The GTX 480/470 performance thread

[KainXS]

They use the same tactics depending on which one is on the bottom, the way I see it since the G80's release, and going by release year with the first top dogs

The 8800GTX destroyed the 2900XT and 3870 but costed so much more than the 3870 is was crazy. Then the G92's came out and the 8800GT was also faster than the HD3870 but the HD4850 outperformed it and the HD4870 solidly outperformed the 8800GTS512 and the new 9800GTX but was still being outperformed by the more expensive GT200's. Just before that Nvidia released the GTS250 which still lost to the HD4870 just barely(but nvidia still claims its faster) but at the time the GT200's were all faster but costed more. Then came the HD4890 which outperformed the GTX260 but couldn't outperform the GTX280, then the GTX275 and GTX285 came out which both outperformed the HD4890 but costed more(the GTX285 costed alot more). Then we come to the current craze, the HD58XX both smoked every single gpu card on the market as expected, then 6 months late came the GTX480 and 470, about 40% off from the performance expected and with a disabled SM on the 480 and both running much hotter and both using about 40-50% more power like the GTX280 vs the HD4870 for their respective competitors.

This is just like the the GTX280 vs the HD4870, theirs just a much smaller performance gap this time.

 

[shevanel]

also the hottest sound card ever.

if they can get the temps lower so its not as noisy, hot and costly then this could be a great card...

$500 would be nothing if it were only not so hot.

$400 would make it reasonable. but $500 is just rediculous for this blow dryer

heat/noise is truly the only 2 major cons i see. and they are what makes the 15% gain seem like -15%

 

[Super XP]

Whether he was right or wrong.One thing remains the 480/70 was a balls up as its too hot,too inefficient,not much faster than 58xx.

Agreed. I can see ATI sales going crazy now. Who on earth want's a GTX 480 or 470 in a notebook? No Thank You

also the hottest sound card ever.

if they can get the temps lower so its not as noisy, hot and costly then this could be a great card...

$500 would be nothing if it were only not so hot.

$400 would make it reasonable. but $500 is just rediculous for this blow dryer

heat/noise is truly the only 2 major cons i see. and they are what makes the 15% gain seem like -15%

I would never want to hook the 480 on a water cooling setup, it would heat up the water way too much and screw with all other components temp's. GTX 480 & 470 is an epic failure. Maybe Fermi 2 will be much better.

 

[Super XP]

Darn, the ATI Radeon Hd 5870 1GB still costs $450 here in Canada. That sucks, the price went up ever since Fermi's release

 

[KainXS]

I think if things went right for nvidia and the card actually didn't have many problems, I think we would be looking at a card with all 16 SM's active and the core at 1.05V with a clock of 750/1300 with the same temps they get now, hopefully they will get this design down by the HD485's release but by that time, . . . . who knows what AMD will have up their sleeve,

This is the only time I can remember for a while that AMD has been in lead with a slower card.

 

[qubit]

I would never want to hook the 480 on a water cooling setup, it would heat up the water way too much and screw with all other components temp's. GTX 480 & 470 is an epic failure. Maybe Fermi 2 will be much better.

I think the only thing 'wrong' with it is that the chip design is too advanced for the process tech it's manufactured on. 3.2bn transistors is an awful lot and it appears to be maxing out the 40nm process tech it's built on. I reckon the die shrunk version will be much better.

I'll definitely be waiting for the 'GTX485' version. In the meantime, I'll likely be getting one of the lower-end Fermi derivatives in a couple of months, so that I can play with DX11 & 3D Vision together. (I already have 3D Vision).

 

[Super XP]

All I have to say is AMD is not going to give NVIDIA a chance to recover. AMD's got a HD 5800 re-fresh on the way which should wipe out the current Fermi's along with the HD 6800 series (New Design) right after in around late Q4 2010. With a re-fresh of the HD 6800 series approx: 6-8 months after its original launch.

NVIDIA's problem is heat, power and more heat and power. they are in big trouble, they got no back up solution, they put too much time and effort in Fermi and it failed. I am sure they are working day and night to get something out like a Fermi 2 just until they can figure out what the hell to do and how to make this work. We are talking about 6-12 months away.

Until then AMD/ATI is king of the Graphics. HD 5970 2GB rocks anything right now.

 

[Super XP]

I think the only thing 'wrong' with it is that the chip design is too advanced for the process tech it's manufactured on. 3.2bn transistors is an awful lot and it appears to be maxing out the 40nm process tech it's built on. I reckon the die shrunk version will be much better.

I'll definitely be waiting for the 'GTX485' version. In the meantime, I'll likely be getting one of the lower-end Fermi derivatives in a couple of months, so that I can play with DX11 & 3D Vision together. (I already have 3D Vision).

Don’t get me wrong Fermi is one hell of a complex design. Dam innovative to say the least. NVIDIA actually did a great job, but like you said the manufacturing process is most likely the issue. It’s too bad NVIDIA couldn’t just wait it out a little more so they can get that thing on a 32nm instead of the already faulty 40nm. It probably would have worked out better for them.

But I would like to stress that NVIDIA sort of messed up, you never release a high end part on a new manufacturing process in this case 40nm and hope for the best, you always release your low end cards on something new and learn from them, correct what needs correcting then you release your high end parts.

ATI’s been doing this for a while now and they’ve rectified any possible issues before the real high end GPU launch.

 

[shevanel]

i wouldnt call it fail for nvidia, they will sell all they have.

it's fail for us. i'd buy one if it were worth it but it's not... doesnt mean others will not.

 

[qubit]

Because he is not right at all. He is semiaccurate at best and that works both ways, he can be right or not and most of the times he is not.

When Charlie writes one if his "I told you so!" articles, he always backs it up with many, many links to his previous articles that show him predicting all that has come true. This time is no exception. He's not "semiaccurate" at all.

 

[Super XP]

When Charlie writes one if his "I told you so!" articles, he always backs it up with many, many links to his previous articles that show him predicting all that has come true. This time is no exception. He's not "semiaccurate" at all.

Very true indead

i wouldnt call it fail for nvidia, they will sell all they have.

it's fail for us. i'd buy one if it were worth it but it's not... doesnt mean others will not.

If the heat was right and the price was right they would have had a compeling product. Yes it's a fail for us.

 

[phanbuey]

of course he does!!!... he got it right. Does he ever write "I was way the hell off" articles, when he is wrong as hell? NO! So yeah... of course he links all of the things that say "i told you so" haha.

If you throw enough spaghetti at a wall, you will get some that sticks, thats for sure.

 

[Super XP]

No Charlie admits when he's wrong but really lets people know when he's right.

 

[phanbuey]

No Charlie admits when he's wrong but really lets people know when he's right.

Ive seen the Inquirer admit that theyre wrong. But i have yet to read a Charlie article saying "sorry for being way off".

Charile = Bill O'Reilly

 

[EastCoasthandle]

Because source is biased ,dont want to say more and derail thread .

Edit : I saw a lot of sites tested dirt2 demo gtx480 running at Dx 9.0 , actual game runs at DX 11 (but not demo ) at gtx480 .Check the TPU dirt2 numbers and compare with others .

Hey thanks for the info.

 

[RoutedScripter]

Charlie was right with the nvidia abandoning high and mid market but it wasn't so real looking since the stocks were still some and he was right about that in january we will se some low end cards.

From what I seen , he is not 100% proof , but he tries to think up the closest probablitly , while it might not be so bad or true , it's not totally off. And certainly what nvidia does and what you are getting shows that he's not saying bullshit after all.

For the linking , it's real alright , some guy complained about something , and they replied with a perfect example of one other site posting on 15march 2010 an article , what S|A talked about already in July 2009.

 

[Benetanegia]

Don’t get me wrong Fermi is one hell of a complex design. Dam innovative to say the least. NVIDIA actually did a great job, but like you said the manufacturing process is most likely the issue. It’s too bad NVIDIA couldn’t just wait it out a little more so they can get that thing on a 32nm instead of the already faulty 40nm. It probably would have worked out better for them.

But I would like to stress that NVIDIA sort of messed up, you never release a high end part on a new manufacturing process in this case 40nm and hope for the best, you always release your low end cards on something new and learn from them, correct what needs correcting then you release your high end parts.

ATI’s been doing this for a while now and they’ve rectified any possible issues before the real high end GPU launch.

Before you start reading this, please keep in mind that only the first sentence is really the reply to your post. The rest is my view of everything related to Fermi release and it's "fail", for all to see.

They were working on smaller chips indeed (GT200 derivatives), but they were cancelled, probably because 40nm was so fucked up that they were worse than the euivalent 55nm chips. Remember what a disaster RV740 was for AMD at that time despite the fact the chip was so small. AMD fixed the problems with 40nm, although they are still below 60% yields and for AMD and a medium sized chip, 6 months after release, that's still very poor yields. There's no denying that 40nm is totally fucked up and as with everything, that will affect some chips more than others and luck is always a factor too.

It's not Nvidia's fault, TSMC should have a mature process by now and they don't, and that's the end of it. Pretty much like an architect that bases his bridge on steel beams of X strenght and the contractor uses cheaper/weaker ones and the bridge collapses, Nvidia was promised a mature and good 40nm process and got what they got. When you design you always calculate a margin of error, but 40nm failed way beyond any margin of error. Nvidia has been making big chips since the beginning, and has never had such problems. Ati did the same until very recently and never had such problems either. The only time was R600 and that was a much bigger failure than Fermi. For instance, it had more transistors than G80, consumed much more power and most importantly it was 40% slower. I don't know how people are even comparing Fermi with R600. Sure it didn't meet it's performance goals, but it's still 10-20% faster and on some games is actually 60% faster than the HD5870, faster than 5970 actually (Battleforge, HAWX...). TWIMTBP here has nothing to do either since those games have long been released and patched and they were definately not made for Fermi architecture, not to mention that many of them are AMD games. The only difference between the games where it performs faster than HD5970 and the ones where it performs just barely like a 5850 is probably driver optimization. And we are talking about pre-release cards and pre-release drivers ladies and gentlemen. And we are talking about a completely new architecture. Most people are downplaying how different Fermi is, but in some aspects is more different than a Cypress and a GeForce 2 are between each other and the reason is that this is the first time in history that a GPU has parallelized the part of the chip dedicated to geometry and rasterization.

A lot of the silicon, a lot of what it makes Fermi so big and so hot and so unmanufacturable is because of that fact and you are seing 0, Zero benefits on current games, except when Tesselation is used and when it is used, it's undeniably much much faster than AMD cards, to the point that even the 470 can decimate the HD5970. OpenCL/CUDA, Compute, F@H... is another place where Fermi decimates Ati cards and previuos Nvidia cards. So it's not as if all that silicon is useless after all.

But are we going to see that advantage only in tesselated games? Probably no. The reason that Fermi is faster on tesselation and not much faster without it is that its' 4 setup/raster engines are only being put to work when tesselation is working, because rendering is single threaded. With DX11, multi-threded rendering is available, something that no single game or benchmark to date is using, but that will put those units to work for a fairly greater performance. Right now they are only there consuming power. Fermi is like a Intel Core i7 in year 2004, it can be seen as a Quad Core GPU+ hyper-threading in fact. Remember when nothing could take advantage of a dual core? We are there with rendering. It's been long since pixels have been parallelized, but rasterization is still a single threaded task.

Nvidia did a compromise, DX11 has tesselation and has multi-threaded rendering and has Compute, and all those things are going to be used sooner or later, so a change in GPU was required. AMD will have to make some compromises eventually and I hope it turns out better for them, but the thing is that Nvidia already made them and the hit has not been so big. The only thing that is terrible about Fermi is power consumption and even that is not so dramatic as people are making it to be. I don't remember anyone making any similar comment about the HD4870X2 and now, oh boy, that is a high power consumption. Compared to the GTX285 that's 160w more for about the same performance difference we see between Fermi and Cypress. And it's not like the X2 is not popular in these forums, it's not like no one had two of them in Quad Crossfire, I mean, please some coherence and don't be hypocrytes. That being said, Fermi in it's current form is not somethig that I'm going to buy or recommend because of it's power consumption (and because since G92/RV670 days, I don't buy or recommend anything that costs more than $250), but it is extremely false to say that Fermi is a failure.

Sorry for the long post and have a nice (rest-of-what's-left-of-the-)day after you finish reading this.

 

[the54thvoid]

lol, I aint even reading that post. I'm too old and might die before the end....

did see this though...

'even the 470 can decimate the HD5970'

can you quote source?

 

[RoutedScripter]

TSMC does not improve the chips , the process is 40nm and therefore same for all chips , they build chips how and what their customers order.

 

[shevanel]

It basically says fermi isnt using all of it's power on what is available and when software can utilize and take advantage of multithreaded stuff then you will know who the boss is and 96c aint shit compared to the joyous amounts of f@home #'s you'll be pulling and hopefully it'll be so fast that video encoding will be done before your balls start to sweat.

http://www.youtube.com/watch?v=CVEhyfbeDSg

 

[RoutedScripter]

Point is , this does not matter at the time were at now.




The GTX200 series seems to not be able to play crysis normally (already too hot)
http://www.youtube.com/watch?v=sFCeR-xJh-M&feature=related

 

[Super XP]

Before you start reading this, please keep in mind that only the first sentence is really the reply to your post. The rest is my view of everything related to Fermi release and it's "fail", for all to see.

In regards to your whole post, the Fermi design better be much faster than ATI's current GPU. ATI's been refreshing the same old chip design for ages now where as NVIDIA's Fermi is based on a "Brand New Architecture" build from the ground up. And new architectures should be extremely faster than what is out today. We are talking about 60% up to 150% better performance improvement over older designs.

What I see is GX 480 regardless of its power & heat issues performs quite miserably overall not to mention it’s a so called Next Gen design. 40nm being problematic? Yes indeed it is but ATI’s HD 5800 series are on the same process and outperforming anything NVIDIA has to offer, so I wouldn’t put the blame on 40nm but rather how dam COMPLEX NVIDIA chose to make Fermi.

Its fine to go nuts with a GPU design but it also has to work right and so far Fermi running 97C and sucking back over 300W of power is not running right.

Personally NVIDIA’s CEO is to blame, because he stubbornly refused to listen to his CTO. But that is a different story.

 

[Zubasa]

I think the only thing 'wrong' with it is that the chip design is too advanced for the process tech it's manufactured on. 3.2bn transistors is an awful lot and it appears to be maxing out the 40nm process tech it's built on. I reckon the die shrunk version will be much better.

I'll definitely be waiting for the 'GTX485' version. In the meantime, I'll likely be getting one of the lower-end Fermi derivatives in a couple of months, so that I can play with DX11 & 3D Vision together. (I already have 3D Vision).

I will say there is more than that.
The GF100 simply have too much junk on the chip for what the card is intended for.
It have loads of GPGPU crap that not even folders will utilize, and remember this is a consumer grade gamming card.

All the GPGPU functions are great for a Tesla but not for a Geforce.
What happens is the consumers are paying for extra transistors which are no more than a lump on a log.

 

[the54thvoid]

Found a good article at Hexus, explaining why despite it's obvious design prowess (heat/power/noise) aside, there is no reason for a Fermi based architecture 'at this moment'.

http://www.hexus.net/content/item.php?item=24017

 

[phanbuey]

if the process did have good yields, a 512 shader 750Mhz card would have been pretty phenomenal. Too bad too bad.

I wish they will do what ATi did with the 38xx to 48xx series and double up... 1024 shaders at 32nm!!!!

or hell... even 960 ... i could live with the gtx 580 with 960 shaders.