I'm getting my Vertex 3 120GB SSD today. From reading up a few forums, am I right in thinking that the ICH10R (Sata 2 3Gbps) will give me better performance than the Marvell 9128 (Sata 3 6Gps) due to the limiting factor of only having 1 PCIe lane (5Gbps)? Cheers, Simon.
Hey Simon, Those vertex 3's, as I'm sure you know, use the updated sand force controller and will happily push above 500MB per second on compressible data. Running it on a 3Gbps (375MB) controller will certainly hold it back. I would try the SATA 3 port for sure. These 3rd party chips on boards haven't bee all that great but it is worth the attempt I'd say. If there's some bottleneck against 1xPCI-e gen 2 lane here, then that's still 500MB per second. Not completely ideal, but significantly better. You should also have a pci-e x4 on your board and could always use an add-in card like I do. The Asus U3S6 has worked a treat for me.
i think you mean 275MB/s for 3Gb/s... & you'll not get anywhere near 500MB/s using that controller - or in fact any 1x pci-e gen 2 port as there are overheads - but sensibly you'll struggle to get much more than 400MB/s sequentials with the shonky Marvell 9218 thing, if that... Yeah, simple way to look at it is that 'if' you want higher large sequential r/w (though still heavily ltd by the shonky 6Gb/s controller) use the Marvell... ...& 'if' you want noticably higher random r/w (though more heavily ltd large sequentials) then use the intel 3Gb/s controller.
yep as i said, before, likeley to hit max of 400mbs, my m4 on old marvel chip gives me 380mbs and ssd capable of more.
1B = 8b so I think 375MB/s is right technically, but agree that there are overheads and that's not a real world speed - which is probably more important. Also that the marvell is a bit shonky unless they've fixed it recently. I would go the add-in card if you have space. The x58 gigabyte boards tend to have that x4 slot doing nothing and it would be a shame to hamstring the vertex 3!
Thanks for the reply's. Anyway, I tried the drive on the ICH10R.....Basically, it sucked! On the Marvell controller, it is limited. Pictured below. But slightly faster than my C300 in raid 0. roosauce; I only have 1x PCIe slot left, as my 3x 580's and sound card take up the rest. Cheers, Simon.
Sorry, but that's a misunderstanding of the basic tech... Whilst it is a 3Gb/s interface which would mean 375MB/s, all of the (actual) data is 8b/10b encoded. This, by itself, takes those figures & reduces them to 2.4Gb/s or 300MB/s as the absolute theoretical maximum for the interface for transfering actual data. You then take off latencies & other overheads &, whilst you'll 'can' have a burst transfer rate that isn't 'that' far off the 300MB/s - the sustained transfer rate is lower - hence something around the 275MB/s mark 'should' be achievable as a sustained transfer rate with a decent 3Gb/s ich controller & rst drivers. Personally, i'd stick it on the intel controller pro tem, & wait until something better comes out in the budget card end.
No problem at all. Yeah, it's the problem that that's they've named the tech after the encoded transfer speed... ...though it obviously sounds faster. it's obviously not the only 8b/10b encoded thing out there (or one that uses an alt encoding), but i can't instantly think of another one that's so obviously mis-advertised. Well, something like pcie being 250MB/s & 500MB/s per lane per direction for gen 1 & 2 is easy to remember & both are the theoretical max speeds after the extra 2 bits that are sent or received from the 8b/10b encoding are knocked off... ...& gen 3 pcie being 1000MB/s per lane after encoding is then obvious - although that instead uses 128b/130b encoding - make it ~18.5% more efficient (if i can add up today).