NVIDIA GeForce 9600 GT SLI Benchmarked

[Jimmy 2004]

CHIPHELL has managed to get hold of a couple of NVIDIA's upcoming 9600 GT samples, and as you can probably guess that means only one thing: SLI. The two regular 9600 GTs were benchmarked in a system running with an Intel E6850 processor at 3.0GHz, an nForce 680i motherboard and 2GB DDR2 memory at 800MHz, using Windows Vista as the operating system. By itself, a single 9600 GT scored 10036 points in 3DMark06, but when combined in SLI the two cards managed an impressive 13080 points - over 900 points above the score of a GeForce 8800 Ultra (12142) on the same system. Meanwhile in 3DMark05 the 9600 SLI hit 18805 points and the setup reached over 55000 points in 3DMark03. Not bad for a mid-range card.

View at TechPowerUp Main Site

 

[Jimmy 2004]

Sorry for the quality of the images - they aren't showing on CHIPHELL at the moment so I'm having to use the copies from VR-Zone which have two watermarks...

 

[kwchang007]

Wow, not much difference for sli....but one by itself is rocking, slightly cpu bottlenecked I think, but this is going to be amazing if it launches at a price around the current 8600s

 

[ShadowFold]

My HD 3850 scores higher with a slower cpu.. If this is around 100$ good deal but anything above 150 is asking too much imo

 

[newtekie1]

I hope some driver maturing improves that SLI score, or it might be a CPU bottleneck also. We'll have to see. As usual, I am waiting for the cards to actually be released.

 

[Exceededgoku]

My theory is to only get a mid range card if it is equal to or better than the previous gen high end (maybe ultra is too far though but you get my drift...), I'm still waiting....

 

[trog100]

2006 is always cpu bottlenecked its useless at its default resolution as regards being a grafix card bench.. we need a new one..

i can score higher than the pair of them and the ultra with my one ati 3870 card.. 13330 in 2006 and 23000 in 2005.. 51000 in 2003.. but only and only cos its a useless grafix card bench and my cpu is running at over 4 gig when i run the benches..

having said that being as i know how it works.. i am quite impressed with what in essence is a mid range card.. over 10000 in 2006 aint bad..

and a pair of em beating an ultra is very good.. say no more.. he he

trog

 

[Steevo]

How much for these cards?

 

[PVTCaboose1337]

If the price is right, this is my card. If not I will get the 3850.

 

[jocksteeluk]

does anyone know the comparisons between these and the 8800gt and GTs?

 

[PVTCaboose1337]

9600 is slower than the 8800gt.

 

[DarkMatter]

2006 is always cpu bottlenecked its useless at its default resolution as regards being a grafix card bench.. we need a new one..

i can score higher than the pair of them and the ultra with my one ati 3870 card.. 13330 in 2006 and 23000 in 2005.. 51000 in 2003.. but only and only cos its a useless grafix card bench and my cpu is running at over 4 gig when i run the benches..

having said that being as i know how it works.. i am quite impressed with what in essence is a mid range card.. over 10000 in 2006 aint bad..

and a pair of em beating an ultra is very good.. say no more.. he he

trog

So right. Anyway 3DMark has never been the better way to compare graphics cards besides those that belonged to the same GPU architecture. The better 3DMark performer rarely is the better gaming card. It's been a long time since 3DMark is nothing more than a showcase for overclockers and people trying to achieve the record. Has nothing to do with games and by extension graphics cards.

 

[Steevo]

Not entirely true, if you have one standard to hold a card to, and that uses common gaming engines, or what they predict game engines will be able to do then you have a benchmark, perhaps a synthetic benchmark, but a standard. Look at games and see what happens when they tweak drivers for games, you get variances, 3Dmark however tries to make the field level by forcing compliance in drivers and what they feel a system will be able to do.

I'm not saying it is perfect, but it is better than arguing about this game over that game, and how you feel about this over that, and the fanboy routines. I used the reviews here, and based some of my information off the 3Dmark scores that a 3870 and CPU combo got to determine where I would end up in performance. It ended up right about where I guessed, and now allows me to play all my games with maximum detail at default 1680X1050.

 

[substance90]

Actually, these are pretty bad results for a mid-range card from the NEW series...
And this is 3dmark, and everyone knows, that 3dmark loves nVidia, the realworld performance would be even lower

 

[AddSub]

Not bad, well for a $200+ card that is. I mean, we all know that these cards are going to suffer from “8800GT syndrome”. A tightly controlled supply and 25% to 30% markups on retail side, and let’s not forget the 10% “GeForce” tax.

And yes, I know, I keep hearing it too: 9600’s will be priced “competitively” vs. 3850’s. (Which can be found for $150 @ Newegg and TigerDirect) Anyone who believes that is naive or in denial on how nvidia operates.

 

[moto666]

Not bad!
If price right...
But I think I buy a HD3870, a little more money but all I need for 2008!

 

[candle_86]

Not bad, well for a $200+ card that is. I mean, we all know that these cards are going to suffer from “8800GT syndrome”. A tightly controlled supply and 25% to 30% markups on retail side, and let’s not forget the 10% “GeForce” tax.

And yes, I know, I keep hearing it too: 9600’s will be priced “competitively” vs. 3850’s. (Which can be found for $150 @ Newegg and TigerDirect) Anyone who believes that is naive or in denial on how nvidia operates.

really, lets see the chip is cheap to make in general, less complex than the 8800GT but shares all its enhancements. Its half the chip the G92 is, which last i checked is midrange, maybe im wrong there. Seeing has how Nvidia can afford to supply 8800GT's for 250, and the 8800GS at 200, this being a cut in half G92 in essance means the chip is even smaller, and cheaper, so 150 isnt far fetched and rumor has it, there is a 9600GTS planned also.

 

[choppy]

If the price is right, this is my card. If not I will get the 3850.

+1 this is exactly what im going to do too, an overclocked 512mb hd3850 is £115 in the uk, so im hoping a similar 9600gt will be lower than that, considering 8800gs are going for hd3850 price

 

[Xaser04]

Actually, these are pretty bad results for a mid-range card from the NEW series...
And this is 3dmark, and everyone knows, that 3dmark loves nVidia, the realworld performance would be even lower

Remember though that this card is nothing more than a G92 8800GTS but in half (128 shaders vs 64) albeit keeping the 256bit memory interface. It is not an entirely new architecture like the name would suggest.

A score of 13.xk for SLI in 06 isn't too bad considering they are using a core 2 duo rather than a quad (the latter would probably add roughly 2k at the same clock speed).

 

[DarkMatter]

Not entirely true, if you have one standard to hold a card to, and that uses common gaming engines, or what they predict game engines will be able to do then you have a benchmark, perhaps a synthetic benchmark, but a standard. Look at games and see what happens when they tweak drivers for games, you get variances, 3Dmark however tries to make the field level by forcing compliance in drivers and what they feel a system will be able to do.

I'm not saying it is perfect, but it is better than arguing about this game over that game, and how you feel about this over that, and the fanboy routines. I used the reviews here, and based some of my information off the 3Dmark scores that a 3870 and CPU combo got to determine where I would end up in performance. It ended up right about where I guessed, and now allows me to play all my games with maximum detail at default 1680X1050.

If that's the case, good for you. But that it's far from reality in my case and I would bet that for most the people.
I will put my last purchase (Gigabyte 8800GT@700Mhz) as an example. In all the reviews i have seen, my card should score near 12000 points. In reality it scores 9200, that's 75% of the performance. Then, it's my card defective? Should I return it? No, it only took an hour playing (with fraps on) Crysis, COD4, UT3 and Bioshock to realizse that my card indeed run faster than in the same reviews where it scored well below in 3DMark. Later I have tried GOW, STALKER, Prey, Jericho, Oblivion and many others and the card still performed a little better than on those reviews (mostly beause it's slightly higher OC), but in 3DMark the same low score. 3DMark06 debuted in late 2005 IIRC and it was supposed to be representative of the games coming in the next 2 to 3 years.
Now tell me is 3DMark being any representative of these games 2 years later than its debut? I guess no, since my overclocked card and my now low CPU, perform better than the rig that they use for benchmarking, but not on 3DMark.
And it's been the same with every single graphics card that I have purchased and tested with 3DMark.

 

[Steevo]

Depends on resolution, graphics settings, and a few other factors. I had some stuttering in Test Drive Unlimited with my X1800XT overclocked and my CPU at 2.6Ghz. New 3870 and no more stuttering, and I was able to turn the settings up to max. Our systems look almost identical minus the video cards, and I score higher than you. That is probably due to this card being slightly better in 3Dmark, and slightly slower at games. Or other factors, ram timings, chipset timings, drivers, etc.....


Anyway, it does let you know a general idea of what the hardware is capable of. And is directly comparable to other hardware, or the same or different class.

 

[DarkMatter]

Hmm. You score higher with your CPU at 2.58Ghz and the system OCed as is stated on your system specs or at stock speeds? Because that would explain a lot...
And yeah the HD2000/3000 series perform better than Nvidia offerings at 3DMark and worse on games. Another reason why 3DMark isn't very useful for those kinds of comparisons.
As I said, it's useful if you want to compare a 8800GT with an Ultra, or you can guess how the 9600GT would perform because its the same architecture, well kinda because since SP/ROP ratio has changed its unpredictable how the card would perform on games based on 3Dmark score.
Of course a better card will perform better on 3DMark but the difference between the cards is not real. Most of the time (not always) you can guess that a card is better than the other, but not by how much.

Another point you have overlooked is that 3DMark's engine barely represents a real game engine now or in the future. This has been said by most respectable game developers and both Ati and Nvidia. And it's no secret: Unreal Engine 3, Cryengine 2 >>>>>> whatever is the name of the engine used by 3DMark.

It's a trait on the human kind to compete each other and 3DMark was just the easy way, not the better way. People will trust wathever tool you throw at them if it demostrates he (or the thing at his possesion) is better than others. 3DMark was successful because of this and not because was really useful for hardware comparison. It does compare the strenght on some aspects of the graphics card, but not all of them. 3DMark is like trying to compare a cars max speed in an engine testbed. Sure the one with more horsepower will be faster, if it's the same car what we are comparing. But between two different models or brands...

 

[Steevo]

Sorry I need to update, I have it at 1.38vcore now and 2.68Ghz
2Gb kit at 2.3.2.5 1T 2.85vdimm 223Mhz