DrMrLordX
Lifer
- Apr 27, 2000
- 22,945
- 13,031
- 136
Lock thread plz, k thx bye
It's a shame it got so out of hand. For what it's worth, Presler is what Smithfield should have been at launch(though the larger cache is totally unnecessary), and Cedar Mill is what Prescott should have been at launch.
An improvement, but too little, too late. Should be fun to overclock em though.
It's a shame it got so out of hand. For what it's worth, Presler is what Smithfield should have been at launch(though the larger cache is totally unnecessary), and Cedar Mill is what Prescott should have been at launch.
An improvement, but too little, too late. Should be fun to overclock em though.
About time, the K8 was thought up in 1999, just a turn of the tide. However, what little lead Intel might have AMD will be right with them.
Either way we win.
Either way we win.
- Aug 22, 2001
- 32,056
- 32,578
- 146
Overclocking is great for the enthusiast, but if Intc doesn't release higher clocked parts than the current line-up soon, the performance gains, and lower heat&power factors, will not be enough to stop AMD's momentum.
Fanboism makes baby jesus cry.
Seriously. Why are we arguing about this when we don't have the most important thing of all? A FRICKIN' BENCHMARK! Goddamn. This isn't the video forum.
Seriously. Why are we arguing about this when we don't have the most important thing of all? A FRICKIN' BENCHMARK! Goddamn. This isn't the video forum.
I can't provide you with all details about throttling. I'm simply not interested in Intel anymore. But it goes something like this:
Intel have progressively introduced a number of techniques to keep heat under control.
Two of them, TM1 and TM2 (temperature monitoring...I think?) qualifies as throttling since they depend on temperature.
But there are others, enhanced halt, speed step and enhanced speed step. Let's start with those. Enhanced halt simply chucks the CPU clock down to 2.8GHz, when it doesn't do anything. It's always active for those P4s that feature it. It simply replaces old halt instruction and system has nothing to do with it. As you might guess by the 2.8GHz number, it uses the multiplier.
Speed step and enhanced speed step is controlled through BIOS and system though. And as you know, it tries to anticipate load and moderates some clocks to keep power down.
It can be disabled in bios, and I would suggest that it is done for any OC-testing. IMO, it should definitely always be active for any Intel system in use. It's actually able to reduce throttling.
Now to throttling:
TM1 simply monitors the temperature at strategic places on the chip, and when it gets too warm, simply stops the internally generated clock from issuing a sync. And as long as it's too hot, it will keep blocking the next sync, thus cutting (50% I assume) internal switching speed by intermittently stopping the clock. It's not visable in any other way than benchmarking performance! ...Or using 'Throttlewatch', which I assume uses some kind of benchmarking mechanism to discern throttling.
Performance during TM1 throttling may very well fall far below core switching speed. The reason is that the rest of the system, like the memory system, doesn't know the core is sagging and thus comes out of sync.
While TM1 throttling becomes active when the core becomes too hot, the new supplementary TM2 throttling tries to brake in good time. Basically it tries to avoid TM1 throttling cutting in. It uses the same clock moderation mechanism as 'enhanced speed step' (and thus, I suppose, only exists on chips that support 'enhanced speed step'). But instead of adjusting speed according to anticipated load, it adjusts speed according to (rate of?) temperature rise. Unlike TM1, TM2 throttling should be readable someway, just like the clock moderation of 'enhanced speed step'.
Legend has it that TM1 kicks in around 60-70 deg C. Thing is, it doesn't really rely on the system's reading of temperature and I've had both a Willamette and a Northwood that apparently started throttling somewhere above 50. Inaccurate readings aside, I suppose it depends on the hotspots on the die. So maybe a bit of luck with the chip, the integral heatspreader and the care in mounting the heatsink also are factors.
This entire P4 throttling technology is only deviced for one purpose: Facilitate high nominal clocks. Clockrates that the clueless customer will happily buy and feel smug about, but that his cheap retail system in reality will not be able to sustain.
An AMD box with inadequate cooling will result in an unhappy customer. The Intel buyer, OTOH, doesn't have a clue.
Like Intel's heavyhanded influence on benchmarking (PC-Mark, SYSMark, 3D-Mark), high nominal clockrate inflates the customer's perception of Intel performance.
Typical retail desktops, like Dell :disgust: ("rhymes with hell"), are spec'ed so badly anyway that they won't be suitable for hard work. And how high is CPU utilization in a surf/email/office use? Just a few %. Even media playback makes very moderate demands.
When it comes to gaming, the (inadequate) videocard pretty much, in reality, flatlines performance across the field of CPUs. Again, throttling may not be immediately obvious.
Only hardcore gamers, who's friends likely have AMD machinery, will realize something is wrong. (...and spend days fiddling with 3D-mark and driver installs, still not realizing it's throttling...)
I'm certainly no OC expert. As you know, I generally don't do that. But I suspect that speed step and throttling in OC, continues to facilitate high nominal clockrates. Clockrates that actually aren't possible to sustain. Benchmark, benchmark, benchmark.
...And give it an hour. It takes an hour.
Markfw
Moderator Emeritus, Elite Member
- May 16, 2002
- 27,274
- 16,120
- 136
Vee, What Duvie and I witnessed, was at 1.5v vcore, and 3.54 ghz supposed speed, after it hit about 55c (maybe less) one core would throttle down to even as low as 2200, while the other core showed full speed, but wouldn;t accept a full load, like 80% ! Then it would go back up. Very stange !
Well, each core has its own throttling and its own on-die heat sensors that control throttling, so the fact that they behave differently is no mystery.
The question is how do they disturb each other. What clock do they have in common? Multiplier or just fsb?
And would TM2 throttling initiated by one core, through some clocking also affect the other core? I have no idea.
And what happens at the FSB when one core starts to retry and retry memory reads due to TM1 throttling? Should affect the other core's performance, I think.
And, what speed step tries to do with a dualcore cpu, again I have no idea.
Taking a wild guess, I'd say one core is TM1 throttling (providing S&M really can detect TM1 throttling?) and memory read retries interferes with the other core's memory access.
Markfw
Moderator Emeritus, Elite Member
- May 16, 2002
- 27,274
- 16,120
- 136
We disabled TM1 throttling in the bios, but this strange behaviour occured anyway. And it was allways core one throttling, and core 0 loosing %load, so it seems strange they could be so different in heat. Regardless, all of this is rediculous IMO. It should be cool and run, period ! (like AMD chips do currently)
I witnessed the same thing duvie and mark did when overclocking my Petnium-4. At stock speeds throttling kicks in at about 72-73c. Overclocked at stock voltage, it stayed teh same, but increasing the Vcore caused the throttling to kick in sooner. I looked in to the throttling stuff quite a bit when I was having so many problems with my P4 550 3.4ghz prescott.
The P4, 5xx series has TM1, but not TM2. The Pentium-D doesn't have TM2 either. Enhanced halt state can actualy be disabled, it's disabled on my pentium-d, but again, it doesn't really do anything for the 2.8, since the lowest multiplier is 14x. TM1 does not affect the actual clock speed, but uses a duty cycle. It's adjustable, and can go anywhere from ~12.5%-86%, basicly it "pauses" the proccessor depending on what the throttling is set to. For example at 33% throttling, it will idle the execution units 1 out of every 3 clocks. TM2 does basicly the same thing, but only lowers the CPU to 2.8ghz and also lowers the voltage. While it does effectivly act like a lower clocked CPU, CPUz won't show it, because it doesn't change the actual clock speed.
Both TM1 and TM2 use the on demand clock modulation feature to control the duty cycle. If things get bad enough, I think it overides the bios settings for the % throttling, and will cause it to throttle even more. There are other programs that can show it, such as Rightmark which will show the effective clock speed, and throttlewatch which seems to just show the % of throttling. Like you said, speedstep will lower the CPU multiplier, and voltage, depending on the CPU load. The lowest multi on 800mhz FSB P4's or Pentium-D's is 14x, so 2.8's can't actualy use speedstep, because they only have the 14x multi.
The P4, 5xx series has TM1, but not TM2. The Pentium-D doesn't have TM2 either. Enhanced halt state can actualy be disabled, it's disabled on my pentium-d, but again, it doesn't really do anything for the 2.8, since the lowest multiplier is 14x. TM1 does not affect the actual clock speed, but uses a duty cycle. It's adjustable, and can go anywhere from ~12.5%-86%, basicly it "pauses" the proccessor depending on what the throttling is set to. For example at 33% throttling, it will idle the execution units 1 out of every 3 clocks. TM2 does basicly the same thing, but only lowers the CPU to 2.8ghz and also lowers the voltage. While it does effectivly act like a lower clocked CPU, CPUz won't show it, because it doesn't change the actual clock speed.
Both TM1 and TM2 use the on demand clock modulation feature to control the duty cycle. If things get bad enough, I think it overides the bios settings for the % throttling, and will cause it to throttle even more. There are other programs that can show it, such as Rightmark which will show the effective clock speed, and throttlewatch which seems to just show the % of throttling. Like you said, speedstep will lower the CPU multiplier, and voltage, depending on the CPU load. The lowest multi on 800mhz FSB P4's or Pentium-D's is 14x, so 2.8's can't actualy use speedstep, because they only have the 14x multi.
That would be the on demand clock modulation. You can't disable it, so it will cause throttling to happen anyway, even if you have TM1 disabled. Took me a while to figure that out myself when my P4 was throttling.
TM1 throttling has nothing to do with bios, any clocks or any external thing at all (except the heatsink). Whatever you did in the bios was something different. Probably speed step, or some throttling performed by the MB itself.
TM1 throttling absolutely CANNOT be disabled. It's entirely contained inside the core itself. It does not work through any clock or manipulation of any clock at all.
And it can ONLY be detected by measuring actual workrate and comparing it to what it should be. Any read of any clock will falsely show cpu to be at any set clock, (set by multiplier, speed step/enhanced speed step, TM2 throttling, whatever) NOT the actual switching rate resulting from TM1 throttling. I don't think TM2 throttling can be disabled either. But I also suspect 800 series CPUs lack TM2 throttling. I think only 600 and possibly 503? have this.
I do not find it strange that one core starts throttling. It's what I would expect. IIRC each core has maybe four diodes. Some single diode is going to trigger first.
And when the first core throttles, it may stop the other core from reaching critical temperature, either by interfering with with its running as I suggested, or by lowering the overall temperature of the package and the heatspreader.
Markfw
Moderator Emeritus, Elite Member
- May 16, 2002
- 27,274
- 16,120
- 136
Well, the bios has a setting for that, and we disabled it (not that it works) All I know is that at 1.45vcore and and open case with fans, and the best air cooling out there, I can only get to 3.43 ghz without throttling occuring, and S&M is the only software that appreas to catch this happening.
Well, I guess stevty2889 is right and there is something called TM1 on demand clock modulation. I'm not up to date on that. It seems to be a way to set the performance degration to kick in earlier and gradually, and sound about like what I understood TM2 functioned like. Apparently, this function now exists for TM1. Anyway, TM1 throttling (remember TM only means temperature monitoring) cannot be tampered with. It's contained inside the core.
Try using rightmark, it should also show the effective clock speed when you are throttling. I haven't overclocked yet, and can't raise the Vcore, so I have had no throttling to verify myself. Even S&M only got me to 63c, which is no worse than my single core @3.45ghz.
As you can see, the dual core algorithms for cooling and throttling are not completely full proof. I would inquire what motherboard you have, as the throttling should be directly controlled by the BIOS and what it reports.
I have Dell Precision Pentium-D systems at work, and contrary to what Vee says, they are put under full load for extended periods (but not 24/7) of time doing various simulations that require a decent amount of FP/INT power. They certainly do not have the "best air cooling" system and they do not suffer from clock throttling (if it did, then the work would be finished at varying times). I was wondering if the more conservative timing/settings of the Dell (eg Intel) board may be attributed to a more stable system. But with that regard, I dont think Smithfield Pentium-D systems should be tinkered with at all, because they are mostly unstable at anything but the most default settings.
Originally posted by: Hacp
Originally posted by: SickBeast
Originally posted by: Acanthus
Originally posted by: SickBeast
Man you need to pay attention to the down low. I do tons of reading online; I don't always remember where I see these tidbits.
Intel's new architecture won't hit mainsteam during 2006 whatsoever though, and that has indeed been widely reported.
The voice inside my head is telling me that you're not very well informed, unfortunately.
So intels roadmap that says "2h 2006" and the fact they already have working 65nm engineering samples (the same process conroe/merom will be on initially) is no indication at all that they might be on track?
So your heavy online research on an architecture that even the engineers at intel havent finalized is obviously going to be:
1. late
2. slower than AMDs offer at the time
Sorry i was skeptical, you seem to know what youre talking about miss cleo.
Ok Beevis, you should really read more, it would benefit your brain which seems to be somewhat lacking as of late.
Everything I have read has slated Conroe for "late 2006" release, and many are speculating that it will not be released in 2006 at all. Even if it *is*, there is no guarantee that it will be faster than the A64, or whatever AMD has up its sleeve at that point in time.
FYI, early 2007 is a year and a half from now, making two years far from impossible.
The voices inside my head are telling me that you own intel stock and that you cannot bear to hear about how horrible their architecture is. Sucks to be you I suppose. :thumbsdown:
Fanboyism at its extreme. Results to personal insults to try to prove his unprovable point.
Hey, he told me I was hearing "voices" and called me "Miss Cleopatra". As others have said, I'm not a fanboy in stating that netbust sucks. Posting a link to an article about Conroe's release date doesn't make me one either. Show me a link that states otherwise.
secretanchitman
Diamond Member
- Apr 11, 2001
- 9,352
- 23
- 91
Well, I believe this to be wrong. I've just read up on the subject, and "on demand
mode" is apparently intended for investigating cooling solutions during testing. For this purpose the TCC can be forced active from external input (instead of being triggered by high temp). By a driver. Also in this mode, the "duty cycle" is configurable in steps of 12.5% from 12.5% to 87.5%.
This is apparently what you're seeing in the bios. I'm sorry, but this does not appear to be able to affect thermal throttling in any way at all. ...Except to delay it since the processor is already throttling, ...if you use an external driver to generate throttling.
Even if "on demand mode" is active when the core reaches critical operating temperature, the automatic mode allways takes precedence, and you get, not the configured duty cycle, but instead the fixed dutycycle. This is fixed and processor specific.
There is no point in having 'on demand mode' enabled, unless you for some reason is running some software, that for some reason uses it. Otherwise it doesn't do anything.
'On demand mode' throttling affects both cores in the 800. The automatic mode, heat triggered, only affects each core individually.
This clearly fits what I, Duvie, Markfw900 and stevty2889 have experienced.
stevty2889 is correct about the duty cycle. Thermal throttling works by blocking the clock, and thus stopping the core from switching, during a time period that is represented by a duty cycle. At the end of the duty cycle the core starts working again. But if the core is still hot, it will shut down again next duty cycle. And so on.
You can't configure this in bios. Period.
(Yes, TCC automatic mode must be enabled, but it always is. It is required that it is. It is enabled during boot.)
stevty is also somewhat correct about TM2 throttling. This simply drops the core down into another operating condition, which includes lower voltages and 2.8GHz.
I don't see this. It may well be that your Dells don't throttle. But I do not see why the work would be finished at varying times if they do. If the load is definitely too much for the Dells' cooling, then they will definitely throttle. And it would be a stable pattern with constant processing times. Wouldn't it?
...And there is no "clock throttling". TM1 thermal throttling does not throttle the clock.
I have no argument with that. I don't OC. Actually that was pretty much the point with my first post, on page one, of this thread.
You've seen dual core P-Ds running at 4.25 Ghz, unthrottled? I mean we can correlate kinda, but using 1.6/1, which is the norm, I'd say, then a 2.65Ghz or so X2 is equal to that. Alot of people (my self included) can hit that pretty easily with AMD's technology. Which is why I say its no big deal.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products- Started by poke01
- Replies: 411
Facebook
Twitter