Via Chipset/GeForce4?

[Gunthr]

I'm running a Shuttle AK31 Rev2/AMD 1800XP, and I'd like to get a GeForce4 Video card, but I'm wondering if I would have the problems you hear so much about.

Also, I can't figure why reportedly some via chipsets seem to work ok with these boards. I'm confused.

I've considered going around the widespread via/nvidia problem altogether by buying an nForce 2 mainboard, but I'm not thrilled by the few choices with the nForce chipset.


I'm also wondering if an ATI video card might be less likely to choke on my Via chipset...

Right now l'm leaning towards waiting for the ATI 9700 to come out and just take a chance on it.

Will the problems with Via ever be resolved?

Any suggestions on what you would do given my situation? Is there reason for me to jump AMD ship and switch to P4?

[SKurj]

i'm runnin a fake g4... gf4 mx with a via KT133A with no probs whatsoever

MSI K7T Turbo 2


SKurj

[Gunthr]

thats good to hear Skurj, although I've got the 266A. Its a good sign.

Even the MX version Geforce4 must kick butt...

[SKurj]

100+ fps in some situations Gunthr
@1024x768x32bpp

SKurj

[bloom25]

There isn't an issue between VIA and the GF 4.  The problem is with one particular Gigabyte board which is popular because of good reviews on Tom's Hardware.  The issue on that board relates to AGP voltage levels, and aren't VIA chipset related.  I haven't looked, but I'll bet the Gigabyte Intel boards (845E) might have the same problem as well.

I'm using the Asus A7N266-C board (nForce 415 chipset) and it's running great.

[Gunthr]

nice Skurj. thats enough fps for anybody I would think...

If any body has a Via 266a chipset using a GeForce4, please advise which card...         thanks

[Gunthr]

Thanks Bloom, that relieves a lot of anxiety. I've got a local seller of GF4's within 5 miles of here. The cash is starting to warm up in my pocket...

[bloom25]

One thing, do make sure you've got at LEAST a 300W power supply.  The MSI G4Ti4200 64 MB card is a good deal at under $190 retail.  I've got mine overclocked to just under Ti 4600 speeds and got it for $159.

[Eagler]

this board:
http://usa.asus.com/mb/socketa/a7v266-e/overview.htm

with this video card:
http://www.pny.com/home/products/Vcard_ti4400.cfm

with this cpu:
http://athlonxp.amd.com/pressInformation/2000Release.jsp

and two 256 PC2100 sticks of this:
http://www.crucial.com/store/listparts.asp?Mfr%2BProductline=ASUS%2BMotherboards&model=A7V266-E&submit=Go

with win98 on one of these;
http://www.wdc.com/products/products.asp?DriveID=11

full tower with 300 watt AMD PS

pushes the fps to a steady 85, 19" monitor res max running 32 bit at 1280 x 1024 <-Vsync on

[AKWarp]

VIA KT333 chipset and GF4 TI4600 here and zero problems.

The early VIA KT266 had a lot of problems, primarily with PCI latency and lack of enough power output to USB devices.  The next revision of that chipset, the KT266a fixed a lot of the problems and made some enhancements.  The KT266a is a pretty good chipset still.  Their current crop of KT333 seems to have solved all of VIA's problems and is a kickbutt chipset.  

With DDR memory recently released in the PC3500 flavor, the real bottleneck on AMD based systems is the CPU's FSB (still at a paltry 166mhz native).  The DDR memory is getting faster to try to fill the void that rambus memory left for pentium owners.  It would make good sense for AMD to follow up with a significant FSB speed increase in order to make use of the newer, faster DDR memory (and of course, to once again trounce pentium systems..muhahahaha!!!).

[bloom25]

AKWarp, don't fall into the trap that a lot of reviewers have these days.  The current Athlons use a 133 Mhz DDR FSB (front side bus) which has the equivilant bandwidth of a 266 MHz FSB.  Moving to a 166 MHz DDR (333 MHz equivilant) FSB will certainly improve performance, but how much so depends a lot on how the CPU itself is designed.  Should future Athlon XPs use a 166 Mhz FSB, they will be incompatible with current Socket A boards and chipsets.  (Just like the 533 Mhz FSB P4s.)  A 166 MHz FSB would also require the use of DDR 333 memory to avoid the additional latencies imposed by running the memory at a different frequency than the CPU.  While increasing bandwidth by moving to 333 Mhz DDR memory will improve performance somewhat, the Athlon design is more dependant on memory latency.  In contrast the P4 requires high bandwidth for top performance.

I could go on and on with this, but suffice it to say that AMD knows this.  If you look at the K8 (Hammer) you will notice that the memory controller has been moved on the CPU core itself, which reduces latency.  This is the main reason for Hammer's performance increase over K7 (Athlon).  The smartest thing for AMD to do with the current Athlon XP design is to maintain compatibility with the current Socket A chipsets and motherboards, rather than force the use of a 166 MHz FSB supporting chipset so close to the release of Hammer.  The Athlon XP would actually benefit more from a move to 512 KB L2 cache, rather than increasing the FSB.  This also maintains compatibility with current boards.  If you look at the Throughbred die photos you will notice that the L2 cache was moved and reorganized to allow for it's expansion.  I'd expect the next Athlon XP core (codenamed Barton) to have a 512 KB cache, but still run at 133 Mhz FSB.  After Hammer's release AMD may move to a 166 MHz FSB, but I don't expect that to happen anytime soon.

[jconradh]

Bloom25, a couple of things:  

1) The P4 requires high frequency; RAMBUS is low bandwidth compared to slower DDR, yet the RDRAM is optimal memory for the P4 currently.

2) You are pretty much spot on about AMD.  W/O blowing NDA I would say don't look for a 166MHz core fsb anytime soon.

Jeff

[bloom25]

jconradh, you are mistaken here: Rambus PC800 has a bandwidth of 3.2 GB/s.  PC2100 (133 MHz DDR -> 266 MHz equiv.) has a bandwidth of 2.1GB/s.  Rambus memory most certainly has greater bandwidth than DDR SDRAM.  Where Rambus suffers is latency.  Without a doubt Rambus memory is the preferred memory type for the P4.  Now Granite Bay (upcoming Intel chipset) may change that, as it supports dual channel DDR PC2100 ram, giving 4.2 GB/s memory bandwidth.  This is the same as the recently released PC1066 Rambus modules.  

How to calculate memory bandwidth:

Rambus PC800 memory runs at a clockspeed of 800 MHz, and transfers 16 bit wide data.  Current Rambus setups utilize a dual channel setup, yielding a 32 bit wide data path.  (32 bits -> 4 bytes).  Multiply 800 MHz by 4 bytes = 3200 MBytes/second -> 3.2 GB/s.

PC2100 DDR SDRAM runs at a clockspeed of 133 MHz but transfers data on a 64 bit wide bus (8 bytes) and is both rising and falling edge triggered.  This is what gives it the 266 MHz equivilancy.  266 MHz * 8 bytes = 2128 MBytes/second -> 2.128 GB/s memory bandwidth.

BTW:  Which site do you write for?  I'm sure we could have some interesting discussions.  (I have a bachelors degree in Electrical Engineering in this area. )

Edit:  I must be slipping, PC2100 runs at 133.33 MHz which would thus yield 2.133 GB/s theoretical maximum bandwidth.

[AKWarp]

Ooops, my bad bloom, I meant 133mhz FSB, not 166mhz.  

AMD should up the FSB's native to somethihg on the order of 266 or more.  Yes, it would require a new mobo, but really, when you upgrade your system every couple of years or so, you end up buying a new mobo anyway.  Mobo's aren't that expensive.

I would be more than willing to dump cash on a new mobo if AMD would come out with a super fast FSB in keeping with the newer, faster DDR memory.  I just hate paying through the nose every year or so for slight incremental increases in power and speed.  I want a BIG jump!!!!!    

I really think we're going to see AMD go in that direction...they really have no choice.  The FSB is the current bottleneck for both AMD and, to some extent Intel.  

The Hammer is a good move with the on chip memory controller, but it's not the complete answer.

Of course, if IBM delivers on its new magnetic memory technology in the next year or so, then who knows what will happen.........