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

try temporary files or the clipboard maybe

I'm happy to hear that
it makes sense because my cpu needed roughly 5W of power which is quite the same as without any tweak

 

2 X160RAID0Intel16K/strp here.

I tried both Stamatisx and JJB tweaks, separately AND together.

Both separately AND together showed pretty much similar results.

Stamatisx's tweaks are much easier to implement, but gave me higher temps ± 2 ~ 4 C.
JJB's tweaks are much more complex to implement, but allow for more fine tuning.

Both showed me 4Ks R&W increase from ± 15MB/s to ± 24MB/s, and from ± 60MB/s to ± 180MB/s.

Both showed my fans kiking off even in Idle Time while they weren't before.

I am so far to notice a real world difference for my brain to justify the noise to my ears...

The BEST tweak I've found so far to be somewhat a kind of compromise is only using the following from JJB:

1. processor idle promote threshold (99% plugged)
2. processor performance core parking overutilization threshold (99% plugged)

This being said, I would like to thanks both of them for their work, and also all the others here who were willing to give it a shot and share their experience, thanks to Phil as well for starting this thread.

Maybe Intel and Cie should try to think about maybe somewhat if not directly hiring people from here, at least, learn from them...

Let's help them guys; JJB and Stamatisx just showed they INDEED NEED us!

 

k I have found out which feature is the one and the only one required to get ssd improvements
and the winner is
processor idle promote threshold
hey eye-I-aie I'v just read your post you have stolen my breaking news lol

 

You are right as this one (processor idle promote threshold) is the most significant one, boosted my 4K from 14 to 19, and from 60 to 150; the other one though (processor performance core parking overutilization threshold) helped me getting from 19 to 24, and from 150 to 180.

Sorry for breaking your news

 

strange I don't get any improvement with the second one

 

it may be because eye'm in raid...

 

have you checked the value of minimum processor state ?
100%? 0%?

 

this value is set to 0 in both plugged or battery.

 

could phill or some mod add JJB's tweak and the other tweak to first post so its easy for us to find them.. this is a much better tweak than directly disabling C-states.

 

I think when it's been tweaked and tested enough (read: more than two days =) and is ready for mass consumption, it probably will be.

 

I agree, but I think we need to wait of few days. First I have not seen a post from Phil for a while and second, at the rate were refining the tweaks it looks like we will only need a much simpler combo of settings. From reading above it may be just a matter of 2 or 3 registry / Power plan settings, although I question the power draw the simpler version may use, need to do some more testing....

 

I agree with JJB
just 2 or 3 registry gives only a basic power efficiency when I logically look at the parameters
for instance : giving less time between the check (processor idle time check) probably make the cpu answer faster to the power requirement and then being closer to what it is asked (an analogy would be that a captain of a ship trying to go strait forward from A to B must turn right and left all the time to stay on track, the faster the captain answer the closer he is to the strait line)

another thing:
a reason I see why idle disable tweak is quite close to just changing processor idle promote threshold is
that disabling idle state is quite close to not letting the cpu goes into a deeper idle state even thought it doesn't mean exactly the same thing
that's what is done by giving processor idle promote threshold a value of 99% or even 100% which is nearly never reached so the cpu should stay to a lighter idle state with this high value

 

I changed to just the two '99' settings and Min CPU to '0', everything else at default. It seems to work but I am getting wild swings in the individual runs in CDM, for example I have got a 4K of 18.1/32 up to 23/59 all with 50MB x 3. I don't recall having such a variance before and not sure what option may have made it more stable.

Either of you noticing this fluctuation while testing? It's most noticable in seq r/w and 4k r/w....

Edit: I just noticed my CPU idle wattage is up a bit with above settings, now ~6.2W and temps are up a bit to 43/33...

 

I noticed that too

 

Found out about this issue just recently, a while after i found this topic so.
I figured i had one of the chipsets mentioned in the posts around this issue and an SSD (4 actually) so had to try this out. Went into bios and turned of the whole ''C-state''. And indeed after reboot speeds went up quite big.

4k values doubled, and others went up by quite a bit (around 50MB/s). Used some of the tweaks here, but can't get the same performance as turning off c-states totally (in BIOS).

I really hope this issue will be fixed in the future.

 

thanks to hxclan
I have found an interesting link about c-state which give me some answers

Everything You Need to Know About the CPU C-States Power Saving Modes | Hardware Secrets

which could explain why we get better performance results by not letting the cpu goes into a deeper idle state

you got better performance but you probably got more heat and power requirement as well as that's what c-state stand for

 

not on my side:

 

eYe-I-aïe...

Interesting, I have no idea what the difference is except your in a RAID0 array, not sure why that would make it more stable though. Mayber it's the CPU or MOBO / Mfg difference. My speeds were more stable with the original tweaks but have not had time to try and figure out which settings may affect this yet.

On a side note I went to the link in your sig and the intel ssd power usage caught my eye, not sure where you got those numbers from but below is from the intel product specs for the X18/25 - M series (G2). Much more power efficient than the Samsungs.



PS: the case rating on the drive is just a 100% max allowable number to get the UL/CSA (and other) agency approvals, it has nothing to due with the actual power draw of the drives.

 

Should "typical" users perform the tweaks right now, or should we wait for more specific instructions?

 

Your call, but I think if you wait a few days we should hopefully have a more simple (less settings) and cleaner way to do this. The current instructions add like 25 new power plan setting adjustables under the 'processor power' heading. We will probably narrow that down to between 3 - 8 that actually need to be added to the plan. The ones that will 'go away' are really items that no one would need to adjust or tweak 99.9999% of the time, that's why they were 'hidden' in the first place.

 

Best way to do the settings would be to export the entire tree as a .reg file once you figure out which ones are minimally necessary, that way people don't have to try to get into the registry manually and try to match the UUIDs. Easy thing to mess up

 

I think it would be a good idea to link to possible tweaks in the start post.

Please let me know what posts (number or permalink) to link to.

When the tweaks get better in a few days I can update the links.

 

Welcome back We've made progress....

OK I just updated the 'Titles' with a "Preliminary" in front of them. Step 1 is permalink #301 and step 2 is #327. Feel free to post them on the OP and also add any additional warnings you think may be appropriate. So far no significant problems other than the hassle of loading the regedit changes to get the options to appear in power plan settings. You may also want to put a note to browse the posts after #327 to see some feedback and several CDM shots showing the improvements over several platforms and SSD brands.

We may be able to reduce the actual settings to just 3 - 8. Currently trying to fine tune things and get the lowest heat / power levels, which are quite good so far. There has been additional benefits for some who have verified better CPU performance and I had a friend who was able to eliminate his annoying CPU whine with the same adjustments. You would think intel would be all over this type of sloution....

 

I think my RAID array might be the cause of the more stable results I'm seeing. However, everyone should note that:

1. CDM is showing somewhat an average of it's tests, and this average is much more closer to the top, to the peek than it is to the average, and I'll prove that later on;
2. CDM makes several reads and writes per each test, for instance, if you choose 3 passes, it will read and write kind of maybe 3 times each pass, so that will be 9 passes total, or 9 rounds of reads and writes total for 3 passes, and I'll proove that later on as well.

As for the SSD power consumption, I simply stated what the RATING says on both the Sammys and the Intels, RATING which as you say appear on the stickers of the drives.

On my Intels, it says rated 5V and 1A, which is if my maths are still correct 5Watts RATING;
On my Samsungs, it says 5V but 0.35A, which if my maths are still correct is 1.75Watts RATING.

The real life power consumption might not reflect those rating numbers, however, I simply compared the numbers rating written on each drive, numbers put there by each manufacturer, right?, so hard to see where I'm wrong...

Besides, it's a known-to-all secret that Samsung's SSDs ALWAYS showed the lowest power consumption on the consumer SSD market. I don't have the link handy, but I remember before I bought my M17X with 2 Sammys SSD inside that I've read over AnandTech that the only advantage of Samsung over their competitors was their lower power consumption...


As for your tweaks, again, on my side, I found that:

1. processor idle promote threshold (99% plugged)
2. processor performance core parking overutilization threshold (99% plugged)

are the only setting that need to be changed in order to get MY SSDs to perform (in benchmark) the best. Again, I am yet to notice a real world difference.


On a side note as well, I can tell you by experience (I've had both Sam and Tel in RAID0 in my M17X) that the Intels use more the CPU than the Samsungs. And I am taking the guess that your tweaks would help an Intel drive better than a Samsung one, all other things being equal furthermore, for this sole reason. And actually, thinking about it, your tweaks just tend to prove that as all we do is keeping the CPU alive, so it's more responsive when the SSD needs a hand from this buddy of his...

And finally, I am happy to report that my temps are back to almost normal, almost where they were before the tweaks, ± 50 ~ 54 Celcius now, where they were before the tweak at ± 50 Celcius, and right after the tweaks at ± 56 ~ 65 Celcius. All those temps are in Idle Time.

Again, thank you for your hard work and dedication.

eYe

 

A dutch user of desktop with Asus p7p55d-e motherboard posted these scores:

ms driver en Wprime


without wprime


Familiar picture I thought.

 

OK, so here's the link to watch CDM in action:

CDM

This video has been recorded with the only two JJB's tweaks discussed above.


Note the following:

1. My temps went up from 56 Celcius (start) to 66 Celcius (end);
2. CPU usage is on average ± 20%;
3. 4Ks are similar in both the two 50MB/3 passes;
4. They are a little slower with the 500MB/5passes (which would be more accurate);
5. Note for every test CDM updates a result, it's almost always higher than what the Drives Meter has just shown during the test;
6. Drives Meter shows in realtime the throughput of the data being read/written to the drive, and this is where you can see that CDM kind of BOOST a bit the results, as if they were on steroïds or something, looks like it shows more the highest count than the average.

I might post a vid of As SSD if you guys feel it'd be nice to compare.

I might have forgot something; please let me know if you need more details or so.

eYe

 

cdm always take the best value of different passes
if your 3 passes are 50 / 5 / 120

cdm will show 120 at the end when the average is 58.33333


look at what I 've found from another website
it's not from me

 

Wow, me wants!
seriously though, what the hell do you do with those rates ?
start windows in 0.3 seconds ?

 

i would say it sarts windows before you push the button

 

These can't be real life numbers, unless there's 10 SSDs in RAID0...

or so...

Phil posted something similar not too long ago...

Photoshop is still used in the market, right ?

For those who'd like to take a look at As SSD...

Here it is !

Funny how when CDM shows the better, AS SSD do NOT necessarily show the lower...

Enjoy, it's free !!!

Sorry Buddy, can't with you here agree.

For instance, the highest count Drives Meter shows for the fisrt 4K only test reads is 23.69, while CDM shows a result of 23.95, WHICH IS HIGHER than the highest Drives Meter showed (23.69).

As well, for the second 3 pass/50MB 4K reads, CDM shows a score of 24.22, while the highest number Drives Meter showed is 28.57 so, again, unless I'm missing something,
CDM does not show the highest, neither the average, but a number located more toward the highest number than the average. Unless I'm out of my shoes, which happens sometimes...

I did not do it, but my guess is that we would see the same pattern with any other test. Unless Drives Meter is missing something on its own...

 

Thanks for the video posts . A couple of things; first I noticed that for some reason your CPU seems to have a much higher load than mine when running CDM and it seems quite a bit lower running AS SSD. My CPU rareley get above 15% when running either benchmark. Yeah I know mine is running at 2.8Ghz but that's only on 2 threads for these tests not all 4, while your hitting like 30% at times and have 8 threads. Not sure if it means anything but wanted to point that out. Secondly I guess I thought it was common knowlegde that CDM only posts the highest of the runs for each test, not the average. That's where the big discrepancies I mentioned earlier shows up with just the 2 (99%) settings, and on 3 passes it would show huge swings in the final scores, especially for the 4K numbers.

Finally, with just the 2 setting adjustments, my SEQ and 4K numbers take a major hit, ~20% lower reads, and 10% lower writes, than with 2 or three of the other setting adjusted, I also verifed the same reduced performance with 2 tweaks with AS SSD. I notice that your 4K read are quite low which will be your 'real world' performance bottleneck. If you have time maybe you could reinstate the other tweaks and double check your 4k read speeds with all other things the same as your other test runs. I believe I saw some posts earlier here that showed much higher 4K read speeds with the same RAID0 array you have....

 

i read from this link something interesting
http://www.hardwaresecrets.com/arti...out-the-CPU-C-States-Power-Saving-Modes/611/6

regarding core 2 duo T9400, T9600 taken as an example
going from C0 state to C4E state save up to 77.66% in current consumption
while going from C0 to C6 state save up to 87.87% when reducing a lot the performance due to higher latencies

T9400, T9600
C0 to C4E : 77.66%
C0 to C6 : 87.87%

X9100:
C0 to C4E : 80.17%
C0 to C6 : 81.35%

I read as well that going to the deeper power down state could impact differently cpu's depending on the kind of cache they have

which could mean we could get different results depending of the cpu we have

I was looking at the PM55 and PM45 datasheet to see if they were any difference which could lead to these different results we get

 

CDM does not show the highest of what it is or what drive meter reads but rather what cdm reads as being the highest

 

My pleasure, thanks for your tweaks !

CPU: I've got 2 Intels SSDs in RAID0, and this could explain the CPU usage discrapency, as far as I understand something about that. See my previous post with regards to CDM and the highest number.

Tweaks: Again, I tried every single of them separately, and combined, never got less CPU usage with any other combination.

4K READS speed: This has always been a bottleneck since I've put my two Intels in RAID0 in my lappy. This was one of the reason I was very unsatisfied with those two drives, as opposed to my previous Sammys (also in RAID0 in the same lappy). It might be because I use 16K stripe size, but 16K is what Intel recommand on one hand and, on the other hand, I also tried 8, 32 and 64K stripe size with no improvement. Thats is why I was most interested in your tweaking to get my 4Ks at least on par with what they should look like considering I AM RUNNING TWO INTEL G2 IN RAID0, and for the price they sell, it's more than just A BIT dissapointing, if you see what I mean... (For info, without your tweaks, I get 14 ~15 4K reads for two Intel SSDs in RAID0; scary to say the least ...)

As the title of this thread states, Intel PM55 and HM55 chipsets prevent SSDs to work at their best, and this could explain that. Gotta see if the other benchies you're talking about (that have better 4K reads than mine) are Intel raided SSDs with Intel PM or HM55 chipsets, just to make sure we're comparing apples w apples hmmm, not a good example, we're comparing bananas with bananas...

Got to point out as well that Intel SSDs use more CPU than Samsung to take an example, and this is surely part of why they do deliver better results. I did not write it IS because, I wrote it's PART OF why... RAID might also be a factor.

Cheers !

eYe

But it seems to read wrong though...

May need glasses...

or a single glass...

of water...

a crystal glass ...

glass of water...

water...mark...

a crystal glass of watermark...

never mind...

 

if paul says A and jack says B how to know who is telling the truth
same thing for cdm and drives meter

 

yup, eye agree; in this case, I would rely on what Pete would say...

...mainly because I know both paul and jack are liars...

 

haha ur right

I think it's me sorry

 

Guys (and girls) please avoid making two consecutive posts. Instead edit your last post. Thank you.

 

..........

 

@erig007

Regarding your above post about the 'hardware secrets' article, it was a good find except that it is from 2008. Since then intel has implemented several new power saving features, hyperthreading and speedsteping. So the numbers and info, while useful in general, IMO have little bearing or our newer iCore CPU's.

EDIT: Does anyone know of a simple app to test just the maximum Turbo Boost speed on only 1 Core for the new CPU's ??

Super pi for me is inconsistent, it kicks in a second thread at times, and the intel peocessor ID utility always says 2.8Ghz which is my max speed with all threads. I am trying to find quick and accurate way to test setting changes with regards to single core max turbo speed.....

 

yeah I knew that i7 can shut down each cpu separatly and other thing like this but as you say it was to give general informations that's why I was looking at PM45 and pm55 datasheet and whatever file to get more accurate informations

 

I sincerely appreciate the way you quoted me, but this way, I cannot quote you easily...
No a real problem though...


Got to point out as well that Intel SSDs use more CPU than Samsung to take an example, and this is surely part of why they do deliver better results. I did not write it IS because, I wrote it's PART OF why... RAID might also be a factor.

Huh, never heard anything about that before, not that it matters to me since I have 5 machines with intel drives (won't be changing brands), just doesn't seem to make sense in that I would think any drive would use the same CPU cycles to do the same task...


Well, this also has to do with the SSD's cache, and the controller of the drive. For instance, Intels tend to get better for 4K writes as queue depth increase; this I assume consume more CPU cycles (to actualy compute) and also, the more channels the controller handles, the more CPU cycles it takes to again compute... And all this at the end of the day (or at the end of the process) translates into faster transfer rates... Thus the Intel faster than the Samsungs, but more power hungry. On top of this, if you add a RAID setup...

 

With processor speeds as high as they are nowadays and multi-core CPU architectures, I'd much rather have a drive that uses more CPU but runs faster. Even with dual cores and especially with quad cores, most of us only use a fraction of the CPUs capacity anyway, so why not utilize some of that to help the performance of the largest bottleneck in our machines, the hard drives. Intel FTW =)

 

I agree. The thing is that the solution to increase drive speeds appears to be solvable without much impact to CPU performance, temps or power draw. Since I am currently close to this solution with fewer adjusments I am just trying to get the best performance with least negative impact at this point. All my tweaking may boil down to being specific to just specific platform / CPU combinations though. For example I am getting poor battery life with just the "2 x 99%" solution, but with some other settings added It appears that battery life can be greatly improved with near full SSD performance....

 

This is done upon booting. You invoke it by pressing F5 or F-something or CTRL+something after the BIOS screen.
But since you disabled your raid pile, no wonder you don't see it in RST...
Besides, if you RAID them, Intel Toolbox won't work.

This will change once manufacturers start to ship ATA-ATAPI- 8 compliants SSDs where, finally, we'll be able to TRIM in RAID.

Which might just be the next Intels; after all, Intel do RAID controllers too...

can't wait...

 

Intel does the ICH10m and the CPU's and IEST etc. too................

 

Maybe by disabling other cores under power management and testing one thread with wprime (I think Hot cpu tester doesn't support iX's)

 

Yeah, tried that and definately don't recommend it, especially changing the last option (forget the name) anything below 100% causes very weird and scary results. I also tried sys config to boot with 1 core (thread) and then 2, with even scarier results. Do not try this unless you want to wait 20+ minutes to boot up.....

 

wow look scary

as much as cutting the cpu in half with a knife to test only one core

wonder if what cause these weird results is Hyperthreading

anyway I have no proof of that but my understanding is that due to hyperthreading and probably other things
testing one core separatly may not give the real performance of the cpu as
core (1 and 2) = core 1 and core 2 and (core 1 and core 2) where (core 1 and core 2) is the other thing like hyperthreading
same thing for consumption matter as some parts like controllers on the cpu may still be working even if one core is shut down

 

What are the tweaks you did to achieve x25 multiplier? Did you also apply stamatisx'?