NVIDIA's GeForce GTX 480 finally unleashed. Reviewed and rated.

[Terbinator]

Tessellation FTW

I was thinking more of possible optimisations to come in future games. I've not really formed an opinion on them overall yet, I'm guessing its going to be a few driver releases before we see some proper performance figures - more so seeing what ATI pulled out the bag with 10.3.

[Zhaoman]

tbh, that was marginally better than I was expecting. I wasn't expecting much, though.

You really mustn't have expected much at all because I felt it was a lot worse than I was expecting. While the architecture is interesting the fact that the power draw, the price and the heat AND the fact that they couldn't even make the product as intended just shows how bad a design it was. The overclock is meaningless imo considering the power draw and heat already: I would not feel comfortable until better coolers get slapped on the card. With all the hype and rollo threads and 'leaked' slides I was half expecting a 5970 killer (half hoping is more accurate). But instead we have a card that is really pushing the thermals with a ridiculous power draw that doesn't even challenge a 5890 and just about bests a 5870. I think it's a no brainer when it comes to 5870 vs 480GTX and that is a disappointment for me :/

[mercyground]

given how much of a Bacon Fryer it is... Guess we know why Rollo was being pimped to hype it up and make sure ppl held on and didnt buy even MORE ATI cards.

Now i like bacon sandwiches at my computer... i just dont like my computer making them...

[ProbablyJoe]

The benchmarks were about what I expected... maybe a little lower, I was expecting the 470 to beat the 5870 too. The performance isn't the issue though. If they'd released these cards at identical price points to the 5850 and 5870, then there'd be some competition. And sure, tesellation, PhysX, and driver updates might well boost them further in the future, but right now the price just makes them hard to recommend.

That, and the fact that I just wouldn't feel comfortable putting one of these on a 650w power supply, or in anything without some high graphics cooling.

[dfour]

What better cooling are you after?? It come with 5 direct contact heatpipe cooler that could cause potential size issues in some cases already. What to you want?? Its own seperate watercooling loop installed??
Think how much that would be

This card would be what nvidia wanted it to be (and rollo) if they could transition to the 32nm which wont be out for a while yet. It seems likely ATI's fermi killer will also have to be done on 40nm so there is potenetial 'fermi' problems looming for ATI

[GoNz0]

i just hope EVGA dont take the piss to much with the water cooled version as i wouldnt buy a stock card anyway, i also hope Nvidia manage to build a better driver as for now they seem to have shot themselves in the foot keeping to a release deadline that they cant even sell a card at.

i have not dissed it yet, and i wont jump ship to ATI just yet either!

[kalniel]

Depending on where the newer games go in a design sense Fermi may end up blasting the ATI cards due to the superior tessellation offered.

I think by the time games come out that will really need that kind of tessellation both vendors will have had several new generations of products out and the relative performance of the 480 will be insignificant.

This card would be what nvidia wanted it to be (and rollo) if they could transition to the 32nm which wont be out for a while yet. It seems likely ATI's fermi killer will also have to be done on 40nm so there is potenetial 'fermi' problems looming for ATI

Why do you think that? ATI cards have been using 40nm for ages and it seems they are able to design cards that mitigate many of the problems with the TSMC manufacturing process. If they go back to a huge chip design (which they nearly did) rather than the smaller efficient chips that they've been doing recently then maybe, but I would hope that someone at ATI recognises the success their approach has brought them.

As for a 32nm transition of fermi.. don't hold your breath - 40nm is new enough for nVidia.

[DR]

One thing which is always interesting to see is driver performance after 3 to 6 months.

AMD have released Cat 10.3 - Link: http://www.hexus.net/content/item.php?item=23028

This unleashed a ton of performance on their SKU's - makes me think how long will it be until NVIDIA unleash/find the extra performance.

All I can say for sure is this is excellent for the consumer - you've got another choice and even if this isn't what you wanted it to be, imagine it with a die shrink etc.

[kalniel]

One thing which is always interesting to see is driver performance after 3 to 6 months.

AMD have released Cat 10.3 - Link: http://www.hexus.net/content/item.php?item=23028

This unleashed a ton of performance on their SKU's - makes me think how long will it be until NVIDIA unleash/find the extra performance.

The cynical/realistic among us might argue that AMD deliberately held some performance back in drivers because they didn't want nVidia to know the exact performance to aim for with the 480/470, so while they might have found a few percent (about the amount the 4000 series improved by perhaps), the rest could have always been there.

nVidia might be doing the same - controlling performance just enough to beat the competition, however I think that's less likely to be the case as with the cost/power/heat disadvantages many people were looking for more performance to justify it. But we'll see.

I'd like to see if nVidia responded to AMDs 4000 series success - the response might be the next gen of cards, or maybe the one after, but I'd love to see them turn their talents to smaller more efficient chips.

[cjs150]

As I am stuck on 1680x1050 resolution I see little point in changing from the 9800 I have.

This 480 is crying out for watercooling - problem is that you will need a 240 rad just to cool this baby, but does mean someone could come out with a single slot design which would be nice.

[scaryjim]

Funny, I don't remember typing

Tessellation FTW

...

But I was impressed by the huge difference in image quality you get from a small amount of tessellation

You really mustn't have expected much at all ... The overclock is meaningless imo ... I would not feel comfortable until better coolers get slapped on the card. ... I was half expecting a 5970 killer ... I think it's a no brainer when it comes to 5870 vs 480GTX ...

Well, they got the card out with more shaders, at higher clocks and with lower power draw than most of the internet rumours were pointing to, which suggests they were sneakily hiding away the top binned silicon. To get the 480 shader card working @ 1400MHz while "only" pulling 250W is a reasonable bit of engineering. As to OC, sure it'll need better cooling, but there's > 10% more performance sitting there within the 300W PCIe power envelope - again, internet rumour suggested they were running on the ragged edge and I was expecting it to fall over within 5% of the stock clocks.

A couple of weeks ago I expected Fermi to struggle against the 5870 across a range of games, winning in a couple but being behind in more. When the specs came out - i.e. 480 shaders @ 1400MHz - I felt that they'd probably done enough to stay consistently ahead of the 5870, but I anticipated a power draw in the 290s and no overclocking headroom. So overall, they've done a little better than I thought they would. But yes, I had pretty low expectations ...

And as to 5870 vs GTX480 - as you say, complete no-brainer. Even if someone gave me a GTX480, I'd probably sell it and buy myself a 5870 instead

[kalniel]

To get the 480 shader card working @ 1400MHz while "only" pulling 250W is a reasonable bit of engineering.

The problem is they haven't. They've lowered the TDP figure arbitrarily, as the testing seems to indicate it can pull much more than that.

It's another of those 'how do you compare different TDP quotes from different companies' thing, but nVidia have changed their system compared to how they used to do it. When a 480 with '250W' designation pulls more power at both idle and load than a 5970 with a '295W' designation it tells you something is up.

[GheeTsar]

One concern I have about the noise/heat issue with the 480 is that it's hard to imagine scope for aftermarket air coolers to outperform the stock cooler given its size. I guess it's watercooling or bust if you want it to use it at a sensible volume.

I feel slightly vindicated in jumping ship to ATI now - I feared I may have leaped to early (based on Charlie's articles, which have been wrong in the past).

[TooNice]

The cynical/realistic among us might argue that AMD deliberately held some performance back in drivers because they didn't want nVidia to know the exact performance to aim for with the 480/470, so while they might have found a few percent (about the amount the 4000 series improved by perhaps), the rest could have always been there.

I find that hard to believe. By the time the 58xx was released, Fermi must've already been so far into the development that it's not really feasible to go back to the drawing board. I do not doubt that they they intentionally put a lot of effort to ensure that the 10.3a would be out on time for Fermi, but I don't think they deliberately throttled back performance for the first 3-5 months of the release. nVidia was most likely shooting for 512 cores long before the 10.3a (and I am sure that scaling back to 480 is not intentional either).

Remains to be seen how long it will take for Fermi's refresh to be out. The 2900XT lasted 6 months (May 07 - Nov 07) before AMD came up with something that was not really better, but at least competitive with the 38xx series. If nVidia can get Fermi to 512 cores, and decrease power consumption by a chunk, I foresee the headlines 'Fermi, the way it was meant to be' being used by the end of the year. It doesn't have to be lower than AMD to be competitive, just as the 3870 did not have to be faster than the 8800GT to be considered competitive, and what the 2900XT should've been.

What would be interesting to see too, is what AMD will bring to the table. They've had more time to work on something even better, but they are for the time being, under less pressure.

[TooNice]

You really mustn't have expected much at all because I felt it was a lot worse than I was expecting. [...]

Well, if you were going by Rollo and ignoring every other rumours on the GTX480, it's not surprising that you would be disappointed. It's a big chip, so 'hot' was the expectation. The performance 'leaks' we've seen showed a fairly wide range, from very little (Dirt), to rather impressive (FC2). People were also saying that if the Tesselation intensive benchmark nVidia showed was the best they can do, then they are in a whole load of trouble. I am on the same page as scaryjim, I expected that it would be underwhelming overall, and it did not disappoint

[dfour]

Why do you think that? ATI cards have been using 40nm for ages and it seems they are able to design cards that mitigate many of the problems with the TSMC manufacturing process. If they go back to a huge chip design (which they nearly did) rather than the smaller efficient chips that they've been doing recently then maybe, but I would hope that someone at ATI recognises the success their approach has brought them.

Exactly my point. The next chip from ATI will be a new archtecture and its unlikely that it will be the same size or smaller than this one (it would be awesome if it was though). they will have to deal with the problem of bigger chips producing more heat/requireing more voltage especially as they will (most likely) be stuck on the 40nm for at least some of the chips life. Look what they did with the 5970 in scaling the speed of it back to remain in the 300 watt envelope.

It has to kill fermi v1 dead as it will have to be competative v's fermi v2.