Re: Linux VIA Chipsets

[Bryan J. Smith <b.j.smith@ieee.org>]

Definitely _incorrect_ on Intel.
They've had their share - especially on the i865.
The i865 is just the i875 that tests to lower tolerances.
In fact, the Intel GTL platform (from Socket-8 to LGA-775) was never designed for more than >4GiB.
But that's another story.

For the most part, the chipset itself isn't an issue.
All conform to Intel APIC which acts like Intel's GTL for the PC platfrom
(even Athlon's EV6 and Hypertransport, which has no legacy PC lines).
There have been issues with some AGPgarts (n/a to Athlon, which puts it on-CPU, so the problem moves there).
And some have have buggy PCI support, which results in bridge/I2C and PCI DMA issues.

The problems people have are with on-chipset _peripherals_.
There are many reasons, but in a nutshell, variants in the "southbridge/Legacy PC/peripheral" logic seem to creep in.
This is especially the case when it comes to ATA, NIC, Audio and other logic.
Even if it is based on an existing peripheral solution with a driver, it never seems to be exactly the same.

ALi, AMD, Intel, nVidia, SiS, ViA - there are _no_ exceptions.
Anytime a new logic is introduced, there are issues with compatibility.
E.g., when Intel switched from its established PIIX to ICH, all sorts of incompatibility arose.
Same deal with newer ICH5+ southbridges.
ViA's transition from the 686 to 82xx did the same, and one has to watch out for newer ViA southbridges as well.
AMD has been fairly solid since the 751, as the newer 761, 766, 768 and 8111 all use the same logic,
but at the same time, AMD doesn't put much on-chipset.
The newer the logic, the less likely a driver will work, or work reliably.

Heck, even the brand new nForce4 looks to be pretty good as of kernel 2.4.23+/2.6.5+.
NVidia has contributed the required GPL code to the i810, OHCI/EHCI, 1394, etc... for various components.
It's largely due to the fact that the nForce4 uses the same logic as the nForce3 did that the nForce2 first shipped with.
It sucked when the nForce2 first came out, but years later the GPL support is very good.
In fact, the only thing that is proprietary in the nForce is the NIC (nvnet),
although the GPL driver (forcedeth) is good for 10/100 in 2.4.23+/2.6.5+
(the new 1000 speed is new to the newer nForce3/4 and not reliable yet).

In reality, the new v. mature argument isn't limited to Linux.
Many mainboard manufacturers these days are shipping with _both_ an on-chipset and off-chipset ATA and NIC,
because it's likely at least one will work well.
Especially when a slight chipset revision outside of the control of the mainboard manufacturer causes disk or network to be inaccessible,
the two most crucial peripherals.
The mainboard manufacturer has more quantity and control over ATA/NIC ICs the can add for peanuts to any board, than they do with the often model-specific chipset.

Which brings me to "brand loyalty."
It's a joke to even try, you have to take products on a per-_model_ and per-_revision_ basis.
Because outside of packaging, marketing and servicing warranties,
many products from dozens of vendors come from the same 4-5 factories.
And vendors change their manufacturers _regularly_.

I wouldn't have bought a nForce2 chipset years ago due to immaturity on Linux,
but I'm eyeing several nForce2 (Athlon XP), nForce4 and even one nForce4+AMD8131 chipset mainboards now.
Especially for Athlon64/Opteron, where the chipset is nothing more than a peripheral interconnect,
because memory, CPUs and I/O all use different "front-side busses."
Mainboard designers can mix and match HyperTransport components to get  what they want in a A64/Opteron system.

In fact, I wouldn't buy anything Intel _except_ Socket-604 with a 7500 (ServerWorks) series chipset.
Digital Semiconductor designed nearly all current PC interconnects,
including the GTL to PCI bridge, various other bridges, memory logic,  etc...
And most of those guys now work at API Networks, fka Alpha Processor Inc., now a R&D arm of AMD.


-- 
Bryan J. Smith (currently mobile)
b.j.smith@ieee.org

-----Original Message-----
From:  NZG 
Date:  05-1-5 11:01
To:  silug-discuss@silug.org
Subj:  Linux VIA Chipsets

> Intel doesn't make chipsets for AMD processors.  :-)
O, yea guess they wouldn't, lol
>
> I don't think VIA has been a problem for almost 5 years.  (I know it
> has been >3 anyway.)
I'll have to pass that on, lot of VIA prejudice still floating around here, 

In general though, you would say that an Intel processor and chipset is going 
to have better general Linux support than a AMD/VIA type motherboard though 
right?

Or is that too broad a statement?

NZG.

-
To unsubscribe, send email to majordomo@silug.org with
"unsubscribe silug-discuss" in the body.

-
To unsubscribe, send email to majordomo@silug.org with
"unsubscribe silug-discuss" in the body.