GeForce GTX 480 And 470: From Fermi And GF100 To Actual Cards!

[Guest]

Physix was disabled. These are not benchmarks that will accurately represent the overall real-world experience.
Though the reviewer stated they just wanted to benchmark pure graphics performance, it would have been very useful to run Physix enabled to see what real world performance would look like.

I'm not sure if it was mentioned in the review (I may have missed it), but one of the big changes was a lower penalty hit for GPU to GPGPU context switches.

Personally, I'm thrilled with it. I Have zero intention of buying one for gaming. However, I've got a c2070 Tesla card on pre-order and Fermi is everything we hoped it'd be. I've got a research who is just starting to port his Matlab code to Jacket on the c1060. I can't wait to see what the performance delta will look like. Our hope is that he'll be able to do real-time analysis on a dozen data streems on a single CUDA workstation. We had previously planned to do on a compute cluster.

This is where NVidia is going. It's still a small market compared to desktop video boards but consider this. I've got 200 cluster nodes coming in for one researcher. He's interested in GPGPU coding. If he goes CUDA (openCL) that's over $2000 per card. We're talking $400,000 in sales, the vast majority of which is profit.. for one researcher. Urbana has thousands of GPGPU compute nodes.

ATI has totally failed in this area with FireStream. It's not their cards or their pricing that's a problem as much as their implementation. All our people run Intel again. Core was a game changer and nehalem put the nail in the coffin. Everything we're buying is Intel Xeon.
Firestream is implemented as an extension to the AMD compiler which generates binaries optimized for Opteron. Firestream has been buggy and it runs like crap on Intel systems.

OpenCL still has the potential to reverse AMD's GPGPU fortunes but, for now, they're on the outside looking in.

 

[jdp245]

After reading this, I'm still happy with my 5870 purchase. But, I am impressed with Nvidea's tessellation and AA performance. I just don't think it would have been worth the $100 more to get a 480.

 

[randomizer]

At any rate, with 15 pages now, I'm starting to feel Tom's should find a way to re-vamp their comments section, 'cuz this is ENTIRELY too long to browse through page-by-page... Either options for skipping multiple pages, options for larger pages, or perhaps a "view all" option would be nice.

Click the "Read the comments on the forums" link at the top of the comments section. You'll get 50 posts per page instead of 20, and you can edit.

 

[Guest]

I was waiting for the fermi cards to come out before my next high end build( looking for price drops), but I actually did not expect this card to be this fast. The GTX480 is ~15% faster than the 5870, but for $100 more, it is just gonna be a Nvidia loyal card, and the 5870 will probably drop just a little if at all.. The 5850 and and 5830 should drop $25-50. Now, would I like to have a fermi, well yeah for sure, but I would much rather have a 5870 and down the road add another. A GTX 480 still uses the same, if not more power than (2) 5870's. Now this reminds me of the last gen of the P4's. or as we know em, the Preshots. Basically, Nvidia's idea of a huge chip, with yes impressive performance, was just the wrong approach. I mean, their next-gen, if based on this same doubling architecture would draw 300w+ easily and almost require water cooling because the next TSMC process is going to be 32nm and that will not allow them to "cut the chip in half." ATI's theory started with the 4000 series has proven to be a much better/efficient design. I think they could make a 6870 with 40nm TSMC right now, but ofcourse it would be a hot chip. Now when they get the 32 TSMC FABs running, Nvidia has got to re-design their chips.. And with how hot the GTX 480 is, I dont see how they could make a GTX 495. Also, the 5890 is right around the corner and that should put the final punch to KO Nvidia in this GPU generation. On a side note, Thanks " " that there is some healthy competion or AMD might pull what Nvidia did and rebrand the 8800 5 or 6 times.

 

[dragonsqrrl]

umm the new ati cards are coming vary soon......what is nvidia trying to do anyway......2 simple little words can be used to describe this "EPIC FAIL"

How Sad.... all my previous GPU's were Nvidia starting with the Fx's looks like Iv'e been holding my breath for nothing, oh well... I'm gonna be seeing red on my next build. maybe AMD will be rolling out the HD6000's before Nvidia get their low range 400's out.... if they plan on making them, or can make them after this blunder.

So with only 6 months left before ATI's next model cycle, I guess I'm stuck waiting for that to help me out.

This response is meant as food for thought for everyone who is waiting for ATI's next gen cards, or who thinks there's only "6 months left before ATI's next model cycle".

ATI has been concentrating on producing smaller and more scalable GPU's since the Radeon HD3000 series, but because of this I simply can't see ATI's next generation architecture debuting before the end of the year. ATI's business model has shifted drastically as a result of their current generation architecture, that of the R600 (HD2900 XT). The primary focus in designing and implementing their hardware is now centered around 'being on time' and 'keeping on schedule' rather then the introduction of new features or even higher performance. Any new feature or expansion on an existing feature that would delay the introduction of a product generation is either dropped or stripped down to conform to this schedule. ATI's Carrell Killebrew explained this new approach to GPU production in a recent interview, summing it up in two broad areas of focus, smaller/scalable GPU's, and 'showing up to the fight' on time.

What this basically means is that ATI's next gen cards, AKA HD6000 series, won't be manufactured on the 40nm process, and therefore probably won't come to market this year. Even if ATI were to go against their new business model and produce a GPU with a large die area, there's still basically no way it could be manufactured on the 40nm process given the boost in transistor count. This is unless ATI comes across a breakthrough low transistor count architecture that either preserves or slightly increases the GPU die area while providing next gen levels of performance, however as you might have guessed I find this extremely unlikely. The 40nm process is already at it's limits in terms of transistor count, and therefore die size. Observing transistor counts basically doubling from generation to generation means that a next gen architecture on 40nm is impossible. Gone are the days of the G71 (Geforce 7900GTX) and G80 (Geforce 8800GTX), separated by nearly a year but both produced on the same 90nm process.

Why such a concentration on 40nm? Because TSMC's 40nm process is still young, still ramping up production, and as a result still hasn't reached it's full yeild and volume level potential. The current generation manufacturing process began back in 2008. It took over half a year before the first low/mid range GPU came to market, albeit plagued by manufacturing issues, and over a year before we saw the first high end GPU's come to market, those found in the Radeon HD5000 series. Even over a year and a half after TSMC introduced 40nm, it still hasn't reached production levels seen in past 65nm and 90nm generations, both of which also saw long life spans. TSMC's 28nmHP process is their next "full node" production tech (widely thought to be the next gen step for ATI/Nvidia), and while it's scheduled for introduction in H2 2010, it's highly unlikely we'll see a high end, complex, and high transistor count GPU produced at 28nm coming to market by the end of the year, much less one that's built on an entirely new architecture. It's far more likely we'll see either current gen revisions or low/mid range GPU's first, before a high end GPU is ever produced at 28nm (much like the 40nm, 65nm, and 90nm generations).

Factor in all the aspects of the situation, the fact that this will be ATI's first new architecture since R600, that ATI needs to integrate enhanced GPGPU compute functionality if it wishes to remain competitive with Nvidia and even Intel, that new GPU architectures often require much more R&D, more time, and larger amounts of resources then architectural 'refreshes', that the 28nmHP process will still be very immature even by the end of the year, that ATI's new business model prioritizes smaller die sizes, and you have a situation where ATI (as well as Nvidia) needs to use the 28nm process for it's next gen GPU's. Except Nvidia's next product generation will likely be a refresh of the Fermi architecture, a major but important 'future oriented' shift in design that had to be implemented sooner or later. And as anyone who has read in-depth articles on the Fermi or Larrabee architectures knows, making a powerful gaming card and a powerful GPGPU card require two entirely different sets of strengths, and it's extremely difficult to develop an architecture that can excel at both.

So no, I don't think there's any way ATI's next gen GPU will debut just one year after the HD5000 series (6 months from now), and I think it's highly unlikely we'll see it by the end of 2010 either. It could realistically come out in Q1 or Q2 of 2011 at the earliest.

 

[azgard]

Wow, Tom's readers are a bunch of clueless kids now. Sigh, use to have rather intelligent people who used to discuss rationally but it's now populated by fanbois, gamers, and just stupid people.

Let's get this straight, looking at the architecture of this card and driver/software support that Nvidia has been heading for, this card is NOT focusing on the gamer side, this card is geared much more towards the professional/scientific community. If you think otherwise, then you don't belong on here posting as you cannot think past what only matters to you. At $500, that's a bargin price for these communities to try out and use. You can bet that Nvidia is eyeing various aspects of the supercomputing market along with other profession applications such as from Adobe.

I myself was hoping for better game results, however I also see where they are heading in the future. And currently, they are quite a bit ahead of AMD/ATI and Intel for that matter, a rack of these chips with mature drivers will beat anything those 2 companies could come up with for straight computational power.

That aside, they also did a whole new rework for dx11 which ATI will have to do to keep up in the future.

Toms, games and most of your benchmarks are NOT the end all be all of computing. I understand you don't do alot of enterprise hardware, but you should have realized what this card represents and as such included that in your review. As it sits, you have nearly everyone saying it's an epic fail when in fact it holds tremendous potential. This isn't Tom's Games Site yet.

This is just a cop out of Nvidia's failure. According to Nvidia they seem to think this is a gaming card and that this is the computing card. Keep your fact's straight and don't try to skew them.

 

[dragonsqrrl]

I was waiting for the fermi cards to come out before my next high end build( looking for price drops), but I actually did not expect this card to be this fast. The GTX480 is ~15% faster than the 5870, but for $100 more, it is just gonna be a Nvidia loyal card, and the 5870 will probably drop just a little if at all.. The 5850 and and 5830 should drop $25-50. Now, would I like to have a fermi, well yeah for sure, but I would much rather have a 5870 and down the road add another. A GTX 480 still uses the same, if not more power than (2) 5870's. Now this reminds me of the last gen of the P4's. or as we know em, the Preshots. Basically, Nvidia's idea of a huge chip, with yes impressive performance, was just the wrong approach. I mean, their next-gen, if based on this same doubling architecture would draw 300w+ easily and almost require water cooling because the next TSMC process is going to be 32nm and that will not allow them to "cut the chip in half." ATI's theory started with the 4000 series has proven to be a much better/efficient design. I think they could make a 6870 with 40nm TSMC right now, but ofcourse it would be a hot chip. Now when they get the 32 TSMC FABs running, Nvidia has got to re-design their chips.. And with how hot the GTX 480 is, I dont see how they could make a GTX 495. Also, the 5890 is right around the corner and that should put the final punch to KO Nvidia in this GPU generation. On a side note, Thanks " " that there is some healthy competion or AMD might pull what Nvidia did and rebrand the 8800 5 or 6 times.

Please read my previous comment for further insight into this situation... as it is right now, ATI/Nvidia's next gen cards will more then likely be manufactured on TSMC 28nm process.

 

[cmartin011]

i betcha after a couple drivers they may better harness the power of this GPU like they always do. Perhaps throw more power for FPS instead of the tessellation wizardry when needed. though agreed to little to later o well.. sticking with 295 gtx till they build one twice as fast for same power or less from ati or nvidia. Kinda a bummer they thought high power was the way to go with all this heat i can't see them lasting unless in a extremely ventilated enclosure. luckly my antec 1200 is perfect for my 295

 

[luke904]

nvidia has brought cards that perform how they should considering their prices and the current market of video cards (using recommended retail pricing, 5870=389, 5850=289)

the problem is that they are just releasing them while AMD has had theirs out for ~6 months and thus are 6 months ahead of nvidia to work on the next generation

also- the temps and power requirements are insane


overall
i would buy a 470 over a 5870 because its cheaper (350 vs 389) and has roughly the same performance, i dont care about power usage (im not the one who pays for my electricity..lol)

 

[ohim]

[citation][nom]luke904[/nom]nvidia has brought cards that perform how they should considering their prices and the current market of video cards (using recommended retail pricing, 5870=389, 5850=289)the problem is that they are just releasing them while AMD has had theirs out for ~6 months and thus are 6 months ahead of nvidia to work on the next generationalso- the temps and power requirements are insane overalli would buy a 470 over a 5870 because its cheaper (350 vs 389) and has roughly the same performance, i dont care about power usage (im not the one who pays for my electricity..lol)[/citation]
sorry to interrupt you but which reviews did you saw ? since the 470 is under the 5850 ... that means nowhere near the 5870 .. and given the fact that the 470 alone eats about 100 W more than the 5850 ... the 470 makes no sense at all in buying.

 

[scrumworks]

[citation][nom]kutark[/nom] Now, i don't wanna screw with it. So i buy an nvidia card, and yeah i pay a bit more for it, but it just works.[/citation]

Or those drivers might just fry your card. You never know after it hits you.

 

[socal6801]

100 bucks more than a 5870? What's 100 dollars. Nvidia has always had great driver support and good overclocking ability. I will easily shell out the extra cash for the fastest single GPU card available. Actually Two. For anyone in the conversation about high end graphics, worried about power draw. Go buy a Mac.

 

[carlos0248]

我觉得，这个产品就就像编辑说的那样，性能很强劲，功耗很大，价格比较贵，不用希望这样的产品能够导致ATI的显卡降价。

I thought the GTX480 just like editor said that the best performance but the price and power consumption was higher. Don't count it can cause ait drop their price.

 

[brisingamen]

i see within 3 months a revised 5870 with better yeilds, and a 6xxx possible this year because global foundries is producing perfect cpu wafers at 40nm, thats down right amazing, with acceptable 32, and 28nm chip yeilds, to be honest the 5870 doesnt even need a revision, just more hardware and clock speed and the sky is really the limit,

i also read on the amd website im quite sure that the 6xxx is scheduled for august, with a few hiccups, october.

until then you will have higher memory and superclockable 5870's. revised or just awsome yeilds.

ati is on a yearly schedule for thier graphics program,
october 08 for the 4870
october 09 for the 5870
october 10 for the 6870????

i dont see them softening up just because nvidia is having trouble, the gtx480 and subsequent parts will continue to get faster and run cooler.

here is an article on global foundries showing how much better they have things going than tsmc,

http://www.brightsideofnews.com/news/2010/3/17/globalfoundries-produces-zero-defect-wafers2c-10025-yield.aspx

 

[XxOsurfer3xX]

Ill wait for the price cuts and buy a couple of 5870s, better performance than the 5970 and less heat, i now have a 4870 X2 and want a less power consuming, less heat card, i do not like the new nvidias just because of that, is not cool to be playing and worrying about the GPU running too hot or the fact that the fans are really loud and you get annoyed by it because no one wants to hear them over the game, so AMD for me is the way to go, and if a 5890 comes out even better, lower the prices even more...

 

[dragonsqrrl]

i see within 3 months a revised 5870 with better yeilds, and a 6xxx possible this year because global foundries is producing perfect cpu wafers at 40nm, thats down right amazing, with acceptable 32, and 28nm chip yeilds, to be honest the 5870 doesnt even need a revision, just more hardware and clock speed and the sky is really the limit,

i also read on the amd website im quite sure that the 6xxx is scheduled for august, with a few hiccups, october.

until then you will have higher memory and superclockable 5870's. revised or just awsome yeilds.

ati is on a yearly schedule for thier graphics program,
october 08 for the 4870
october 09 for the 5870
october 10 for the 6870????

i dont see them softening up just because nvidia is having trouble, the gtx480 and subsequent parts will continue to get faster and run cooler.

here is an article on global foundries showing how much better they have things going than tsmc,

http://www.brightsideofnews.com/news/2010/3/17/globalfoundries-produces-zero-defect-wafers2c-10025-yield.aspx

hmmm... lol, I think we have to clear a few things up here.

ATI GPU's are manufactured by TSMC and not Global Foundries.
ATI has not been on a 12 month cycle for either of the existing GPU generations you've listed, 4870 and 5870. Both the dates you provided are incorrect, the HD4870/4850 were released in June 08, while the HD5870/5850 were released in September 09, and as for the rumored HD6000 series, well... I suggest you read my previous comment for more information.

 

[brisingamen]

maybe you should re-read my post then,

i never said global foundries made ati gpu's, i posted a link, showing how well they are doing in the 40nm, 32nm and 28nm process vs tsmc, which is alot, and in buisiness things move fast, things are very competitive,

i was implying that, . .

since 6xxx series is a completely new architecture they could switch,
i dont work for ati so i dont know how thier contracts work but, things can change, if tsmc continues to struggle they wont just be handed contracts forever,

but you were are little to dense and self absorbed in your own posts to pick up on that.

my money is on 6xxx being on the market before jan 1st 2011. how about that? thats my opinion, and your entitled to yours.

i did read a press release from an ati exec stating 6xxx series was due out in august, yes that could change but, why?

alot can happen in six months with the 28 and 32nm yeilds, is it really so hard to believe that global foundries to have really great yeilds on the 6xxx series design by august? not at all.

and yes your entitled to your prententious conjecture, as distasteful as it is.

 

[Dan__]

Sapphire 5970 OC 4GB

Cebit 2010: Claims its the fastest in the world

Pictures...

http://www.fudzilla.com/content/view/17873/1/


With the launch of NVIDIA’s GF100 series cards drawing near, ATI really doesn’t have anything to quell the hype of the release, or do they?


At CeBIT this morning, HEXUS.net managed to catch a quick glimpse of a graphics monstrosity by Sapphire, a 4GB 5970 OC. The card features more than just doubled ram however; whereas the current 5970 utilizes only a dual 5850+5870 GPU mixture (each GPU has 1,600 shaders but only opperates at 725MHz/4GHz), this beast uses two true 5870 GPU’s on a single PCB, meaning 3,200 combined shaders and each GPU running at 850MHz core and 4,800MHz memory.


Sapphire has also managed to keep the full load just shy of 300W but the report indicates that 400w draw – given the extra overclocking headroom – wouldn’t be out of the question. The card requires two 8-pin PCI-E power connectors.


Sapphire has also teamed up with Arctic Cooling to provide a beefy triple cooler design, which manages to keep load tempuratures hovering around the 90C mark.


The rumoured street price for the behemoth, a paltry $1,000 USD.

http://www.hardwarecanucks.com/news/video/sapphire-radeon-hd-5970-4gb-oc-graphics-monster/

 

[notsleep]

intel will probably buy/takeover nvidia when nvidia file for bankruptcy with the failed fermi launch.

 

[booseek]

[citation][nom]Dan__[/nom]The rumoured street price for the behemoth, a paltry $1,000 USD.http://www.hardwarecanucks.com/new [...] s-monster/[/citation]

I'd rather do two 5870s in crossfire for around $800.

 

[snqwerty]

I think that all the disappointment is useless.
We are talking about new architecture, new API ( directx 11 ).
If you want performance gain, then you should wait for the updated revisions of this architecture, like the 8800GTS vs 8800GT.

I believe that we should stick to what we have for a little bit more. Competition will not let us down in the coming months

 

[phantomtrooper]

do remember that this card is more about the other technology it offers and not so much about its performance in current games. the programs that would take advantage of its technology are not out yet.

 

[gbean02]

When are they going to use a larger fan on the video cards.. My system is easy on the ears except for the small fan on my video cards that are screaming like a banshee. How about this?

Dear video card vendors,
For once, put a bigger fan on a card for less noise.

Thank you

 

[Guest]

Could you please verify the idle temps in a multi-monitor setup!

According to this review it raises your idle to 90C: http://www.legitreviews.com/article/1258/15/

If that's true those cards shouldn't have been released.

 

[surda]

from how hot it is, i think i can bake eggs on it...