50:TTH_EN and C3A50:TTH_RATIO. The southbridge accepts a signal from a chip like the MAX6512 and will throttle the CPU when it asserts. But only if the motherboard has this hardware AND the firmware configures the southbridge to do so

Re:But in the end

[ahertz]

But I can't believe they managed to play Quake on a CPU without a heatsink unless the mobo they used was doctored in some way to throttle speed based on the thermal diode reading.

Just to clarify, maybe you should take a look at the video again. (You did watch the video before posting, right? :b ) Despite what the blurb on the previous article would have you believe, the didn't play quake on a CPU without a heatsink. Without a heat sink, the system immediately shut down. They managed to play Quake for nine minutes on a system with the heat sink fan disabled.

More believeable, isn't it? Admittedly, I don't know enough about the subject to judge whether the video is fake or not, but it's not as wildly unbelieveable as all that.

Good Work Tom's

[MrBlack]

I say good work to 'em. It seems THG was correct all along, but I couldn't detect the faintest sense of gloating or grandstanding in their article. They've pointed out a fairly big problem with some of AMD's chips, AMD have responded with new info to MOBO manufacturers (who will no doubt consider this new feature fairly important after the attention the orriginal article on THG received). Great. No threats of legal action by either side (AFAIK), no finger pointing, no FUD. Now if everyone could behave like this......

and in the end it doesn't really matter

[hattig]

Basically, the XP in combination with all currently available motherboards will kill itself if a disaster happens (the heatsink falling off of a CPU happens very rarely). AMD responded and designed a circuit so that the CPU will shut itself down if it gets too hot, causing data loss. In this rare case, data loss isn't an issue.

In the more common case of the CPU fan failing, the CPU will heat up more slowly. Hence the other protection mechanisms in the CPU will be used, and the user will get a chance to save their data.

However, AMD should have designed the safety circuit they have shown off in the article INTO the actual CPU itself, so it can save itself. And it should save itself by basically clocking itself down to 100MHz or slower, so that data loss does not occur and the user can save their data. Hopefully this will be implemented in a future revision of the CPU.

However, the instances of CPU heatsinks falling off are extremely rare, and probably attributable to either a poor initial fitting of the heatsink, or a bad socket with a weakened retention mechanism. In a tower case, the heatsink would probably fall onto the graphics card or spring onto the memory and damage these anyway...

Re:and in the end it doesn't really matter

[jhoffoss]

Something like this will be crucial if AMD ever hopes to break into the server and workstation market in a big way...If a company is going to run a db server, they're not going to use AMD for reasons like the heat protection that will be back-patched on new mobos. A company needs the data to be there whether the cpu is gonna fry or not. Granted, frying the cpu is worse than losing a db, but not by much (either way, you may have to go to the tapes...)

On the other hand, if this company chooses an intel system, they know (as illustrated by Tom's video) that the system will survive, though performance will die. But the integrity of the data remains.

This, as far as I can see, is one of the main things keeping AMD processors out of higher end systems. Even though the piii/p4 performs better in some areas, a person can put a dual athlon board on the desk for around the same cost, which would more than make up for it.

Re:Quick Summary

[darkonc]

SO the good news is that AMD is now specifying a crude thermal protection for boards using their CPUs. The bad news is that

• This protection doesn't exist for old (current) boards
• It just shuts down the whole system (i.e. crash)
• AMD didn't bother to mention that this is a patched board that they're using.

You could probably do a board design that, instead of shutting down the whole machine, switched it to 100MZ (or some other 'safe' clock speed), but AMD hasn't quite gotten around to that, (yet).

better than a shutdown

[Cynikal]

it should be integrated into the cmos to choose if you want to shut off the power instantly, or throttle the cpu to a lower speed and managable temp to let the system shut down properly. I'm sure a 1.5GHz cpu could function at a low temp if reset to a 66mhz bus (or even 33) and lower multiplier, just enough power to save any data and exit gracefully.

then again, isnt that what intel got bashed for doing with their p4? i'm not sure on the specifics of the why's how's and when's of p4's power throttle, all i remember is people pointing at p4 and saying "bad". Doesnt sound so bad now tho.

I imagine the most important point to any failsafe like that is letting the user know clearly why exactly their computer shut down, or is running incredibly slow. maybe having the soundcard play some 70's pron music in the background would be a good enough indication that something is getting too hot? :P

No big deal

[Mathness]

Several things strike me as odd about this "news" (2nd article) and the first article which spawned it.

-"Siemens assured us that the thermal protection circuitry is definitely working on their motherboard." (1st article)
Well, it did not work anyway, whoops. And therefore AMD must be at fault here?

-"We rushed to the telephone to confer with Siemens. The engineers assured us that what we had seen was for real. The thermal diode of Palomino is unable to react quickly enough. Only 1 degree/s is what the thermal diode is able to handle." (1st article)
Okay, they check with Siemens. Why not have a chat with AMD? If the thermal diode can only work this slow, why can the AMD engineers make a working shutdown with a common electrical component?

-"AMD showed us how all Palomino CPUs could be protected against overheating with relatively little effort." (2nd article)
Wonder why Siemens could not make this? They where the ones who claimed that the MB would protect againt meltdown, and that it worked. AMD said that the thermal diode worked. And showed it on a modified MB, which have no protection from the start.

THG is making a big fuss, about a somewhat minor problem. And THG was too biased against AMD, I can only hope I was biased enough the other way ;)

And what is up with this "We saved the hardware manufactors and you" theme in the 2nd article?
Are they losing commercial revenue?

More on amdzone. P4

[leuk_he]

AMD zone also has an update on this story. The most interesting thing (to me) is that they have a 2nd video that show the P4 shuts down. With possible data loss like the modified XP tom was shown in munich.

I still think it's not quite AMD's fault

[Strepsil]

OK, I suppose I want to see AMD as the good guys here, since I love their products. Even so - I can see how things could spin a little more in their favour if you're inclined to be generous.

Firstly, let me say that I belive that thermal protection integrated into the CPU, like to P4 has, would have been the best way to go - but then we'd all be complaining about how expensive the new Athlons are, wouldn't we? AMD give us lots of grunt for our dollars, and we can't expect them to pack every little feature into the CPU for the great prices they give us, any more than we can expect Apple to sell us an iPod for 50 bucks.

Now, the original article at Tom's has the following interesting quote ...

We rushed to the telephone to confer with Siemens. The engineers assured us that what we had seen was for real. The thermal diode of Palomino is unable to react quickly enough. Only 1 degree/s is what the thermal diode is able to handle.

I pointed this out the other day, too. A Siemens Engineer was consulted. Excuse me? An AMD CPU just fried itself on a Siemens-made board, and they don't ask the CPU manufacturer why it happened - they ask the board manufacturer. That's not where I'd be directing my enquiries if I was doing the test.

Now AMD prove that with a simple external board, everything shuts down and saves the CPU - just like we'd want to happen. To me, this suggests that the thermal diode has no problems reacting in time, and that maybe the board manufacturers screwed up or cut some corners when they were developing support for it.

It's understandable, for the same reason AMD didn't pack the chip full of power management goodies. Keeping costs down on goods that are bought by very price-conscious buyers looking for maximum performance at the best price. They implemented support that was adequate for a fan failure, which is the most likely thing that'll happen in real life. They could well have decided that a simpler circuit was all that was necessary, since a heatsink isn't likely to fall off.

Most likely we'll never know the full sotry. Everyone will blame everyone else, and in a year or so we'll forget all about it because the hardware will be obsolete. We'll have new problems. In the meantime, everyone has the message that they should be careful that they install their CPU cooling devices properly. AMD will recover from any negative press. Hell, Intel put out a bunch of CPUs that couldn't do Math at one point, and they survived. :)

Why both are correct.

[Chas]

Tom's shows what happens if you use any Athlon
with any currently available motherboard.

While it's correct, it's like griping about no RDRAM support on an Athlon board. The feature isn't supported by the motherboard yet. End of story.

The AMD video shows what happens when you use an Athlon with a thermistor in motherboard that has the Power Now thermistor support..

And while it's also correct, it's basically vaporware until such a board debuts.

Underclocking to the rescue?

[Anonymous Bullard]

While I've followed the Athlon performance discussions with no little amount of awe, I've been frustrated by the heat issues and the related power consumption. Due to reasons beyond my control, heat, relatively speaking, is an environmental issue where I work. Humongous HFC's aren't the ideal solution to this problem IMO.

Recently I've started pondering whether it would be possible to take one of the newer Palominos (eXPee+ series Athlon) and underclock it (e.g. from 1,333Mhz to 1Ghz or even lower) via the multiplier and voltage while keeping the external bus speeds (which aren't part of the heat problem) intact or even slightly overclocked.

Has anyone studied the underclocking opportunities of Palominos with the latest mainboards? Waiting for the move to a 13 micron silicon-on-insulator process is starting to get on my nerves. ;-) VIA's latest C3 (800Mhz) looks technically "cool" but it understandably lacks DDR support and tick-for-tick its performance still sucks compared to the K7 designs by AMD.

Also, does anyone know if the upcoming "finer" Athlons will be compatible with the mainboards on sale today?

Or should I just give up on AMD and go for a P4 when the DDR-boards become available?

Re:Underclocking to the rescue?

[RainbowSix]

The XPs have the problem that their L1 bridges (which are the ones controlling the multiplier lock) are not only cut but have small troughs between them where they are cut. I will assume that the Palominos have the same problem, otherwise it would be a trivial pencil trick to unlock.

1. Clean out the troughs. They have scorch marks
2. Use tape to cover the bridges, but expose the troughs, and fill them with some glue of sorts, then sand them down level
3. Use pencil trick or conductive ink.
4. Change multiplier as you like.

If you use one of the VIA boards (not sure about the DDR ones, but the SDR for sure) you can run the CPU at 200mhz bus, down from 266, but then bump the memory bus back up to where you want it since they are seperate.

There are several guides online on how to do this. A quick search should yield several.

Re:I still wouldnt get an Athlon or any AMD chip.

[jmichaelg]

Intel stuff is just better engineered.
That statement is not supported by most benchmarks I've seen (excluding quake demos).

I have both Intel and AMD based machines and I'm not seeing any discrepancy in stability as measured by crash rates - Windows will as happily crash on the Pentium as it will the Athlon. As long as you cool the AMD properly and feed it clean power, it does just fine. Cheapo cooling and/or power supplies will break any cpu.

The bottom line for me is stability.
The bottom line for me is stability and performance. In my business (I grade children's arithmetic tests), I need all the fpu performance the Athlon can deliver and I'm not particularly inclined to pay 3 times the price for equivalent performance. Nor am I about to rewrite my code to take advantage of proprietary sse2 instructions to get that performance either.

Often enough

[Tridus]

"Yes, existing Athlon owners are at a SLIGHT risk of failure if their heatsinks fall off (I'd love to see REAL stats on how often THAT happens)"

This happens a lot with computers that are shipped (eg from Dell, etc). Thats probably one of the reasons why Dell doesn't do AMD, replacing all the chips that get fried by a heatsink coming off in shipping would be a lot more expensive.

(and I've had this happen to me before actually, although it was with a P3, which of course didn't fry because Intel put some thought into the design)

Response time of thermal diode is less than 1ms

[De Lemming]

On Ace's Hardware, this subject is discussed thoroughly the last days. An interesting article can be found here. Some thoughts from this article:
- the component needed for proper protection of the cpu costs $0.85 (in quantities of 1K).
- "Obviously, Siemens used an external temperature probe and tried to pass it off as using the internal diode."
- The reaction time of the thermal diode is certainly not 1 degree/second: "At our worst case rise of 300C/sec, that translates to a response time of less than 1ms. No way would this result in a fried CPU if power off is immediate upon the signal occurring."

For a matter of fact, an engineer told a friend of mine it would be very difficult, if not impossible, to develop a thermal diode with such characteristics (1 degree/sec response time).

You can find the main thread of this article here.

another experiment.

[lophophore]

When I removed all 20 lug nuts from my car, I could only drive it for about 1 seconds before the wheels fell off. The brake discs got smashed on the pavement, the road got scraped up, and worst of all, I spilled my margarita!

The car is clearly defective since the wheels fell off when I removed the lug nuts.

And what's worse is that my drink tasted awful when I slurped it out of the carpet.

Just as it should be

[gotan]

I read the first article and was quite impressed by the P4's performance of just going on at a slower clockrate. I also found it quite informative, that an Athlon might not only fry itself, but also the motherboard or even start a fire. The point is quite valid, that the heatsink might fall off during transport (although i suppose one would hear the loose heatsink rattling in the case) and will take a little more care with present CPU/Mobo combinations.

Let's not forget, that the throttling technique got some critique when it became known, mainly because the user wouldn't know about it, and it seemed a cheap way for intel to resolve heat problems with the P4. Toms article showed, that the throttling is quite a good feature to have (it would be even better, if there was a way of notifying the user without counting clockticks). So the first thing i thought after reading Toms first article was: "will AMD something similar, and when?". Now the followup shows, that AMD is aware of the problem, and is willing to spend some effort to offer solutions.

While the Motherboard-solution is more like a quick hack useful enough for present boards and AMD-CPUs, i expect something built into the CPU next time (at least a sudden shutdown should be possible, altering clockspeed at runtime might be a different story, and probably takes some haggling about patented technology with intel).

In conclusion i think things are as they should be: the test has made AMD aware of the fact, that protecting the CPU from overheating is a feature customers want, and they set to work on it and offered a quick solution after a few weeks. Noone tried to ban the story from toms site and people actually got together and talked about the problem without shifting responsibility around.

I still think the Athlon has the better bang for the buck, but i won't mind shelling out a little more money for a mobo with overheat protection.