Voodoo eXtreme -- Logo by |2| (2@2design.org)


Guillemot Cougar
EMS PC133 HSDRAM
Montego II Quadzilla
Monsoon MM-1000's
i-Art Eye3D glasses
3dfx Voodoo3 3500TV
AMD Athlon 600Mhz
MSI MS-6163 Pro
P3-600 Quickie
Asus P2B-F
Kryotech Renegade
Wicked3D eyeSCREAM
3dfx Voodoo3
EverGlide


Reference driver list
S3D tweaked drivers
Stealth 3 Xtreme drv.
Stealth 3 Xtreme ICD
Asus 3100 drivers
Athlon mobo drivers
ALi IDE drivers
Spectra 2500 drivers
V3 Overclocker 3.1
ATI Rage 128 drivers
Wicked3D drivers
WickedGL download
GLSetup 1.0.0.106
CL's Unified Driver


Savage2k+ preview
Quadro preview
Filtering methods
AA Explained
Glaze3D preview
Lighting article
ATI Rage Fury MAXX
Making T&L; Work
3dfx FXT
TNT2 vs. M64
The Hardware Pimps
Asus TNT2
Celeron Cooling

Chipsets Impact
February 21, 2000 - Ga'ash Soffer
   
  I had just became aware, after working with a fast P3 on a VIA chipset, and reading up about the problems with AMD's Irongate chipset that only Intel seems capable of making good chipsets. Ok, it's not that extreme, but Intel DOES make the best chipsets. But now, Intel is killing off the BX chipset in favor of i820, a solid chipset, but one which requires very very very very (very) expensive RAMBUS RAM to work well. This leaves the consumer searching for a fast P3 system in somewhat of a catch-22 situation. Either go with a good chipset, the i820 and pay a premium of around $1000, or pay no premium and go to a poor performing VIA chipset. Even if one opts to ditch the P3 and go for the Athlon, there are still problems; the AMD Irongate chipset has problems with AGP 2x. What it all boils down to is that chipsets are making a larger impact than ever. Unfortunately, it's a bad impact.

Chipsets, Only Intel?

Whether it's because Intel is the one who designed AGP and most of the other protocols used in x86 systems or not; Intel remain the leader in high performance chipsets. All Slot 1 chipsets not released by Intel were and still are   failures in the performance market. For example, the VIA AGP2x chipset (Apollo 693) took the performance of a P3/800 from around 100fps in Quake3/640x480 (w/GeForce DDR) to under 53. With various BIOS hacks, the performance went up to ~67 which is still far from the target of around 100 and the stability compromises of the "hacked" BIOS are serious. In business applications and similar non-intensive computations, the non-Intel chipsets can compete with Intel counterparts but the purpose is lost; high performance systems aren't used to word process.

Ditching the P3, Still Problems

Even if one chooses to go with an Athlon solution over the P3/Coppermine, the only chipset readily available now, the AMD Irongate chipset cannot handle AGP properly. This causes cards to resort to AGP 1x "command mode" which somewhat slows down the card. Realize that AGP 1x "command mode" is NOT AGP 1x, textures and other data is still retrieved at AGP 2x speed. However, even AGP 1x "command mode" still has a problem, and that problem becomes evident when we look at what's wrong with the Irongate chipset.

Essentially, the Irongate chipset has a bug in it which corrupts data/etc. sent through the AGP bus. In the case of things such as texture data, the number of corrupt bytes is very small; in actual gameplay, one or two pixels which are not what they are supposed to be will never be noticed. However, if a graphic command is passed through wrong, the effects can cause even a system crash. For this reason, BIOS fixes for Irongate chipsets play it safe and resort to AGP 1x when sending commands. It is interesting to note that once applications begin to make extensive use of T&L, the sparse data errors can cause hugely noticeable effects, for example, entire polygons may be severely skewed or completely missing if simply one bit is flipped. Hopefully VIA's recently released Athlon chipset will bring about good things for Athlon.

[UPDATE: Irongate DOES NOT support AGP 4x ... Sorry about the confusion ... should be AGP 2x ]

 

And Now We Can't Count On Intel....

So, Intel can make chipsets, but, unfortunately, they have ditched the BX chipset and are now fully promoting RAMBUS, and only RAMBUS, for the high end. Unfortunately, the price premium is way to extreme to even be considered by 99.9% of all mainstream users. This isn't to say i820 is a bad chipset; quite the contrary, i820 benchmarks have shown that it is faster than BX and much faster than non-Intel solutions in CPU intensive graphical applications, a subset of the few types of applications which really need top of the line power. Regardless, the price makes it unattractive and recent attempts at using SDRAM on i820 based solutions have been largely unsuccessful, both in terms of performance and stability. The most interesting part about this whole deal is that Intel is shooting themselves in the foot, temporarily at least. i820 won't be selling huge quantities anytime soon, especially since most systems manufacturers have opted to use much cheaper SDRAM based solutions which use chipsets not manufactured by Intel. OEMs would like to use BX based solutions but the BX is discontinued and getting large quantities of them is impossible. While there are other solutions available from Intel (other than i820), such as i810e, it is ludicrous to suggest running a $700 Coppermine processor on a cheap i810e board intended for low end Celeron processors. The other solution is i840 which absolutely screams in graphic apps but the cost is even crazier than i820. 

And this leaves us where?

The end result is a giant loss for the enthusiast consumer. Newer Coppermine systems are either extremely expensive or hampered in graphical applications by a poor chipset while current Athlon solutions have troubles running boards in AGP 2x mode, and who knows how fast the Athlon would be if Intel made the chipset. Until RAMBUS ram goes down in price or until there is a true quality chipset for Athlon processors (the latest VIA looks like a step in the right direction), we're left with nothing which works right. Maybe in a several months we'll see a reasonably priced platform, which, ummm, works? Let's hope so.

 
     

Copyright 1999, Ga'ash Soffer and Billy Wilson (Voodoo Extreme).  All rights reserved.
This document may not be reproduced without express written consent of the aforementioned parties except for educational, etc. (non-public) use.