I strongly suspect my Geforce2 is going faulty, and it's a little sluggish in EQ at times anyway so I figure it's about time for a new card.... I'm looking for suggestions.

I'm not exactly rich, so I'd sooner go for a £/$100-ish card than do the rather expensive early adoption thing.

It does not need to run smoothly in the Bazaar (guffaw!) but smoothly in most other places would be good.

Doom 3, I suspect, will require yet another upgrade when it comes out... that's another reason I don't want to spend much on this one. But if a slightly (i.e. not too much) more expensive card will run it, then great :)

As for the last criteria, I don't know if anyone's up on it here... I'll be searching/asking in other places... but basically Linux support is another thing to look out for. I run Everquest using Transgaming WineX (which actually seems to run it at equal speed to Windows itself), and read somewhere that said product is written with NVidia cards in mind more than ATI, despite the lack of documentation from NVidia. But I'll dig deeper myself on that one.

I understand that some of the cheaper FX cards are rather sub-standard... especially to those with a taste for high-end speed. However in terms of "bang for the buck" could they be a reasonable option? Or something else?

Cheers.

I would expect that a GeForceFX 5200 would blow your GeForce 2 out of the water even though it is slower than the rest of the current lineup. It also provides support for DirectX 9 which EQ2 will use and it can be had for around $100.

The really tempting one (in terms of value) seems to be:
128M Abit Siluro FX5200DT TV/DVI @ £55.22 inc vat (around $55.22 given the computing industry's exchange rates!)

But at that reasonable a price I want to be sure I'm not buying a total lemon. I've read plenty of times about how the Geforce4-MX is exactly the same speed I have already, and if this is barely any faster than that...

128Mb Gainward (4769) FX PowerPack! Pro/760 XP "Golden Sample" (FX5200 Ultra) - at £152.46 this one's a little steep, but if the Ultra model is *considerably* faster then it could be an option.

("Golden Sample" though? Sounds like something you'd give to a doctor!)

This is a confusing, transitional time for graphics cards, because there are three generations of cards out there with confusing model names, and the newer ones aren't always better than the older ones. But basically anything on the market will blow away a GeForce 2. <img src=http://lag9.com/biggrin.gif>

If you're looking in the $150-$200 range, I'd consider a Radeon 9500 Pro (good and fast, but being discontinued) or 9600 Pro (the newer model, slightly slower), or maybe a 9700 non-pro at the high end (faster than both, but being discontinued).

If you're going Nvidia, my best advice is to hold off right now, since Nvidia will soon come out with the 5900 series, and there should be a good budget model in the mix within 2 months.

If you MUST buy Nvidia now, avoid the 5200 Ultra, it's a dog. It's slower than a Ti4600, yes the same card that's been around for ages. You're better off going with an aging Ti 4200-4600-4800 (ranging from about $80 to $140), which will be slightly slower than a Radeon 9500 Pro/9600 Pro but in the same ballpark. If money is really tight you could probably pick up a used one on eBay, as people will be dumping them in favor of the newer batch of high end cards.

Either a 9500pro/9600pro or Ti4600 will be adequate for EQ at lower resolutions (e.g., 1024x768). You should, for example, be able to go through the Bazaar with medium clip plane, some new models, and not be seeing a slide show, so long as your CPU is decent. At 1600x1200 they're going to struggle though -- you'll need either a 128Mb FX5900 (when it comes out) or 128Mb Radeon 9800 Pro, both around the $375 mark, plus a good CPU to really make some headway there.

Be sure to factor in the "outside-of-U.S. ripoff markup" when you make your buying decision.

Thanks for the input...

Confusing! And this is coming from someone who's normally a bit of a local PC tech support...

The "outside-of-U.S. ripoff markup" appears to apply quite well here - the cheapest Ti4200 is at £90, everything below that mark is an MX440.
Ti4600s don't seem to be sold at all on either Dabs or Scan.
For a 4800 I'm looking at £140.

64Mb ATI Radeon 9500 x8 AGP DDR/DVI/TV Out weighs in at £90 as well, which is bearable and probably nicer than the Ti4200, but further research confirms that ATI are not playing nice when it comes to Linux support. I'd actually prefer to keep the good OS and ditch the dodgy manufacturer here :)

So far I'm inclined to grab the cheap 5200 I mentioned earlier then upgrade to something decent (i.e. a 5900 series) when better money comes in and Doom 3 / EQ 2 come out... unless that would be a really stupid move despite the low cost... in which case I'll rethink. If the Ti4200 is faster than the 9400 is it considerably faster, as in worth the extra 35 quids?

Either way it needs to be ASAP. My current GF2 has startd locking my machine up several times per day (in Windows and 2 different Linux distros). It started happening after many months of stability and no tweaking... and I've so far proven its guilt - touch wood - after putting a spare Geforce256 in. So even if it's an interim card that's no faster than what I have now, it'd still have to do.

Otherwise that seems the best option for me at the moment. If it still blows the GF2 out of the water than it will do for now. Just want to be sure I'm not going to kick myself after finding something for £2 more (aside from ATI) that does 10x the performance :)

FWIW, just found a Ti4600 on eBay at £125.

Again, worth the difference considering that a simple 5200 card is only £55?

FWIW, just found a Ti4600 on eBay at £125.
Again, worth the difference considering that a simple 5200 card is only £55?

You have to watch the model names -- "Pro" versus non-pro, "Ultra" versus non-ultra, and changes in digits -- as each can make a big difference in the performance of the card.

The 5200 (non-ultra) has performance similar to a Radeon 9000 or GeForce 4 MX440. It's not really suitable for serious gaming. It's not worth wasting money on a really bad card just to replace it in the near future; you're probably better off RMAing your current card and just using whatever standby card you have on the closet shelf while you wait.

A Ti4200 (that's the low-end model from the 3-generation-old GeForce 4 line) is roughly double the performance of a 5200 non-ultra, and is <$100. It's also pretty overclockable. It's definitely good enough to tide you over with EQ until Doom III comes out, and the prices of the 5900 & 5900 Ultra come down from the stratosphere (~$400-$550 now). Just overclock the Ti4200 as much as you can with the stock cooling, and keep the EQ resolution to 1024x768, and you should be OK.

OK, I'm pretty much convinced that Ti4200 is the way to go for the time being.

One more thing... 64MB vs. 128MB - massive difference?

If no - I'll probably settle for a 64MB Ti4200

If yes - Given the cost of the 128MB Ti4200 models I might as well pay the little extra for a Ti4600. Also a comparable price is an FX5600 (non ultra)... but from what you said about non-ultra models I'm guessing the Ti4600 is the better option.

I can't recall reading much about the 5600 non-ultra. I know the 5600 Ultra, which in Nvidia-speak means a 5600 with a faster core/memory clock, has performance in line with the Ti series when you're talking about older games (such as EQ) and no high AA/AF settings. So I really doubt you'll do better with a 5600 non-ultra versus a Ti4200.

As to 64 versus 128mb, it won't make a "massive" difference, it's more like 5% not 50%. If this isn't a card you're going to keep a long time anyway, you might as well save some money.

Think of the 64 vs 128 in EQ terms as "low-medium quality textures" and "high quality textures". For the most part, resident RAM is mostly used for texture caching. The more often you have to swap out textures from the HD to your card, the longer it'll take to do stuff like zoning. Now if you're still using a slower card, you'll still want to settle for low/med quality textures anyhow to eek out that last iota of performance.

I've been running on high quality textures on the Geforce2 actually... never really thought about bringing that setting down. It does explain a few things though...

Naturally the graphics REALLY slow down in very crowded or graphically intensive zones (1 frame per minute in the bazaar practically). But with an Athlon 1800 and 512MB you have to wonder at it.

The big giveaway of the "texture cache" problem is presumably the bit of graphical "thinking" - a very brief but noticable pause - that happens on this system when a new luclin race comes into view. Whilst it doesn't really affect gameplay once everything's settled down with everyone in the group loaded into the texture cache (hunting, the important bit!) it can get annoying whilst travelling.

If I get a faster card I'd be inclined to keep the same quality rather than downgrading it :P But I'll give it some consideration given that both the GF2 and the prospective new card have 64MB.

I've only ever run in 1024x768 (except for the early days of EQ) and have no desire to run any higher.. or the monitor to cope =P (It can manage 1280x1024, but only at 60Hz which hurts my eyes). So higher resolutions are not an issue.

I'm basically settled on a 64MB Ti4200 therefore, and will order it v. soon.

Thanks for the advice... helped me to make an unavoidably badly-timed, but at least as well informed as possible, interim purchase :)

Cloudien says: The big giveaway of the "texture cache" problem is presumably the bit of graphical "thinking" - a very brief but noticable pause - that happens on this system when a new luclin race comes into view.

These are two issues. The video card's memory only determines what is going to be processed locally versus being dragged back and forth from system memory. This shouldn't create any noticeable pauses.

The pauses you see are usually caused by hard drive activity. This is because some data is being loaded from the hard drive, or because you are running out of memory and Windows is starting to use virtual memory. EQ is an enormous memory hog. If you look at your hard drive activity light, you'll probably see it flickering occasionally as you travel though a zone.

Naturally the graphics REALLY slow down in very crowded or graphically intensive zones (1 frame per minute in the bazaar practically). But with an Athlon 1800 and 512MB you have to wonder at it.

EQ really likes 1Gb memory if you're going to turn on any of the new models. With 512Mb, EQ can easily gobble up all available memory in a place like the Bazaar. If you can't get 1Gb I suggest limiting the new models to only a few and turning on texture compression.

Also, don't forget that EQ loves raw CPU power. Overclock that 1800+ and squeeze some more performance out of it.

Now that you mention overclocking...

The darn thing runs up to 72C in EQ without overclocking. It has a fairly expensive coolermaster heatpipe on it, rated for about a 2600+, carefully applied thermal goo (*not* one of those solid pads), a fan at the front blowing in and a fan at the back blowing out to create airflow, and a fairly decent airflow path. Checked voltages etc, all settings seem ok... It just doesn't seem to want to cool!

I'll add another 512MB though. Memory is dirt cheap at the moment anyway.

The darn thing runs up to 72C in EQ without overclocking.

Are you sure? Motherboard sensors, even the ones that claim to read the internal diode, are sometimes way off. You need place a thermal probe right up against the core to get an accurate reading.

A real reading of 72C is extremely high if you have normal room temperature (15-25C). Beginning at about 65C you run the risk of having processor lockups. I'm not sure which heat pipe you're using, but it's obviously not doing its job if that reading is real. Even the stock AMD heat sink/fan should keep you well below 72C, considering an 1800+ doesn't generate a ton of heat to begin with.

Come to think of it, maybe the CPU is related to the source of your trouble. Did you try a stability test, like Prime95, over the course of a few hours?

Might just be the motherboard sensor then. For what it's worth, the board is an Asus A7V-333... a quick google turned up possible concerns about the reliability of the sensor on that board.
Actually it could also be "Operator error" :) I'm getting the temperature from an Asus iPanel and just assumed it to be the correct reading. It's a horrible piece of hardware anyway - whenever you switch modes on the iPanel it pauses all processes on the motherboard causing it to "lock up" for a split second.

Do you put a thermal probe by the die under the heatsink? I'm just a little wary of poking things under there in case it interferes with heat transfer. I have a thin.. what looks like a paper-thin thermistor.. thermal sensor that used to plug onto a header on an older board. Back then I placed it in the middle of the heatsink fins. I could try placing that somewhere and plugging it into a sensor header.

I'll try a stability test also. But after replacing the "offending" Geforce2 with an older Geforce256 for a few days it's been absolutely stable. With the GF2 in it was locking randomly, usually every few hours.

The heatsink is a HHC-L61 "silent" heat pipe (not actually silent as branded, just a relatively quiet fan). I'm well aware that quiet heatsinks generally aren't for overclocking, but I'm sure it'd at least be as good as the stock AMD one for running normally.

Just to confirm...

Went into the BIOS and that reading is definitely the CPU one, the iPanel is working properly.
Checked the BIOS revision and it's 1011, the version that fixed an earlier temperature monitoring bug.
Fan is rotating at 3391RPM, which is about right for a quiet fan I think.

Do you put a thermal probe by the die under the heatsink? I'm just a little wary of poking things under there in case it interferes with heat transfer.

Yes, try to nudge it as close to the side of the core as possible (not on top). Although the actual core is in contact with the bottom of the heat sink, there should be a thin space all around that's empty. There's actually an inexpensive device called Compunurse which is ideal for this task. You don't leave it there, just take the reading and remove it.

If this proves too difficult, you can try to take a reading off the top side of the heatsink, as close as possible to the core, such as the middle of the bottom plate. This will give you a reading that's lower than actual but can give you some idea if that 72C figure is accurate.

LOL.... it finally crashed. Same symptoms as before under Linux, flashing caps+scroll lock. (Again it's not just this Linux distro - the system is generally unstable)

Well at least it's not the graphics card, so I'm free to put the GF2 back in and save up for whatever's hot when Doom3 and/or EQ2 come out.

FWIW the CPU was only running at about 55C at the time of the crash.

Looks like it's my old favourite task of lengthy elimination.... replace a card, run it for a few days, crash, rinse, repeat. Once the likely suspect is found, run without it for a few weeks to try and confirm it.

Fun!

Bought a temperature monitor (LCD panel with 3 sensors) and tucked a sensor under the heatsink as far as it will go.

I tried taping the sensor right next to the die and then putting the heatsink back on, but this made it uneven and the CPU started overheating quite quickly... it seems more safe and reliable to tuck it under with the heatsink already firmly installed.

At lower temperatures the BIOS and the probe remain equal.
Under full load using Prime95 the probe reads 52C whilst the onboard sensor reads 62. Certainly interesting... I don't really know whether the increasing difference is a sign of a bad sensor, or a sign (given the probe's position) of one that's working perfectly.

---- Addition

Ooh!
"FATAL ERROR: Rounding was 0.4999997308, expected less than 0.4
Hardware failure detected, consult stress.txt file.
Torture Test ran 30 minutes - 1 errors, 0 warnings."

Sounds like it's time for someone to get a Vapochill! (http://www.vapochill.com/main/page.asp?sideid=391)

<font size="-2">Disclaimer: No, I didn't really read any of the above threads and I don't know if it's needed or not, I just wanted to add a link to Vapochill cause I thought it was sweet!</font>

Looks a sweet piece of kit :) I'd be wanting to store beer in there (not for long of course, though it rarely lasts long here!)

Of course for the cost of one of those beasties to overclock it a few MHz, I'd be able to get *checks* an XP 3200+ 400FSB which would be nowhere near as cool (pun intended) but a few times faster!

---- Another addition

Just switched the 2400RPM "silent" fan for decent 5000RPM one (which, I might add, seems no louder) on the same heatsink.

Load temperature with mprime now 53C by BIOS (45 by external probe) which is certainly more comfortable! Moral of the story, I guess, is to avoid these silent/v.quiet fans.

mprime aborted after 6 mins however with exactly the same error:
FATAL ERROR: Rounding was 0.4999997616, expected less than 0.4
Hardware failure detected, consult stress.txt file.

Not good! The memory checks out ok too, with memtest86 so I guess something more serious is unhealthy.

One of the websites that did a review on the Vapochill said they were able to successfully overlock a AMD 2400+ to 3GHz and keep it stable. Granted, the case cost $5-600+ (you gotta figure that's at least $400 more than even a good aluminum case) for which you could have gotten the 3000XP, but hey...

All joking aside (well not all...) I am seriously considering getting this mod (http://www.thinkgeek.com/pcmods/cases/6169/) for my next case. Now that water cooling kits have become packaged solutions, I'm anxious to get rid of the loud fans for a water cooled solution. My next box will be water cooled... and I kinda want fishies.

Ah yes, I saw that one locally... very sweet indeed :D

I'm tempting fate by saying this, but I think I've tracked down the problem.

It's definitely tempting fate because it's kinda the opposite of what was expected...

After searching around on messageboards and google for the error returned by prime95, many people seem to have narrowed that error down to RAM. Whilst the RAM itself checks out OK after a good 8 hours of memtest86, some suggestions included increasing the CAS Latency in the BIOS.

Well, it was already at maximum, so I decreased it instead.

Already passed two self-tests (1024K and 8K), and half way through the third run :) Makes no sense to me, but if it works it works!

As an encore I may have fixed my father's too. His was resetting all over the place and prime95 was dropping out at the first test. Increased vcore by 5% (crap PSU methinks) and so far so good *touch wood*

Thanks Tudamorf for pointing out prime95, what an invaluable tool!

----- Edit
Several tests later, still going strong. Looks very hopeful... just hope it's that which was causing all the instability (probably not haha)
Dad's was both a hardware and a software problem. Increasing vcore seems to have made prime95 stable in a clean Windows, but in the "active" windows it still chokes. At least it's solved one out of two problems :)

Cloudien says: Under full load using Prime95 the probe reads 52C whilst the onboard sensor reads 62.

A reading of 52 sounds more reasonable. You really don't need better cooling, because I have yet to see a situation where an Athlon is unstable at 52 but perfectly stable at 45C. Now, 65+ is another story, but obviously your motherboard sensor is way off. Save your ears and go with the fan that's the least noisy.

After searching around on messageboards and google for the error returned by prime95, many people seem to have narrowed that error down to RAM.

Yes, RAM is very often a cause of system instability. Prime95 stresses both the CPU and memory. For optimal performance, set the memory usage (in the CPU... menu) to 2/3 of your total system memory and don't run anything else at the same time.

The easy way to diagnose a Prime95 failure is to underclock your CPU by lowering the multiplier while keeping FSB and memory parameters constant. If the Prime95 failures go away, it's almost certainly your CPU. If they remain, it's almost certainly your memory. You could also do the opposite -- keep the CPU clock the same but lower the memory timings -- but as you just noticed, memory timings don't always act as you would expect, because there are 6 different memory parameters (clock, 4 separate timings, and VDIMM).

P.S. Prometeia &gt&gt Vapochill <img src=http://lag9.com/biggrin.gif>