TO: Rosetta@home project team...

As a long time Seti@home (classic) and now BOINC participant in both Seti@home and Einstein@home, I was very excited to participate in Rosetta@home using a Power Mac G4 Cube with OSX 10.3.9.

Problem is the Rosetta@home client drove the cpu to overheating - twice. Each time an internal sensor (presumably) forced the Cube into OSX's "sleep" mode after less than an hour of Rosetta client runtime.

I read in the "Rules and Policies" section when signing up that "some computers may overheat". But, since both Seti and Einstein BOINC clients run flawlessly, I had no idea that would actually occur -- I felt the comment must be directed to "older" PCs/Mac's.

This second time (ealier tonight), I came back to find an unmistakable hot smell of electronics coming from the center vent. Amazing! I've taken such great care of this computer for over 2 1/2 years AND NEVER had ANY issues with ANY piece of software at ANYTIME -- it has run FLAWLESSLY from the start.

As a professional software developer (for about 13 years now, and U of W graduate), may I suggest that the Rosetta@home client design team take a page from either Seti or Einstein and *fix* your app!!

A cryptic reference 3/4 of the way down a "Rules and Policies" section, followed by a disclaimer that neither the U of W, etc. etc. is responsible for any damage to computer equipment... etc. etc. is just NOT acceptable -- if you must, put the warning clearly visible on the main project Web page, so that people can actually see it. Better still, *fix* your app! Adding "Read This First" in parens after a section named "Rules and Policies" doesn't cut it, either.

The Rosetta@home BOINC client drove a perfectly fine, well maintained computer to effectively shutdown due to overheating in less than 1 hour -- is THAT good low-level software design?? -- neither Seti@home, NOR Einstein@home, NOR anything application I've ever used on this computer has exhibited that behavior.

"Adding a cpu limiter utility is also NOT an acceptable solution"... designing the Rosetta@home client to behave properly as a BOINC client IS an acceptable solution.

IF you want people from "around the world" to participate in Rosetta@home, the client is going to encounter a wide mix of hardware... DO YOU REALLY WANT TO FRY IT??

-- looking forward to participating... someday.

---

How? You say you are a professional software developer. Okay, so am I, in fact, specifically on Macs, although not part of the Rosetta project team. HOW would you "fix this app"? Exactly what code could cause a G4 to overheat, beyond "any code that causes 100% utilization"? And once you have told us what that code could be, how would you prevent it from causing the overheating? The Rosetta application is not Altivec-specific, indeed is considerably slower on Mac than on Windows, so the problem isn't over-use of the Altivec portion of the chip...

---

How? You say you are a professional software developer. Okay, so am I, in fact, specifically on Macs, although not part of the Rosetta project team. HOW would you "fix this app"? Exactly what code could cause a G4 to overheat, beyond "any code that causes 100% utilization"? And once you have told us what that code could be, how would you prevent it from causing the overheating? The Rosetta application is not Altivec-specific, indeed is considerably slower on Mac than on Windows, so the problem isn't over-use of the Altivec portion of the chip...

---

How? You say you are a professional software developer. Okay, so am I, in fact, specifically on Macs, although not part of the Rosetta project team. HOW would you "fix this app"? Exactly what code could cause a G4 to overheat, beyond "any code that causes 100% utilization"? And once you have told us what that code could be, how would you prevent it from causing the overheating? The Rosetta application is not Altivec-specific, indeed is considerably slower on Mac than on Windows, so the problem isn't over-use of the Altivec portion of the chip...

Hello, Bill - your questions should be directed to the Rosetta project team, as was my post. If you had taken the time to actually read my post, you would have noticed this statement, "...neither Seti@home, NOR Einstein@home, NOR anything application I've ever used on this computer has exhibited that behavior." That specifically states a processing "track record" for the machine over several years AND specifically with "grid" computing clients -- both Seti@home Classic and more recently the BOINC clients.

Seti@home drives the cpu at 100% utilization - no problem. Einstein@home drives the cpu at 100% - no problem. So do rendering apps, and math modelling solutions packages I run -- all no problem. The better question would be, how does the Rosetta client differ in implementation over the Seti and Einstein BOINC clients?? And specifically does it employ assembly language/low-level cpu techniques in the code??

You say you too are a software developer, then you ought to know that certain low-level (as in assembly language) techniques can be implemented - and co-compiled into source languages like C - that can drive a cpu at the register level -- perhaps that would be a starting place for a code review.

When a standard higher level programming language is compiled, be it C, C++, Pascal, Fortran, etc., the compiler makes use of "standard" cpu specific instructions sets. I don't claim to be an assembly language programmer, my work is in mathematical modelling and numerical analysis techniques, yet what I've heard in the past, is that register level assembly language techniques CAN implement non-standard processor calls. Perhaps something like that could bypass cpu overutilization safeguards (?? - conjecture) -- any assembly programmers out there care to correct/comment on that?? Again, that's conjecture. What I can tell you is that assembly language routines written properly are highly efficient and ideal for certain scientific computing applications.

As a point of interest - Overclocking techniques can/do cause a target cpu to overheat, while showing 100% utilization - so your assertion that 100% utilization is effectively all the same, doesn't apply in an overclocked setting -- which may or not apply to the question at hand.

So, the point remains: other BOINC clients don't exhibit this behavior. Why does the Rosetta client? And why then the warning? I've gone back to running "full 100% cpu utilization" on current Seti@home and Einstein@home work units - again, no problem.

The scientific method "says" that if something is repeatable, then there is probably something there. The overheating within 1 hour of running the Rosetta client, as stated in my post, IS REPEATABLE - and hasn't occurred before or since. You figure it out.

Since, you're not part of the Rosetta team, why are you referring to yourself in the plural "we" and "us"? I don't know if the Rosetta client is open source or not. So, it might be difficult to comply with your requests...

"Exactly what code could cause a G4 to overheat, beyond "any code that causes 100% utilization"? And once you have told us what that code could be, how would you prevent it from causing the overheating?" -- I'd start by answering the question if low-level techniques are implemented in the code base.

However, until either a Rosetta project person weighs in on this OR an assembly programmer weighs in on this, the question is still open. May I suggest that you have something constructive to add IF you choose to reply -- maybe YOU have access to the Rosetta client code base. Maybe you're an assembly language programming and can weigh in on the above. Maybe you'd care to bring the weight of your "professional experience" to bear on this question. I tell you what won't work, though -- sarcastic questions.

---

The scientific method "says" that if something is repeatable, then there is probably something there. The overheating within 1 hour of running the Rosetta client, as stated in my post, IS REPEATABLE - and hasn't occurred before or since. You figure it out.

I'm not trying to be sarcastic, but I have trouble believing that this is really a Rosetta problem. I say that because I have run Rosetta on a G4 Mac Mini for many hours without the temp rising above the same 44C that it sits at for SETI, Einstein, Predictor, SZTAKI... plus have run it on a G3 iBook, likewise no excessive temps recorded. Plus, while I have not done assembly specifically on PowerPC chips, I have done assembly on older architectures. I can imagine only one way to cause a CPU to overheat, and that is by writing code such that it and the attendant data fits entirely within the CPU's cache (no external memory accesses) and runs in a tight loop for an extended period of time. Even then, the effective load on the processor would not be _significantly_ higher than any other "100%" situation, and I would expect something like the Team MacNN Altivec-optimized SETI application to be much worse than Rosetta or any other. I know of nothing 'unusual' about the G4 used in the Cube - I would imagine that it is a Motorola 450MHz 7400 or 7410? Can you verify that? Mine being newer would be a 7450/7455 "G4e" chip... if it would help anything, I would be happy to run Rosetta "only" on the Mini, suspend other projects, while monitoring the temperatures. There is one difference - the cooling on the Mini is more "active" than the passive cooling on the Cube, unless you have your Cube in an enclosure.

My overclocked AMD 3700+ is my primary Rosetta box; again, I've noticed no difference in the temps between Rosetta and any other BOINC project, and I have to watch temps VERY closely there... and there are no less than seven fans...

We do need to hear from someone on the Rosetta team, but my impression from these boards and the performance of the software, is that the Mac version is a straight recompile of the Windows version, with little or no change - thus the slower performance on a Mac. Any problem would be more compiler and chip-dependent than C code related. I have emailed David Kim to direct his attention to this thread.

It may well be that some kind of front-page warning DOES need to be issued saying something like "Certain older G4 Macs have been known to significantly overheat while running Rosetta" - but please realize that you are the FIRST to report this situation, so it's really not fair to blame the project for what is likely a very rare and unusual occurance that hasn't happened on any other Macs, and they DO have the standard "BOINC Warning" about overheating.

The plural "we" and "us" is "those of us who frequent these message boards and attempt to help with problems", not the "royal we" of the project team.

---

feynman, I am sorry you are having overheating problems when running Rosetta. Bill is correct in that the code used for the mac version is identical to that used with windows and linux. We have not added altivec optimization code and do not use altivec yet. We and most all other BOINC projects do warn people of the possible overheating issues in the general BOINC rules and policies and in our recommended system requirements.

I am not aware of any other users having similar issues so your case may be an isolated one. If/when the code becomes open source, we encourage you to look at it for possible causes in overheating your comptuter (we do not have a similar computer to test). I have been running rosetta on my power book laptop for a while with no overheating problems. In the mean time, please dedicate your computer cycles to other BOINC projects.

feynman, I am sorry you are having overheating problems when running Rosetta. Bill is correct in that the code used for the mac version is identical to that used with windows and linux. We have not added altivec optimization code and do not use altivec yet. We and most all other BOINC projects do warn people of the possible overheating issues in the general BOINC rules and policies and in our recommended system requirements.

I am not aware of any other users having similar issues so your case may be an isolated one. If/when the code becomes open source, we encourage you to look at it for possible causes in overheating your comptuter (we do not have a similar computer to test). I have been running rosetta on my power book laptop for a while with no overheating problems. In the mean time, please dedicate your computer cycles to other BOINC projects.

Thank you, David - I appreciate the feedback. Rosetta@home is a great addition to the BOINC projects, so I hope with different hardware (soon) to be able to contribute to Rosetta along with others.

And, I do hope it is isolated to my machine, or maybe to the 500MHz ppc processor class found in Power Mac G4 Cubes of a few years ago -- though, I don't understand why that would be. Anyway, since it is repeatable (and not with other BOINC clients), at least on this "stock" 500MHz Cube, It might be a good idea for any other G4 Cube owners reading this to simply watch the first run of the Rosetta@home client on their Cubes.

Given that no one else has posted an overheating problem, and that you and Bill are also Mac guys running on G4 ppc's without any problems, It's got to be isolated to my machine - but, I'm at a loss to understand why, but will look forward to running Rosetta once I get different hardware.

Anyway, to help conserve forum space, I won't post on this again. But, I do thank you for your feedback and I encourage anyone reading this thread to participate in Rosetta@home -- Rosetta is a valuable addition to the BOINC family of projects. -- feynman

---

The scientific method "says" that if something is repeatable, then there is probably something there. The overheating within 1 hour of running the Rosetta client, as stated in my post, IS REPEATABLE - and hasn't occurred before or since. You figure it out.

I'm not trying to be sarcastic, but I have trouble believing that this is really a Rosetta problem. I say that because I have run Rosetta on a G4 Mac Mini for many hours without the temp rising above the same 44C that it sits at for SETI, Einstein, Predictor, SZTAKI... plus have run it on a G3 iBook, likewise no excessive temps recorded. Plus, while I have not done assembly specifically on PowerPC chips, I have done assembly on older architectures. I can imagine only one way to cause a CPU to overheat, and that is by writing code such that it and the attendant data fits entirely within the CPU's cache (no external memory accesses) and runs in a tight loop for an extended period of time. Even then, the effective load on the processor would not be _significantly_ higher than any other "100%" situation, and I would expect something like the Team MacNN Altivec-optimized SETI application to be much worse than Rosetta or any other. I know of nothing 'unusual' about the G4 used in the Cube - I would imagine that it is a Motorola 450MHz 7400 or 7410? Can you verify that? Mine being newer would be a 7450/7455 "G4e" chip... if it would help anything, I would be happy to run Rosetta "only" on the Mini, suspend other projects, while monitoring the temperatures. There is one difference - the cooling on the Mini is more "active" than the passive cooling on the Cube, unless you have your Cube in an enclosure.

My overclocked AMD 3700+ is my primary Rosetta box; again, I've noticed no difference in the temps between Rosetta and any other BOINC project, and I have to watch temps VERY closely there... and there are no less than seven fans...

We do need to hear from someone on the Rosetta team, but my impression from these boards and the performance of the software, is that the Mac version is a straight recompile of the Windows version, with little or no change - thus the slower performance on a Mac. Any problem would be more compiler and chip-dependent than C code related. I have emailed David Kim to direct his attention to this thread.

It may well be that some kind of front-page warning DOES need to be issued saying something like "Certain older G4 Macs have been known to significantly overheat while running Rosetta" - but please realize that you are the FIRST to report this situation, so it's really not fair to blame the project for what is likely a very rare and unusual occurance that hasn't happened on any other Macs, and they DO have the standard "BOINC Warning" about overheating.

The plural "we" and "us" is "those of us who frequent these message boards and attempt to help with problems", not the "royal we" of the project team.

Bill, thank you for the additional info - I just got to your post after David's post above earlier.

Anyway, the short version is, I'm going to follow David's suggestion and wait to run Rosetta on different hardware. I too have a really hard time believing it has to do with the Rosetta client specifically, or with software in general, causing a cpu to overheat.

Thing is, what keeps coming up is that it is repeatable - run Rosetta, get the problem, overheating within space of 1 hour. Run other BOINC clents (Seti or Einstein here) or any other cpu intensive math-stuff apps (100% util showing) and don't get the problem - those apps sometimes running overnight into the next day.

Anyway, not to be repetitive, but when trying to figure this out the (flawed? - put here again for review) line of reasonng went like this... the only thing I could come up with was that... I knew cpu's can overheat if driven outside of normal operating range. And I knew that, as an example, overclocking a cpu can do that - but, that's typically a hardware mod and not software related. And I knew that compilers compile down to executable code using standard platform (cpu) specific instruction sets, which don't contain any non-standard machine-level instructions - so a standard compiled/optimized executable should be "well behaved" with respect to a specific cpu instruction set, and not present a problem.

So, I figured, unless there's a hardware issue -- but then that should show for any app loading the cpu at 100% overtime, right?? -- that placing a 100% prolonged load on the cpu would not drive it out of normal operating ranges given standard compiled/optimized code.

Okay, so trying to think of how a cpu could be driven by software out of normal operating range. Only thing I could come up with was some non-standard machine-level, assembly language instructions. (Any other ideas here?) And since Rosetta is a sci app, perhaps assembly programming techniques were used to improve computational efficiency - just conjecture here, of course. But that gets back to how those modules might be implemented, if they exist. Okay, that's all I've got by way of conjecture re: how an app could cause cpu overheating.

Thing is, if it was due to something hardware related - triggered by prolonged high cpu load - wouldn't it occur when any app loads the cpu at 100% overtime?? -- that seems logical enough.

Okay, so I'm at a loss, that's the "best" I can come up with from a software perspective. And, of course, if they're not using any assembly techniques, or if the 500MHz PowerPC G4 (v11.3) processor class doesn't support "non-standard" calls, then that whole line of reasoning just falls apart.

Anyway, as I also mentioned in the reply to David's post, in order to conserve forum space and not make too much of this, I won't post on this again. But, thank you for the additional info, just the same. -- feynman

---

Thing is, if it was due to something hardware related - triggered by prolonged high cpu load - wouldn't it occur when any app loads the cpu at 100% overtime?? -- that seems logical enough.

Don't worry about forum space by posting - much more space is lost to flame wars and such, a thread that is actually possibly accomplishing something isn't an issue...

Let me back up and say something for the record - I will reply to _almost_ any Mac question where I have a clue about the answer, because I've been a Mac fanatic since my first Lisa. The last couple of months, I've been working with BOINC projects, the Wiki, etc., learning and answering questions. I am however human and tend to respond "in the same tone of voice" as the original poster. Had your original question been in the same tone as all your later postings, instead of "FIX YOUR APP", we wouldn't have wasted the first couple of messages getting beyond that. I don't know if you spend any time on the SETI boards, but after a few hours wading through unwarranted attacks there, you get a little sensitive... so, my apology for the tone of MY original reply.

Meanwhile, this situation has been driving me nuts. I can see _NO_ way that this can be happening, and yet I fully believe you that it is. I'd like to get more information just to try to get to the bottom of this. First off - is your Cube showing as a "PowerMac5,1" or 5,2? Were you able to determine the "74xx" model of G4 chip used? I can pull up service specs and schematics on just about any Mac, but even within fairly short-run models like the Cube, there are often considerable variations. How much L2 cache do you have? (And I'm sure there are 1000 other questions I should ask, but that's what I can think of right off...)

In trying to postulate some, ANY, mechanism that would cause a specific chip to overheat with only one specific program, I've come up with a few obscure possibilities. For example, if there were some flaw, either in the chip design or in this specific chip, where one particular trace was "thin" and had higher resistance to current, any process that puts a load on that trace would increase the temperature in that spot. If the flaw was such that, say, an integer add would hit it, the chip would likely never leave the fab. But if it were such that it is only hit on something like a missed branch prediction, and if the compiler put the most-likely branch for some heavily looping part of Rosetta in the least-likely spot in the binary code, then that missed prediction would happen every time through the loop - this would explain the Mac version being slower, AND would explain why on your specific G4 chip, you'd get overheating...

David, is there a way to run a SQL query against the hosts table for "PowerMac5"? I'm curious if feynman is the ONLY Cube owner running Rosetta. If you could locate another, we would at least have two examples to look at. Failing that, I know a couple of Cube owners - but I suspect as soon as I mention "check for overheating", they'll run the other way rather than sign up to help. They're justifiably proud of their Macs. :-)

---

How? You say you are a professional software developer. Okay, so am I, in fact, specifically on Macs, although not part of the Rosetta project team. HOW would you "fix this app"? Exactly what code could cause a G4 to overheat, beyond "any code that causes 100% utilization"? And once you have told us what that code could be, how would you prevent it from causing the overheating? The Rosetta application is not Altivec-specific, indeed is considerably slower on Mac than on Windows, so the problem isn't over-use of the Altivec portion of the chip...

Hi
I'm not a software engineer, but i do know when a piece of software is causing my new Macbook to overheat. Everytime BOINC is activated, the fans on my Macbook run full tilt and the CPU heat sensor begins to register temperatures of 77 C or more - after BOINC CPU usage has been limited to 10%! I won't use BOINC until your lab fixes the problem.

---

How? You say you are a professional software developer. Okay, so am I, in fact, specifically on Macs, although not part of the Rosetta project team. HOW would you "fix this app"? Exactly what code could cause a G4 to overheat, beyond "any code that causes 100% utilization"? And once you have told us what that code could be, how would you prevent it from causing the overheating? The Rosetta application is not Altivec-specific, indeed is considerably slower on Mac than on Windows, so the problem isn't over-use of the Altivec portion of the chip...

Hi
I'm not a software engineer, but i do know when a piece of software is causing my new Macbook to overheat. Everytime BOINC is activated, the fans on my Macbook run full tilt and the CPU heat sensor begins to register temperatures of 77 C or more - after BOINC CPU usage has been limited to 10%! I won't use BOINC until your lab fixes the problem.

---

hmmm. have to admit that the heating issue concerns me. i'm running rosetta@home on my intel mac (2.8ghz dual quad core), getting pretty good results i think (9000 credits/790avg score over three days since joining). but. i'm also getting very high core readings while running boinc - as high as 170ºF. knocked the percentage down to 95% usage on just 7 cores (way too slow for anything else if use all 8 processors). normal temps have dropped to 142ºF since then. i'd like to help with this project, but cannot afford to see my mini nor my G4 drop dead. until the heating issue is resolved, think i'll keep the intel mac running for awhile and not use the others. sorry people.

If your machine is not designed to provide enough cooling for all of your CPUs to run, you can control the percentage of CPU utilized by BOINC. It is true that a machine doing useful work produces more heat than one that is idle.

Do not look to Rosetta to make changes, they're job is to fully utilize the resources you provide. You already have everything you need to control BOINC and the amount of heat produced.

---

Rosetta Moderator: Mod.Sense