Sapphire 9800pro - Overclocking worth it?

[Stewart]

Hello all,

Recently, I purchased a new system, complete with a shiny new Sapphire 9800pro 128mb.

Now, I've tweaked my RAM, and nicked a few 100Mhz from my CPU, so I was wondering, if it was worth overclocking the card?

It's running at 378.0Mhz Core Clock, and 337.5Mhz Memory Clock, at the moment. Standard settings.

Is it worth trying to squeeze a bit more out, or is the card, already running about as fast as it will, without going up the wall?

Anyone clocking their Sapphire 9800's higher than this?

The system it's in is an AMD Athlon 2600+, 512mb of DDR400 Crucial RAM, NF7-S Motherboard. Nice and cool at the moment, about 45c, with my Coolermaster running about half as fast as it could, if I could stand the noise.

Cheers,

Stu.

[d_fresh1]

unless ur laggin in games yeh oc it

[SilentDeath]

o/c it, definatly

my 9700np goes from 276/270 to 390/313.5 when clocked, on air, no voltage increase yet (which i plan to do when i get w/c). i think i could get to 470/350 with volt mods etc...

[Stewart]

Someone, who may or may not have known what they were on about, told me that the 'pro' versions of the 9x00 cards, were already clocked at about the speeds they will run correct at.

Is that a load of tosh?

Also, I got RedLine packaged with my card, is that what you lot are using to mess about with your 9800's?

[Austin]

Considering how much that card's worth, it's stock perf and how little you'll gain from o/c'ing it is plainly not worth it at all. Gains will be small as it is VERY near the technological limit as are practically all of the top end cards of their respective ranges. GF4TI4600, GF3TI500, GF-FX5900ultra all gain VERY little ... that's why people get the 4200, TI200, FX5900, 9800/9700 etc in the first place ... it saves them a lot of money and they can often o/c very close to or sometimes beyond the top-end card anyway. I'd say you'd be lucky to get much more than 400/350 o/c'ing and I doubt you'd notice any diff from the stock 380/330 outside of benchmarking. Enjoy your 9800PRO for what it already is ... the best gfx you can currently buy.

[fondie]

mine runs at 460mhz (a 9800pro but not a sapphire), which by my calculations is a bit over 22% faster, a worthwhile gain in my opinion!

[BUFF]

Originally posted by fondie
mine runs at 460mhz (a 9800pro but not a sapphire), which by my calculations is a bit over 22% faster, a worthwhile gain in my opinion!

OK, you have a 22% core overclock - but what extra performance does that actually give you in game?
I would hazard 2-5 % & certainly nowhere near 22%.

[Lexeus]

Originally posted by BUFF
OK, you have a 22% core overclock - but what extra performance does that actually give you in game?
I would hazard 2-5 % & certainly nowhere near 22%.

And anyway, surely once you reach the greatest gameplay possible in a game (which im sure the best gphx card available can achieve) there is no increase to obtain ...... i mean even if a game could push an extra fps, it wouldnt effect the look at all .... an less you were some sort of eye test machine

[Stewart]

Well, I can run the games I have at the moment, in perfectly playable FPS, at 1600x1200, with all the eye candy turned on, so that's no problem.

I thinking more of Half Life II - that will give my 9800pro a run for its money, at higher res, with all the nice stuff on, and then, a few extra frames per second will be welcome.

Of course, some in this thread say they get 22%, some say that only means 2% in games. So I've still no clue if it's worth it or not.

[steve threlfall]

In english extra game performance would be unoticeable

[fondie]

Originally posted by BUFF
OK, you have a 22% core overclock - but what extra performance does that actually give you in game?
I would hazard 2-5 % & certainly nowhere near 22%.

if the graphics gpu is the bottleneck, which it often proves to be in graphically demanding games, then increasing the amount of work it can do by 22% can indeed increase performance by 22%.

following your 'logic'. the 9500pro and the 9800pro would be almost identical in performance (apart from a minorly revised core the only difference between them is clock speed), which clearly isn't the case.

[Agent]

I don't believe clock speed is directly proportional to output though ?

[Austin]

It does get quite complex, modern cards (ie since GF2) have gained in perf quite equally from both RAM and core increases, so a 22% core o/c would equate to no more than 11% extra perf as the RAM is responsible for as much of the perf as the core. However when you bring in AA, AF and DX9 (or even DX8) things get more complicated as some are more dependent on the core while other(s) are more dependent on the RAM, also the actual implimentation varies too complicating matters further.

Therein lies some of the puzzle, if the core and RAM aren't well balanced gains will be much smaller (or zero) unless you raise the one which is effectively causing the bottleneck. So is the RAM and core well balanced in the 9800PRO or is one holding back the other? Best way is to benchmark in what you intend to use the card for, IIRC 5% actual perf gain is the minimum you need to stand a chance of seeing a perceptible difference.

As for the 9500PRO vs 9800PRO thing there are far more factors involved. The 9500PRO is using 128bitDDR vs the 256bitDDR of 9800PRO so it inherently has double the bandwidth from the start. Next the 95P's RAM is clocked at 270 compared to the 330 of the 98P, again that gives the 98P a 22.2% advantage BUT that's actually doubled because each 1mhz counts twice due to 256bitDDR. So it's clear the 98P gets most of its advantage from better RAM (wider and faster). Next the cores, both are virtually identical but 95P runs at 275mhz while 98P runs at 380mhz, a 38% gain for the 98P. If the 95P used 256bitDDR and had clock speeds ot match you would get 98P perf, all you'd really lack are slight optimisations most notably in AA+AF and of course the mysterious T-buffer.

How does clock speed effect the technology behind the Rad9500 to 9800PRO? Well we can see most clearly when looking at 9700 vs 9700PRO. Both use the same core and 256bitDDR ... only the clock speeds vary and we see 97P holding 18% core advantage and 15% in RAM (it's 275/270 vs 325/310). So you'd expect the 97P to be 16.5% (average of 18 and 15) faster right?

http://www.anandtech.com/video/showdoc.html?i=1735&p=5

UT2003 FLY: 5.5%, 14% & 16% (3 diff res)
UT2003 BOT: 3.2%, 0.3% & 5%
Commanche4: 0.6%, 1.3% & 4%
JediKnight2: 0.6%, 0.6% & 1.6%
SeriousSam2: 5%, 12% & 14%

4xAA+16xAF
UT2003: 16% (fly) & 13% (bot)
Comm4: 14%
JK2: 2.5%
SerSam2: 14%

So we see an 18% core increase with a 15% RAM increase yielding between 0.3% and 16% but averaging 5.6% (83.7/15) or with 4xAA+16xAF 12% (59.5/5).

Under o/c'ing in the same review we see what clock speed increases can do for the 9500PRO (it is a 9500PRO review). I'd expect the RAM increase to be the most important as it's half of what the core was designed for (128bit vs 256bit). Increased core of 28% and RAM of 18% would lead to an estimated 23% speed increase overall. This works out in practice to 18% in UT2003 FLY, 6% BOT, 5% Comm4, 5% JK2 and 17% SS2. Interpret it all as you will.

[Tom]

bloody hell austin u are a reply machine, you should get paid. :O

[Stewart]

Cheers for that Austin. Thanks for making the effort mate.