Fermi GTX480 is broken and unfixable

You can look at Fermi being 8 parrallel sections, if you take a bad chip that has 1 bad section you seal it off and make a bad "480" into a good 470. Seal off another section and you've got something that is 75% the preformance. Rinse and repeat until you get all the way down to something that is 1/2-1/8th the preformance based on actual yields. Doing so lets you reuse the "scrap" chips and suddenly your 480 dies don't cost as much or you say they do and your derivitive dies are "free".

Interesting. I didn't know Fermi was designed like that. If that's true then we should expect lesser cards in the mid-range line.
 
Interesting. I didn't know Fermi was designed like that. If that's true then we should expect lesser cards in the mid-range line.

Correct me if I'm wrong, but to a certain extent that's how all modern GPUs are designed. Fermi's design using only 8 X 64 groups of shader cores puts it a disadvantage because losing only one group to a manufacturing defect reduces processing power by 12.5%. Whereas if it were designed in a 16 X 32 layout, losing one group to a defect would only reduce speed by 6.25%.

12.5% wouldn't be so bad (it would put it in GTX470 range possibly) but what if that single defect overlaps multiple shader groups? You could lose 128, or 192, or even 256 shaders to a single defect.
 
You can look at Fermi being 8 parrallel sections, if you take a bad chip that has 1 bad section you seal it off and make a bad "480" into a good 470. Seal off another section and you've got something that is 75% the preformance. Rinse and repeat until you get all the way down to something that is 1/2-1/8th the preformance based on actual yields. Doing so lets you reuse the "scrap" chips and suddenly your 480 dies don't cost as much or you say they do and your derivitive dies are "free".

And, like you said, it's the lower end, "bad 480s" that will make the most money for Nvidia, just as AMD with their Radeon line of graphics cards. These will be the "470s" like you mentioned, the 450s, 420s, 410s from Nvidia's Fermi. But, they do have to worry about losing a bit or a lot of money on their high end GPU due to low yields. In the end, everything takes time to get things right. I'm sure AMD had similar problems when the 2000-series were released.

At this point, I'm not sure how much patience is left for a lot of the Nvidia loyalists to wait and hold out until these are released. I'm sure the non-diehard fans don't care that much-- "it comes, it comes; if not, then I'll go to AMD", right? Or, whoever has the best performing GPU at that time and/or at a good price, people will flock to it.
 
Correct me if I'm wrong, but to a certain extent that's how all modern GPUs are designed. Fermi's design using only 8 X 64 groups of shader cores puts it a disadvantage because losing only one group to a manufacturing defect reduces processing power by 12.5%. Whereas if it were designed in a 16 X 32 layout, losing one group to a defect would only reduce speed by 6.25%.

OMFG, if I have to correct one more person on something this obvious I'm going to scream. GF100 uses 16 clusters of 32 shader cores each, not 8 clusters of 64. Don't believe me? Then read the whitepages before you post (or the anandtech article). I sware 80% of the questions and insults aimed at fermi in here are talked about in the whitepages, I don't think anyone has actually even looked at them other than me. /rant
 
what is this thing that it will come out after cebit?
Five or six cards for a few reviewers? I hope I'm wrong, but I'm not holding out hope for much more than a paper launch. If GPUs are already in production, isn't awfully strange that there is NO firm list of current specifications provided by Nvidia and virtually no positive leaks from any source whatsoever? When was the last time a major new video card was supposedly close to release but there was almost a complete vacuum of verifiable information? At this time we have doom and gloom from Charlie (who is right more often than wrong), we have compelling analysis from Anand Lal Shimpi, boatloads blather from Nvidia PR/marketing, but almost NOTHING in the middle. No leaks, no secret videos, no journalists given a few minutes to play a game. This is either the Great Awakening and Renaissance for GPU technology or something very dark and distressing for Nvidia.

Time for some crow
Yes, probably, but for whom? It ought to be interesting.
 
supposedly close to release but there was almost a complete vacuum of verifiable information

OMG!!! READ THE WHITE PAGES READ THE WHITE PAGES READ THE WHITE PAGES READ THE WHITE PAGES READ THE WHITE PAGES READ THE WHITE PAGES READ THE WHITE PAGES!!!!

Seriously every bit on info on the specs that could be released has been released (back in December I might add), except for clock rates since those aren't finalized until the last minute. And I don't see how they could have given us benchmarks without a working card.....
 
Crow anyone?

crow1207.jpg
 
I haven't read this entire huge thread but I'm willing to bet its mostly "Hai guyz don't believe this guy he's a ________ (<insert ad hominem here)." Take it with a grain of salt"

That should be enough to discredit Charlie amirite?
 
I guess this thread has run its course? Time for some crow ;)

maybe, this is going to be interesting here. Fud isn't as rabid as Charlie but they are still pretty pro Nvidia. The key is what the Launch is going to be, and if he is right and there is money to be made or lost. Myself I still think we are really going to see fermi come into its own if they can go to the 28nm.
 
maybe, this is going to be interesting here. Fud isn't as rabid as Charlie but they are still pretty pro Nvidia. The key is what the Launch is going to be, and if he is right and there is money to be made or lost. Myself I still think we are really going to see fermi come into its own if they can go to the 28nm.

Hah, only if they dump TMSC and go with Global Founderies.
 
Seriously every bit on info on the specs that could be released has been released
No. The glorious white papers spell out the projected, planned, specifications. What I asked for is "current" specifications, what is really coming of the production lines. Real, not theoretical. The only thing current we have is Nvidia's vapid marketing and Charlie's and Anand's articles. It's rare when there is no leaked information or at least grainy videos of benchmarks just before non-disclosure agreements opened up.
 
Hah, only if they dump TMSC and go with Global Founderies.

I highly doubt its TSMC issue for delay, its like a rush technology right there that cause the problem...

and right now seems to be the design problem.....
 
I highly doubt its TSMC issue for delay, its like a rush technology right there that cause the problem...

and right now seems to be the design problem.....

I'm sure vick meant GF will get 28nm way before TSMC which is what is projected.

We should all camp out f5 all the etailers when Fermi ever hits. Turn it around and mark it up real high for the fanbois:D on Ebay. Who's with me?
 
I highly doubt its TSMC issue for delay, its like a rush technology right there that cause the problem...

and right now seems to be the design problem.....

Nvidia has blamed TSMC for problem after problem since 2008 and there are many public news stories of Nvidia execs calling out TSMC and saying they needed to get their vias etc. in order
 
What are the conditions for Crow? Better to lay it out in advance or forever hold your peace. I don't see how Fuad saying we can expect full availability somewhere late summer suddenly makes the OP-article wrong.

Some (possibly) testable assertions on the GTX480 launch:
"Initial production is going to be limited to essentially a PR stunt"
"Every card made will be sold at a huge loss"
"This puts any hope for Nvidia out until 2011 for anything but PR stunts."
"an ATI HD5970, a card that will slap [first batch of fermi cards] silly in the benchmarks."
"Fixing these problems requires Nvidia to [..] double up on the vias and also change the circuits in a non-trivial way. [..] it will still be more than six months before it can bring a fix to market."

So it's essentially saying that: nVidia won't be able to have an economically viable GTX480 part available till around august at the earliest (due to req. 6 month respin). The first batch will be limited in numbers and very expensive (sold at a loss), and draw huge amounts of power. We're not going to see good low-mid range (non-crippled GTX480) parts this year.

I'm sure there's more, but that's what I found to be the core of the original article.

Just to be clear; "unmanufacturable" is clearly not used literally, since the article itself talks about cards being released. I take it to mean "unmanufacturable with respect to sound economics". Therefore, cards being released is NOT IN ITSELF enough for "crow", IMO.
 
Who is this Fud guy you speak of guyz?

http://www.fudzilla.com/content/view/16544/1/



Multiple sources now claim that Nvidia has promised to ship Fermi in January 2010. We are talking about GF100 Fermi, the GPU that is expected to defeat ATIs Radeon HD 5000-series offerings. To make matters more clear, January is the time when you can hope to buy a high-end Geforce desktop graphics card based on Fermi.

Nvidia Fermi-based cards in the enthusiast market have been pushed from a late November launch to a January launch mainly due the fact that the A2 silicon chip revision was not good enough for volume production, so A3 is about to ship to partners soon. With this in mind and if all goes well, Geforce graphics cards based on Fermi architecture are expected to ship in January.

We have also heard that the launch won't take place at CES 2010, which starts on January 7th and ends of January 10th. Instead, Nvidia might launch it later at some point, but nevertheless still sometime during the month.


This latest from Fud means Charlie eats crow? Yes?


Do the green fish ever stop taking the bait?
 
on monday they will have a press conference.

Nothing more than that is said, although BSN believes NVIDIA will announce a March 27th launch date for GeForce GTX 480, with reviews hitting the web on the 29th.
Similarly Fudzilla believes GTX 480 is on tap for a late March launch date.

so will it be march 27th to tell us that the cards are in production. May 27th to see one at egg for $699 oos and by july 5th in stock at only $799 limit 1 per state.
they do have digital downloads of full color pictures thru your iphone for only $2.99
 
^^^^ lol

Even people with money won't be able to buy one because they will produce limited quantities of this card. Most people on the Internet who talks about this card will never own one. Nvidia can't make a mainstream card out of Fermi because it's so broken. Once the 6800 is released in Q3 they will quickly announce Fermi 2 in Q4 thus wasting everyone's time and money on Fermi 1. Good luck with this card folks, Nvidia needs more beta testers for only a $600+ membership fee.
 
^^^^ lol

Even people with money won't be able to buy one because they will produce limited quantities of this card. Most people on the Internet who talks about this card will never own one. Nvidia can't make a mainstream card out of Fermi because it's so broken. Once the 6800 is released in Q3 they will quickly announce Fermi 2 in Q4 thus wasting everyone's time and money on Fermi 1. Good luck with this card folks, Nvidia needs more beta testers for only a $600+ membership fee.

If it's faster than the 5870 then I'm in. If the Fermi 2 is even faster than the ATI offering at the time then I'm in again. I like to have nice cutting edge vid cards.
 
^^^^ lol

Even people with money won't be able to buy one because they will produce limited quantities of this card. Most people on the Internet who talks about this card will never own one. Nvidia can't make a mainstream card out of Fermi because it's so broken. Once the 6800 is released in Q3 they will quickly announce Fermi 2 in Q4 thus wasting everyone's time and money on Fermi 1. Good luck with this card folks, Nvidia needs more beta testers for only a $600+ membership fee.

you mean like the 5970 and the drivers from hell?
 
^^^^ lol

Even people with money cant buy the 5970 because they are produced in limited quantities of this card. Most people on the Internet who talks about this card will never own one. AMD can't make a mainstream card out of Hemlock because it's so broken. Once the Fermi 2 is released in Q3 they will quickly announce 6970 in Q4 thus wasting everyone's time and money on 5970. Good luck with this card folks, AMD needs more beta testers for only a $700+ membership fee.

fixed.
 
you mean like the 5970 and the drivers from hell?


Here comes the driver argument again, laughable.:p

Where are the Fermi drivers? Non-existent!:eek:
You think the drivers will be perfect when released? Get real.:D

Fermi is going against the 5870, not 5970 which it can't touch.;)
I would like to see two GF100s on one PCB, which looks almost impossible.
 
Here comes the driver argument again, laughable.:p

Where are the Fermi drivers? Non-existent!:eek:
You think the drivers will be perfect when released? Get real.:D

Fermi is going against the 5870, not 5970 which it can't touch.;)

you got owned.
 
Even people with money cant buy the 5970 because they are produced in limited quantities of this card. Most people on the Internet who talks about this card will never own one. AMD can't make a mainstream card out of Hemlock because it's so broken. Once the Fermi 2 is released in Q3 they will quickly announce 6970 in Q4 thus wasting everyone's time and money on 5970. Good luck with this card folks, AMD needs more beta testers for only a $700+ membership fee.
you got owned.


Wow, you are so wrong if you think you owned that guy. The changes to the paragraph you made don't even make sense? Calling 'Hemlock' aka the 5970 so broken they won't be able to make mainstream versions of it?

Ummmm.... I'm not sure if your just not following reviews on hardocp.com, haven't been to new egg in months or are just having difficulty accepting reality these days, but, ATI has had mainstream cards for the 57xx and 58xx series out for months. I'm afraid you owned yourself. See: 56xx, 54xxx, 5750, 5750, 5850, 5870. These are all mainstream versions and entry level versions of the hemlock.

Futhermore, if ATI does annouce the 6970 in Q4 2010, it would probably not come out for a while after. All the 5970 adopters would have gotten more than a years worth of gaming bliss or more /w great image quality and eyefinity. That would be much much better than those who waited for Fermi would have gotten what...4 months? From Q1 2010 > Q3 2010? Once again, you kind of owned yourself in your statement above by implying that Q3 2010 is when the Fermi 2 would be announced. Think about the implications of editing someone else paragraphs first before editing them?
 
Last edited:
dont matter about the little details i missed, its the big picture that matters here.

the 5000 series is useless and not needed at all today until the real dx11 games come out, crysis 2 for an example.

the 5000 series is a half baked product compared to what amd is really building for dx11 and tessellation for the 6000 series.

i returned my 5970 2 days ago, it cant even work in 2d mode even with all the manual fixes and the latest drivers, the thing was a joke.

they are beta testing people with these cards, i fell for it but i was lucky to get the money back for it even though i lost $100 bucks for the restocking fee and shipping.
 
Wow, you are so wrong if you think you owned that guy. The changes to the paragraph you made don't even make sense? Calling 'Hemlock' aka the 5970 so broken they won't be able to make mainstream versions of it?

Ummmm.... I'm not sure if your just not following reviews on hardocp.com, haven't been to new egg in months or are just having difficulty accepting reality these days, but, ATI has had mainstream cards for the 57xx and 58xx series out for months. I'm afraid you owned yourself. See: 56xx, 54xxx, 5750, 5750, 5850, 5870. These are all mainstream versions and entry level versions of the hemlock.

Futhermore, if ATI does annouce the 6970 in Q4 2010, it would probably not come out for a while after. All the 5970 adopters would have gotten more than a years worth of gaming bliss or more /w great image quality and eyefinity. That would be much much better than those who waited for Fermi would have gotten what...4 months? From Q1 2010 > Q3 2010? Once again, you kind of owned yourself in your statement above by implying that Q3 2010 is when the Fermi 2 would be announced. Think about the implications of editing someone else paragraphs first before editing them?

Save your breath. The nvidia fanboy is obviously not listening.
 
The 5970 was certainly launched prematurely. That's why they are still limited in quantiy. ATI still has to work out the kinks but they needed the performance crown. But your argument was a about Fermi which is going against the 5870. I can't see how the 5970 dooms the entire line of 5000 series cards. It's difficult to put two GPUs that requires much more than the previous gen. Most people would agree that a single GPU config is better than a dual GPU config. If you knew that you probably would have saved yourself the trouble and purchased two 5870s.
 
i just returned the worst video card(5970) i ever purchased and that's all you can think of to say?

It sounds like you never actually owned a working 5870. Going from gtx260s(3 which I still have) to tri-fire 5870s was one of the more substantial upgrades I've made. I found there were driver problems, but I took time and reported a rather obscure one (eve online had black boxes over items) and it was fixed next rev.

I'm betting you either a) had a defective card and gave up immidiately, or b) never had one and are making up stories.
 
It sounds like you never actually owned a working 5870. Going from gtx260s(3 which I still have) to tri-fire 5870s was one of the more substantial upgrades I've made. I found there were driver problems, but I took time and reported a rather obscure one (eve online had black boxes over items) and it was fixed next rev.

I'm betting you either a) had a defective card and gave up immidiately, or b) never had one and are making up stories.

i got over 15 years of experience in this field, dont worry, the card has and will have major driver issues and some hardware issues.
 
Last edited:
i just returned the worst video card(5970) i ever purchased and that's all you can think of to say?

How about a 5770, 4 gigs ddr 2 667 ram, and a pentium d [email protected] running fallout 3 at 3600x1600 at 40-60 fps?

http://i247.photobucket.com/albums/...inity Setup/Fallout32010-01-0302-32-50-69.jpg

http://i247.photobucket.com/albums/...inity Setup/Fallout32010-01-0302-32-16-84.jpg

Or Left 4 Dead 2 at 50-60 fps?

http://i247.photobucket.com/albums/gg151/Zellio2008/Eyefinity Setup/c1m1_hotel0000.jpg

Maybe you'd like to actually add something substantial to the argument instead of the verbal manure you've done?
 
Lol, at the experience comment. This is the internet, experience means damn all. How many times have I heard the old "I have been gaming since the 70's, I have been in I.T for 20 years" crap.

None of it matters here.
 
Lol, at the experience comment. This is the internet, experience means damn all. How many times have I heard the old "I have been gaming since the 70's, I have been in I.T for 20 years" crap.

None of it matters here.

What counts is being able to evaluate information logically and critically without emotion or bias.

I visited here six months ago and counted the anti/pro Charlie~data posts, they ran 10 to 1 against Charlie~data, many slinging ad hominems with abandon, with only a few lonely voices pointing out what he was ACTUALLY SAYING had factual solidity, was logical and MADE SENSE.

That's now mostly flipped.

What I would point out is the 7 out of 10 of you who reflexively attacked Charlie THEN, because you didn't LIKE what he said, but have swung around to accepting his positions NOW, because he has been PROVEN correct on so many points, are not as clear thinking and logical as you no doubt assume you are.

~ If you attacked Charlie six months ago, your mental faculties and critical thinking skills are, to some degree, retarted by your emotions ... nothing to be ashamed of, lord knows it's an inherent and endemic human condition, unless you fail to LEARN from it of course.

Yeah, I know, I'm an insufferable little know-it-all prick.
 
Last edited:
Back
Top