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

Although this slow down is linked to the chipsets being discussed here, the chipsets themselves are not inherently 'defective' - instead, the aggressive implementation of their power saving features is.

These three benchmarks show this easily:

The first is simply running CDM by itself, the second is stressing the CPU with IntelBurn Test 2.50 at very high settings (4GB RAM used - I have 8GB in my U30Jc) and with the MS AHCI v1.0 (default) Win 7 driver. The third run is also with IntelBurn Test running but this time with IRST 9.6.

All of these are using the most compressable data CDM can give the SandForce controller - so this is the very best case scenario for these drives (in other words, not 'real world' at all). I'm also using this mode of CDM so that I invoke the 'Lifetime Throttling' of this SandForce drive as little as possible.

All tests were run with the power mode in High Performance mode and it doesn't matter if I am plugged in or not, the differences are almost exactly the same.

Maybe others will want to try stressing their CPU too to see if their CDM scores improve too - and, the IRST 9.6 drivers are definitely the ones that give the best performance (compared to the v1.0 MS AHCI (default) drivers.

I am sure that all we need is like what Chastity has suggested elsewhere: the ability (in the BIOS) to disable the chipsets power saving features (specifically disabling power state CE1, if I remember correctly) and therefore letting us choose if we want maximum performance or maximum batttery life. This option is available to desktop MB's users so it should be very easy to implement for us too.

I would be very happy to finally see how an SSD is supposed to perform on one of my own systems!

To Phil: I have stated before it is/was not my system (that was holding the Inferno back); I was wrong. Not only is it my system, but everyone else's too that I read was having problems with their SSD's!

Hopefully, with each of us making more and more people aware of it, this will be resolved very soon.

 

Interesting stuff tiller.

Could you try if running something as simple as SuperPI or WPrime also sets the storage performance to the highest setting?

 

Those are the best speeds i could get on my X25-M 160 with a PM55 chipset on my alienware M17x-R2 :



Those results are in safe mode :

 

thanks mfractal. Could you run this: in high performance mode, SuperPI largest iteration and then start Crystal.

 

Phil:

I downloaded SuperPI Mod 1.5 from here:

See:
techPowerUp! :: Download Super PI Mod v1.5


For the 'before' score, see my post above. Running SuperPI for 32M the CDM score is almost as high as the IntelBurn Test running at very high settings.

If my notebook was a true quad core - I would have SuperPI running constantly (it only uses 1 core) for the performance increase I see from the Inferno with it!

mfractal and anyone else, could you post your scores (not in safe mode) with either IntelBurn Test or SuperPI running in the background?

Oh, and indicate if you're using the MS AHCI SATA drivers or the IRST 9.6 version (or, something else).

Thanks!

 

The following results are performed under a normal start up into Windows.

SSD:
Crucial M225 128GB (CT128M225)

IDE ATA/ATAPI Controller:
Intel(R) 5 Series 6 Port SATA AHCI Controller (IRST 9.6.0.1014)

Operating System:
Windows 7 Professional x64

Windows Power Plan:
Balanced

Test 1: Default (No Load)


Test 2: SuperPI Mod 1.5 (32M)


Test 3: IntelBurnTest v2.50 (Stress Level: Maximum)

 

Here's my Corsair Force 120gb in Safe Mode:

 

Jakeworld, nicely done those benchmarks.

KolosoK, how does your SSD do when Super Pi mod is running in the background?

 

Here's the result (not in safe mode):

 

The following results represent the previous tests performed under Safe Mode.

SSD:
Crucial M225 128GB (CT128M225)

IDE ATA/ATAPI Controller:
Intel(R) 5 Series 6 Port SATA AHCI Controller (IRST 9.6.0.1014)

Operating System:
Windows 7 Professional x64

Windows Power Plan:
Safe Mode

Safe Mode Test 1: Default (No Load)


Safe Mode Test 2: SuperPI Mod 1.5 (32M)


Safe Mode Test 3: IntelBurnTest v2.50 (Stress Level: Maximum)

 

Here are my results with SuperPi in the background running and msahci drivers :
http://forum.notebookreview.com/attachment.php?attachmentid=55050&stc=1&d=1282992846

Second screenshot is with SuperPi in the background running and Intel RST 9.6.0.1014 (23/3/2010) :
AS SSD score of 539 !!!

(All at XPS 1530 with ICH8M, 4GB ram and Windows 7 x64 ultimate)

 

I don't think the difference is as significant as it seems, my CrystalDiskMark random 4K write scores are usually between 44MB/s and 33MB/s. Even with back-to-back runs, just switching from "High Performance" mode to "Power Saver" mode can make the random 4K write scores drop by as much as 7MB/s.

Perhaps he just needed to let GC run.

 

I have an update, I am still running benchmarks with a couple of settings.
I will report back as soon as possible.

 

Here's something I hope you all might find useful. The MSI GX740 uses the HM55 chipset. Its BIOS is quite limited, but one of the few settings you can change is whether the CPU will go into its power-saving C-state. Testing was done on a 240 GB Patriot Inferno:

With C-states enabled:



With C-states disabled:



I thought having C-states disabled would have a pretty negative effect on my temperatures, but that's not really the case. Idle temps with it off only seem to be about 2-5 degrees higher. That 5 degrees is the absolute maximum. It's closer to 2-3 most of the time.

 

Thanks a lot for this very valuable input, that really supports our theory about the power saving features and the BIOS
+1 from me ASAP

 

@stamatisx what's that update you were talking about ? found anything ??

 

Yeap, I am going to create a new thread at the same time, that's why it's taking me so long

I will post most of the benchmark results here though

Sorry for the delay

 

so i did 3 runs, all three at high performance mode.
First idle, second with super pi and third with prime95 in the background.
The reason i added prime95 is because i didn't see much performance increase with superPI running..

Prime95 seems to get the drive to Safe Mode numbers.

 

Running Intel burn does increase the read/write by a lot. I have PM55

 

Ok I can confirm that tampering with the BIOS settings on my laptop changed my SSD's performance significantly. Especially the 4K random reads and writes!
To see how (what drivers, what settings, etc...), check this thread:
http://forum.notebookreview.com/ali...ttings-partially-restore-ssd-performance.html

Now it's time for the benchmarking results:

Before BIOS changes:
Normal Mode / CPU 1% Load


After BIOS changes:
Normal Mode / CPU 1% load / Write Cache On


Normal Mode / CPU 13% load / Write Cache On


Safe Mode / CPU 1% load / Write Cache On


Safe Mode / CPU 13% load / Write Cache On


*EDIT*
It wasn't the BIOS, it was a registry tweak had I applied and didn't mention
Sorry for the confusion

 

Okay, after applying the SSD tweaks from Ashley and rebooting, I get these results from CDM with and without IntelBurn Test 2.50 running in normal mode (not safe mode) and with the High Performance plan selected as the power option.

I don't know what the tweaks did (the interface is more than confusing), but since I'm going to be rebuilding this computer soon enough I had to see what would happen.

I'm impressed! The CMD results are almost the best I've had (especially with Reads and double especially with IntelBurn Test running).

Where is my new ASUS U30Jc BIOS to disable CE1 states!!!

Image on left with SSD Tweaks applied, Image on right SSD Tweaks applied and IntelBurn Test running at very high settings.

stamatisx, you're getting closer to 'perfect' - congrats! You must have a real 4 core CPU, right? that 13% utilization with a single threaded cpu stressing program mirrors my results of 25% utilization with my i3 (two 'real' core) cpu.

newsposter, don't over analyse the data - the aggressive power saving modes of these chipsets is what is causing all the havoc. Doesn't matter whether in safe mode or normal mode - making the cpu work while benchmarking the SSD increases the scores dramatically.

Where are the programmers that can make a utility that allows us to control the chipsets power states in Windows?

 

Image on the left looks like normal Sandforce performance. The image on the right looks like extremely good performance. So what did you do to get this?

PS. update, the Apple notebooks do not seem to be affected by this.

 

What things did you change exactly? Sorry, I don't have access to two computers or a printer right now. (and no pen/paper ) Was it the virtualization technology?

 

I was experimenting with a new VBIOS from Dell and their beta display drivers that allow the laptop to sleep/hibernate and wake up properly. The only way for me not to have issues with those power saving features, was to change the default values under the BIOS, it's not a specific one (it's a bug of our BIOS that when people change its default values it causes problems so I played around with them and found those that didn't). Namely, I enabled the ethernet and disabled the USB emulation, firewire, USB wake up support and USB powershare. The only thing that troubles me is that people with the same laptop tried to reproduce it in order to get my increased performance but they can't so far

*EDIT*
The same registry tweak here, that's what made the difference...

 

It was one of the possible solutions as discussed in the SSD problems in G73 but its not the best solution..

where are these tweaks? Basically we need new chipset software...

 

Does this only affect specific laptop chipsets. What about modified desktop chipsets such as the P35/G33/G31 (as stated in CPUz) found in my C90P? Southbridge is ICH9. Would SSD performance be likely limited by this chipsets now 2+ year age?

 

But it's a solution, at least. With C-state off, the results I achieved seem to match right up with the "normal" Sandforce benchmarks in the opening post to this thread. I notice minimal change in temperature and no other apparent demerits regarding system performance.

Yes, new chipset software would probably do the trick, and from what I've read here, Asus and Alienware have copped an ear. As such, they could very well have broached this subject with Intel.

Until then, this is a perfectly satisfactory workaround, IMO.

 

Is there a way to disable C-state on the laptop without that BIOS option?

 

I have been having problems with my Samsung SSD only when it is in my laptop with the PM55 chipset.

here is a crystal disk mark run


Uploaded with ImageShack.us

 

0.7 is way too low. You can get a much higher writing speed even on PM55.

 

Phil,

sorry, Ashley Maple's SSD Tweaker can be downloaded here:

See:
Guide SSD Tweak Utility


That is what I applied above (auto button) rebooted and ran CDM with out IntelBurn Test, then on the right; with IntelBurn Test.

BTW, those are not 'typical' Sandforce scores on the left: this is the most compressable data CDM can give the drive and are the Maximum this drive will do. With normal data - it will be significantly less (the more the SandForce controller can compress the data - the closer to these artificial scores it will achieve).

 

I Just read through this thread and did some tests of my intel X18M G2 160 Gb drives. Results below are from a post i made on the Hp Envy 15 owners lounge since we have been having slow 4K random R/W issues and HP said it was the controller and there was no fix available. Well thanks to this thread it proves HP tech support was wrong and resonable 4K speeds are possible, unfortunately we have no BIOS options to change the power states.

Below is a copy of my other post with CDM results in normal and safe mode and I added a run at the bottom with super PI running in the backround. The Envy 15 uses the PM 55 chipset, basic spec in my sig...

"For those of you with SSD's, I just found THIS thread on slow SSD performance with the PM55 and HM55 intel chipsets. In it I found some interesting posts of SSD 4K speed comparison while booted in safe mode vs. normal bootup.

I did some quick tests and was floored at the improvement on my 4K random R/W speeds (and overall speed improvements) while in safe mode. This to me seems to point to some driver, process or services causing the poor 4K R/W numbers with the computer booted normally and not the controller as previously thought.

Anyone have any ideas on how to determine what is loaded (or running) when booted normally that is causing the reduced R/W numbers?? It appears while in safe mode the 4K numbers are almost where they should be so what is causing the bottlerneck ??

Here are 4 crystal disc runs, first 2 booted normally with and without write cache on, the second 2 are when booted in Safe Mode, with and without write cache on. Look at the vast improvement overall in safe mode..... (I have the intel X18M G2 160Gb drives with the older 2CV102HA firmware)"

View attachment 55118 normal boot write cache on

View attachment 55119 normal boot write cache off

View attachment 55120 safe mode boot write cache on

View attachment 55121 safe mode boot write cache off

normal boot (cache off) with Super PI running

Edit: Just ran it again with Everest 100% burn in test running in backround, this maxes all 4 threads to 100% at 2.8 Ghz each. Results are even faster than safe mode. CDM with normal bootup and write cache off below:

 

What I would like to know is how this slowdown affects realworld use. What will actually cause this slowdown to occur? I have an Alienware M15x and I purchased a Corsair Force 240 but have not installed it yet. Can you physically see a difference using these drives or is this slowdown something you will only notice while running a benchmark. Would these chipset owners be better off with a Vertex drive instead of Vertex 2, for example?

 

Looking back through this thread again it would appear that the Crucial c300 plays better with pm55 than the Sandforce drives at the moment. Is this a correct assumption? If it is I will most likely return my Force 240 for the Crucial drive and save some coin. How about the drives with sf-1200 like the Agility 2 that are a bit cheaper? Would these drives also be a better option at this point since performance will be limited anyway?

 

I didn't see any results that would confirm that.

 

Intel X-25 M 160G in N61JQ-X1

For some silly reason my NIC driver updated while I was running this, probably didn't effect anything. In my real world usage estimation, this thing flies. My own personal conclusion is that I'
m not overly concerned, but it is an interesting discovery. Hopefully something they can rectify in a driver, but it doesn't sound like it.

 

I just saw your post after posting mine..

Maybe something having to do with a driver being loaded that supports TRIM is causing the slowdown?

I just got done installing Intel's latest RST drivers in order to get ACHI driver in there so I could stop running their SSD Toolbox every week. I'm a little hesitant to start ripping out drivers as I've seen others get to a state where they can't boot. It was a hold-your-breath enough for me when I installed the new driver.

Interesting finding there in safe-mode.

 

Here is my Intel 160G G2 SSD under no load

[/URL][/IMG]

Here is the same drive with the CPU under heavy load from IntelBurnTest. Note I only performed the 4K test on this run. It shows a nice improvement.

[/URL] Uploaded with ImageShack.us[/IMG]

The final test is my standard Hitachi 500G 7200RPM drive. I ran this test under no CPU load and high CPU load and had no difference. This gives you an idea of how much faster a SSD is to a standard HDD.

[/URL] Uploaded with ImageShack.us[/IMG]


This thread shows that there is a problem with 4K read/writes, but as artompkins asks is this a real world problem. I would say unless you have some kind of program that saves a VERY LARGE number of small file sizes you would not notice this. I would also say the only way we even know this is a problem is with benchmarking tools. Never the less this is an interesting problem.

I also tried a few tweaks (really all I have with my BIOS) and was unable to effect the results of testing. I did change out the Microsoft driver for the Intel storage driver and found the Intel was actually a little bit slower, again no change to the above results. It might be a driver issue or possibly a BIOS issue or a combination of the two to fix this.

 

I think it was Carlos that posted the C300 scores and his seemed a bit better than Sandforce scores overall, not just the 4k scores. Again, is this a real world problem where I will notice the difference using my Office 2010 and engineering software(which dosen't use very large files that often) or is this a difference that can only be seen by benchmarking? I think its just as valid to want to know if anyone has actually experienced a noticeable slowdown while using these drives and chipsets together under normal use as it is to want to see the benchmarks.

 

Trim support is actually built into Win 7 and there or no extra drivers necessary for it to work properly and you don't need to run the SSD toolbox app... Unless of course you don't have Win 7...

With regards to the real world performance question above, I agree that the reduced 4K numbers have little, if any, noticable slowdown for typical day to day useage....

 

Thats what I was thinking. It was also the answer I was hoping for. I think I'll be keeping my force 240 unless this is proven otherwise.

 

Edit: I would say with confidence that you would only see this in benchmarks.

I have had my laptop since april of this year and never benchmarked the drive. I felt the difference in hard drive performance was nothing less than amazing compared to a standard HDD. I knew the way that I had it tweaked that it should be performing as good as it could, and my first benchmark shows that it was (other than the 4k problem). Myself I am not going to worry at all about this problem. If a driver or bios update fixes it fine, if not I will not be losing any sleep over this.

 

I appreciate your input. Its a wonderful thing having people run benchmarks to test stuff and find problems when I don't have the time or know how to do it properly but sometimes it just as good to have real world, real time info as well. +1 rep to you.

 

Excellent response. You are absolutely spot on here. The difference when going from a mechanical hard drive to an SSD is like going from a 486 to a Core i7.

Aside from synthetic benchmarks, how does this issue truly impact our overall computing experience? I'm not seeing it. Don't get me wrong, I'm not excusing something that by and large should be fixed, but as said above, I don't think we should be losing sleep over it.

 

is trim support 100% built into win7 or it is partially in the OS and partially in current level chipset and disk controller drivers that run on win7.

big difference.

 

I had installed intel's matrix storage driver some time ago, which ended up overwriting the Microsoft achi.sys with iastor.sys. I didn't know that the MS driver that comes with Win7 supported everything necessary. I went hunting around on Intel's forum after realizing this and ended up installing their latest RST driver which apparently does support TRIM automatically. Things seem to be a little better than before, and I would get these strange lockups once in a blue moon that have not happened anymore since installing the new driver.

 

Of the metrics coming out of CrystalDiskMark 4K random performance is most predictive of real world performance.

By the looks of it there's a good chance that people who paid a premium for a high performance SSD are only getting medium SSD performance.

For example: by the looks of it a capped Vertex 2 is not faster than a normal performing Vertex 1. It might even be slower.

 

It is a write issue and IOPS are still so high you would never really notice the difference real world. It is just that you paid big bucks compared to a HDD and either suffer the lower storage space or way bigger bucks outlaid so you want what you paid for. This is especially true when you supposedly have paid for the latest and greatest chipset too..................

 

Phil,

Those same people are also seeing the limits of Sata 2 as well. Unless you are opening a large number of small files, or writing a large number of small files, where are we seeing this supposed slow down. I know the pagefile is probably the one windows system file usage that has a good deal of small files within it, but on my system i keep it on my HDD drive, and windows seems pretty darn fast to me.
Edit:
Tanware I just saw your post.You are right it really gets down to getting what you paid for more than really having serious performance issues.

 

Exactly, it will still feel fast, you're just not getting the performance you paid for. That is, if the 4K hit influences real world performance. Which I would consider very likely.

As far as I know, no one has measured it yet. If I would have access to a PM55 notebook it would be the first thing I did.

What I saw was that his Crystal results showed 20% lower 4K random reads and 50% lower 4K random writes than normal. Add to that that the C300 is not performing that well on SATA II, I doubt the C300 will do better than a Sandforce drive, also on PM55 chipsets.

It's unlikely you'll notice it, just as it's unlikely you'll notice the difference between a Vertex 1 and a Vertex 2 (uncapped). Yet if I paid to get Vertex 2 performance and would only get Vertex 1 performance, I would not be happy with it.