S-XPS 1645 Throttling Info. and Updates

So today I hibernated my XPS instead of sleepmode which i've been using for the last week.
The last week has been throttle-free, i've had TS force 100% CKMOD and max turbomode for the whole week and it's been 100% throttlefree.

Today, after resuming from hibernation throttling has returned in the first game I tried. It's back to the normal throttling i've been seeing before.

I rebooted but I now cannot get back to being throttle-free and I've no idea how to reverse the hibernation-problem.

For me it's obvious that this is software related more than anything..
I haven't touched my laptop because it's on a desk where I've left it all week long. I have not touched the AC adapter and the temperatures are the same.
Yet, today, it cannot go without throttling where as 3 days ago it was fine running prime95, furmark AND wmp at 90c without throttling.

Yes the XPS has heat problems and yes 90W was not enough but the remaining throttling issues are definitely software related.

 

I agree, mine doesn't get too hot.

I'm not much of a gamer (I use the system for CAD and photo editing), but dragged up a copy of Half-Life 2 I had laying around.

Played for 1.5hrs with every video setting on its maximum. According to HWMonitor - CPU Max: 79/67/72/67, GPU Max: 68.

The room I was in was probably 21°C, the 1645 was sitting on my lap partially blocking the underside vents. No cooling pads or anything.

 

I think i've found the solution, i'm gonna do some tests tomorrow to recreate/test.

 

That might not be a good game to test since it is so old. If I play Biohazard 4, with max resolution and graphics settings (not that there are many), I get a max GPU temp of around 70C, but if I play Left 4 Dead, I go above 90C for my GPU max temp (used to be 94C before I removed the vent covers, which is a ~25C difference between games). This is with my 4670, which runs "a little" hotter than the 5730.

 

Very true. Downloaded the demo of Mafia II and played that, temps were the same as for HL2. Thinking of grabbing a copy of Mafia II, so will report back on how that goes.

(I am assuming Mafia II is actually going to test it. I figure it is seeing as I can't run it on flatout without dropping to 15fps.)

Currently run the mafia II demo at:

Global Settings – Custom
Resolution – 1280x720
Full Screen – On
Vertical Synchronization – On
Brightness – 3
Anti-aliasing – On
Anisotropic Filtering – 4x
Shadow Quality – Med
Ambient Occlusions – On
Geometric Detail – Med
APEX PhysX – Off

 

Yep - all fixed



Uploaded with ImageShack.us


Re-creating throttle problem - check



Uploaded with ImageShack.us

....and back to being fixed



Uploaded with ImageShack.us


In the Prime95 window you can see the times & date of the screenshots.

 

Ok, how did you fix it, Duran?
Just got my new system in. Haven't noticed throttling yet, but curious as to when I will.

Original System:
Scratch and Dent: XPS 16 (1640) Laptop:
P8600, Blu-Ray, 500GB HD, 4 GB Ram, 3670 GPU, 16 inch Edge-to-Edge HD Widescreen WLED LCD (1366x768)

Replacement: Brand New...
XPS 1645 840QM, Blu-Ray, 256 GB SSD, 8 GB Ram, 5730 GPU, RGB LED, WIMAX

 

It's easy..
Hibernate--> throttling, shutdown won't fix it.
Sleep --> fixed

 

Back in the early days of the XPS 1645 throttling problem, some users used to have to go into sleep mode and then immediately come out of it to get rid of a keyboard stuttering/throttling issue if my memory is correct. It sounds like the same sort of thing is still happening many bios versions later.

Great to hear that you finally got this figured out and can enjoy your laptop now.

When you are running ThrottleStop in monitoring mode only without the clock modulation boxes checked, are you still able to maintain the full 12.00 multiplier and 100.0% clock modulation reading when running Prime95 + Furmark? I'm just curious if you need ThrottleStop at all now or can retire it. I don't mind. It is still very useful for the Core i7-920XM/940XM and Core 2 Extreme owners.

What GPU do you have and what driver are you using?

If you still need to use ThrottleStop, your laptop uses clock modulation but does not use chipset clock modulation. There is an option that you can add to the ThrottleStop.ini configuration file so it doesn't monitor for chipset clock modulation since that feature is not needed on these laptops. Just add this:

NoChipset=1

 

This is what happens.



Uploaded with ImageShack.us

Not perfect but still good enough IMO, it's furmark + prime95 after all.

I have the 4670 and went back to latest GPU driver from dell. I went back to see if it was GPu related some weeks ago. The drivers are fine IMO and include the HDMI audio driver which is nice.

 

The Dell engineers decided a little bit of throttling during this stress test should be acceptable to most users because games and other applications typically don't fully utilize the GPU and all 8 threads of the CPU simultaneously. It looks like your laptop is finally running as designed.

Now that you have clearly shown how this throttling problem happens and how it can be solved, maybe someone at Dell can fix this with a bios update someday.

The multiplier throttling by 0.5 is a hell of a lot more tolerable than clock modulation killing you and dropping your CPU down to 25% of its rated performance. Thanks for posting the details and it's good to see that ThrottleStop may not be a necessity but at least it is still a little useful to maximize performance.

 

I almost never use hibernate on my XPS 1645, so I decided to give it a try to see if I could recreate Duran's Throttling problem. It turns out, hibernate doesn't cause any additional problems for me, so it must be on a case by case basis. I have i7 720QM with 4670 with BIOS A09, and otherwise all the latest Dell drivers.

 

gpig: I guess the only other interesting test would be to see if using stand by mode makes any difference to the average multiplier when running the Prime95 + Furmark test. You could run this test for 5 minutes after doing a clean boot and then stopping the test, go into stand by mode and then come out of stand by mode and test again.

 

Should I try high or low screen brightness? If I put it on lowest I can probably go 5 minutes with a steady 12.00 multiplier.

 

Any test you can think of that shows if this using stand by mode trick still makes a difference. I think you were around during the early days of throttling where it seemed like a keyboard driver issue or something weird like that. There must be something odd like this still going on. DuranXL's laptop might be worse than most others but I think there is a bug that is common to all of these laptops. Finding and getting Dell to fix this and the GPU temperature based throttling properly would be a good thing for all users.

 

I ran it for 20min but it changes nothing.. After a couple of minutes my temperatures reach their max and it doesn't consume more Watts after 5min than after 20min.

Before I got 1 FPS in any game after playing for 10min.
Now, none of my games throttle.

So ye, good enough for me

 

LMAO!! A09 does not remove GPU temp limit...it's at 101C!

Code:

Date        , GPU Core Clock [MHz] , GPU Memory Clock [MHz] , GPU Temperature [°C] ,



2010-08-26 15:11:05 ,              675.0   ,                796.5   ,               98.0   ,                      

2010-08-26 15:11:06 ,              675.0   ,                796.5   ,               99.0   ,                 

2010-08-26 15:11:07 ,              675.0   ,                796.5   ,               99.0   ,                        

2010-08-26 15:11:08 ,              675.0   ,                796.5   ,               99.0   ,                          

2010-08-26 15:11:09 ,              675.0   ,                796.5   ,               99.0   ,                          

2010-08-26 15:11:10 ,              675.0   ,                796.5   ,              100.0   ,          

2010-08-26 15:11:11 ,              675.0   ,                796.5   ,              100.0   ,      

2010-08-26 15:11:12 ,              675.0   ,                796.5   ,              100.0   ,       

2010-08-26 15:11:13 ,              675.0   ,                796.5   ,              101.0   ,         

2010-08-26 15:11:14 ,              300.0   ,                499.5   ,               99.0   ,        

2010-08-26 15:11:15 ,              300.0   ,                499.5   ,               96.0   ,      

2010-08-26 15:11:16 ,              300.0   ,                499.5   ,               94.0   ,         

2010-08-26 15:11:17 ,              300.0   ,                499.5   ,               97.0   ,        

2010-08-26 15:11:18 ,              300.0   ,                499.5   ,               94.0   ,        

2010-08-26 15:11:19 ,              300.0   ,                499.5   ,               97.0   ,        

2010-08-26 15:11:21 ,              300.0   ,                499.5   ,               92.0   ,                     

2010-08-26 15:11:22 ,              300.0   ,                499.5   ,               96.0   ,          

2010-08-26 15:11:23 ,              300.0   ,                499.5   ,               92.0   ,        

2010-08-26 15:11:24 ,              300.0   ,                499.5   ,               92.0   ,       

2010-08-26 15:11:25 ,              300.0   ,                499.5   ,               93.0   ,         

2010-08-26 15:11:26 ,              300.0   ,                499.5   ,               93.0   ,         

2010-08-26 15:11:27 ,              300.0   ,                499.5   ,               95.0   ,         

2010-08-26 15:11:28 ,              300.0   ,                499.5   ,               92.0   ,          

2010-08-26 15:11:29 ,              300.0   ,                499.5   ,               92.0   ,          

2010-08-26 15:11:30 ,              300.0   ,                499.5   ,               91.0   ,          

 

Here is my log. First test is regular, 2nd test is after hibernation (I labeled the tests in the log). The first test results are a little better, but I think it's just because the laptop was a little cooler (I thought I waited long enough, but by looking at the temps I guess not). In conclusion, I don't think hibernation affects the performance of my particular laptop.

Edit: note that in the second test Prime was running for 10-15 seconds and Furmark was running for 5-10 seconds before I hit the log file button in Throttlestop.

 

That's weird... I can re-create the problem over and over by just hibernating
Is that with TS off?

 

Yes, it's off.

 

I have a 1647 and here are the logs from SC2. I'm not sure if it counts as throttling, but it never seems to go past 20x... so there isn't any turboboosting.

Edit: Log added. Currently my power usage is 83-85W in SC2.

 

What happens if you turn it on forcing 13x multi and 100%ckmod?

I see no log

 

hmmm the throttlestop log doesn't give you results per core.... so I guess it may be working properly then. It seems that if you shift your focus to a windows that requires less resources i.e. cpu-z or throttlestop, the processor doesn't use turboboost anymore.

However... it is definitely throttling back to the standard level when running Super Pi (51W)... so maybe it's some sort of software issue.

 

The very first line in your log file shows the average multiplier going past 20. You didn't mention what CPU model you have but so far it looks OK.

The amount of turbo boost you get is constantly changing in these CPUs based on how many cores are in the active state and whether the CPU is operating above or below the TDP/TDC turbo limits. That's normal. When you are below the TDP/TDC limits, you will get full turbo boost. If you go over that limit, you will get zero turbo boost. When hitting this limit, the CPU will rapidly cycle between full turbo boost and no turbo boost. ThrottleStop will accurately show you the average multiplier when this is happening.

Post some more details so I can have a look.

 

The laptop has a core i5-520M (22x multiplier maximum), Radeon 5730 and a 15.6" WLED screen.

 

I think this is a misunderstanding of how Intel designed your CPU to operate.

A Core i5-520M has a default multiplier of 18. The only time the multiplier goes higher than this value is when turbo boost is active. Each and every line in the log file you posted shows values greater than18 so your CPU is constantly using turbo boost and is working as Intel intended.

The maximum 22 multiplier is only possible when a single core is active and the other core is asleep. If you open up the Windows Task Manager and go to the Performance tab, you will typically see 500 to 1000 threads running on your computer when your computer is idle. Each thread gets scheduled on both cores so the second core ends up being constantly woken up to process some background activity. The maximum multiplier when both cores are active is 20. The game you are playing combined with the various background activity going on is keeping both of your cores active, the majority of the time. That's why you're reported multiplier is generally around 20. When it goes a little above 20, that's a sign that the second core was able to spend some time asleep so the first core could occasionally use the 22 multiplier.

When the average multiplier starts dropping below 20, that's a sign that both cores are active during the sample period and that you are reaching the TDP/TDC turbo power limit. When that happens, turbo boost will start to rapidly cycle on and off to keep the CPU at or below the TDP/TDC limit. The result is that the multiplier will be rapidly cycling between the default of 18 and the turbo boost multiplier of 20.

If you right click on ThrottleStop, you can bring up a Turbo TDP/TDC adjustment window. Does that window let you adjust these values or are the adjusters locked? The Extreme CPUs are unlocked but so are some of the dual cores, depending on the motherboard. If you were able to increase the turbo TDP/TDC limits then you would be able to increase the amount of turbo boost even when the CPU is fully loaded. Instead of cycling between 20 and 18, the multiplier would never go less than 20 if you were able to increase the TDP/TDC values high enough. On the dual cores it usually only takes a very small increase in these values. If this works, keep an eye on your core temperatures. When your CPU is encouraged to use more power, it's going to create more heat. Some users with Core i7-920XM/940XM CPUs are increasing these values by 60% to 70% which is resulting in a significant increase in full load performance as well as a lot more heat.

If your CPU was dropping down below the default of 18, then that would be a sign of throttling but that's not happening in your log file. Your CPU is running 100% normally and exactly the way Intel intended it to run. It's also not using any of the clock modulation throttling that many other Dell laptops have used and continue to use.

 

Wouldn't that have to not work, since the highest multiplier is 12.00 when running Prime+Furmark (no turbo boost)?

 

Well it just tries to use maximum turboboost and prevent it from going to 11.5 like sometimes happens

 

The 720 can boost to 13, even with all 4 cores active, though it usually won't when working at a full load because it'll hit its TDP fairly quickly. You should still be able to 'force' it to 13 with throttle stop though, just expect it to get a bit toasty, lol.

 

No at full load it won't go above 12

 

Somewhere in this mountain of a thread Unclewebb explained why it can't go above 12x while running Prime+Furmark. You're right though that it can go to 13x with all 4 cores active, but something about the 100% load of Prime on the CPU prevents any turbo boost.

 

If a CPU exceeds its rated turbo TDP or turbo TDC then the CPU turns off turbo boost. That's why when these CPUs are fully loaded with 8 threads of Prime95, you get zero turbo boost and a Core i7-720QM will be running at the default 12.00 multiplier. There is nothing ThrottleStop can do about this since the limitation is within the CPU itself. If Dell decides that the CPU deserves some throttling because of an inadequate power supply, it will be throttled down to less than 12.00 when it really shouldn't be.

On the 920XM and 940XM CPUs, you can adjust the turbo TDP/TDC limits as high as you like using ThrottleStop. They are unlocked on the Extreme CPUs and some dual Core CPUs too. This enables turbo boost to keep going even when the CPU is fully loaded. If you can keep one of these Extreme chips from going nuclear, it's no problem running all 8 threads at a 26 multiplier or maybe 27 if you like doing your computing in a freezer.

Here's a pic of 5150Joker's incredible 940XM to give you an idea of what's possible.



When fully loaded with the TDP/TDC tweaked, these Extreme CPUs are beasts compared to the i7-720QM and i7-820QM and are in a league of their own.

 

Damn... Too bad Intel prices the extreme edition CPUs higher than the price of my entire laptop....

 

Wow... why is this limited to the 920 and 940?

 

They lock the multiplier on their CPUs as anti-counterfeitting and security measure (so they say) like to prevent someone from buying a lesser chip and then falsely marketing it as a faster higher performing one. The 920 and 940 CPUs are part of their Extreme Edition line CPUs which people pay a premium for a higher quality chip capable of running at a higher clock and has more intact and functioning cache, plus it has some of the extra features of the server/workstation class chips including multi-CPU support and unlocked multiplier (which happens to aid in overclocking).

The other piece to this is that the way some CPUs are manufactured is they build one silicon designed to operate at a certain performance level, but due to manufacturing flaws and other defects that can occur, not all are able to operate at full capacity, so they salvage those by reducing the clock (to perhaps reduce operating stress) and/or disable some of the damaged features (like cache, etc), and sell it as a lesser chip. You might even find chips intentionally being crippled/designed to run at less than capacity in order to fill in the supply gaps, and they certainly don't want you to know or take advantage of that (hence the locks and deliberate physical damage you might find on some CPUs, cut pathways or even just simple BIOS/software ignoring/disabling the extra hardware).

 

To put how expensive the extreme edition is into perspective:

i7 620m - $332
i7 720qm - $364
i7 820qm - $546
i7 920 xm - $1054

For an extra $508 over the 820qm, all you get with the 920xm is an extra 266MHz, 10W higher TDP, and an unlocked multiplier.

The only people who buy the extreme editions are people looking for overclocking records, or people with WAY too much money to spend....

 

seeker moc: I think you need to rethink your math.

ebay i7-920XM = $415

Then you can sell off your i7-720QM to recover $250 of that and you are left with a net cost of $165.

This gets you a CPU that can run 8 threads over 100% faster than a Core i7-720QM can run 8 threads. Go run 8 threads of wPrime on a Core i7-720QM and you'll get a 12.00 multiplier on all threads compared to 5150Joker running all his threads with a 26.00 multiplier in the exact same test. Just think how this extra muscle is going to improve every application you run, including games. A typical game might average a 15 multiplier if you're lucky on the 720QM compared to 26 on a 920XM or 940XM.

The 10 Watt higher TDP rating is misleading. With the Core i7-920XM, you can use ThrottleStop and crank up the turbo TDP/TDC rating to whatever you like. Some users are benching with this set to 105W or more. The adjustable TDP/TDC setting can be used to smooth out game play when a laptop is heavily loaded.

This is the first Extreme CPU that Intel has built that is a night and day difference between it and the regular 720QM and 820QM processors. For approximately $165 bucks, no one should be too disappointed.

 

EDIT: All those chips on Ebay are either qualification or engineering samples (beta versions if you will), NOT retail processors. That means there is NO warranty. It also means that they may still have some bugs in them that were later worked out before the retail release, and may have some features disabled. They are also gray market goods, not legal, as they were given by Intel to reviewers and product testers, and were not authorized for further resale, they were supposed to have been returned to Intel. Notice how all their serial numbers are blanked out in the pictures, so Intel can't track them down. Legal release version i7 920xms go for $800-900 used / $1200 new on Ebay. See:

http://www.intel.com/support/processors/sb/CS-030747.htm
http://www.techarp.com/showarticle.aspx?artno=407
http://www.tomshardware.com/forum/262335-28-intel-help

 

Have you ever had an Engineering Sample?? Besides, it doesn't change the fact that you'd be purchasing (technically and legally) stolen goods.

 

The Intel website says that ALL engineering samples are property of Intel, and are covered by NDAs (the paperwork you mentioned), so whichever company they were originally loaned to could be sued by Intel for re-selling them. That's why the serial numbers are removed when you buy them.

Now, I agree that if you bought one of them, nothing would probably happen to you. They're more worried about catching the sellers than the buyers, but you still shouldn't. I don't know how Ebay gets away with allowing their resale.

 

seeker_moc: You are absolutely right that ES processors are stolen property from Intel. I won't argue that point and like you, I would never buy or use an ES CPU. You could say that buying a 920XM ES off of ebay is like selling your soul to the devil. For some users, it's worth it.

If you look at the CPUID, you will see that many ES processors sold on ebay are the exact same stepping as the final retail CPU. When a new CPU is released, all review sites that post a review on the day the CPU is released are all using one of these very late ES processors because they tend to be functionally equivalent to a retail CPU. You never hear about any problems during their reviews. Look closely at the CPU-Z screen shots at Tom's Hardware, etc. of any new CPU being released and you will usually see the ES letters in CPU-Z. Intel wouldn't release these versions for review if there were still major bugs needing to be fixed. You need to watch out for early ES processors by looking at the CPUID number in CPU-Z.

The other thing to be aware of is that even retail CPUs have a long list of bugs and issues which Intel tries to fix from one stepping to the next. There is no such thing as a perfect CPU.

You can download a Spec Update sheet from Intel for the Core i7 mobile CPUs. The long list of bugs in a retail product was shocking the first time I saw it, especially the bugs that they have no intention of fixing. Head to the Errata section.

http://www.intel.com/Assets/PDF/specupdate/320767.pdf

 

Like the hibernation issue, whenever I play a game after entering standby, my the framerate cuts in half.

 

Hi,

I have A11 Bios running. Is it true that the GPU (4670) do not throttle with A09?

So I am thinking about flashing back to A09, but with the Dell Flash Programm its not possible.

Is there a easy way to flash to A09?

 

I heard that there is something specific in BIOS A09 which results in less throttling issues with the HD4670 though I can't give you the details as to what allows this. I can't even call upon personal experience as yet but I have heard of a number of people flashing back to A08 to get back to A09.

I am sure someone else will jump in with more information.

 

Basically A09 had a higher GPU temperature threshold limit before throttling than other versions. People with 4670 might suffer from throttling on other version because it generates more heat than 5730. I think the common knowledge these days is A09 for 4670 and A11 for 5730

 

Hi,

flashed back to A09 now. GPU Throtttling starts at 100 Degree now. Just wondering that the GPU gets so hot.
Testet with Prime95 and Furmark. Now I have to test if the GPU goes to 100 Degree in a Game.

 

Furmark stresses the GPU beyond what a game will, but even so, 100 is pretty high. Have you tried dusting out your heatsink and/or removing the dust filter from the fan intake? That lowers GPU temp around 5c when under load.

If you hit 100c while playing a regular game (not furmark) then you may have a problem with your cooling system, and should give Dell a call.

 

Yes,

2 days ago I removed also the dustfilters, the Laptop is not old, so there was not much dust in the headsink.
But still 100c with Furmark and Prime95.

Any other possibilities to reduce temp?

Also I still have to check Temp in Games

 

up.........................

 

What are your specs, and what is at 100c, the CPU or GPU?? Either way it shouldn't be that high though. You could try replacing the thermal paste. If not, time to call Dell...