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

would this work with other SSD's or only sandforce ones?

 

+1 to that Phil, why pay for an X25-E if it is to perform the same as an X25-V which costs 7x less (in everyday usage nobody would feel the difference)?

 

My main deal was I was starting to go back and forth on the Force 240 drive I just bought but you all put my mind at ease a bit. I get what your saying about wanting what you pay for though. Its a lot like the underclocking deal with the gtx 260 and M15x.

 

I have an Intel G2 in MSI GX-740 laptop with HM55 chipset. Here is the performance of my SSD when the system is in idle:



Here is the performance of the same SSD in the same laptop, but while system is fully loaded (100% on all threads and GPU, using BOINC):



I assume this is somehow related to the differences in benchmarks between normal system mode and safe mode. Does anybody know why is this happening and how to fix it? Does Intel know about this?

 

That's interesting, I used that SSD Tweaker last week myself.

I'm still puzzled by the 30MB/sec 4K random read in Crystal. I've never seen that on any SSD on SATA II.


Any idea to why you're getting higher 4K random reads than normal performing Sandforce SSDs?

Edit: OK I see. Well for simplicity, let's keep CDM at default settings.

 

Thats because CDM was set up to use Compressible data. You can change that in CDM if you click: file>Test Data>All 0x00, 0fill.
Okay mine is a wee bit lower but still about 30 MB/s 4k read as well if I run Intel Burntest in the background. Weird stuff huh.
I have the 120 gig OZC Agilty 2.

iemie



Uploaded with ImageShack.us

 

Well for simplicity, let's keep CDM at default settings.

 

Phil,

I have explained elsewhere why I'm testing/benchmarking my SandForce drive with CDM and non-default settings.

But to quickly recap: I don't want to kill the SSD with needless benchmarking - I simply am using benchmarking to see a difference (if indeed there is one) between two 'states' like cpu loaded vs. cpu idle.

I'm not worried about 'killing' the SSD completely with running benchmarks - I know that the nand chips can take a lot of runs before they will begin to fail.

I'm much more worried about the 'Lifetime Throttling' that the DuraClass technology does and brings the write speeds to a crawl. When running AS SSD a few (very few) times, I noticed it more than once that whle running it's 4K iterations, it would briefly post 0.7MB/s - yes, half of a mechanical Raptor's 'score'.

This is the reason I'm using the fully compressable data set CDM provides and ever since I've stopped using the U30Jc to convert and edit RAW image files (which generated 4 to 20+ GB's of uncompressable writes in as little as 1/2 an hour or less) I haven't noticed the SSD to 'pause and stutter' either. Please note that 'pause and stutter' for me is not like what the original SSD's did with the bad controller (what was that controller's name?) - to me, 'pause and stutter' is in relationship to how the SSD makes the O/S feel compared to the 'feel' of my Raptors.

By not using the computer as I intended (because it so negatively impacts the SSD's performance), the computer does seem very snappy and lively - but my hope is that Patriot will release a firmware (like Mushkin has for its SandForce based SSD's) that is less aggressive with the 'Lifetime Throttling' it enables by default and allows us to use the SSD in a real world productive environment.

I will continue to use the fully compressable data set that CDM offers - while my 'scores' won't be directly comparable to others that use the CDM defaults, they are still useful because they show a difference (or not) between two states on the same computer/setup.

 

Ok.

But for the screenshots posted in this thread, for simplicity, let's all keep default settings for CDM and AS-SSD.

If someone is worried about performance degradation I suggest to do fewer runs and only do 4K read and write.

Also let's standardize on Intel RST drivers so we're comparing apples with apples.

 

I'm not sure what chipset resides in my 2010 MacBook Pro and I'm having a hard time finding a hard drive benchmarking tool to test it with.

help?

 

It's a modded PM55 chipset.

Would be very interesting to see your CrystalDiskMark result in bootcamp.

 

@Phil

I just came accross a very odd issue. I re-ran a default CDM run with a 100% Everest burn in test running in backround and here are my 'new' results with write cache off vs on;

cache off
cache on

As you can see my 4K numbers crashed to single digits I have rebooted and ran it several more times with similar results, if you look at my CDM runs on page 9 the 'cache off' numbers were significantly higher...

Went back to cache on and results are back to normal, try cache off again and same low numbers.... AFAIK all setting are the same as before and no idea what may be causing this? Any ideas?

PS: I have disabled and re-enabled superfetch but the results seem to be the same with it on or off.... I normally have it enabled because it reduces my boot times by ~6 seconds

 

I'm not sure I'm understanding you. It's normal to get better performance with write caching on.

 

Look back at my CDM results on page 9, all my 'write cache off' runs were higher than with it 'on'. Also none of my results were anywhere near the single digit level.... I just can't figure out what could have changed to have that big of degredation in 4K performance. Also I just ran it again on my other drive (have dual 160G G2 drives) and similar 4K results with cache off...

Edit: Just ran CDM again on both drives and get the same ~ 4Mb/s 4K speeds with write cache off, and normal results with write cache on (while running 100% CPU load) so at this point since I keep the write cache on I am not to worried with the other low numbers, I just can't figure out why my earlier results were so much higher with cahce off (unless I had the cache on by mistake)...

 

Any SSD should work.

 

Phil, sounds good.

I did run the standard tests and concentrated on how long the CDM runs took (cpu idle vs. running IntelBurn Test too).

What I found was very interesting, I think.

Except for the Sequential scores, every other test took 2 minutes to run (I was using 9 runs @ 4000 MB in the options to be able to time it more accurately). So, what is this telling us? CDM is only running a test for a limited time instead of running it for the size (in MB's) we're asking it to?

I think that I should have seen a time difference between a 22MB/s 4K run compared to a 55MB/s 4K run (the 55MB/s run should have been less than half the time, right?).

Instead, only the Sequential scores differ by about 30 seconds (the IntelBurn Test running shortened the time).

When running the compressed data sets - the reverse occurred: with IntelBurn Test running, it took 6 seconds longer for the Sequencial tests to complete.

Can't really say I trust CDM (not that I trust benchmarks in general anyway) to report accurate SSD 'scores' anymore.

To anyone who can make a benchmark (or, if CDM and/or AS SSD creators read this post) this is what I would consider a proper benchmark:

To test each parameter to completion: if we specify 9 runs at 4GB size, then write 36GB of compressed or uncompressed (as chosen by the user) data to the drive.

Give us the maximum performance of the run: as CDM now does in the end.

Give us the minimum performance of the run: only shown 'live' but not recorded anywhere that I can see.

Give us the time needed (mm:ss) to complete the run: this will allow us to verify if the average performance is indeed the median value of the run, or, if the drive is performing better/worse than the median value indicates.

Give us the total time to complete all the tests: so we don't make stupid mistakes adding 1+1.


If anybody has any input as to why the test times are consistent for the 512K, 4K, 4K Q32 in Random as well as 0 filled modes (compressable), I am all ears!

 

@tilleroftheearth

Those are all good points and I don't really disagree with any of them. But IMHO I think your missing the point, Phil wants to compare results from a given baseline test so we are comparing 'apples and apples' and using the default CDM runs makes perfect sense, if we all use it then the results will show the true overall trend. I have noticed that what is running in the backround (super pi, burn test etc) makes as much or more of a difference in the results than the variance between runs at default CDM settings, but again if we all use the same thing for loading the CPU then we have an equal baseline test for comparison purposes.

 

My vertex 2E results are fine with my P45M. Get up to 277MB/sec.

 

Meaker,

Could you post a CDM 3 'defaults' result that:

1) is using the IRST 9.6 driver
2) is compared to idle cpu usage vs. running IntelBurn Test 2.50


What we're comparing here is not the sequential reads/writes - but rather the 4K Random R/W's and the 4K Q32 R/W's.


Thanks!

 

I'm using Intel's latest rapid storage drivers.

The above is idle with the drive as main OS drive and 9.2GB free (60GB edition).

If you want me to use another benchmark tool while loaded vs idle then which one?

Crystal bench does not agree with this SSD so I will not use that.

 

From everything I've read ATTO is the worst benchmark tool for SSD's - but I'm only repeating what I've read elsewhere...

CDM does not agree with your SSD? In what way? Do you mean the fact that is is writing truly random data that a SandForce based drive can't compress and thereby gives bad 'scores'?

Well, if you want, try AS SSD benchmark - before IBT and during - this too will show a difference in the 4K R R/W scores; if your chipset suffers too.

Thanks!

 

If memory serves, Phil said Sandforce based drive don't do good with CDM. Phil?

And for AS SSD, I get very low scores with my Vertex 2.


on a side note, anyone with Intel 160GB on PM55 can test their Acc.Time using AS SSD for me? (just want to compare against the XPS 1340 +160GB)

 

I get very similar scores on both AS SSD and CDM.

 

Here are my AS SSD runs on intel X18M 160GB G2, surprising difference in access times with CPU load and no load (as well as speeds).

with CPU idle

with Everest stability test running (100% load all threads

 

Can you post the results? I would be interested to see those since I get low scores from AS SSD vs CMD.

 

Thanks for taking the time to test.

Its really sad to think M15x (PM55) has this issue

Here's the AS SSD from Intel 160GB + XPS 1340 (NVidia chipset)
Fresh install (cpu - idle)

 

very good. will do the CDM in bootcamp tomorrow.

 

I'm looking forward to it. If Macbook Pros are affected too I expect there will be a lot more attention for this issue.

 

I'm not really up to date with this thread, but has anyone find a permanent tweak to lift the cap? I am hesitant to buy another SSD.

 

Not that I am aware of.

To recap what I tried with no luck in a permanent fix:

Tried MS AHCI and Intel AHCI driver (MS driver was a bit faster in all the tests by the way) So if it is a driver issue it manifests itself in both drivers.

Switched my power options mode to the maximum performance setting, Along with disabling any power down stuff, and putting min/max processor states to 100%.

Tried realtemp and throttlestop to try and force C1E to a disabled state (still not clear if they actually work at disabling C1E).

My Sager BIOS has minimal settings to change, one that earlier in the post that helped someone was disabling Intel Virtual support, I tried that with no luck as well.

As far as buying a SSD, this problem does NOT rise to the level of a show stopper, you still have a significant upgrade in speed with these drives. And I will state again that in real world use you would not even realize there is a problem, you will see this in benchmarks, but I cannot get anybody here to provide evidence of actually seeing real world normal everyday use of your laptop slowdowns that would be of any significant impact.

Also I want to point out I want to see a fix for this, as I want to get what I paid for as much as anyone else around here.

 

I don't have the time to read it now but is this article relevant?

Does Power-Saving Technology Kill SSD Performance? : Flash SSDs Can Suffer From Power Saving Features

 

Phil's original post in this thread contains my scores. I have since gotten better ones without changing any settings, but they are still on the same level. I haven't attempted running the tests while under load except once - I ran CDM while SuperPi was running in the background. No improvement

 

I just ran a quick scan of the article Phil linked too, and it concludes all these power saving options are to blame for the performance drop.

The bad news is this seems to be BIOS related issues, so the level of support to get fixes will vary by manufactures.

With Sager I am not holding out much hope for this. (Sure glad, as I said earlier I won’t lose sleep over this issue, cause I would never get any if I was up waiting for a fix from them.)

 

Has anyone though of repeating all of these so-called tests with a wide variety of conventional hard drives and other OS and filesystem combos?

Among the untested variables here is the native performance of the file system, NTFS, at the driver and OS level.

Also, what about testing with other operating systems and file systems?

You've still got Windows (desktop) in the mix here, Linux and Windows Server need to be 'tested' in the same way as the Windows client has been.

It's possible that MSFT has performance-limited their desktop client at the raw I/O level as well as certain r/w profiles of the filesystem.

And no, one person doing a short round of tests on other OS and filesystems is not sufficient. You're going to need as many people who point and click a windows benchmark to do the same work on other OS and filesystems.

 

There may be a way to change the CPU power state settings in the registry editor. I remembered coming across the section below when I was trying to change a battery charge parameter. I would think that one of these controls could change the policy on the C1E power state but it is way beyond my level of knowledge and comfort to play with these settings.

Maybe someone with programming experience can figure out a way to change the settings to get full performance out of our SSD's.... Ideally I think I would want full performance when plugged into AC and not when on battery so that battery life would not be affected when unplugged.

The image below is a 'REGEDIT' screenshot (Win 7 Pro) that shows where the ASPM (active state power management) controls can be found. I think this is where the C1E state may be adjusted. Caution DO NOT play around with these settings if you do not know what your doing....

 

Unfortunately for us, C1E and C6 states are controlled via the bios, which we can't edit or really do much about. Both states (c1e is power savings, c6 is core parking/ht control) contribute to this ssd throttle, but hellcry on the asus forums found the only workaround we can deploy is to have the cpu under load when you want max ssd speed.

So if anything you'd want some programmer with too much spare time to come up with an otherwise fairly useless program to put a high load on a single core... when it comes down to it, you'll notice the loss of a cpu core more than the loss of some ssd speed I think

This is 100% up to intel to fix, asus/sager/aw/apple ect do not have the tools to fix this, pretty much all they can do is bother intel about it.

 

Ah, thought you had it sorted after I suggested you to do the secure erase.

Well, your CDM score is only equivalent to my AS SSD score which is very low and that is the reason why I said I get low scores from AS SSD

 

I'm wondering if disabling the processor's C-states will solve this issue? Would any of you care to try the following? I don't have a SSD yet so can't verify myself.

Cheers,

Sander Sassen - Hardware Analysis

 

No, unfortunately, I still do get low speeds. I am fine with them because they are faster than any HDD I've used, but I don't feel that my family, who bought me the SSD as a gift, got their money's worth. How would I perform a secure erase? Can I back up my data, perform the secure erase, then put my data back without doing a reformat?

 

Actually, if you look at the 'descriptions' in the regedit section I highlighted above, it appears to have several core parking and power state control options that can be adjusted (maybe this is an HP only thing?). Anyway it would take someone who knows the available parameters that can be input to edit these settings to work on a solution.....

 

Is this what you are looking for? (its the default %)

 

No, I have those settings all set to 100% and the issue is still obvious: running IntelBurn Test while running CDM shows a marked improvement in 4K and 4K Q32 runs.

 

EDIT! SOLUTION FOUND!

+REP FOR SSASSEN!

Here is my CDM run after following ssassen's instructions on the previous page. I initially forgot to complete his last line of instructions about disabling the 'idle state' in the power plan (advanced) setting that appears after the regedit change. Now in my 'High Performance' setting I get full SSD performance when there is no load on the CPU. There is a downside in that the idle temps for my i5 have doubled to ~63C but when I switch to 'balanced' all goes back to normal temps (you can choose which setting to change the idle state).

speeds at no load (0%) CPU (new high performance power plan)

For reference here is my 'balanced' power plan CDM run:

balanced PP no load on CPU



Ssassen, could you maybe suggest a way to just disable the idle state on just 1 core (or ideally just 1 thread) ? That may still provide the performance boost while keeping temps reasonable.....



Also could you please confirm the default values for the 3 highlighted lines in the screenshot below? I want to be sure I returned the settings to the correct values (forgot to backup registry ). From the top down I have them at 1, 0, 1



Maybe some combination of the 3 settings would work, I don't have time to try right now, let me know if you have any other ideas to try out.

Thanks.

 

@JJB

Very interesting find.

How's the idling temp after running the CDM?

 

After 10 minutes at idles the CPU temps seem to stabilize at 65C max on both cores. looking at ThrottleStop, the CPU idles at X21 multiplier which is 2.8Ghz on all 4 threads while showing 0% usage in Task manager. Since the Envy 15 has such a great cooling system it is barley warm at these temps but I would bet any i7 quads will be much hotter. Looking at the power draw, at idle, it is now 19W vs. ~6W when in 'normal' balanced mode power setting.

This is not a perfect solution but it gives me the option of getting full SSD performance when I want it at the expense of higher temps or regular (slower 4K) with a simple change to balanced power plan. I have already noticed a big improvement in my Nikon photo software when scrolling through very large (400+ 20MB) image files, the images now scroll rapidly at full res with no hesitation or pauses which always happened before.....

I think that an ideal solution would be if we can figure out how to 'disable' the idle state on just 1 core (or 1 thread) to help lower the temps. Hopefully sassenn can figure this out as I see some 'thread' and 'core' idle state and 'parking' options in regedit, I'm just afraid to mess with them....

Edit: Here are my idle temps after 30 minutes at this setting while doing light browsing, I think most of you will not be as low as this due to the Envy 15's cooling system, it's actually running the fans only 1 notch up from the lowest speeds, very quiet and almost cool to the touch on the chassy:

idle temps only, not actually running stability test, normally 32C and 22C.

 

My benchmark also boosts to normal SSD 4k read/write after the registry tweak!
Thanks!

 

I will try this when I get home from work, I dread the thoughts of the temps with my CPU i7 920XM

 

Okay, ssassen, you're a genius!

Just like JJB, I too tried this except my regedit setting was set to 0 already, so, I set it to 1 and rebooted. I knew that something was up - the fan stayed at full speed - cool!

I tried CDM with no load and with IntelBurn Test running, see for yourselves below:

I also tried some 'real world' tests:

Launch PS CS5: 5 seconds (previously, up to 20 seconds!)

Update MSE and do a quick scan: 98 seconds (120,000 items scanned) (previously 3 or 4 minutes and occasionally up to 7-8 minutes!).

Conveniently, CS5 had an update today: download and install less than 1 minute 26MB download (previously, 4+ minutes for a much smaller download size). Note, I'm not counting the time to download: merely indicating how big the 'update' was.

Along with CS5, LR3 was also upgraded today: not including time to download 185MB installer (this is not really an update - it is a whole new program install actually) extract files 30 seconds, install new program 15 seconds!

Are 4K R R/W's important? Yes! Now my notebook is what I expected it to be with 8GB RAM and a $400+ SSD inside.

Thank you ssassen! +rep from me too!

Image on left: no cpu load. Middle image: IBT very high settings. Image on right: fully compressable data, no cpu load.

So, my take on this SandForce based SSD, is that depending on the actual data the SSD is handling, my performance is actually somewhere between the image on the left and the image on the right. More importantly, if my clients are considering an SSD, then I can confidently recommend them one after almost a year of disappointments from personal experience.

The link Phil posted in the last page or two specifically states that HD's are not affected by these settings: only SSD's show subpar performance with power saving settings enabled.

One Happy Camper!

 

This registry change actually works for me. Upon disabling the idle processor in the power settings, I obtained read scores for the 4K assessment that were nearly 10-12 MB/s faster. My write scores are unaffected, but in my own observation (at least on my specific drive and computer), the Crucial M225 does not benefit from increased load on the CPU for write performance, which in my testing, has never exceeded 11 MB/s on the 4K assessment. Please see the attached images for the before and after results. Note that this was performed with no load on the CPU.

Before (Processor Idle Disable: ON)


After (Processor Idle Disable: OFF)


Of course, a disadvantage to this resolution is that the CPU always runs at full throttle, and typically idles at around 75 to 80°C for my i5-540m. Still, with C-states disabled, that is to be expected.

 

Can someone look into how this affects battery life? Is there a way to have two different "High Performance" for when performance is needed at the expense of heat and when it's not?

EDIT: Will the battery life be worse than on chipsets that aren't PM55 or HM55?