How To Improve SSD performance on Intel Series 4, 5 and 965 chipsets (Stamatisx Tweak)

Here's a nice overview of 4K write speeds:
OCZ Vertex 2 100GB Solid State Drive Benchmarks – Crystal Disk Mark :: TweakTown USA Edition

Only the ADATA performs similar to TANware's.

 

You have to remember that this is compressed data, if I do the aligned data the numbers are higher.........

 

Sure. The benchmarks in that chart are done with compressed data too.

I know you were quite sceptical about Sandforce SSDs and performance degradation. How do you feel about it now?

 

Oh boy TANWare you should stop using CDM it's not made for too fast ssd

The Adata S599, the OWC mercury extreme pro and the OCZ Vertex 2 use the sandforce SF-1222

interesting link here as well
http://www.storagereview.com/owc_mercury_extreme_pro_ssd_review_120gb
http://www.overclock3d.net/reviews/storage/mushkin_callisto_120gb_ssd_review/3

 

Just read through this thread and wanted to clarify some things.

1) My tweaks leave Turbo boost fully functional

2) Music - While you may joke about this as a fix it is really just a way to load the CPU with another thread which in turn allows my tweaks to provide full SSD speeds accross the board during CDM runs (much higher than without the tweaks). Anything that places a small load on the CPU seems to work with the same effect. Keep in mind that you can get the same accross the board improvements with my tweaks by changing the two 95% settings to 99% or 100% and then you don't need to 'play music' for the gains in CDM on all R/W speeds and Turbo still works.

3) Without 'music' the 'real world' results for me (@95%) show vast improvements --

MSE full scan ~8 min faster

Adobe full CS4 install 3 min. faster

Reboot time 3 sec faster

Loading 22GB to iPod touch 9 min faster

Bottom line is that I think that 'music' while just running CDM shows the real world speed potential since most things you do that need high speed R/W's are going to place the additional load on your CPU without the need to for adding 'music'...

 

Actually those are 1 fill, CDM was set back to default before I took the Screen Shot.

I am happy with the drive until lifetime write throttling enables then is it time for a Secure Erase.

CDM is fine, just prepare to do a SE sooner than you normally would...............

 

Stamatisx, how come you are still recommending both lines, while only the ppm line seems to work fine?

Are there any laptops that need both lines?

 

Not really, it works just fine with only the intelppm.
It's that I have always being using it with both (due to overclocking) so I posted it like that.

I also did some testing, with that tweak I can run the wprime 1024M using 4 threads 10sec faster than without it. The max temperatures are the same with and without the tweak. Just mentioning it if somebody is interested to test it.

 

Even works on the Intel PM 45 Chipset, Ran CDM, 50MG 4K test
Before: 19.56 / 48.05
After: 22.73 / 71.07
Nice work, rep added.

 

Thanks man, good to know that!

 

So it works on 5, 4, and 965. Wow time to change the title.

How To Improve SSD performance on Intel Series 5, 4 and 965 chipsets (Stamatisx Tweak)

Do you agree stamatisx?

 

Be my guest Phil

 

I run a PM45 chipset as well. I can also tell you with just the PPM at 4 I get 68 MBs on 4KQ 1Q but with proccessor at 4 as well the 4KB jumps to over 90 MBs..............

 

Good observation! Thanks

 

I was curious as to why my clock seemed to matter so much on the throughput and seem to have found something interesting. While my system runs fine the tweak had an effect to DPC latency.

While I was under 2000us at 3.2Ghz at times it was close and almost always at 500us or greater. With the tweak off my DPC Latency average is about 1/2 or less of that. My best guess here is that the CPU cycles dedicated to the hardware bus is greater with the tweak therefore higher CPU clocks avail even more computing cycles............

 

Yeah, run a wprime with and without the tweak to see that with, it's faster

 

@Stamatisx and Phil, Wouldn't it be a good idea to include a link or write up that explains the fundamental tweaks that need to be done for a SSD (disable prefetch, superfetch, enable TRIM, etc). The reason is over time people coming to this thread from 'Google search' will just blindly do the things mentioned in the thread and forget the fundamentals.

 

There is a thread already for that
http://forum.notebookreview.com/har...sd-performance-tips-underperforming-ssds.html

 

There is , but just post a link to it for folks who come straight to this thread.

 

Well I've tested a bit more;
When I go from 3 to 4 on the "Start" item in either Intelppm or Proccessor I go to 68MBs on the writes of 4K 1Q with CMD. The difference is with the tweak in the Intelppm the DPC Latency is greatly increased and with the proccessor DPC Latency is actually slightly lessened.

Also with only the Proccessor tweak the CPU in device manager is still recognized properly.....................

 

This is while running the benchmark


No latency in my laptop, it is exactly the same with or without the tweak

 

Very nice, seems a low DPC Latency for those systems..............

Edit;

DPC.................. DPC_Pro......... DPC_PPM_Pro..

 

Here are my benchmarks for my 2 SSDs in my AW M17x-R2

After putting the regedit changes in I have my best results to date!

 

Neutron, the intel numbers are pretty low, specifically for 4k R/W.
I get about 22/45 there with JJBs tweak applied.

 

Ugg... as you can see i've never had good performance on those numbers

Where's JJb's tweak? been looking, dont see it

 

Can you make the picture any bigger? i am having a hard time seeing what kind of numbers you are getting.

 

i can try... maybe you should decrease your resolution

 

Decided to give the tweak a run, and the 4k's have improved nicely. Left is before the tweak, right is after.



I got to hand it to you Stamatisx, you're definitely bringing the 4k's back! +1 Rep.

 

It's a little messed up.. Read the posts that are linked to here http://forum.notebookreview.com/har...-full-advantage-fast-ssds-51.html#post6684914

 

@ Mr. Neutron, Hearst, CDM with 5 runs of 1000MB file size writes several gigabytes to your SSD. If you set the file size to 3x 50MB you can still measure improvements while causing less wear.

 

Here

 

my only issue with this tweak is that turbo boost goes off with it. I wouldn't mind doing it.. has anyone with a 128GB C300 SSD done stanimax's tweak? I want to see what improvements u get compared to JJB's..

 

Sean, you do know that the tweak only affects installation performance?

All other heavier tasks will automatically make the CPU come out of the power saving states that are causing the problems.

 

uhh i did the tweak and can't tell if my turbo boost is malfunctioning or not.

820qm, i'm using argus monitor to look at the cpu speeds. looks like all 4 cores are boosting on regular use from stock 1.73ghz to 2.0ghz but it never boosts to faster speeds on a lesser amount of cores. if it's already boosting to 2.0ghz then it's working right?

 

i've tried it again... this time with JJB's tweak and turbo hit 1.87GHz max... i was running super pi which should stress 1 core... i wonder how stanimax is hitting his turbo so high... i've installed 10.0 RST drivers which seem to help too...

 

As I have mentioned before, this is due to the BIOS, I can overclock the CPU and hit 4.355 GHz across all cores


and maintain a multiplier of x30 for single threaded applications like superpi.
http://hwbot.org/signature.img?iid=448480&thumb=false&iehack=.jpg

but if I run a wprime with 8 threads simultaneously, the multiplier goes down to x23 for all the cores due to the temperatures

 

unfortunately this tweak removed my turbo boost which is a shame because i got way better SSD scores with it as opposed to just using JJBs tweak.

 

ok i answered my own question above. it did cause the turbo boost to malfuntion on my i7-820qm.

 

how did you guys test to see if the turbo was working? Using the intel turbo boost monitor and running superpi with stamatisx's tweak, I was getting cores Oces to around 2.00 and 2.4ish, but only for a second

 

run superpi that will use only one core and check your turbo boost

 

A dutch user posted his results of applying the tweak with a Seagate Momentus XT and an Intel G2.

The Intel doubled it's 4K random performance in both read and writes.
The XT more than doubled it's 4K and 4KQD32 random performance in both read and writes.

Is there any way to make a switch or a command line that changes the Intelppm from 3 to 4? So we don't have to go into the registry every time?

 

Phil, I think that dutch user's results are a little misleading.

The basic test was at 5 run at 1000MB size and the 'better' results were 5 runs at the 50MB size.

This basically invalidates it for me.

 

Yes you are right. I asked him to re run.

 

Here you are Phil, I have attached a zip file with two registry files inside, double click on the one that you want in order to apply the changes. (Reboot is required though after every change)

 

damn. thanks to the OP.

just got me a 60gb G.Skill Phoenix Pro Series (Sandforce 1200 Controller) for $60 bucks. 15-750 + p55 mb + 4gb RAM

here are my results before and after doing the registry tweaks.

Before



After

 

After some troublings, here there are my results after changing the registry

Code:

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Intelppm

on Start from 3 to 4.
Thanks to Phil and of course stamatisx for the hint

These are after a fresh install of Win 7 Ultimate 64 bit and SSD tweaks.
The HD is A-DATA S599 64GB

Before:



After:



Moreover, Turbo Boost is still working!!!:laugh: :laugh:


WEI for the SSD passed from 7.5 to 7.6.



I'm wondering if keep the change or go back to "factory" settings.
My concern is don't overstress the CPU!

Any advises?

 

What upsets me is that, if you compare the following benchmarks with my former post (A-DATA S599 results after registry tweak)


The intel is faster!!!
Isn't suppose the Intel to be slower than the new SF1222 controller?
If you look ATTO benchmarks, it is so:


This SSD is a Intel X25-M 80GB that I bought a year ago.
It's installed on my game machine, a i7 920 socket 1366 overclocked to 3.6GHz, ASUS P6T DeLuxe V2, Mushkin Enhanced Blackline 6GB SDRAM DDR3 1600, and two fanless Gigabyte HD4850 in CrossfireX.

I trust more to CrystalDiskMark than ATTO benchmarks...

Is it normal that the Intel is better than the A-DATA SSD?
Thanks for any replies

 

ATTO benchmark on any Sandforce based SSD is basically meaningless. It only measures the interface speed, not the real usage speed(where data are actually read from/written to the NAND).

If you want to measure the true sequential speed (read) of a Sandforce drive, use the plain old HDTune on a populated drive(say a typical Windows installation). That would give you an idea of how fast it can read from real data(forget about the average or the max, just look at the slowest one that is where data are actually read from the NAND).

 

ATTO measures speeds with compressible data. CDM on default setting works with incompressible data.

Sandforce drives work with real time compression, hence the better results in ATTO than CDM.

In CDM with default settings? Should be similar. Try running CDM with 0 fill data and the Sandforce will be faster.

How does the ATTO look for the Sandforce drive?

In reality the Intel and Sandforce drive will perform nearly the same.

PS. running CDM with 5x 1000MB isn't good for longevity. Instead run 3x 50MB.

 

Thanks Phil, thus, here there are 0 fill data (3x50MB, okey dokey) :wink:

A-DATA S599:
[/URL][/IMG]

Intel X-25M 80GB:
[/URL] Uploaded with ImageShack.us[/IMG]

Chimpanzee (nice ID, really), :laugh: I agree with you.
Found HDTune download and there you are!

A-DATA S599 64GB:
[/URL] Uploaded with ImageShack.us[/IMG]

Intel X-25M 80GB:
[/URL] Uploaded with ImageShack.us[/IMG]

These data suggest that there is something odd with the A-DATA...

In fact, I found this:
[/URL] Uploaded with ImageShack.us[/IMG]

I mean... 192 reallocated sectors is not good, right?

Instead, the Intel was fine:
[/URL] Uploaded with ImageShack.us[/IMG]

What is the best thing to no now????:confused2:
PLEASE, HELP!!!:cry: :cry: