Is the onboard Marvell SE9128 really *THIS* bad?

[Soulkeeper]

I have one of the 88SE9123 addon cards

My guess is you are being limited by the pcie lane/lanes dedicated to this "onboard" chip.
I'm getting similar speeds as you are with my M4 128

If we think about it, it's kinda sad:
SSD controller -> ssd host controller -> pcie "bridge" chip -> pcie bus -> pcie host controller -> southbridge -> link to cpu/nb ... can vary of course
each step limited by latency, bandwidth, and potentially os/software factors.

Also worth noting, asmedia is a subsidiary of asus. They apparently are very small chips, but the performance is comparable to marvell. I noticed the asmedia product website has mention of SSD controllers. This makes me wonder if they might dive into the SSD game at some point.

 

[Yellowbeard]

If I were you I would use the SATA2 native port until the next mobo upgrade. Make sure to get an intel SATA3 native port when you get your next mobo.

^This^

It's been quite a while since I benched Intel SATA2 vs Marvell SATA3 but, when I did, the Intel benched slower than the Marvell in most drive benchmarks. But in real world tasks such as LARGE file transfers and also using PCMark Vantage the Intel controller mopped the floor with the Marvell.

 

[groberts101]

the Marvell chip ain't all that bad when you think back to what it was originally designed for.

With sata3 HDD it's perfectly fine even in raid 0. They just forgot to tell all the C300 owners what would happen with raiding those SSD's, is all. They(and all the other raid-happy SSD users) of course found out the hard way quite quickly. lol

The best thing I can possibly say about the Marvell chip's is that.. because they answer through the PCI-E bus(whether card or onboard).. I always get decent improvements with PCI-E overclocks on any system I've ever run that Marvell chip on. Even my 2720 shows marked improvements from OC with single drives and even moreso with larger raids.

you'll rarely see a a Sandforce 2281 based drive score 700+ on a Marvell chip and it's all thanks to the PCI-E OC.

 

[wpcoe]

I've come to believe that the Marvell chip, per se, is not so bad. It's just that in my case it's a double-whammy of being (a) on PCIe 1.x and (b) 1x lane.

If it had a 2x or 4x lane and/or if it were on a PCIe 2.x channel, I suspect it would be much better.

I have a PCIe 2.0 1x add-in SATA-3 card on order and should have it in a few days and will place it in a PCIe 2.0 16x slot. I'm very curious to see if it pans out like I think (hope!) it will. <fingers crossed>

 

[groberts101]

at this point in time(no PCI-E x2 lane Marvell chips used on cheap cards)?.. I would lean towards the ASMedia 1051 controlled cards. They'll get you 300MB/s or beyond write speeds.

be sure to post back and let us have a peek at your results when you get your card set up.

 

[Old Hippie]

I always get decent improvements with PCI-E overclocks on any system I've ever run that Marvell chip on.

So now we need to overclock? LOL!

That chip is pretty much crap and you'll get much better all around performance with an Intel SATA2 port.

But some will do anything to prove a point.

 

[wpcoe]

The card I have on order is an ORICO PAS3062-2S SATAIII PCI-E express card. Orico is a Thai company, and from the web site's slightly awkward English I wasn't sure of the rating of the card, so I e-mailed them and they assure me it is a PCIe 2.0 x2 card. That should have 1GB/s throughput, right?

I moved my video card to a PCIe 1.0 x4 slot and so far have not noticed any downgraded performance, so I will put the new card in the only PCIe 2.0 slot on the mobo -- the 16x slot intended for the video card. Where there's a will, there's a way.

 

[Old Hippie]

Where there's a will, there's a way.

Yep!

And you can lead a horse to water, but you can't make them drink.

And you can't make a silk purse out of a sow's ear but Good Luck!

 

[groberts101]

The card I have on order is an ORICO PAS3062-2S SATAIII PCI-E express card. Orico is a Thai company, and from the web site's slightly awkward English I wasn't sure of the rating of the card, so I e-mailed them and they assure me it is a PCIe 2.0 x2 card. That should have 1GB/s throughput, right?

I moved my video card to a PCIe 1.0 x4 slot and so far have not noticed any downgraded performance, so I will put the new card in the only PCIe 2.0 slot on the mobo -- the 16x slot intended for the video card. Where there's a will, there's a way.

there are only a few sata3 chips being used on cheaper cards right now and neither the Marvell or the ASMedia will deliver any more than the inherent PCI-E revision/lane allowance architecture will allow. So running devices off a chip that supprts single lane architecture will be no faster even if it's put onto a x16 card.

The chips can only do what the raw capability of the chip itself will allow and increasing the lane allowance of the cards interface speed will do absolutely nothing to raise that chips raw speed capability.

Will be nice when Marvell starts using the 92xx's with their x2 lane implementation. Still won't run 2 fast SSD's to 1 gig speeds.. but it'll be far better than the current ones allow. The 9182 chip uses x2 lanes and it still tops out at about 700MB/s in raids. That's Marvell for ya.

 

[wpcoe]

The only reason I'm going to use an x16 slot, is because it's the only PCIe 2 slot on the mobo. Since the Orico card is PCIe 2.0 x2, I want to give it the benefit of the PCIe 2.0 speed.

I wonder if they use their own (Orico) controller or a Marvell (or another brand)?

I assumed (dangerous, I know) that when they said the card was PCIe 2.0 x2, that the controller chip was x2. If the controller chip is NOT x2, then how can they say the CARD is x2?

 

[bobni]

I have read the Marvel ports will not allow the Windows 7 trim feature to be utilized by a ssd. Any thoughts or experience with trim and Marvel ?

 

[MeSoSneaky]

I have read the Marvel ports will not allow the Windows 7 trim feature to be utilized by a ssd. Any thoughts or experience with trim and Marvel ?

Interesting
Just upgrade mobo. Would probably cost the same after all the trouble. Sell old mobo too.

 

[Soulkeeper]

I have read the Marvel ports will not allow the Windows 7 trim feature to be utilized by a ssd. Any thoughts or experience with trim and Marvel ?

trim works fine with my SSD and marvel controller in linux.
It should work in windows too i'd think ...

 

[groberts101]

It wasn't until more recently that Marvell rolled trim pass-though capability into their drivers. Before that you had to run MSAHCI drivers to allow pass-through.

 

[Old Hippie]

It wasn't until more recently that Marvell rolled trim pass-though capability into their drivers.

This is the first time I've heard that.

Got a link?

 

[groberts101]

This is the first time I've heard that.

Got a link?

yeah.. somewhere. lol

IIRC it was posted on Marvell's site(rather obscurely) after much concern/speculation on various forums.

Personally?.. I'm still doubtful that they would have figured out scsi related trim pass-through quicker than Intel or other various raid controller mfgrs did. To be honest though, I haven't done much trim related testing since I rarely care about it when leveraging dedicated idle time GC anyways.

 

[Old Hippie]

yeah.. somewhere. lol

IIRC it was posted on Marvell's site(rather obscurely) after much concern/speculation on various forums.

Personally?.. I'm still doubtful that they would have figured out scsi related trim pass-through quicker than Intel or other various raid controller mfgrs did.

If you find it please let us know but I have the same doubts as you.

Do you happen to know if this is true.......

And when a trim command is sent and received, nothing is forcing the SSD to act on it immediately, it can put in a table that this space is now available to the GC, and wait till it's appropriate (idle time, or every X hours, or every XX GB written) for the collection.

 

[Soulkeeper]

If you find it please let us know but I have the same doubts as you.

Do you happen to know if this is true.......

In linux there is a way to verify if blocks were trimmed
i've verified that it works immediately on my particular hardware/software
with this method

 

[groberts101]

If you find it please let us know but I have the same doubts as you.

Do you happen to know if this is true.......

Can dig into some older favorites files on my other system later on tonight.

As for that trim recovery protocol?.. yes.. that is in fact how Sandforce controllers work. There are also other controllers that when tasked hard enough, will forego the actual recovery process until disk activity has settled down to the point that overall performance will not be affected.

And as far as forcing any Sandforce controller into recovering trim marked blocks?.. only one way to do that. Write to all physical space and force the controller into on-the-fly recovery.

Many have tried(with Linux as well) and it simply cannot be done. That's just the way they work and seeing that the logical markers/notification was sent is one thing.. but actually accessing the controllers physical mapping/space to confirm it is entirely another.

As it sits right now, no end-user available utility(or OS) will be able to do that. If it was possible?.. you'd see it plastered all over the reviewers sites by now for actual proof/measurement of trimmed blocks. Hell.. for that matter they'd even be peeking into the physical data structure to see what the various mfgrs are doing to optimize it(including wear leveling). All they can do as of now is to look at speed increases in the assumption that it has occurred. Logical data is not the same as physical data and is not going to tell you what the controller has done with that trim command.

 

[Old Hippie]

Thanks for your insight Groberts. :thumbsup:

 

[groberts101]

LOL.. always keepin me on my toes around here, hippie. Half the time I can't tell whether you're serious.. being sarcastic.. or just plain yankin my chain.

Of course, I'm the same in that regard and is probably why I like ya.

 

[wpcoe]

The PCie SATA-3 card arrived earlier this week. I had been assured via e-mail with Orico that the card was a PCIe 2.0 x2 card. However, upon boot, the Orico AHCI boot screen says "Asmedia 106X SATA Controller," and checking the Asmedia web site it appears it is PCIe 2.1 x1. That cuts the PCIe speed by half -- 500MB/s vs 1000MB/s. The card is in a PCIe2.0 x16 slot.

Comparing the 120GB Intel 520 SSD on (a) the motherboard (Intel P55A) chipset SATA-2 port, versus (b) the PCIe SATA-3 card:

Note: Tested using compressible data

Note: Tested using 46% compression.

I will keep the SSD on the PCIe card, at least for now. I'm disappointed to be only getting about 75% of the drive's advertised sequential speeds, but there still is a slight improvement overall on the PCIe card. If the card has been x2 I think I would be seeing 500+ on the sequential read/write speeds. Oh, well.

 

[Old Hippie]

LOL.. always keepin me on my toes around here, hippie. Half the time I can't tell whether you're serious.. being sarcastic.. or just plain yankin my chain.

I was serious.

The thumbs-up icon shoulda given ya a clue.

 

[groberts101]

I know I tend to ramble a bit too much at times, but I tried to tell you. And now that means I have to fall back on the old "I told ya so" here.

there are only a few sata3 chips being used on cheaper cards right now and neither the Marvell or the ASMedia will deliver any more than the inherent PCI-E revision/lane allowance architecture will allow. So running devices off a chip that supprts single lane architecture will be no faster even if it's put onto a x16 card.

The chips can only do what the raw capability of the chip itself will allow and increasing the lane allowance of the cards interface speed will do absolutely nothing to raise that chips raw speed capability.

Oh well.. you didn't spend an arm and a leg and your better off then you were. Hey.. at least you didn't try to raid SSD's off that chip! That woulda had you really upset.

 

[jacktesterson]

For a few days I was running a OCZ Octane SSD in my Laptop. (before dying and being sent back for a refund, thanks Newegg as always)

I put my old 64GB Kingston V100 back in (one of slower SSD's out there) and I noticed no difference for everyday usage.

Still WAY snappier than a mechanical.


Mechanical to SSD = Awesome

Avg SSD to Super Fast SSD = No real noticeable difference unless benchmarking