Intel "Eaglelake" Chipset

[nmX.Memnoch]

DailyTech had some good information on Intel's next chipset series posted.

http://www.dailytech.com/More+Intel+Eaglel...article8077.htm

This bit sounds really interesting:

Intel pairs the Eaglelake-family with a new ICH10 south bridge. Intel has put the ICH10 on a diet. ICH10 only provides Serial ATA, PCIe, USB and high-definition audio – essentially legacy free. New to the ICH10 south bridge are FIS-based port multipliers. Multiple SATA devices can piggyback off a single SATA port with FIS-based port multipliers, without sacrificing I/O performance.

Following their link on the FIS-based (Frame Information Structure) port multipliers for the SATA controllers provided some cool info:

FIS–based switching offers high performance storage connections to multiple drives simultaneously. The host issues and completes commands to drives at any time. The port multiplier will direct data to any drive ready for I/O. An arbitration algorithm ensures a balanced data flow. Unlike Command-based switching, FIS-based switching allows aggregation of reads to fully use the higher bandwidth of the 3Gb/s host link and takes full advantage of the performance benefits of Native Command Queuing (NCQ) on the port multiplier, resulting in aggregated throughput of up to 300MB/second.

[ripken204]

well that all sounds cool but i really hope that the sata thing works.. they say "An arbitration algorithm ensures a balanced data flow." what i'm getting out of this is if you have 5 disks on 1 port, that all 5 will transfer at about the same speed, but that they won't transfer at their full speed. 2 or 3 drives will probably work fine tho. i think sata3 is coming out next year and that would work nice too if intel puts this feature in those mobos too.

[nmX.Memnoch]

It's a known fact that a single SATA drive doesn't use the entire bandwidth of a single channel anyway. I don't think having four drives on a single channel is going to hurt the performance of a desktop computer. The only time lots of data will get written to multiple drives is with RAID arrays, and most enthusiasts (small part of the market) generally only have two drive arrays (whether it be RAID0 or RAID1). You could eliminate any sort of problem by splitting the array drives up between channels. For example...I have two arrays in my system. I could put one 250GB drive and one 400GB drive on one multiplexed port, and the other 250GB and 400GB drives on another. Then the arrays would be split across channels.

There's 1000 different ways it could be setup so that it wouldn't impact performance at all.

Also, please let's not start calling it SATA "3" since there's no such thing as SATA "2" or SATA "II". Calling it SATA3 will only confuse people into thinking "3Gb/s".

[puntoMX]

Interesting as always...

But why not put an extra controller by it self on a motherboard?

I see most newer chipsets with 4 ports and up already.

eSATA can handle 5 devices with a port multiplier so now the wan´t to do it too for the "normal" SATA.

For the signal strength they just rised the voltage on those ports a bit like they did with eSATA and use almost the same "protocol" changes as they did for eSATA to use the port multiplier.

Most people will call SATA 150 as SATA 1(I), SATA 300 as SATA 2(II) and upcoming SATA 600 as in logic SATA 3(III) so we will understand each other altrough it isn´t 100% correct in tecnical words .

I personaly would never hook up more HDD drives on 1 bus because of burtsmode (Cached) that could get higher then 150MB/s of each drive. I see it as a good option for optical drives...

Other then that, just marketing as always, but good if it´s almost for free .

EDIT: just a picture and a link.

Edited July 19, 2007 by puntoMX

[Zxian]

Cooooool. I'm assuming that the Intel Matrix RAID is also supported on this chipset?

I love Intel. They know when to cut their losses and make something great. I really haven't heard anything new or exciting from AMD lately, and Barcelona is starting to get the feeling of vaporware...

[nmX.Memnoch]

But why not put an extra controller by it self on a motherboard?

I can only make assumptions, but my first guess would be power usage. Second would be silicon space in the ICH.

Most people will call SATA 150 as SATA 1(I), SATA 300 as SATA 2(II) and upcoming SATA 600 as in logic SATA 3(III) so we will understand each other altrough it isn´t 100% correct in tecnical words .

Most people do it because it's a marketing term...and then they see other hardware enthusiasts/professionals (who know better) use the same terms.

I personaly would never hook up more HDD drives on 1 bus because of burtsmode (Cached) that could get higher then 150MB/s of each drive. I see it as a good option for optical drives...

Ever use SCSI? It never hurt the performance there. SAS is already using the same type of technology as this. What you're seeing is a convergence of SCSI and SATA technologies to get the best of both worlds (which is what SAS already is).

Other then that, just marketing as always, but good if it´s almost for free .

I think there may be a bit more to it than just marketing...but no one has ever said that Intel wasn't a marketing machine.

Cooooool. I'm assuming that the Intel Matrix RAID is also supported on this chipset?

I would assume so but it's way too early to tell yet. This is post X38 stuff for next year.

[jcarle]

I would assume so but it's way too early to tell yet. This is post X38 stuff for next year.

X38... now that's what I'm just itching to buy. I can't wait for it to come out. It's eating away at me like mesothelioma.

[nmX.Memnoch]

That's what I'm waiting on as well...even though I can't justify replacing my X6800 yet.

[puntoMX]

I personaly would never hook up more HDD drives on 1 bus because of burtsmode (Cached) that could get higher then 150MB/s of each drive. I see it as a good option for optical drives...

Ever use SCSI? It never hurt the performance there. SAS is already using the same type of technology as this. What you're seeing is a convergence of SCSI and SATA technologies to get the best of both worlds (which is what SAS already is).

Yes I do use SCSI/SAS my self here for servers and workstation, I don´t sell only low-end systems . I started with those 50 pin SCSI drives long long ago (yeah I´m older then 16...)...

300MB/s Isn´t good enough for 2 drives at full burst speed, okay, not that you have it a lot, and yes, for just storage it will be okay.

I think it would be great when SATA 600MB/s comes out .

[ripken204]

It's a known fact that a single SATA drive doesn't use the entire bandwidth of a single channel anyway. I don't think having four drives on a single channel is going to hurt the performance of a desktop computer. The only time lots of data will get written to multiple drives is with RAID arrays, and most enthusiasts (small part of the market) generally only have two drive arrays (whether it be RAID0 or RAID1). You could eliminate any sort of problem by splitting the array drives up between channels. For example...I have two arrays in my system. I could put one 250GB drive and one 400GB drive on one multiplexed port, and the other 250GB and 400GB drives on another. Then the arrays would be split across channels.

There's 1000 different ways it could be setup so that it wouldn't impact performance at all.

Also, please let's not start calling it SATA "3" since there's no such thing as SATA "2" or SATA "II". Calling it SATA3 will only confuse people into thinking "3Gb/s".

ya thats why i was saying 5 drives. nontheless there would eventually be a certain number of disk that would cause a decrease in performance. ya that stupid SATA debate, what should i say then? that the ports max transfer speed is 6GBs? it really has nothing to do with the disks.

[nmX.Memnoch]

300MB/s Isn´t good enough for 2 drives at full burst speed, okay, not that you have it a lot, and yes, for just storage it will be okay.

You do know that Ultra320 SCSI is 320MB/s per channel and not per drive...right?

The introduction of SAS made it 300MB/s per drive.

ya that stupid SATA debate, what should i say then? that the ports max transfer speed is 6GBs? it really has nothing to do with the disks.

It's 6Gb/s...bits, not bytes. And it has everything to do with the drives. The actual throughput comes from the drives. The controller provides the bandwidth for the drives. But the drives, being mechanical in nature, simply cannot spin the platters or move the heads around fast enough to read/write the data fast enough to fill the available bandwidth. That's why the other technologies like NCQ and massive cache buffers are there to help with the read/write speeds. It's also why you see faster spindle speeds and larger cache sizes give better performance.

[ripken204]

no, i know the drives can't fill the available bandwidth. but is'nt it that the controllers max available bandwidth is 6Gb/s?

[puntoMX]

The introduction of SAS made it 300MB/s per drive.

SAS indeed gives 150/300 and even 600MB/s for each device. If iNTEL does it right it should be 300MB/s for each device as well if they do it like SAS witch is very simular to SATA. But they use port multipliers, and that´s in my eyes a downgrade to your harddrives (not optical drives).

UW SCSI 320MB/s needed to replaced, altrough they came up with 640MB/s it didn´t give enough troughput to each drive so they went with SAS. Now they want to use SATA as PATA again, and using port multipliers, so "multidrop" like and not point-to-point. This will give more signal clutter and lowers the performance of a drive, eg 2 drive on each line vs 2 drives on the same line will be won by the 2 drive on each line.

But this is just theoretical, I think we have to see the benchmarks/tests first next year .

[puntoMX]

no, i know the drives can't fill the available bandwidth. but is'nt it that the controllers max available bandwidth is 6Gb/s?

No, it´s still 300MB/s, and in your words 3Gbps

[ripken204]

no, i know the drives can't fill the available bandwidth. but is'nt it that the controllers max available bandwidth is 6Gb/s?

No, it´s still 300MB/s, and in your words 3Gbps

sata 3, not 2