Please register or login. There are 0 registered and 768 anonymous users currently online. Current bandwidth usage: 326.30 kbit/s October 21 - 05:33pm EDT 
Hardware Analysis
      
Forums Product Prices
  Contents 
 
 

  Latest Topics 
 

More >>
 

    
 
 

  You Are Here: 
 
/ Forums / Clock throttling, a hot issue?
 

  Clock Throttling Controversy - Hardware Analysis' Methods called into question 
 
 Author 
 Date Written 
 Tools 
Scott Tribe Jun 10, 2001, 12:29am EDT Reply - Quote - Report Abuse
Private Message - Add to Buddy List Replies: 8 - Views: 567
Sander.. with regards to your attempt to clear Intel of any wrong doing in the clock throttle issue.. please read the following comments on your article from Aces Hardware:

>I see that this "study" was limited to one P4 processor.
>To compare it to Van's study (Editors Note: Van Smith of InQuest) is insidious at best, since even Van noted that clock
>throttling did not occur on every processor or system. Instead, he carefully stated in his article both the situations and the >circumstances under which tne phenomenon occurred, and opened the floor for debate- so that the full
>dimensions of the problem could be determined.

> No debate has occured only an obvious "cover-up" using one processor.
>Any one who swallows this dreck as "proof" has only the interests of Intel in mind. Not
> the interests of the consumer.
> If I were a scientist or businessman about to buy a large order of P4 based systems for
> some particular purpose that REQUIRED full performance, I'd be watching my checkbook REAL
> careful about now....
> Of course few applications really REQUIRE full performance, and that, I think, is what
> Intel is counting on.

The link is here if you wish to defend yourself over at the Ace Forums

http://www.aceshardware.com/board/general/read.php?message_id=35006601



Want to enjoy fewer advertisements and more features? Click here to become a Hardware Analysis registered user.
Darren Krape Jun 10, 2001, 02:42am EDT Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> [No Subject]
Good lord... they are rather brutal over there.

Darren Krape

Chief Graphics Designer
dkrape@hardwareanalysis.com
http://www.darrenkrape.com
#hardwareanalysis on newnet.net
Raymond Armstrong Jun 10, 2001, 05:47am EDT Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> [No Subject]
Yes they are.
Last time I looked they were talking about HAVING to write SPECIAL programs to flood the CPU with instructions to attempt to make it throttle.
Hey. Makes sense to me. Guess I'll have to run one of those SPECIAL programs each time I use my computer just to try and make my CPU throttle back to satisfy those idot's predjuce's.
Or maybe I could remove my heatsink/fan. That might do it.

Sander Sassen Jun 10, 2001, 06:21am EDT Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> [No Subject]
I couldn't resist doing a posting, as me and Mr. V. Smith have had previous engagements that were of similar nature. The post can be found here:

http://www.aceshardware.com/board/general/read.php?message_id=35006640

Have a good weekend,

Sander Sassen

Chief Executive Officer, Hardware Analysis
Email: ssassen@hardwareanalysis.com
Visit us at: http://www.hardwareanalysis.com

Sander Sassen
Editor in Chief - Hardware Analysis
ssassen@hardwareanalysis.com
Scott Tribe Jun 10, 2001, 10:53am EDT Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Aces is a good site
Hey Sander.. my origianl posting I quoted wasnt even from Van Smith.. but from a different user.. I'm not really too concerned about the feud you may or may not have had with Van in the past.. I was more interested in seeing a posting of a rebuttal to see what your answer to this poster was... You should know that you are regarded with some suspicion over there (myself included I admit) over whether you treat Intel with more kid gloves then you should be... and whether you give AMD a fair shake or not (By the way.. that brings up another point - you editorialized over why you thought the Athlon MP wouldnt do well against the Xeon - did you do a benchmark testing on here against it to see how it would do? Other sites show that it stacks up very well against its higher Mhz competitior)

As to the other couple of comments in this thread from a couple of other users . Aces HArdware is regarded as a very technical site.. it has a good reputation around as such... there ar some highly partisan discussions in the forums .. but thats common at every site, not just Aces... the technical message board is very good reading for those who want a more systematic looks at how things work or may work...

But.. I am sure Johan and Brian dont need me to defend their site



Scott Tribe Jun 10, 2001, 02:16pm EDT Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Van Smith replies to Sander's Rebuttal

Paul Tiseo Jun 11, 2001, 12:26pm EDT Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Testing Is Empty
This editorial is rather pathetic. Note I am not saying you are, but rather that the article is lacking. Personally, I am no conspiracy theorist and will not level anything against you or Van Smith.

This article you wrote is empty of worth as far the issue at hand is concerned, namely that of possible throttling on the P4 affecting performance. Basically, you misdirect the reader. Why you do this I will leave to you to explain. Perhaps you didn't think your test through?

There are two issue: thermal throttling and thermal shutoff. The latter is one that serves as valuable protection of the CPU and Intel has had for a long time. AMD has finally come on board and added it, seeing as it wants its AthMP to get beyond the desktop market.

However, by looking at correct and incorrect heatsink installations, the crux of your article, you solely tested the thermal shutoff and never addressed the thermal throttling issue. I admit that this must be a hard setup to create, but you go about it all wrong by allowing the CPU to heatup via bad heatsinking in certain scenarios, undoubtedly triggering the shutoff. You compare bad heatsinking to good, thus it is an article addressing the shutoff, *not* the throttling. To test throttling, you must undoubtedly walk a fine thermal line that doesn't allow the CPU to overheat and shutoff or act so erratically that it fails the benchmark run. The pass/fail test of yours is too black-and-white.

Also, is Sysmark the best benchmark for this? I guess fundamentally, what you are showing is that with a good heatsink well installed, a desktop user might not have anything to worry about as any difference will be imperceptible, but one would think that you'd want to test for different situations that possibly max out CPU utilization where throttling might occur. Server load sims, or scientific calcs would be an example...

But, the fact remains that this test is not conclusive nor does it explain away what Van Smith has shown.

Also, I'd try not to brag about your patents and degrees in this or other forums. It seems that whatever education you have, which seems extensive, still allows you to make mistakes in others.

ádog

Sander Sassen Jun 11, 2001, 08:38pm EDT Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> [No Subject]
Paul,

I guess the real question you’re asking is why I didn’t step up to the plate to meet InQuest on their own ground and prove that their conclusions are flawed. And you’re right, maybe I should have, instead I choose to take a look at the matter in a different light, please let me explain why.

I did look at the applications and conditions under which they claim the throttling occurred, I however failed to comment on and properly address the issues raised by my findings therein. I realize now that that is what I should’ve done. Their Quake scores I could not reproduce, whatever I tried. I researched the ATLAS benchmark problem with the P4 only to find out that the results could not be easily reproduced as the problem may have been due to a cooling issue, a fan duct misaligned or something.

The other benchmarks they used did nothing for me at all, or they must have had a cooler that was hanging by one clip or something. So what then? Should I comment on my findings and come up rather empty handed, with actually doing no research of my own into whether the clock throttling occurs or not. All I’ve done sofar was look at what they’ve done and trying to reproduce their findings, which I was unable to do.

At that point I decided to not try to prove that they are wrong, but rather take a fresh, new approach. I decided to investigate whether I could prove that clock throttling occurs or doesn’t occur under normal circumstances, for example when applying a Photoshop filter, rendering a video sequence or some other pretty average, yet high-stress, desktop application. If I can show that clock throttling is actually happening with these apps then Intel has got a serious problem, as normal desktop applications such as these should not trigger the clock throttling feature.

In that light I’d like a benchmark with apps lots of people use, something that is a cross-section of a general user. As frankly any CPU can be brought to its knees with some custom benchmark or some other program 99.99% of the users wouldn't care less about. So I'd like something that an average user would feel comfortable with. So I thought of using SYSmark, SYSmark 2001 is too new though, I don't feel comfortable using it yet, as I don't know what is optimized for P4 or not and it has some quirks. SYSmark 2000 is old, but isn't optimized for P4 for sure and it is a good cross-section of popular apps. So I decided to use SYSmark 2000 as benchmark or rather collection of apps to stress the CPU with and determine whether throttling occurs or not.

I tried very hard to see whether it did, with a variety of thermal conditions that I feel are representative of the conditions under which you’d think throttling would occur and couldn't see it happening. I think it is a fair choice I made both in benchmark app. and testing environment, if you feel differently then I’d like to hear your solution as how else I should’ve gone about triggering the throttling.

Another good point, and one I failed to mention in the article, is the simple fact that there’s a huge amount of P4 systems already sold, the majority by Companies such as Dell, HP etc. And I’ve not heard of any of them being returned due to lacking performance or throttling issues. Most people that buy a P4, and are of an inquisitive nature would’ve looked into the matter when the rumor surfaced. But then again, they might have run into the same problems others have, how do you go about testing?

Others have suggested writing custom apps or utils to try and trigger the feature. The danger of doing so lays within the task you set yourself out to do, do you WANT to trigger the feature? Or do you want to SHOW that the feature can be triggered? These are two entirely different things and under the control of the programmer/reviewer, if he’s not careful in describing these, or rather make the distinction himself that could lead to all kinds of flawed conclusions. Disclosing a program’s source code is a way to circumvent some of that, but that makes it more difficult as only those with an understanding of the coding used will be able to grasp the underlying code. Most people will glance over it and accept it for what it is, not a desirable situation, and certainly one that I will avoid.

So yes, I’ve focused heavily on the cooling issues, what I feel to be the core of the problem. I think the clock throttling feature is not an issue if you have proper cooling, both on the CPU and in the system. High thermal ramp rates and hotspots will become a bigger issue in upcoming chips from both Intel and AMD, and any other manufacturer putting lots of transistors in a small die. Actually AMD has also included ‘clock throttling’ or a similar feature in their Athlon CPUs just look at page 10-11 of this tech document, http://www.amd.com/products/cpg/athlon/techdocs/pdf/23792.pdf

Anyway, I think we’ll need some further investigation to determine whether the clock throttling feature really is a problem, or just a fail safe device. I think I demonstrated that it is not an issue when running desktop applications, which would be most people’s concern. It might be when running high-stress scientific applications for longer periods of time, but the evidence for that is lacking, or at least I couldn’t find anything that is conclusive.

Hope this answers some or rather all of your questions/concerns.

Best regards,

Sander Sassen

Email: ssassen@hardwareanalysis.com
Visit us at: http://www.hardwareanalysis.com

Sander Sassen
Editor in Chief - Hardware Analysis
ssassen@hardwareanalysis.com
Paul Tiseo Jun 12, 2001, 09:47am EDT Reply - Quote - Report Abuse
Private Message - Add to Buddy List  
>> Reply To Sassen
Hi Sassen,

Thanks for taking the time to reply to me. You are right in that the issue here is how does one test this feature out to see if it is a potential problem in a way that avoids bringing in the personal and the subjective that simply clouds the issue.

The issue is two-fold. How do I see if it [thermal throttle] is being triggered and how do I know if it impacts my operation in a negative way? I did try to think of something before posting this reply. I don't want to come off as one of these posters who complain just to complain because their pet hardware is a target. :)

IMO, the only way I can that one can test this as conclusively as possible is to knock it out selectively.

According to Intel spec sheets[1], the throttling feature is BIOS-enabled into an automatic mode. Also, there is an on-demand mode that can be triggered via bit 4 of the ACPI Thermal Control Register.

So, I will assume that all P4 mobos have this "auto mode" hardcoded to be on.

Also, I'm assuming the OS you'd want to use doesn't turn on the On-Demand mode on unless you activate some software that will do this for you upon your (or the software's) instruction. (IOW, that the OS is running a silent monitoring of some sort.)

So, first, one would have to re-engineer the BIOS feature so as to make it user-selectable. I'm sure this is not easy. Someone knowledgeable would have to putz around with mobo BIOSes to do this in a stable manner. (Perhaps you could get with your contacts at mobo makers to do this for you?)

Once that's done, run a varied set of benchmarks on a set of P4s with the feature enabled and disabled. And, not just on one P4, as you'd want to get a representative sample that will allow one to address the possibility raised by some that some P4 might be more sensitive than others. Run any and all benchmarks with the feature on or off. See what changes, if anything.

To get more thourough, create a small app that allows you to actively turn on the On-Demand Mode while operating, rather than via a reboot and BIOS change. Run a constant bench without it on and then turn the On-Demand feature on at some time into the bench. See what happens.

This type of setup will allow you to test any benchmark, from a purely synthetic, contrived one to any app you [the user] commonly uses, thus addressing the problem of applicability and veracity all in one.

Basically, neither you nor Inquest have done anything conclusive to elucidate if this is a problem for all P4 consumers, some or none. Inquest provides circumstancial evidence to raise questions and you try to introduce doubt into that. What made it worse is you and Van Smith took it as an opportunity to p**s on each other rather than build on each other.

But, then again, if I was a real good conspiracist, I'd say you were both in league together to generate hits on your sites via scandal and hoopla... ;)

adog

PS: While AMD may have added a thermal shutoff feature, I haven't heard that they have added a throttling feature. If they have, please point out specifically where they document this. I looked through your link and it didn't say anything of the sort.

[1] See section 7 of http://developer.intel.com/design/pentium4/datashts/24919803.pdf


Write a Reply >>


 

    
 
 

  Topic Tools 
 
RSS UpdatesRSS Updates
 

  Related Articles 
 
 

  Newsletter 
 
A weekly newsletter featuring an editorial and a roundup of the latest articles, news and other interesting topics.

Please enter your email address below and click Subscribe.