Either will do just fine, though GF 4 Ti may have the advantage when it comes to using older drivers, though to this day I still have no need for that. Maybe someone else could elaborate.

Other than Splinter Cell 1 I don't know of any other title with major issues on geforce FX though. Splinter Cell Pandora Tomorrow, on the other hand, may be best played on FX 5900/5950 (I read it has troubles with GF6/7), but this is getting beyond win9x era. These two are the exception rather than the norm so no need to get fussy about it.

If you go FX, best is to get 5900 / 5900 Ultra / 5950 Ultra of course. Also I'd still rather get a 5900 XT than 5700 Ultra. That said, even FX5500/5600/5700 will be just great for most win9x games. Plus good 2D as well for DOS.

archsan wrote:

Either will do just fine, though GF 4 Ti may have the advantage when it comes to using older drivers, though to this day I still have no need for that. Maybe someone else could elaborate.

Other than Splinter Cell 1 I don't know of any other title with major issues on geforce FX though. Splinter Cell Pandora Tomorrow, on the other hand, may be best played on FX 5900/5950 (I read it has troubles with GF6/7), but this is getting beyond win9x era. These two are the exception rather than the norm so no need to get fussy about it.

If you go FX, best is to get 5900 / 5900 Ultra / 5950 Ultra of course. Also I'd still rather get a 5900 XT than 5700 Ultra. That said, even FX5500/5600/5700 will be just great for most win9x games. Plus good 2D as well for DOS.

Agreed with everything. If you already have a working Ti 4600 I'd just leave well enough alone - the 5800/59x0 series can be faster in some situations, but unless you need very high resolution + very high AA levels, it probably won't be noticeable. For a 9x build the Ti 4600 is already well into the land of overkill, and it should do just fine as a result.

Splinter Cell, IME, has issues with the later GeForce FX cards (NV35+), but it isn't unplayable on them - if you disable the advanced shadowing thing it runs just fine. As far as support beyond that I think the GF4 Ti and GF FX should be pretty equal, with the higher-tier FX cards being somewhat faster in certain events.

I'd agree with 5900XT over 5700 or any lesser card, and I'd probably even extend that to vs any FX card given the prices on higher-spec 5900 and 5950s I've seen recently compared to the XT (the Wiki isn't entirely accurate - the XT isn't a crippled core, it just has somewhat slower memory).

Another consideration here is what CPU is this going with? If you're putting this with a K6, Pentium II, low-speed Pentium III, etc even the Ti 4600 will be limited by that more than anything else.

Thanks for the info guys. The CPU is a AMD 1800+ so its 1.53ghz. The ti4600 hasn't given me any issues but you know how it is. Its like putting a V6 in a kids go cart and then being like "hmmm, it could really use a V8"

The 5900XT can often be found for real cheap yes (let them think it's crippled -- it helps!). 😀
I'm actually using one instead of the ultras to save them from wear. Also the Ultras are much rarer where I live.

In my searches it seems like the ti4800 is kinda rare. I hardly ever see them and that partly the reason I grabbed a 4600. The 5900 and 5950 seem more common, at least on eBay but yhea for an old card they can be pricey it seems.

soviet conscript wrote:

In my searches it seems like the ti4800 is kinda rare. I hardly ever see them and that partly the reason I grabbed a 4600. The 5900 and 5950 seem more common, at least on eBay but yhea for an old card they can be pricey it seems.

I have two Quadro versions of the 4800 😀 Nobody wants them, they are much easier to find.

But you're not missing much. It has AGP 8x over the 4600 but the clocks are identical.

soviet conscript wrote:

Thanks for the info guys. The CPU is a AMD 1800+ so its 1.53ghz. The ti4600 hasn't given me any issues but you know how it is. Its like putting a V6 in a kids go cart and then being like "hmmm, it could really use a V8"

The 1800+ should be a good mate for the Ti 4600 - the only limit will be SSE2, but the games that require that will also require features the GF4 (or FX) won't do, or won't do very well, so it isn't a big concern. I wouldn't worry about the 4800s - like Mai1wurf1977 observed, they're identical to the 4600 except for supporting AGP 8x, which does nothing for performance (there's a number of review/comparisons you can find on the web if you're really that interested - honestly it's about as entertaining a topic to test or read about as watching paint dry though).

archsan: Fair enough. I know my 5900XT clocks up to 5900U specs pretty easily, at least on the core, but the RAM doesn't want to come up that much. It does help performance somewhat (most of the later GF-FX cards have (relatively speaking) silly amounts of memory bandwidth - iirc the 5950U isn't far off from the 6800U, despite the 6800U having something like 2-3x the processing power). I'm not sure if this is common for XTs or not - they are all full NV35 cores though. OFC it isn't a huge clock jump - stock is 390-400 depending on which manufacturer, and the 5900U is only up at 450.

Out of curiosity are there even any windows 98 games that use sse2?

I have both a Ti4600 and a Ti4600-8X (aka Ti4800) that I've tested in my ABIT KT7A v1.0 motherboard with an Athlon 1400 Thunderbird and 512mb PC133 RAM running Win98SE.

(Please keep in mind that the KT7A only has AGP 4x)

They both perform identically and can get me ~9000 in 3DMark2001SE on a good day.

I also have an FX 5950 Ultra that I run in my ABIT KT7A v1.3 motherboard with an Athlon XP 2100+ Palomino and 768mb PC133 RAM running Win98SE. Interestingly, it only gets me ~8900 in 3DMark2001SE. I do not know why and I haven't swapped in my Ti4600 cards for comparison. Why? Well...

... as others have mentioned, the Ti4600 and FX 59xx cards are serious overkill for any Win98SE/DX8 system. There isn't a single game that suffers under either of those configurations. So, I just haven't bothered troubleshooting to find out why my supposedly superior system benchmarks lower. Maybe one of these days I'll figure it out.

(As somebody wiser than myself once said: "You don't build the sytem for benchmarking, you build it to play the game.")

Until then, here's my opinion on the Ti4600 vs the FX 59xx cards:

You can find both of these on eBay for under USD$25 if you're very patient. If you're in a hurry, you will pay more for convenience.

The Ti4600 and the FX 59xx both offer very similar levels of performance. The difference is that the FX 59xx card gives you better FSAA options.

(The FX 59xx cards will probably benchmark higher with CPUs and RAM faster than what I run.)
(While I haven't compared an FX 5900 Ultra to my FX 5950 Ultra, I expect the performance difference would be negligible.)

If you're strictly after a high FPS count, then stick with the Ti4600 and the greater flexibility it offers with legacy drivers. If you want better FSAA, then get the FX 59xx card and accept that your driver options are slightly more limited.

If you want the FX 59xx card and run anything less than a P4/Athlon64 ~3ghz, get the 5900 Ultra. It runs a bit cooler and you won't notice a difference in performance anyway. You can probably find it cheaper too.

Note:
Ti4600-8X (aka ti4800) and FX 5900/Ultra requires drivers v45.23 or newer.
FX 5950 Ultra requires drivers v56.64 or newer.

(The earliest version I run nowadays is v45.23, so I don't mind very much.)
(I find v45.23 to be the ideal Win98SE driver.)

Regarding the limited availability of Ti4600-8X/Ti4800 cards:
Don't search eBay for Ti4800 cards. You'll likely only find the silly Ti4800SE, which I myself was scammed by at one time many years ago.
(The Ti4800SE is merely a Ti4400 that can run at 8X AGP. But, why bother?)

Expand your search to include all Ti4600 cards and look at the pictures carefully. Look for "Ti4600-8X" printed on the stickers on the card. That indicates that it's actually a Ti4800. Even though the card's sticker identifies it as being a Ti4600-8X, the NVIDIA drivers will detect it as a Ti4800.

I obtained my Ti4600-8X/Ti4800 by accident. I thought I had purchased a regular ol' Ti4600 with a broken HSF. At the time, I was running v30.82 drivers and it wasn't detected by the system correctly. Upon further troubleshooting and upgrading to v45.23, I found it was detected as a Ti4800. More research revealed that the Ti4800 was usually marketed and sold under the standard Ti4600 moniker and sometimes as "Ti4600-8X". A sticker on the back of my card showed "Ti4600-8X", thus confirming my card as a Ti4800.

Edit:
While I own a copy of Splinter Cell, I've never played it. I have no idea if it would work with the FX cards or not.

Edit edit:
The Ti4600 also has an advantage in that it only occupies a single slot, unless you get crazy with aftermarket coolers. The Evercool VC-RF and its various rebranded clones offer great performance and will only use up a single slot if you don't bother with the ridiculous cover for the HSF.

soviet conscript wrote:

Out of curiosity are there even any windows 98 games that use sse2?

Does Oblivion or Hitman 4 run under 98? Neither game explicitly requires SSE2/3, but I noticed a fairly dramatic improvement with both moving from AthlonXP to a Prescott-based Celeron D (which, otherwise, was about equivalent in terms of performance (e.g. WarCraft III or 3DMark03 or something)). Those were the "earliest" games I remember noticing that with. I think FEAR also benefts. But basically, my point was, it's unlikely you'd come across such a thing with 98, and even if you did, the game would also require DX9 features that the GF4 Ti and GF FX would be inadequate for.

KT7AGuy:

If I remember right, the "Ti 4800" was sold as "Ti 4600-8x" in some parts of the world; it's kind of an odd card in the history of nVidia as the "8x" variants were released mostly as stop-gap products more than anything else. As far as the FX 5900 scoring slightly lower, it's probably a combination of three things:

- The Athlon is a bottleneck.
- Using ancient drivers (for FX cards you should really start with 53.03 and go up from there - the original release drivers with any nVidia card will always offer worse performance/compatibility/etc).
- 3DMark is generally unrealiable/unpredictable and does random nonsense like this from time to time.

In real-world you won't likely notice any significant difference, as you've observed, and the fact that the 5900 allows you to enable more AA without much loss indicates it's got performance left to spare vs the Ti. 😀

The only thing I'd even bother with would be updating the drivers for the FX card; there may be bug-fixes in there that could influence some games (ofc if you never knew the games existed, let alone had bugs, who cares?), and it should improve performance.

As far as Splinter Cell, when I compared FX 5800 to 5900, the 5900 had the "shadows flickering problem" while the 5800 did not. The assumption made in the Splinter Cell thread was that nVidia broke or removed the capability for the ultrashadow effect either in drivers, hardware, or both. The 5900 and later do have hardware differences from the 5800 and kin (it's been theorized that the NV35 and later lack the integer ALU units in order to make room for an additional FPU stage, for example). I did my comparison with 71.89 drivers, and the difference still existed, which either means it's a hardware difference or the game is actually checking device_id strings to enable/support certain features. With the shadowing feature set to "normal" (where there is no chance of it flickering/bugging out), both cards ran the game just fine in terms of things looking fluid.

thanks for all the great info guys. I think I may stick with my 4600 though by chance I did throw out a random best offer for a 5950 ultra last night and it was accepted so I scored one for $20 shipped which i'm thinking is a decent deal either way.

also when you guys say your driver options are more limited with the 5xxx cards does that equal game incompatibility or do the later drivers just offer less option for tinkering with the card settings?

obobskivich wrote:

In real-world you won't likely notice any significant difference, as you've observed, and the fact that the 5900 allows you to enable more AA without much loss indicates it's got performance left to spare vs the Ti. 😀
The only thing I'd even bother with would be updating the drivers for the FX card; there may be bug-fixes in there that could influence some games (ofc if you never knew the games existed, let alone had bugs, who cares?), and it should improve performance.

Thanks for the info regarding the low benchmark. Like you said though, real-world performance is way more than adequate. I've kept my FX5950 running v56.64 for European Air War. I know from the experiences of others that it can get flakey with drivers newer than v61.76.

soviet conscript wrote:

thanks for all the great info guys. I think I may stick with my 4600 though by chance I did throw out a random best offer for a 5950 ultra last night and it was accepted so I scored one for $20 shipped which i'm thinking is a decent deal either way.

also when you guys say your driver options are more limited with the 5xxx cards does that equal game incompatibility or do the later drivers just offer less option for tinkering with the card settings?

$20 for a FX5950 is a very good deal, IMO.

When I said that driver options were slightly more limited, I only meant that you will need to run v56.64 or newer for the FX5950. You won't be able to run v45.23 as you could with an FX5900. However, as obobskivich wrote, you probably want to run the newer drivers anyway as they'll give you better performance. Unless you have a specific need for an older version of the drivers, I agree with him completely.

soviet conscript wrote:

thanks for all the great info guys. I think I may stick with my 4600 though by chance I did throw out a random best offer for a 5950 ultra last night and it was accepted so I scored one for $20 shipped which i'm thinking is a decent deal either way.

$20 is great as long as the card works! 😀

Just make sure the system with the 4600 has a sufficient PSU - if I remember right the 5950U is 75W TDP (it's the highest power draw GeForce FX afaik, and it's not far off from the 6800 Ultra in that respect either).

also when you guys say your driver options are more limited with the 5xxx cards does that equal game incompatibility or do the later drivers just offer less option for tinkering with the card settings?

IME I haven't run into game incompatibility running 71.89 (which are VERY late for FX cards), but I have a fairly narrow range of games that I run on my 5800. The only setting that I don't see available is the "6xS" AA mode, but I do have 8xS which reviews of 4x.xx tend not to mention or show as available, so I'm thinking nVidia just replaced/upgraded one with the other at some point. Everything else is available as it should be, but do note if you move into the 9x.xx drivers and beyond (I forget if 8x.xx made the switch) you will get the "new" nVidia control panel, which is a much heavier UI and installation, and while I'd suggest that for a GeForce 7/8/9 I wouldn't suggest it for an FX as most of it's extra features aren't supported by the FX (e.g. FX does not do CUDA, SLI, PhysX, PureVideo, etc).

I'd also suggest nVTweak if you're going with the "old" control panel, as it will open up some additional options within the drivers (like overclocking, AGP information, etc).

KT7AGuy wrote:

Thanks for the info regarding the low benchmark. Like you said though, real-world performance is way more than adequate. I've kept my FX5950 running v56.64 for European Air War. I know from the experiences of others that it can get flakey with drivers newer than v61.76.

How odd - it's probably limited to that game or games based on that engine I would suspect, yes?

As far as the drivers for the 5950 - I don't know where the "break" for improved performance is. Here's an example with the FX 5800 U in 3D03:
http://techreport.com/review/4966/nvidia-gefo … 00-ultra-gpu/11 (I wish they had tested both drivers in all games, but we'll just have to settle for the synthetic)

I know other articles have also demonstrated as much with FX - due to how weird CineFX is, nVidia was in the business of re-writing or hand-optimizing shaders to improve performance in a variety of "popular" applications over the life of the FX cards. Of course also they improve things "in general" (stability, extra features, etc) as is normal for nVidia driver releases over the life of a card. However this is all mostly applicable to PS1.4 and higher; for DX5-7 era the FX cards tended to have less/no handicap compared to Radeon, and are generally very good performers. You can see this in the linked TR review if you look at Quake 3 and other older applications.

Something else to consider regarding AA/AF/IQ (and note that I don't own a 5950 and don't remember if they support power/clock management like the 5800 does) - IME running with 4x AA mode tends to allow the card to drop its clocks (and power draw/heat production) in older games, where an 8x mode tends to force it to run full-out. It's certainly capable of running 8x without any performance hits IME, but I tend not to notice a huge difference between its 4x and 8x modes, and would rather it run lower power/cooler/quieter/etc in that situation. I've observed a similar thing with my GeForce 7 SLI cards, where an old game may not be "enough" to tax it out of an idle state continuously, but cranking up AA/AF to the max tends to increase power draw pretty seriously. Just something to keep in mind - it isn't wrong to max out AA/AF, but otoh I certainly think there's a "break" whereafter you run into diminishing returns, despite increasing power draw and so forth to provide the feature.

obobskivich wrote:

How odd - it's probably limited to that game or games based on that engine I would suspect, yes?

The problem is actually limited to the GF6 cards with drivers newer than v61.76. However, because of that I've been hesitant to upgrade my drivers even though I run an FX5950. Maybe I should try it out and see what happens...

Question for you about older games. Something I always struggled to understand.

The Voodoo uses dithering to display more than 16 bit colours. Now somewhere I read that there are games that only support 16 bit colour and will look better on Voodoo cards because of dithering. Is there some truth to this? Basically if a game does not support 32 bit rendering, is there some benefit of going with a Voodoo over a GeFore, or is it all the same?

I made sure I checked the description carefully so it was posted as used but guaranteed working. I'll double check but the PSU should be up to snuff. when it gets here I'll do some testing with both cards and see how it goes.

KT7AGuy wrote:

The problem is actually limited to the GF6 cards with drivers newer than v61.76. However, because of that I've been hesitant to upgrade my drivers even though I run an FX5950. Maybe I should try it out and see what happens...

It should be okay I would guess - the GeForce FX and GeForce 6 are quite different. I'd say it's worth a shot if you think the newer drivers might benefit some other game(s) on the system. 😀 I wouldn't, however, tinker with things just to tinker with things (in general I'm not a fan of constantly updating drivers as new versions come out - if/when you have something that's stable and performs well, leave it be imho).

Tried Oblivion in WinME earlier today.
Doesn't launch on the FX 5800 Never tried to run it in 9x before 🤣.
Surprising it installed fine, from disc with no issue tho.
56.64 drivers.
Does it Need SSE2? to work?

AlphaWing wrote:

Tried Oblivion in WinME earlier today. Doesn't launch on the FX 5800 Never tried to run it in 9x before lol. Surprising it insta […]
Show full quote

Tried Oblivion in WinME earlier today.
Doesn't launch on the FX 5800 Never tried to run it in 9x before 🤣.
Surprising it installed fine, from disc with no issue tho.
56.64 drivers.
Does it Need SSE2? to work?

I've run it on AthlonXP, so it doesn't explicitly require SSE2, but I do know that when I switched from AthlonXP to Celeron D (same graphics card (6800GT), OS, etc) it saw a very respectable performance improvement (basically went from periodic "slowdowns" to running smooth AND doing it with more IQ enabled). Given that my two systems in that "comparison" benched pretty similarly in 3D01, PC Wizard, etc EXCEPT for the Celeron D putting up much bigger numbers in SSE-supported tests in PC Wizard, I'm assuming SSE2/3 was the contributing factor to the performance increase. After observing this I went "back" and tried Hitman 4 (it was the other brand new game I had 😊) on the XP with the same graphics card, and it was a slideshow unless I turned the settings way down (on the Celeron D it ran at either 1024x768 or 1152x864 (monitor limitation there), and medium settings; on the XP it was at 800x600 on low and still would be sluggy at parts (especially levels with a lot of water from what I recall)).

Of course, there's also oddball stuff like this:
http://www.youtube.com/watch?v=s6lEpgNMCBo
(Fallout NV running on an AthlonXP 2000... 🤣)

On Oblivion - I remember it running on my FX 5900 in XP but not performing very well. I'd honestly consider GeForce 7 or higher to be a reasonable cut-off for that game, especially if you want a lot of IQ.

Did a quick search for Oblivion on 98 and found this: http://www.oblivionmodwiki.com/index.php?titl … _for_Windows_98

If you test it out, do let us know if it works. I probably wouldn't download the DLL from that page, and would instead just extract it from a working 2000/XP build, but that's me. 😊