SSD Speeds low?

[RampantAndroid]

Hi all,

Just grabbed an Intel 510 250GB drive...realizing it would not be as fast as a Vertex 3...I just wanted to avoid OCZ. Now that I have it installed, perf seems low compared to even what reviews were showing.

I'm using an Asus Sabertooth X58 with the Marvel SATA3 controller onboard. Here are my results, while the drive was not in use by windows (tests run from a different windows install.)

Do these seem low? Is this because of the motherboard I'm on? Alignment looks good...so I just don't see what else it could be...

If it IS t he motherboard, is there a PCIe SATA3 controller that would offer better performance?

 

[Echo147]

I have a similar issue (and near identical speeds), 510 250GB on Asus P6T (x58) - have tried both Asus' U3S6 controller and Startech's PEXSAT32 (both marvell-based) and can't get near the advertised speeds.

I'm pretty much stuck with it until switching to a motherboard with onboard 6Gb/s which isn't Marvell-based

 

[RampantAndroid]

Well, I was looking at the U3S6 as a possibility...guess not.

Oh well...no native 6Gbps X58 mobos out there, that I know of...

 

[RampantAndroid]

I have a similar issue (and near identical speeds), 510 250GB on Asus P6T (x58) - have tried both Asus' U3S6 controller and Startech's PEXSAT32 (both marvell-based) and can't get near the advertised speeds.

I'm pretty much stuck with it until switching to a motherboard with onboard 6Gb/s which isn't Marvell-based

This will be of interest:

http://forums.storagereview.com/ind...tel-510-ssd-on-x58-ich10r-marvell-quick-test/

Seems it's bound by it's PCIe link...a single lane. Pretty sad.

Doesn't explain why add in cards don't work (could be the SAME reason though...)

Oh well, the drive is capable of SATA3. One day I'll be on SB/IB/LGA2011 or something else with proper support. And I'll buy an Intel motherboard with no bells or whistles... :hmm:

 

[Echo147]

Interesting read, thanks.

Does seem idiotic to use a single lane when that's way short of SATA 6Gb/s limits, talk about false advertising

Roll on LGA2011 and fingers crossed both USB 3.0 and SATA 6Gb/s are finally native!

 

[greenhawk]

Does seem idiotic to use a single lane when that's way short of SATA 6Gb/s limits, talk about false advertising

would not be the first time something stupid has been done.

was a poster on another forum that spent extra to get a pci-x slot so they could run their scsi raid card, back when pci-e started comming out.

turned out, while the pci-x solt worked at full speed (64bit @ 66Mhz), the controller chip was limited to 1xpci-e slot instead of the 4 the controller could handle. so ~150MB/s max instead of the 500MB/s+ that the slot should have been capable of.

 

[sub.mesa]

Also i'm not very sure whether the Marvell drivers you're currently using supports TRIM. You may want to use only drivers that are confirmed to pass the TRIM command. Those are the stock Microsoft drivers (AHCI & IDE) and Intel/AMD drivers released recently. There was some talk whether or not recent Marvell drivers support TRIM, but so far I've not seen anything substantive.

 

[RampantAndroid]

Also i'm not very sure whether the Marvell drivers you're currently using supports TRIM. You may want to use only drivers that are confirmed to pass the TRIM command. Those are the stock Microsoft drivers (AHCI & IDE) and Intel/AMD drivers released recently. There was some talk whether or not recent Marvell drivers support TRIM, but so far I've not seen anything substantive.

Reasonable point - though crystalinfo reports trim capabilities. I'll move back to ich10 I guess...just as fast anyway.

 

[thescreensavers]

I dont get why people are still using Cyrstal DiskMark for SSDs. OP you just read/wrote 20gb of data to your SSD in your simple test.

Use ATTO(uncompressed data) or AS SSD(compressed data) some drives handle compressed and uncompressed data differently.

 

[Phynaz]

Yes, how dare he shorten his drive life by an entire day!

 

[Yellowbeard]

would not be the first time something stupid has been done.
.

LOL....so true. I remember back in the old 875 chipset days that one unnamed MOBO maker put a SCSI 320 controller on the PCI bus. Can you say bottleneck? On top of that, they never released a working boot ROM so if you did choose to use this controller, it could only be for attached storage. You couldn't run the OS on it.

 

[stahlhart]

Iwill?

 

[Yellowbeard]

LOL....no and since I'm a rep I should not publicly say who. It might come back to byte me ;-)

It was an enthusiast board, not a server board.

 

[sub.mesa]

Reasonable point - though crystalinfo reports trim capabilities. I'll move back to ich10 I guess...just as fast anyway.

What CrystalDiskInfo reports is what the SSD tells it to; the SSD reports having TRIM capability, but that does not mean the TRIM commands effectively reach the SSD. For TRIM to work:
1) you need filesystem support which actually generates the TRIM commands (NTFS on Windows 7, Linux, BSD)
2) you need a storage driver that accepts and passes TRIM to the SSD (Microsoft AHCI, Microsoft IDE, modern Intel and AMD drivers while the SSD is not part of a RAID array)
3) you need an SSD which accepts and implements TRIM commands

There is no way to actually measure that TRIM commands each the SSD. All you can do it ensure the TRIM is not explicitly disabled on the command line (deletenotify setting) - something that should never happen unless the user changes this setting. You can see that the SSD reports having TRIM functionality, but you cannot see whether the driver actually passes the TRIM command.

So in the above three steps, the second step is crucial and that's the one the least amount of information is available on. Pity; since i guess alot of people use a TRIM-compatible SSD on say a Marvell controller with Marvell drivers and think they have TRIM, but they don't have TRIM since the driver does not pass the command (it silently ignores the TRIM commands it receives).

I dont get why people are still using Cyrstal
Use ATTO(uncompressed data) or AS SSD(compressed data) some drives handle compressed and uncompressed data differently.

Intel 510 uses a Marvell controller and that should not give any difference between compressed or uncompressed data. Only Sandforce controllers employ compression and de-duplication techniques so that easily compressible data such as the zeroes that ATTO writes would yield totally different scores than real random data like CrystalDiskMark and AS SSD benchmarks generate.

Though in most cases; benchmarking with 1x 100MB would be sufficient for decent measuring. So you don't have to benchmark at 5x 1000MB. But even if you do, it should not be of major concern to drive longevity. It would be bad for performance if you do it on an SSD which cannot receive TRIM commands though, such as when using a non-TRIM compatible storage driver.

You can see which driver you're using with AS SSD. It displays it at the top-left corner where it says "OK" or "BAD".

 

[hdrive]

...
There is no way to actually measure that TRIM commands each the SSD. All you can do it ensure the TRIM is not explicitly disabled on the command line (deletenotify setting)...

Google "SSD TRIM Monitor", which shows a tool (hIOmon) that might be helpful in this regard.

 

[groberts101]

while he's at it with that tool.. he should check his writes(or smart data) to see the nearly 100 gigs of data written by that 5 x 4000MB test size. 5 x 1000MB default tests write about 20 gigs per test.

it should not be so much about lifespan(as most are now learning), but more about fresh block reserves. Not too familiar with that drive but being a Marvell controller it may not be the greatest at immediately trimming and returning those dirtied blocks right back into the fresh reserve pool.

Think of them like coffee pot sizes for a maker. You shouldn't pour it all off in one shot.. or you'll need to allow some time for it to be replenished.

TRIM commands are one thing. When the controller actually returns those blocks back into the fresh pool will vary between systems and controller types. Dedicated garbage collection time never hurt SSD performance either.