We all knew VIA has had a bad rap, but Damn...

[Dr Mordrid]

Exactly.

To my knowledge the nForce hasn't been tested with the RT.X100 and I'm on the beta team. It'll be interesting.

This card is a real test for this kind of thing since the RT.X100 itself can easily push 80+ mb/s with a deep effects stack, and that's not counting any other hardware being active.

The SiS 735/745 chipped boards I've used didn't even twitch. With the MuTIOL bandwidth being doubled to 1066 mb/s for the SiS 648/658/746 they should be even more solid.

Dr. Mordrid

[Ryu Connor]

Given your description I could already see why the VIA chipsets were inadequate. The crux of their PCI problems were due to a broken PCI arbiter.

[Dr Mordrid]

Yup, and it's still broken so they fell flat on their arses. Realtime exports to HDD or IEEE-1394 are impossible and stability is...well....

Even using an AMD 760/761 with a VIA southbridge (686b) on this GA-7DX+ things are a bit twitchy when I test the RT.X100 with it, which is why most of the time it lives in the K7S6A.

Just waiting now for SiS 746 board to hit the streets

Dr. Mordrid

[Ryu Connor]

Even using an AMD 760/761 with a VIA southbridge (686b) on this GA-7DX+ things are a bit twitchy

The 686B was a poor pairing for that Northbridge.

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/24952.pdf

Actually that southbridge in general is piss poor. I'm still surprised that ATI actually decided to use it along side their IGP320.

The manufacturers really managed to short change everyone by doing that.

[K6-III]

Originally posted by Ryu Connor


Actually. I find the coming Asus A7N8X infinitely more interesting.

http://www.tbreak.com/forums/showthread.php?postid=105654

I will be shifting over to it upon release.

You'll never get Epox's overclocking options and support from an Asus.

IE, you can actually talk to an Epox Tech at www.aoaforums.com

[Schorsch]

Originally posted by Dr Mordrid

The SiS 735/745 chipped boards I've used didn't even twitch. With the MuTIOL bandwidth being doubled to 1066 mb/s for the SiS 648/658/746 they should be even more solid.

Dr. Mordrid

The SiS 735/745 chipsets even have 1.2 GB/s of MuTIOL bandwidth so the SiS746 probably won't be an improvement regarding PCI bandwidth. However, the first SiS648 reviews indicate that SiS seems to have improved its memory controllers so we might see better overall performance with the SiS746 as well.

If nVidia manages to their improve their PCI bandwidth with the nForce 2, it should be an excellent chipset, too (since that seems to be the only serious problem with the nForce so far).

[Greebe]

I dissagree Schorsch. The seperate 963 southbrige that will be mated with the 746 northbridge will have an improved PCI bus arbitor doubling it's bandwidth. They have already proven bandwidth at 533. Given this, moving to 1066 will no doubt have it's benifits too. SiS is on the right track. I'll be staying with them for the time being because they are headache free, work great and are extremely cost effective.

............

Even if nV's chipsetted products work to the degree that is neccessary of such high bandwidth demands I do believe Matrox would be hard pressed to state that their products are compatable given the history M vs nV have (nV's bad). I would expect nothing mentioned at all would be most likely Matrox's feelings on this issue.

If nVidia wants to work with manufactures they will have to stop their corporate sabotage/technology theft (not to mention all their other shifty business practices).

[Marshmallowman]

I think nvidia missed the boat somewhat.
They have dual bank board , but you can't use to banks of cheap(slow) ram

..for example

166 bus and two pc1600 dimms, asychronous

200 ddr and two pc1600 dimms(a bit of an extreme long shot ) but if you can pull 3.2 Gb/s of synchronous bandwisth with CAS 2 parts your woul be flying.

eg..use the twin bank to allow for reliable low latency "cheap" memory instead of expensive ddr333 or ddr400.


instead you have 3.2 GB/s of bandwidth that is only partially useable by the Integrated video and some busmastering DMA on PCI cards

[lecter]

Marhsmallowman, i think that the Nforce to buy is 415D, not 420D with integrated video...

[Ryu Connor]

A few points here.

MuTIOL does not increase the PCI bus bandwidth (in the traditional sense any ways). It's still a 32bit @ 33MHz bus with 133/MBs worth of bandwidth. All MuTIOL does is provide for all the integrated material to be removed from the PCI bus. So you can use the onboard USB and PCI RAID card at the same time and the USB not impact the RAID cards PCI peformance.

Intel's PCI bus is hella fast as well and Intel chipsets hubbed connection only provides 266MB/s of bandwidth in total. So one doesn't need that huge northbridge to southbridge pipe to be successfull. You just need a good PCI bus implementation. If you hammer the hell out of integrated stuff and the PCI bus at the same time though, MuTIOL and the like will be a big help.

SiS PCI performance will likely remain the same. They don't tend to change the design of that bus unless there is a problem.

nForce and nForce2 do not require the use of DDR 333 or DDR 400. nForce offered all the necessary divider adjustments to run faster RAM slower processor or faster processor slower RAM. 100/100, 100/133, 133/100, 133/133. nForce2 will be no different. So using slower RAM will remain an option.

The bandwidth of Twinbank is used by more than just the integrated video. The AGP slot and MCP southbridge through it's hypertransport connection have full acess to it. It's design allows for one bank to be feeding the processor fully while the other bank manages to deal with heavy PCI transfers/integrated material demands.

[Dr Mordrid]

If you hammer the hell out of integrated stuff and the PCI bus at the same time though, MuTIOL and the like will be a big help.

Very easy to do, especially since all the RT.x100's effects are resident on the C:\ drive. This means you have the primary master IDE, often a RAID and/or a drive on the secondary IDE ('tis often good to put audio and video on separate HDD's), the RT.X100, a sound card (or integrated audio) and whatever other integrated stuff you're using all going at once.

It's pure sadistic fun to watch a VIA try to do this

Dr. Mordrid

[Technoid]

It would on the other hand be easy for sis to make it so that each PCI "port" has a full 133mb/s at the same time!

Wich would mean that it wont matter if you try to hammer a SIS board to death with a gigabitNIC, raid and a RT.X100, neither of these would run into bandwith problem since they would have a "reserved" bandwith!

I815 (just as an example , because I own a board with that chipset ) has two PCI busses with one for the ports and one for the integrated stuff!
Thats why it sais it has 266mb/S

[Marshmallowman]

yes lecter 415 would be the older nvidia chipset to buy,
(if you really wanted to)

BUT 4.2G of memory bandwidth does not fit through a 2.1g bus, so the bandwidth is wasted unles you use the integrated video...which is not a candidate for me.
(I was thinking more about the newer nvidia chipsets eg..ddr 400)

but they had a great optunity to use 2 lower speed memory dimms to get the same performance as a single high speed (expensive) memory/bus...ala some server-works chipsets and the new intel P4 ddr server chipsets.(prestonia?)

[Dr Mordrid]

Technoid;

To resolve the PCI-133 bandwidth bottleneck, 3 technologies are utilized in SiS961 (and 962 and 963), namely multiple DMA buses, Multi-threaded I/O Link, and MuTIOL Connect.

Instead of connecting all the I/O bus masters to the PCI bus, each integrated I/O bus master as well as the PCI master clusters is offered a dedicated DMA bus with separate address bus, input data bus, and output data bus that features pipeline & split transaction.

The integrated Multi-threaded I/O Link (MuTIOL) further buffers and manages these multiple DMA buses to ensure concurreny of multiple upstream and downstream data transfers.

Sounds like they went out of their way to keep devices from stepping on each other, and with 1066 mb/s woth of bandwidth in the SiS963 it'll take a lot to saturate the new SiS based boards.

Dr. Mordrid

[K6-III]

You could theoretically give each PCI slot its own bus and connect them with HyperTransport. I think we'll see this eventually with HyperTransport connecting multiple PCI Express busses...