'Laptops w. Intel Series 5 chipset can not take full advantage of fast SSDs'

Hey guys,

Firstly huge thanks to JJB and Stamatisx, i have a feeling you guys are going to save my hearing

Im looking at applying the tweaks simply for the reason of getting rid of my CPU whine. Im also running a SSD, but the performance hit doesn't worry me too much.

Im wondering if you guys are able to come up with settings that would be specifically targeted at keeping the CPU out of that C1 state that we assume is causing the CPU whine? Would be handy for a bunch of laptop owners, and might possibly be able to avoid any significant battery life loss and CPU temp increase (not that you really seem to be having that problem anyways).

Unfortunately i dont understand anywhere near enough of it to try and come up with the settings myself, but im happy to let my laptop (Envy 14, i5 520m, 160GB Intel G2) be a tester for when the whine goes away with whatever settings (if it does go away at all).

That way im hoping we can come up with a solution of bare minimum changes that can help people with a CPU whine problem without worrying about the SSD performance (and with minimum loss to battery life and increase in heat).

Watcha think ?

 

Until one (or some) of those wonderful folks come with a specific solution regarding the CPU whine, you might meanwhile want to take a look here just to see if it could not help you a bit...

Cheers !

eYe

 

Right !

I was talking about RAID controller because since the RST version 9.6, this controller now DO INDEED support the TRIM command, but in order for this command to work with raided SSDs, those have to support it, which is going to be implemented in the next ATA standard, the ATAPI-8.

The chipset (ICH10m) and the processor (CPU), while possibly made by Intel (other manufacturers make some too) have nothing to do with the TRIM command which, to be working, needs together the 3 following:

1. An O/S that supports the command;
2. A Controller that can rely the command from the O/S to the drive;
3. An SSD that supports the command.

We have everything at that point except for the drives to be able to swallow the command from a RAID controller, and this is where the ATAPI-8 will save our life...

So basically, we first had SSD and O/S that did not support (not even knew) the TRIM command;

Than we got an O/S and SSDs that support this command;

Now, we're only awaiting a RAID controller that will transmit it, which we got (IRST v9.6 and above do supports TRIM in a RAID setup, but ONLY for SSDs not being part of the RAID array).

Therefore, for TRIM to work in RAID, we need the drives to be able to get it from the RAID controller, which the standard ATAPI-8 will provide us with.

Amen

 

I was being satirical, meaning Intel had the CPUs and the ICH10 etc., in essence design control of the entire platform, and their own SSD's and we have this issue with the platform and 4K writes……………..

 

yep, I get your point

On my side, this issue is not with the 4K WRITES, but the 4K READS...

But I agree with you: Intel needs to fix this as they are the only one who can, in my case, since, as you say, they control the whole platform, as I got:

Intel Board based PM55;
Intel Storage/Raid controller (IRST 9.6.3.1001);
Intel SSDs in RAID0...

So ya, Intel needs to fix this, intel-ligently and dil-ligently

ok bye...

 

Tried that and many more solutions. I have even narrowed the noise exactly down to the CPU, and almost exclusively to some form of low power state.

There are good words about this registery / power options fix, but i just dont want to waste any battery if its not needed. Im sure there are multitudes of others like me (and not just Envy owners).

Either-way im at the mercy of the more brilliant individuals at the forum, hopefully they can commit some more time and branch into this particular problems aswell

 

Does anyone know if using the "restore plan defaults" button in the advanced power plan settings will still do that since we have added all the options from the registry?

I really messed up last night when I moved my computer with the power plan window open, like an idiot I didn't notice that it had switched to the 'power saver' plan and then it took me 30 minutes of tweaking the wrong plan before I noticed it.

To be more specific, will the 'restore plan defaults' button on the Advanced plan settings window just restore settings for the plan that is open for adjustment, or will it restore all of the 6 power plans at once????

 

It will restore only for the current power plan opened (not activated).

ie,
Power option > change plan setting > change advance power setting > High performance > Restore plan defaults = will restore High performance advance settings to default

 

interestingly enought, in my newly created power plan, the option to restore to default is disabled, can't use it...

 

there is no tweak
133 x 25 = 3325 Mhz
in theory the i7 620M goes up to 3.3 Ghz

some others could have 266 x 12
a x12 multiplier

 

Thanks for the quick replies, even though they conflict with each other, LOL.

Actually I noticed a 'custom plan' I made has the 'restore defaults' button greyd out but it seems the Win 7 plans have the button active. Which is good I really didn't want to have to go through all the 30 some new options we added for each plan setting again

 

But you had both cores running at those speed, didn't you?

EDIT:
Post #362

 

I also noticed that with certain settings my 540M would actually run up to 2.93Ghz in more than 1 thread when it is supposed to max out at 2.8 when two or more threads have a load. I also noticed that when in this 'mode' the max turbo of 3.06Ghz never worked (as far as I could tell) and when I loaded all 4 threads with a stability test they all maxed out back to 2.8 which is what it's supposed to do.

 

I'm not sure if it's that
I'm going to change the post 362 that u're refering to in order to avoid confusion

I was using HWINFO32 which is not very good to show the real cores states
It kind of looks like they where both boosted at the same time but that I can't be sure until further investigation using the right tools for that

what I can say is that with the idle disable tweaks coresmark shows a max frequency of 3,067 Ghz
and with the JJB tweaks coresmark shows a max frequency of 3,333 Ghz a boost of x2 multiplier
which doesn't show how, how long and if both or only one core reach that frequency (resource monitor not being precise enough)
both ways show a significant increase in ssd 4k speed

all my important results are here
Post #395

 

When I said:

interestingly enought, in my newly created power plan, the option to restore to default is disabled, can't use it...

It was more an irony as you can't set back to default a brand new power plan you just created, as the way you did create it become de facto the default for this particular new power plan you just created...

Sorry for any confusion...

 

Here is the results I have with what, to me, appears to be the best tweaks to improve 4K read numbers while having little to no impact on power / heat. This version does have lower Seq and 512K read speeds but not below where they were without tweaks. I think I like this option since in real world use the Seq and 512K read speeds were blazing fast to begin with, the bottleneck was the 4K read numbers. It takes 5 or 6 adjustments to get to these results and it appears that changing just 1 or 2 of them will allow full Seq and 512K speeds at the expenses of slightly higher (+4 -5C) temps. I may use the higher performance settings for 'plugged in' and the lower seq speed version for 'on battery'.

I think this may be the best all around option with the the least power impact and so far it seems to have no battery life impact.

Your Thoughts?

Here are the results, CDM 5 x 50MB, CDM 5 x 1000MB and AS SSD. The temps shown are after waiting 10 minutes after test runs so should be full idle temps (or close):



Setting changes from 'High Performance' defaults for above runs:

NOTE: Make sure to start with the default settings for the: High Performance -- Plugged In -- Processor Power Management -- Advanced Power Plan settings. (if you use 'balanced' or 'power saver' other default settings will be different)

Proc. performance increase threshold = 10%

proc. perf. decrease threshold = 85%

Proc. perf. decrease policy = ideal (not sure if needed, seems to lower idle wattage)

Proc. idle demote threshold = 85%

Proc idle promote threshold = 95%

Proc. performance core parking overutilization Threshold = 95%

I have the 'minimum processor state' = 0%

These changes also seem to make the CPU idle speeds and the individual CDM runs more stable (less variance).

Please give these a try and provide feedback when you get a chance. If you want to switch to full performance just change the 'idle promote...' and 'core parking overutilization...' from 95% to 100% (or try 99%) and you should see full max speeds with slightly higher temps.

BTW my CPU idle wattage with all above settings is ~5.75W vs. ~5.25W in power saver with no tweaks, not bad for the increased performance.

 

wow impressive
with your tweaks my gain is 4k : 0.8/6 MB/s
for only roughly 0.4W more

happy to hear that you finally found the best solution among 192 possibilities if we take into account only the 7 variables and only 2 states for each one except one

Have you thought about the variablilty ?
in order to find the best combination I was running wprime with different values
To avoid variability due to some background processes I started windows in safe mode
even tough when I started wprime several times without touching anything else i noticed a variability in the results of 2.6%
when opening windows explorer this variability increased up to 6.9%

making it harder to know exactly the impact of the variation of each variable on the performance

 

from the looks of this my guess is that each person's laptop is going to need different settings to keep power consumption as low as possible when this tweak is done i guess..

 

Your'e propbably correct.

I tried to take this into account by changing settings that had 0 to 100 ranges and dialing them 'down or up' to the point just before performance started degrading repeatably, instead of dialing them in to max performance gains. My thinking is this should give a greater overhead for all the various CPU / SSD combinations, this is most likely still going to have a big variation between platforms but it will be interesting to see how it works on other systems.

 

Yep, you can get a headache just thinking of different ways to try to test it, and you forgot that at least 4 options have a 0 to 100 (or more) range, so a lot more than 192 .......

I have decided for my own test runs (CDM) to perform them with no other apps actively running, some are open like HWmon, IE8 etc. but not actually using the CPU when testing. My reasoning is that I have found that when running any CPU intensive app in the backround (super pi, Wprime, stability test etc.) the CDM scores always are better than the idle numbers. So by testing at idle I am making the assumtion that performance should be better when running other things at the same time. I even noticed just listening to iTunes in the backround improved the numbers a bit, especially the seq & 512K read speeds. This is another reason I don't mind seeing lower numbers on the seq & 512K read in the 'at idle' test runs, I figure if I'm doing something the demands max read speeds that there will be some CPU activity to keep it out of idle, and if not I can live with 225MB/s vs. 250MB/s which you probably wouldn't notice anyway....

 

Which one are you referring to.? There is 3 options that begins with the mentioned name.(weighting, threshold, history threshold).

 

Threshold. Good catch, sorry about that, I am going back to my post and correcting that.

Thanks

 

How to prevent SSD degradation without TRIM... ​

 

One more q'n...Is this setting for 'Plugged In' or 'On Battery'?

 

Here is an example of what I mentioned on the previous page regarding gettting even better performance with these settings when you have anything with even a light load on your CPU. In other words my settings seem to give (almost) full 4K performance at idle with no CPU activity and shows a slight, ~10%, reduction in Sequential Read and 512K reads. Now when you have even have a minimum load on your CPU your peformance in Seq and 512K reads jumps back up to maximum. As you can see below the first run is at full idle and the second run is with just iTunes playing in the backround which shows a variable 0% to 3% CPU load:

CDM at idle vs. with iTunes music

PS: Above iTunes results were with 'Life in the Fast Lane' by the Eagles playing in the backround. Seemed an appropriate choice

 

I am currently running these settings for the 'High performance' power plan and the 'plugged in' settings. Make sure you use the High Performance power plan with the default settings to start with before making the changes otherwise other 'processor power management' settings may be different and get you different results.

Another good catch, I guess I was being lazy when I typed the post, I'll go add a note about this also. Thanks again, + rep

 

Thanks for clarifying. Here is my results on a QM57 chipset. By the way the i created a new plan off the Default High Performance. The default for the 'Min Processor State' when 'plugged in' is 100%.

The temps are between 54-58C
Here are the results with CDM 3x1000MB run.

 

OK I'll change the 'minimum processor state' on my post also, i did note that I forgot what default was on that one... Thanks once again.

For reference could you maybe add what normal CDM numbers were without the tweaks (maybe just run in 'balanced' power plan defaults) and also try it with the tweaks and while playing itunes music (or another music player) during the CDM run...

 

in fact the value is more like 7.5690896395201006508054459784927 e+62 possibilities

 

Well with that many possibilities I am sure there is much room for improvement over my settings then, or I got really, really lucky

 

Here is my updated results with various plans and load, etc.

 

Hmmm, I noticed that you ran CDM with the 1000MB setting, I used 50MB and have noticed a big variation between the two. Also it's odd your performance dropped so much with the load, I also use Capture NX 2 and I'm not sure if that's a good choice for the test run, I have found it to be a very poorly written program with regards to fully using system resources, the same batch files with Adobe CS4 equivalent adjustments always runs much faster, I know this is not an identical comparison but adobe is a much more Win 7 friendly program.... Good news is the 4K numbers improved for you either way but you took a hit in the other areas...

 

Ok..Looks like NX2 really sucks. I don't see any difference in usage feel, but the way it utilizes the resources is so bad.

Here is my updated results with Photoshop CS5. Much better.

 

There's only one option with overutilization. Others are over utilization (with space)


Processor performance core parking overutilization threshold

Processor performance core parking over utilization his...
Processor performance core parking over utilization we...
Processor performance core parking over utilization history

 

On a side note I read that with a quad core and CS4 when doing large batch files; If you just select a folder to process it will only use 2 cores, but if you open the folder and 'select all' then start the batch process it will use all 4 cores. This may be fixed with CS5 already but you may want to try it....

Back on topic. @LOUSYGREATWALLGM, Have you tried the new settings yet?

 

Yep. Actually just finished testing with different settings and found one good.

I'll have to sort them first then post

Okay here it is:

EDIT: Its 100MB not 50MB (need to run at 100MB to compare with my previous bench tests b/c all were ran at 100MB )


New settings: (in red text)
4K is more stable (individual CDM runs)
No changes on the CPU idle wattage & CPU temp

JJB's latest settings
Proc. performance increase threshold = 10%

proc. perf. decrease threshold = 85% (changed to 10%)

Proc. perf. decrease policy = ideal (not sure if needed, seems to lower idle wattage)

Proc. idle demote threshold = 85%

Proc idle promote threshold = 95% (changed to 99%)

Proc. performance core parking overutilization Threshold = 95% (changed to 99%)

I have the 'minimum processor state' = 0% (changed to 5%)



@JJB
Can you also try changing the following and see if it will give you more stable 4K read/write?
proc. perf. decrease threshold = 85% (changed to 10%)

 

My above changes in blue actually give me much more stable 4K R/W's. I will try your version also but I think we may have different goals. I am trying to boost 4K R/W with the lowest possible impact on wattage and temps, if you look back a page or 2 you'll see I found that with any load at all on the CPU (I used itune music for posted CDM test run) then full Read / Write speeds are intact. My thinking is that I really don't need the max numbers at idle, any time I will want full speeds it will be when some program is putting a load (however small) on the CPU, so as long as the 4K numbers are immediately available (at idle) then all should be good and no watsed energy or extra heat.

Try my revised (blue) settings above at idle and then run it again with a simple app running (play some music) and I bet you'll get the same results as your settings but with much lower temps.

That is the messiest reply i've ever made, LOL, with all the color changes etc, i just previewed it and I hope it make sense....

 

I didn't see any increase on my CPU idle wattage and CPU temp after I did my new settings (red text)

Do you?

No prob. Got your point anyways

 

Yeah as soon as I changed to 99% the temps jumped about 5C and idle went up 1W.....

I also noticed that the idle CPU clocks were jumping around all over the place (in HW32) with my settings they seem to stay much less active until they see a load.

EDIT: here are is a CDM comparison of my defualt 'balance' power plan (no tweaks) and my new revised High performance PP tweaks above (blue) both at idle and with iTunes playing music. Note the added music with my settings increases SSD performance while in default settings it decreases it. I think in general what I've done is get the CPU to leave low idle state with a much lower CPU load (1 to 2 %) which should provided full performance with any real app that may require full R/W speeds.

Balanced PP default, Idle vs. with iTunes

High perf. PP with tweaks, Idle vs. with iTunes

 

Okay. I'll use 95% then.

 

To get one thing clear; If i change the values in regedit that Static suggested (to disable cpu drivers), will turbo boost still work?

 

yes it will

 

Then what do those drivers actually do?

 

It will go up to a minimum guaranteed of 2.8GHz, but it can hit up to 2.93 if the cooling/power will support it. The base frequency is 2.533GHz, and I believe it can do up to 3 bins if the cooling supports it and both cores are loaded. But it won't do 4 bins unless it's just a single thread.

 

@Pitabred

I thought it was 2.8 max if both cores (1 thread or more each) had a load. But I do recall the speedstep info talking about allow max speeds based on the thermal readings so i guess that makes sense.

Another question, I noticed in the HWinfo32 app (just installed) that it has a temp sensor for the intel IGP. Never seen this reading in anything else and my question is why does it seem to change temps independantly from the CPU cores if the IGP is disabled???

I would expect the IGP temps to change but basically in parralel to the CPU cores due to thermal tranfer if it was truly disabled, but I am seeing it jump around more like the IGP is 'active'. Do you know of any way to check if the IGP is truly disabled? I am wondering if this may be why the idle power draw (in battery bar) always seems to be 3 to 5 watts above what I would expect the the system to be using....

 

I have no idea how to check if the IGP is disabled, unfortunately. And I believe that the 2.8 is actually the minimum, and the dual-thread max is basically as conditions dictate but still less than the single-core max. But I don't know for sure. I do know that I routinely see both cores peg at 2.93 under heavy loads such as transcoding via Handbrake.

 

That sounds good to me, I guess I never really have anything open to show me what the speeds are when under a heavy load, so it just caught my attention when testing these new processor settings and appeard to be higher than I expected.

I'm not sure if I included the most recent setting info on the latest PP tweaks for SSD 4K speed improvement when I PM'd you. Look at page 47 for what we have narrowed things down to....

 

JJB have you run wprime?
I get better results with CDM but not with wprime
never mind I get the same results as with default settings must be something else

 

I'm having trouble finding all the various posts that shows the steps involved with tweaking the different settings to enable ssd performance.

Can the OP put up a rundown of what to do in the first post so that all the new guys coming into this thread don't have to scour through 50 pages to find 3 posts that pertain to the solution. If not no biggie, I'll keep digging.

 

i second this.
phil could you, pretty please