SSDs: Performance ok? (1 Viewer)

[Luca Brasi]

Hey guys,

don't ask me why but I thought I would check my ssds performance. just to clarify: theres nothing wrong, everything is running fine and fast... but out of curiosity I would like to know what you think about the following results:

My htpc: ocz vertex 2 60GB, 43GB used, around 2 years old.

compared to a review in the net (ok, this is 120GB):

My pc: crucial m4 128GB, 20GB used, 8 month old.

compared to a review in the net:

Thoughts:

- Reviews are done with empty drives. Are SSDs significantly slower when some data is on them? (I know they need a certain amount of free space). Where's the critical point?
- I know I could have done the alignment and partitioning manually to have 1024k but does this explain a drop in performance? Can't believe that...
- The M4 is connected to a additional marvell chip not the onboard intel. I will have a look at that tonight...
- Benchmarks are bs either way...

What do you think?

Luca

 

[Scythe42]

Use the Intel. The Marvell is way slower in real life application from my experience (I don't care about benchmarks) with SSDs. Using M4s as well in several recent builds for friends. Also make sure you have recent chipsset and SATA drivers from Intel installed. Benchmark results can vary depending on the chipset of you mainboard. The bottleneck doesn't need to be the SSD at all.

Free space is no issue for a recent SSD as there is "reserve memory" to map defective cells. When a disk is full (doesn't matter if HDD or SDD), the more likely you encounter more defective sectors and the drive needs to re-map them. This is the reason for a speed drop for consumers storage when disk become full as they do a lot of tries before remapping.

1024K alignment is automatically done by Win anyway when partitioning/formatting in case of an SSD unless you use an ancient OS like XP in which case SSD performance will always suck.

 

[Luca Brasi]

hi,

thanks for your thoughts!

Use the Intel. The Marvell is way slower in real life application from my experience (I don't care about benchmarks) with SSDs. Using M4s as well in several recent builds for friends.

right. as mentioned above everything is running fine in real live. but will go for the intel to see if there any improvement in real or bench performance.... maybe later...

Also make sure you have recent chipsset and SATA drivers from Intel installed. Benchmark results can vary depending on the chipset of you mainboard. The bottleneck doesn't need to be the SSD at all.

They are, had to set up my sys a few weeks ago.

Free space is no issue for a recent SSD as there is "reserve memory" to map defective cells.

That's good info. So I don't have to think about that...

When a disk is full (doesn't matter if HDD or SDD), the more likely you encounter more defective sectors and the drive needs to re-map them. This is the reason for a speed drop for consumers storage when disk become full as they do a lot of tries before remapping.

Did a check how many sectors have been remapped. According to the vendors tools both drives are fine...
OCZ says

Reserve blocks remaining: 100%

Crucial: same value in SMART

1024K alignment is automatically done by Win anyway when partitioning/formatting in case of an SSD unless you use an ancient OS like XP in which case SSD performance will always suck.

Sure. Still there is some advice in the net to do the partitioning/formatting manually so you won't get the system volume before c:/

I still wonder why the values in the benchmarks are so significantly lower than in the reviews....

 

[Luca Brasi]

Hey again,

so I couldn't resist to change the sata port to a intel one.You were absolutely right Scythe42.
Way better! Feels better, results are better:

Now:

Before:

Especially write performance improved a lot.
Additionally boot up is faster as I disabled the marvell chip in bios.

I remembered why I went for the marvel in the first place: It is 6G, the onboard intel is 3GB/s.
Either way it seems to be faster. But might this explain the still existent backlog compared to the review?

So, any more good thoughts to speed up the ocz?

 

[Scythe42]

That looks more like it for an M4!

Actually I would need to see your BIOS/EFI settings and what mainboard/CPU/RAM combo you are using. As well as the Win7 settings (in general the defaults should be perfectly fine). There is still room for some improvement. 15-20% should be possible easily. Maybe more on sequential reads. But hard to say from here.

With various "tweaks" on the net you won't get much more. Avoid them, they are all bogus! Win7 knows what SSDs are and handles them correctly.

If you benchmark make sure Virus Scanners and disabled to make sure they don't have an effect on the benchmark. For real live later they shouldn't be a problem. Also make sure Win7 has indexed everything so basically there is not much activity in regards to IO. Just let the system run a bit. Doesn't take too long.

So it comes down to Intel Drivers and BIOS/EFI settings, which are often very conservative. Basically: set everything to Plug and Play 64Bit OS, so the OS can control anything. All AHCI/ACPI stuff to the newest version. Disable any legacy support unless some hardware requires it. And check if the RAM timings are correct. Make sure chipset and sata drivers from Intel are current (download them directly from Intel).

That's all I can help you with from here.

PS: Also check if your M4 has a recent firmware. Updating it is usually no biggie. But you don't need to. But it can't hurt also. If you feel comfortably with such stuff. Same for you mainboard BIOS. Just the usual maintenance you do when upgrading hardware.

 

[Scythe42]

In regards to reviews: They usually don't boot the OS from the drive and just have it connected as another one for theoretical performance. If you like I can do a benchmark on my Core i7 with an M4 for reference of a real system that's perfectly setup running MP.

 

[Luca Brasi]

Actually I would need to see your BIOS/EFI settings and what mainboard/CPU/RAM combo you are using.

I'm gonna stick with the pc for the moment....
I'm still running bios here although i think (??) the board of the pc supports uefi. Didn't have the time and/or mind to get into that...
Bios settings are quiet standard: AHCI on, all devices not needed disabled, .... wait, some pictures on their way....

(too lazy for all the typing )

Sys like that:

As well as the Win7 settings (in general the defaults should be perfectly fine)

Some tweaks are done: prefetching off, hibernate off, pagefile set to 2048/2048MB, indexing still on.

With various "tweaks" on the net you won't get much more. Avoid them, they are all bogus! Win7 knows what SSDs are and handles them correctly.

Yepp, think so too. The tweaks above I consider reasonable though.

If you benchmark make sure Virus Scanners and disabled to make sure they don't have an effect on the benchmark. For real live later they shouldn't be a problem. Also make sure Win7 has indexed everything so basically there is not much activity in regards to IO. Just let the system run a bit. Doesn't take too long.

Already took that into account before doing the first bench....

So it comes down to Intel Drivers and BIOS/EFI settings, which are often very conservative. Basically: set everything to Plug and Play 64Bit OS, so the OS can control anything. All AHCI/ACPI stuff to the newest version.

I activated acpi 2.0 while taking the pics. Newest result is this:

Little bit lower then before but this is no reliable measurement. Could be any kind of system activity not active while first run.

One thing came to my mind: SSDs seem to be limited by cpu performance. So I disabled balanced power mode temporarily and went with high performance:

No significant difference in my opinion....

Disable any legacy support unless some hardware requires it.

The usb legacy support is still on as I have some older smartcardreader here. never checked if it runs while deactivated...

And check if the RAM timings are correct.

They seem fine to me. Never been a friend of messing around with it as I don't see considerable performance gain.

Make sure chipset and sata drivers from Intel are current (download them directly from Intel).

Ever did and ever will...

PS: Also check if your M4 has a recent firmware. Updating it is usually no biggie. But you don't need to. But it can't hurt also. If you feel comfortably with such stuff. Same for you mainboard BIOS. Just the usual maintenance you do when upgrading hardware.

Both are running the next to latest version. Don't see any advantage in updating right now as both latest ones seem to fix some stuff I'm not concerned about.

In regards to reviews: They usually don't boot the OS from the drive and just have it connected as another one for theoretical performance. If you like I can do a benchmark on my Core i7 with an M4 for reference of a real system that's perfectly setup running MP.

Thanks! That'd be nice.... I'd like to see some more comparable results...

 

[Scythe42]

Looks OK on the first glance. I'll comment on some stuff later this night and also do a benchmark on a live idle system for you that runs MP and is also used as a full audio workstation, so you have comparable results beside pure review tests.

 

[Luca Brasi]

Great, looking forward to it...