[q][b]Forge said:[/b]
No 200 FSB with 33.333MHz PCI on KT400/KT400A.[/q]well, at least not that we've SEEN. it might be possible, later, with a BIOS update. :)

LiamC - The 'K8' PCI-X bridge actually makes perfect sense, as there are 'K8' 8111/8131/8151 chipsets in large availability... And the Hypertransport bridges on nForce/nForce2/nForceFX are close enough as to make using the AMD Hypertransport chipsets easy.Butch up the northbridge side of the link, connect SPP ->8131 -> MCP, viola.

Besides, high end SMP users are getting PCI-X on the Intel side, it makes sense to have it on the AMD side, too.

I hope it's real, I expect it's not.

AG #64 - Nope. Via apparently still puts PCI on the southbridge, and the NB -> SB link is V-Link.

AG #66 - 1/3, 1/4, 1/5. 100, 133, 166 FSB, DDR 200/266/333. No 200 FSB with 33.333MHz PCI on KT400/KT400A.

what is the max divide for the PCI and AGP bus on this chipset?

Originally Posted by Just Brew It!
[q]hmm isnt the northbridge responsible for the dividers?[/q]
IIRC the K8 architecture still requires an external chip to handle PCI... so that chip would play the role of the northbridge, as far as PCI bus is concerned.

hmm isnt the northbridge responsible for the dividers?

JBI, I'll bite on why no one has jumped on the dual Socket A bandwagon.

ROI. It is an order of magnitude harder to debug SMP chipsets -> drives up cost. The engineering effort required is also significantly larger -> driving up costs.
Support issues are greater.
The volume is small, which means that the cost of each chipset is _much_ greater as there is less volume to amortise costs.

The Taiwanese are noted for cheap volume production - nothing wrong with that BTW - but this means they have a stigma to overcome _before_ they can begin to sell an enterprise class chipset. Already the business case is looking very shaky.

Then there is AMD's perceived lack of interest in supporting the Socket A platform -> we should all be buying/running Hammer's by now. It is only AMD's stumble with K8 that has created this perceived niche. Also, the Hammer architecture overcomes a lot of the difficulties in designing AMD SMP systems - so why go to all the trouble?

As for the NForce FX -> what was a K8 PCI-X chip doing on such a board? - hoax methinks.

Don't get me wrong, I'd love a dually - I have enough frames, I have enough speed for almost anything I'd do, so those frequent pauses while the system does it's thing drives me spare - and this is something that a dual-proc rig _will_ cure.

How in God's name did you come up with these Sandra mem scores? I'm using PC2100 memory on 333 with the slowest timing settings and I get 2553/2419. Surely PC3200 from Corsair can do better than that. Did you pluf the two modules into the wrong slots???

I, too, have a very stable KT133A. It's sitting on a monitor these days, though. It got a little picky about the PCI slots, won't detect anything in certain ones.

Hey Damage, im gonna go dig in that can of worms a bit here...
[q](An exception: the KT133A chipset was a steaming pile of crap.) [/q] Of course YMMV but i have a nice asus KT133A based board, and as you said for your own 266...... it never hangs, it never bluescreens, only times ill reboot is upgrading drivers, or poweroutages(which happen at 1 a year if even that).... and when i go away for more then 2 days.

And thats a Asus A7V133-C ..... with SB Live, gf2ti(at 280/480) + 2 matrox pci cards. Running W2K. XP1800+. Uptime when running 3DMark and Prime at the same time was a week before i got tired of it. And i still run at optimized memory timings.

Although i can agree the KT133(non A) was a steaming pile of crap.. ;)

No lockable PCI on Via this go-round. Maybe on KT8 stuff.

If this theoretical nForce FX SMP board exists, it would be singularly ideal.

Athlons use independant FSBs for each CPU.
nForce motherboards use dual DDR channels.
2 CPU FSB == 2 RAM channels.

I know I'd be in the market for at least two boards, especially without onboard video, as I wouldn't want the system memory cannibalized on my workstation, and my server barely merits the PCI Radeon 7000 that's in it. Command lines aren't hard to draw, and dual CPUs will need every last KB/s nForce FX SMP could provide.

Hmmm... nForce SMP FX. nForce sMPfX, nForce MPX. Heh.

based on the comments it looks like KT400A won't be able to lock the PCI and AGP bus right? i was hoping otherwise :(

Originally Posted by Just Brew It!
[q]Just Brew it, the idea of the northbridge based L3 would be to hide memory latency, and when it\'s chucking data around in large quantities I think 512KB would be ample..[/q]
Provided it is an \"exclusive\" cache, yes that should help. (Aside: Is an \"exclusive\" cache really a cache in the traditional sense, or more like a gigantic posted-write buffer?)

[q]Will the Nforce \"FX\" have SMP support or is that just someones \"wish list\" ?[/q]
I think it is only a \"wish list\" sort of thing... at least for now.

[q]Personally, I couldn\'t give a damn for SMP support until MS figures out a way to make it work on non multithreaded applications via the OS[/q]
Not gonna happen.

[q]OR all (and I mean all) applications use it.[/q]
Well... I think most compute-intensive applications that would get a significant benefit from it will support it eventually, if only because optimizing for Intel\'s Hyperthreading will also have the side effect of making applications SMP-aware.

Will the Nforce "FX" have SMP support or is that just someones "wish list" ?

Personally, I couldn't give a damn for SMP support until MS figures out a way to make it work on non multithreaded applications via the OS, OR all (and I mean all) applications use it.

Sure I multi task, but rarely do I multi task with something in the background I *NEED* churning.

It's just not cost effective for me.... some of you sure - and good for you too :) but ... for most people - no use.

Now nforce fx onboard video might not be TOO bad, and some kind of hacked up method of running the CPU @ 400mhz+ FSB or even with dual channel - instantly doubling the CPU FSB? (dual channel 400mhz ram = 800mhz cpu - that I could deal with)

Just Brew it, the idea of the northbridge based L3 would be to hide memory latency, and when it's chucking data around in large quantities I think 512KB would be ample..

yeah, but if you go from 2200+ / SiS 735 to 2600+ / Nforce2, then the difference between the processors will be amplified by the support hardware.

(I just went from a 1700+ / SiS 735 to a 2400+ / Nforce2, and the difference is HUGE, much bigger than just swapping the processor)

Originally Posted by Just Brew It!
[qimme a pair of Bartons, a 512KB L3 embedded in the northbridge, and a memory subsystem that can keep both processors fed with data, and I\'m SO THERE.[/q]
512KB of L3 wouldn\'t do you much good... that\'s only half the size of the combined L2 caches of your pair of Bartons. A couple of MB oughta do it though... hey nVidia, are you listening? :D

Originally Posted by Just Brew It!
[q]As for \"20%\" not being much, how big is the difference in performance between a 2200+ and a 2600+ ?[/q]
For most people, not big enough to justify the upgrade, if you\'ve already got a 2200+!

Just Brew it,

Good question, I want my Nforce FX >:(

gimme a pair of Bartons, a 512KB L3 embedded in the northbridge, and a memory subsystem that can keep both processors fed with data, and I'm SO THERE.

Just Brew It!
And I still stand by my claim that the 760(MPX) is the only suitable chipset for AMD-based mission-critical servers, since it is the only Socket A chipset with working ECC support.

It's also the only Socket A chipset with SMP support, which is a big thing in servers apparently :)

As for "20%" not being much, how big is the difference in performance between a 2200+ and a 2600+ ?

Well well well another via board............

If it wasn't for the Nforce 2 you guys (and via of course!) would have me almost 1/3 convinced Via is a reliable solution to a chipset need now.

However, with the Nforce 2 out, true dolby support and the fact that the via KT 133 / 133a was a complete and utter abomination AND via handled it badly, I can be confident in saying "no thanks"

There's less and less via "haters" out there like myself and I must say even I dislike via less than I did in the previous 12 months - but none the less I've been EXTREMELY bitten by them and their foul chipsets of a past, to add to this after actually experiencing the SIS 735 (ECS K7S5A) (WOW) (for a cheapie) I can say no no no to via.

Now if VIA was faster than the nforce in ALL benchmarks, had SATA off the bat in the chipset , USB 2 / firewire / AND the dolby digital audio licensed from Dolby hell I'd be almost 2/3 convinced - but the Nforce 2 has me (big time)

I intend to build a GOOD machine shortly - due to the video card industry being topsy turvy and no price drops on Radeon 9700's (yet) I'll be looking at using an Epox 8RGA+ (with onboard video) and using dual channel DDR for plenty big speed! - should make a great temporary machine for 800x600x16bit medium detail for the next 2 months.

I have no doubt some via lovers will be in this little comments thread and maybe some hated replies, but really - just before you flame me, bear in mind some of us had to "beta test" via while you guys were still in diapers ;) - no doubt some of you STARTED on the KT266A so no wonder you claim "via is fine" but some of us know otherwise.

Originally Posted by Just Brew It!
At the risk of taking this discussion off on yet another tangent... would anyone care to speculate on the reason that [b/b] 3rd party chipset vendors have released a dual-capable Socket A chipset? Surely they could make up for the lower volume by charging a significant price premium. And now that most PC users have finally been weaned off of Win9x/ME, the (lack of) SMP support in the OS has ceased to be an issue.

I think nVidia (at least) has demonstrated that they have the expertise to pull something like this off. Why the lack of interest?

Originally Posted by Just Brew It!
[q]The MPX is a very different animal from the old 760.[/q]
Well, obviously the dual CPU support and 64-bit PCI are important differences... but I was under the impression that the memory controller and AGP interface were essentially carried over intact from the 760. They also share the same synchronous FSB/memory architecture, and neither one supports FSB speeds over 266MHz (at least, not officially... apparently the IWill MPX2 is a pretty impressive overclocker).

Can you elaborate further on the \"very different animal\" bit? (I\'m not trying to be argumentative, I\'m genuinely curious.)

The MPX is a very different animal from the old 760. ;)

Originally Posted by Just Brew It!
Damage --

Well, I suppose I [b]am[/b] a bit biased against VIA... but not without at least some justification. I certainly had my share of run-ins with various mobos based on the \"steaming pile of crap\" KT133(A). About a year ago, I decided to give the KT266A a chance, to see if they\'d gotten things straightened out; that system build was problematic as well. So at that point, I simply swore off of VIA chipsets -- quite frankly, I was tired of banging my head against the wall.

I\'m not trying to claim that the 760 is the all-around best platform for everyone -- of course it isn\'t. AMD has neglected it, and doesn\'t even produce the non-MPX flavor any more. But I guess what kinda got me going was the implication (by AG #16) that the only meaningful measure of the quality of a chipset is framerate in game benchmarks.

And I still stand by my claim that the 760(MPX) is the only suitable chipset for AMD-based mission-critical servers, since it is the only Socket A chipset with working ECC support.

#40 (Just Brew It!) -- Ok, I'm going to open up a big can of worms here, but I can't let this go by without comment. Your mention of stability with reference to the 760 chipset and the comment that you almost never have to reboot seems a little off to me. I've been running a KT266A in my main system for months on end, and I nearly never reboot. It does not lock up or crash. Only power outages, hardware upgrades, and new video drivers prompt a reboot for me.

I'm not saying this to proclaim the unique virtues of the KT266A. The KT266A is simply a reasonably competent PC platform, and many other chipsets offer similar stability. But there seems to be some common perception that PCs crash pretty regularly unless they have The Very Best Stuff inside (and that VIA stuff doesn't qualify for such status). That's just not so.

(An exception: the KT133A chipset was a steaming pile of crap.)

Your bold proclamation of the AMD 760's superiority sounds just a little self-centered to me. I'm pretty sure the hardware you happen to own isn't the best of its kind anymore. Not that there's anything wrong with that.

Originally Posted by Just Brew It!
#41, don\'t get me wrong... I\'m mightily impressed with the nForce2. May even get one at some point. However:

- I don\'t think I\'d describe a ~20% performance delta in games as \"screams its way past the other\". 20% is generally not going to mean the difference between unplayable and playable framerates.

- For people with mission-critical servers, the AMD 760 (or 760MPX for duals) is [btil/b] the only Socket A chipset that supports ECC memory properly. AFAIK the 760MPX is also the only Socket A chipset that supports 64-bit/66MHz PCI slots. AMD\'s stated goal of making inroads into the server market is being hurt by lack of suitable chipsets.

Just Brew It, that's the thing.

the Nforce2 _does_ work (some foibles with unperfect memory aside, and you get those with any chipset to some degree, the Nforce2 is just picky on ram) and it goes all the way to 11

given the choice two stable platforms, one of which screams its way past the other performance wise, which do you chose?

Do you then look at overall featureset when balancing the decision?, 'cause the Nforce2 wins there as well :)

Originally Posted by Just Brew It!
#16, \"best\" does not always mean \"fastest\". IMO this obsession that many computer users have with speed is partly to blame for all the crap hardware and compatibility issues we have these days. Manufacturers are being pushed into considering performance and price above all else, with stability and reliability running a distant 3rd and 4th.

I\'m with #15 -- My main rig is running on an AMD 760 chipset... with ECC memory. It [beve/b] crashes (except when I do something stupid, like trying out the latest leaked \"beta\" nVidia Detonators).

AMD\'s chipsets aren\'t the fastest, and never have been (the 750 was a mediocre performer as well). But one thing AMD [boe/b] know how to do is make a stable Socket A chipset that supports ECC memory properly, and is capable of running 24x7 for months on end without so much as a hiccup. I\'m willing to trade off a little performance for that kind of stability.

eitje - Whoops, forgot that.

Well, if the VT8235 is doing the PCI, then there is definately NOT 1/6th PCI or async PCI, so 200 FSB with PCI in spec is not possible.

Guess we'll have to wait for VT8237 and hope it has 1/6 and maybe even 1/7 PCI dividers.