GTX 470 pics *EDIT* possible benchmarks

[GaiaHunter]

exactly.

Not quite.

It seems to be about equal in Crysis with 8xAA and worse with 4xAA.

But it is only a game and with dubious source benches.

The other bench for example still has 5870 somewhat faster with 8xAA (no 4xAA numbers).

DiRT2 DX11 19x12 / Ultra 8xMSAA:
5870: 59.2 / 51.9 (Avg / Min)
GTX470: 53.8 / 41.6

So we can't even talk about the GF100 taking a lesser hit when going from 4xAA to 8xAA compared to 5xxx.

For me the most interesting of all this is that it seems the GTX480 isn't going to have 512 shaders. So it seems GTX480 480(?) cores and GTX470 448(?) cores.

 

[ugaboga232]

Also, you can't forget both the 5870 and 5850 overclock like champs. Signs point to the Fermi being pretty crappy OC'ers. You can even overclock a 5850 to pass a 5870 so I would still get a 5850 over a 470.

Besides, who actually plays crysis with 8xaa. I can understand 4x but 8x seems pretty insane. There are also other configs that give better graphics with less strain.

 

[nitromullet]

Not quite.

It seems to be about equal in Crysis with 8xAA and worse with 4xAA.

But it is only a game and with dubious source benches.

The other bench for example still has 5870 somewhat faster with 8xAA (no 4xAA numbers).

DiRT2 DX11 19x12 / Ultra 8xMSAA:
5870: 59.2 / 51.9 (Avg / Min)
GTX470: 53.8 / 41.6

So we can't even talk about the GF100 taking a lesser hit when going from 4xAA to 8xAA compared to 5xxx.

Yeah, it's no secret that between the GTX 470 and 5870 the 5870 is overall the faster card. I imagine most people already assumed this before looking at the benchmarks.

The relatively minimal loss in performance for the GTX 470 and the extremely low min fps for the 5870 when going from 4x to 8x MSAA is about the only noteworthy thing in these benchmarks. Otherwise, they are fairly in line with the expectation that the 5870 will outperform the GTX 470.

 

[pmv]

The real issue is that minimum means nothing without a plot (or at least more info). Mean represents what it is, and we all understand its significance.

I'm not sure why some reviewers started throwing min and max in there, differentiation from the other guy likely, but it is too vague to be of value unless as part of a set of values with some information on time and method.

Ideally what you want is a graph of "frame rate" on the x-axis against "number of seconds with that frame rate" on the y-axis. So you'd see the true distribution of frame rates across the length of a demo or whatever. Can't remember if I've ever seen that on a review or not.

 

[nitromullet]

Ideally what you want is a graph of "frame rate" on the x-axis against "number of seconds with that frame rate" on the y-axis. So you'd see the true distribution of frame rates across the length of a demo or whatever. Can't remember if I've ever seen that on a review or not.

HardOCP does that. I'm not sure why more review sites don't, FRAPS hands you that graph on a silver platter out of the box.

 

[SlowSpyder]

Not that Fudzilla has been very accurate regarding Fermi thus far, so I don't know how much faith to put into this, but according to Fudzilla the GTX480 has a 300 watt TDP and the GTX470 is around 220 watt TDP. Meh.

http://www.fudzilla.com/content/view/17921/1/

 

[rahvin]

Not that Fudzilla has been very accurate regarding Fermi thus far, so I don't know how much faith to put into this, but according to Fudzilla the GTX480 has a 300 watt TDP and the GTX470 is around 220 watt TDP. Meh.

http://www.fudzilla.com/content/view/17921/1/

Everything Fudzilla is saying right now is what Charlie said a month ago, the only difference is Fudzilla is putting a heavy nvidia friendly (and press department approved) spin on it. They have basically at this point confirmed the GF100 is running low yields, will be very hot, very power hungry, won't overclock and that there is going to be a very limited number of the cards. Recent rumors are there are going to be less than 10,000 of these cards worldwide. Lets face it, the GF100 is going to be a non-existent collector card. If you are waiting for it you are better off waiting for the second version that will hopefully fix all the problems nvidia has encountered with Fermi 1. Charlie, even if he does have an anti-nvidia bias, clearly has a source within the supply chain that nvidia uses that's been feeding him information as he's been dead on with nearly every prediction he's made on Fermi going back to mid last year. Hate him or love him you can't discount his info just because he's biased, everyone is biased to one degree or another. Fudzilla is just on the other side and with Fudzilla giving the Nvidia press department spinned version of the news.

Both ATI and Nvidia blow it about every 3-5 generations and completely blow a release by either completely messing up production of the chip or fail to deliver on time and in quantity or they blow the performance curve and end up with a weak chip. ATI did it 2 generations ago and it's Nvidia's turn to blow the production of the chip and basically miss a generation. Its unfortunate but if anything Nvidia is overdue to blow a chip release, with such tight schedules and such large chips on such new processes it's a given it's going to happen every few generations, don't let your Fanboism blind you to that reality.

We'll see in a fairly short time but I bet GF100 is non-existent in the sales chain. The rumors of BFG and EVGA switching to ATI are also indicators that the GF100 isn't going to be available in quantity. The secrecy and rumors combined have led me to conclude that GF100 has been a huge miss for nvidia and they are simply trying to cover it up as long as they can in the hope they can get Fermi2 out before they do extended damage to their reputation.

 

[Hauk]

Everything Fudzilla is saying right now is what Charlie said a month ago, the only difference is Fudzilla is putting a heavy nvidia friendly (and press department approved) spin on it. They have basically at this point confirmed the GF100 is running low yields, will be very hot, very power hungry, won't overclock and that there is going to be a very limited number of the cards. Recent rumors are there are going to be less than 10,000 of these cards worldwide. Lets face it, the GF100 is going to be a non-existent collector card. If you are waiting for it you are better off waiting for the second version that will hopefully fix all the problems nvidia has encountered with Fermi 1. Charlie, even if he does have an anti-nvidia bias, clearly has a source within the supply chain that nvidia uses that's been feeding him information as he's been dead on with nearly every prediction he's made on Fermi going back to mid last year. Hate him or love him you can't discount his info just because he's biased, everyone is biased to one degree or another. Fudzilla is just on the other side and with Fudzilla giving the Nvidia press department spinned version of the news.

Both ATI and Nvidia blow it about every 3-5 generations and completely blow a release by either completely messing up production of the chip or fail to deliver on time and in quantity or they blow the performance curve and end up with a weak chip. ATI did it 2 generations ago and it's Nvidia's turn to blow the production of the chip and basically miss a generation. Its unfortunate but if anything Nvidia is overdue to blow a chip release, with such tight schedules and such large chips on such new processes it's a given it's going to happen every few generations, don't let your Fanboism blind you to that reality.

We'll see in a fairly short time but I bet GF100 is non-existent in the sales chain. The rumors of BFG and EVGA switching to ATI are also indicators that the GF100 isn't going to be available in quantity. The secrecy and rumors combined have led me to conclude that GF100 has been a huge miss for nvidia and they are simply trying to cover it up as long as they can in the hope they can get Fermi2 out before they do extended damage to their reputation.

Well said sir. Perfect wording for Charlie, that he has a mole within the supply chain. At the TMSC level I'm betting. A clean room employee with access to lot details for example. Someone's been getting off on leaking details; Charlie's happens to be the moron who's receiving it. Like a kid with a gun, he's dangerous with it.

Regarding nVidia, your bolded comments make a lot of sense. I keep thinking about their direction with monolithic gpu's. If Fermi can't compete in areas we scrutinize (performance, power consumption, thermals, cost, etc), it may be chalked up to a painful progression down that path. ATI's gamble has paid off. Even a die hard nV fan can admit the last two generations for ATI have been smart ones.

Who knows though, maybe we're wasting energy. Maybe Fermi will be an available, stellar performer and won't cost much more than competing product. [smiles slightly]

 

[nyker96]

I somehow don't believe these benchmarks. gtx470 should be faster. Anyhow have to wait until they come out with it to know for sure but I do expect 470 to be faster than 5870.

 

[FerraraZ]

I somehow don't believe these benchmarks. gtx470 should be faster. Anyhow have to wait until they come out with it to know for sure but I do expect 470 to be faster than 5870.

Nvidia fanboy in denial

 

[SlowSpyder]

When AMD released the 2900XT, while it was a dud with enthusiasts, it gave AMD a good architecture to build from. We'll have to wait and see how Fermi performs when released, but I have a feeling it'll be similar for Nvidia. It's late, rumors are it's hot and uses too much power, and I'm willing to bet the performance will be underwhelming for the power used/heat and how late it is.

 

[Soleron]

When AMD released the 2900XT, while it was a dud with enthusiasts, it gave AMD a good architecture to build from. We'll have to wait and see how Fermi performs when released, but I have a feeling it'll be similar for Nvidia. It's late, rumors are it's hot and uses too much power, and I'm willing to bet the performance will be underwhelming for the power used/heat and how late it is.

The G80 (8800 Ultra) was about 30% faster than the R600 (2900 XT). The G80 has a die size of about 480mm^2 and the R600 420mm^2. That gives a perf/mm^2 advantage for Nvidia of about 15%.

If a GTX 480 performs 5% faster than a 5870 (and this is the big assumption of course) and I use Charlie's die size of 550mm^2 versus Evergreen's die size of 336 mm^2, that's a perf/mm^2 advantage for AMD of about 50%.

That's a much bigger gulf to make up in a future fix to Fermi, without a completely new architecture. If Fermi is as bad as the rumours say then even if they pull off an RV770 they'll still be significantly behind. If they do two RV770s in a row then maybe, but that would be two years away even for the best engineers. Also it would require an internal workflow change (like AMD went through with small-die in 'the RV770 Story') and there's no evidence Nvidia would consider such a change, or even have the self-reflection to enact it.

 

[Gannef]

For me the most interesting of all this is that it seems the GTX480 isn't going to have 512 shaders. So it seems GTX480 480(?) cores and GTX470 448(?) cores.

I think your wrong on this point, the GTX480 will probably have 512 cores

The chip exists out of 16 parts, each contain 32 cores
Probably the GTX 470 has 1 or 2 defect parts losing 2x32=64 cores
512-64= 448 foila

 

[MrK6]

When AMD released the 2900XT, while it was a dud with enthusiasts, it gave AMD a good architecture to build from. We'll have to wait and see how Fermi performs when released, but I have a feeling it'll be similar for Nvidia. It's late, rumors are it's hot and uses too much power, and I'm willing to bet the performance will be underwhelming for the power used/heat and how late it is.

Agreed, and I like the 2900XT parallel. While the 2900XT was late, ran hot, and didn't perform as well in games, it was a benchmarking king. Fermi might have such a scenario, and it will be the king of F@H or something. It'll be a niche part, and the next generation Fermi might make some improvements in the gaming department. All that said, I still just bought two 5850's .

 

[Hauk]

The G80 (8800 Ultra) was about 30% faster than the R600 (2900 XT). The G80 has a die size of about 480mm^2 and the R600 420mm^2. That gives a perf/mm^2 advantage for Nvidia of about 15%.

If a GTX 480 performs 5% faster than a 5870 (and this is the big assumption of course) and I use Charlie's die size of 550mm^2 versus Evergreen's die size of 336 mm^2, that's a perf/mm^2 advantage for AMD of about 50%.

That's a much bigger gulf to make up in a future fix to Fermi, without a completely new architecture. If Fermi is as bad as the rumours say then even if they pull off an RV770 they'll still be significantly behind. If they do two RV770s in a row then maybe, but that would be two years away even for the best engineers. Also it would require an internal workflow change (like AMD went through with small-die in 'the RV770 Story') and there's no evidence Nvidia would consider such a change, or even have the self-reflection to enact it.

Exactly. And after reading articles like AT's on next gen ION, I can't help but wonder what percentage shift of resources is happening internally.

It's obvious they're eyeing the cheddar: http://images.anandtech.com/reviews/mobile/2010/next-generation-ion/ion-tng-netbook-growth-sm.jpg

 

[zod96]

I smell a nice price war coming If the 5870 will still be faster than a GTX470 and from what I've heard the GTX470 is going to be about $400 then I bet Ati lowers the price on the 5870/5850 cards. You'll be able to get the 5870 for about $280-320 not bad for a card that can play Crysis pretty much maxed out yet uses 50% less power the the GTX470/480 cards and runs alot cooler...

 

[Hauk]

I smell a nice price war coming If the 5870 will still be faster than a GTX470 and from what I've heard the GTX470 is going to be about $400 then I bet Ati lowers the price on the 5870/5850 cards. You'll be able to get the 5870 for about $280-320 not bad for a card that can play Crysis pretty much maxed out yet uses 50% less power the the GTX470/480 cards and runs alot cooler...

Prolly not that much of a price break. They could easily do $359 though. Would be nice to see it, and the 2GB version launch at $419-$449

 

[tviceman]

Remember that the tesselation performance is better in a tessellation benchmark. NV hardware uses shaders to do the tessellation heavy lifting.

Wrong.

 

[tviceman]

That's a much bigger gulf to make up in a future fix to Fermi, without a completely new architecture.

Nvidia did not bring a radically new architecture when going from the fx5800 ultra to the fx5900, yet the fx5900 pretty much "fixed" much of what was wrong with the FX series all in less than 5 months to market.

 

[Acanthus]

Those are speed holes; they make the card go faster.

 

[MagickMan]

Nvidia fanboy in denial

AMD fanboy throwing out bait?

 

[bryanW1995]

Wonder why they would use such a slow CPU and only 2GB RAM. That's of course if any of this is legit.

maybe bfg10k was doing the review and...never mind...