Nvidia's Fermi GTX480 is broken and unfixable

[L1qu1d]

http://www.semiaccurate.com/2010/02/17/nvidias-fermigtx480-broken-and-unfixable/

makes you think?

I personally will hold my thoughts till release

be warned it is a Charlie Demerjian article

And from what I remember it usually is like this:

Charlie :0c==============3 ATI

But like I said I'm going to hold my thoughts for now, just because we have seen Nvidia's constant mistakes (one of which is the laptop line 8600Ms) but nothing as dumb as this.

 

[p1n3apqlexpr3ss]

Kinda sad really, im a bit of a ati fanboy but like to keep a open mind.. would love to see nvidia come out with something great in the next month or so and then see what ati has up its sleeve, competition = lower prices = what most consumers care about

 

[notty22]

Yawn,
I think I read "Nvidia didn't do its homework' at least 3 times. Ok he's got Fermi outperforming Cyrpess. What else matters ? lol

 

[p1n3apqlexpr3ss]

yield issues for a start... and the ability to launch a product within 50$ of cypress that performs similary, i wish nvidia luck... theyll need it i thk

 

[jsc]

This looks to cost nVidia the better part of another year.

 

[Truhls]

man if its true, and i say if lol. Nvidia really shot themselves in the foot, then decided to donkey punch themselves, then decided to put sandpaper on dildos and sit on them.

will be fun to see what happens. But i hope this doesnt stall the 5830 means it wont have competition.

 

[p1n3apqlexpr3ss]

...so much for seeing some samples show up in that convention in a few weeks time...

 

[hallowed_dragon]

On one hand this is Charlie...on the other hand he was right so far about all news regarding Fermi.
I hope things are not that bad for nVidia. We need healthy competition.

 

[4745454b]

Some things I remember about the article. I still believe that the GTX480 (or whatever they are calling it this week) will still have 512 shaders. It came out awhile ago that the workstation cards would have 448. This was assumed because they would be running cuda code and all shaders would be working at 100%. By lowering the number, you'll cut down on the heat.

If the yields are as bad as claimed, this is horrible for Nvidia. You should see more GTS460/440s, and the GTX480 will be mostly a paper product. I am believing more and more that Nvidia will still produce all the cards they have out now. GTX4xx will be DX11, GTX3xx will be DX10.1, and the GTX2xx cards will be DX10. The big question now will be who will be able to afford the GTX4xx?

Also bothersome was the line about AMD simply pricing the 5xxx so that Nvidia sells in the red. This will mean high prices for all of us. I don't think we'll see any price cuts for quite some time this time around. One thing not mentioned is that AMDs costs per card didn't really change much. They doubled the number of shaders, but moved to a smaller process. If you compare 5xxx to 4xxx, the die sizes should be similar. The 5xxx will be slightly bigger, but not by much. This means, assuming equal yields, that 5xxx costs about the same as 4xxx. This is not the case with GTX4xx. They could barely afford to produce the GTX280, how long will the GTX480 stay around?

 

[noob2222]

Every manufacturer knows about defects. The big thing is if you choose to be prepared for those defects or just wing it and see what happens, thats the game Nvidia has been playing ever since thier monolithic designs.

This page is a perfect description of defects on a manufacturing process for computers, be it cpu, gpu, or memory. http://www.anandtech.com/video/showdoc.aspx?i=3740&p=8 It is also a very good article.

Nvidia has been able in the past to dodge the bullets that are involved in a monolithic design, defects. Now with TSMC not being the best at 40nm, its killing them in the worst possible way. Picture this. Right now Nvidia's top chips are about as scarce as the chips AMD is cherry picking for the 5970's, maybe less. This means if Nvidia went forward with the release, they will be higher priced and harder to find than the 5890's.

Lets examine the sorting for a second. ATI makes the 5800s on the same die, same wafer, and bins them. First you apply low voltage to the chips, see wich ones pass and can be used on the 5970 (very low yield). You then up the voltage, and picks the chips suitable for the 5870s (little higher yield). Do this X times ... AKA binning, you get the variations in products from the exact same die, and exact same wafer.

Fabrication processes also applies to cpus as well. Intel has thier 32nm chips going, but what was thier first one? A Tiny dual core chip, rather than trying to make thier monolithic 6 core chip, and they also left the graphics and memory controller off of it (its on the other chip). Improve the process then increase the size. Nvidia has a huge problem on thier hands. ATI and Intel learned from their past mistakes, what will Nvidia learn? More importantly how long will it take Nvidia to have a product to sell?

(hmm sounds like the same boat AMD is in with thier CPU's)

 

[JAYDEEJOHN]

What Intel did may or may not fit your scenario, as the markets usually drive what Intel chooses to do.
Another thing to consider here is, the 40nm process is becoming very expensive, being the yields have been poor on everything, and why we see the lower nVidia cards cut down and priced high as well, including ATIs lower solutions.
Add in the higher problems with yields on such a huge die, and theres room for alot of trouble for the green team.
My biggest concerns are, we havnt actually heard a thing yet, and thats not good, and for those who say this is what they also did with the G80, we knew more then a month before release than we do now with Fermi, and lastly, I finally heard my first rumor about Fermis lil bros coming out, or at least taping out, but only 1 so far, which isnt good either

 

[clement4413]

Nvidia should has improved the GT200 as ATI is doing at the present with R600 -> Rv770 -> Rv870.
The GT300 is remembering the R600, ATI met a lot of problem with the PCB, put back the launch into the market etc...
When the GT300 will be launch into the market : The card will be noisy, high consumption, huge size, with the same powerful than HD 5870. I don't imagine the price.

 

[Mousemonkey]

When the GT300 will be launch into the market : The card will be noisy, high consumption, huge size, with the same powerful than HD 5870. I don't imagine the price.

The GT3xx have already been released but there are no reports of them being noisy, having high power consumption or being a huge size because they are OEM parts.

 

[clement4413]

I don't understand
The GT300 = Fermi = GTX480 ? No ?

 

[4745454b]

Thats a good point Jay. We are around a month from launch and still nothing. No real leaked benchies at all.

 

[Mousemonkey]

I don't understand
The GT300 = Fermi = GTX480 ? No ?

No.

http://www.nvidia.com/object/geforce_family.html

 

[clement4413]

Ha yes, I had forgotten so I talked nonsense sorry

 

[Mousemonkey]

Here's a better equation. Easier to understand.

GT300 = Fermi = 0

How does GT300 = Fermi?

 

[mactronix]

Some people just don't bother reading the whole thread.

I'm waiting for a release, I'm fed up with people claiming this and that based of off a guess that this did this so the new one should do that. I'm talking press here not on the Forum.
I have read articles claiming its the next coming of the silicon messiah and then there are those who claim its a complete waste of silicone.

So I'm just happy to wait now

Mactronix

 

[mindless728]

+1 matronix, with only 6 weeks or so to go, waiting seems like a good idea
though for entertainment purposes, these threads get hilarious after about page 8

 

[KidHorn]

Looks like a complete disaster. Luckily nVidia doesn't depend solely on high end graphics, but I see quarterly losses for a long time.

If their price drops below $7, Intel may look to acquire them. We have a long way to go to get to $7. Intel can make 28nm processors and can probably figure out how to make Fermi, given enough time.

 

[notty22]

Looks like a complete disaster. Luckily nVidia doesn't depend solely on high end graphics, but I see quarterly losses for a long time.

If their price drops below $7, Intel may look to acquire them. We have a long way to go to get to $7. Intel can make 28nm processors and can probably figure out how to make Fermi, given enough time.

Another post full of nonsense. LOL Its ATI thats in danger of financial failure , Nvidia is doing fine. And if people want a example of hardware failure there is one in progress with the continued 5700 debacle going on. Just look in these forums.
Nvidia swings to quarterly profit as sales more than double

http://www.marketwatch.com/story/nvidia-swings-to-profit-as-sales-rise-sharply-2010-02-17
SAN FRANCISCO (MarketWatch) - Nvidia Corp. said Wednesday it swung to a profit for its fiscal fourth quarter, as sales more than doubled thanks to strong demand for chips used in personal computers and workstations.

Santa Clara, Calif.-based Nvidia /quotes/comstock/15*!nvda/quotes/nls/nvda (NVDA 16.85, -0.99, -5.55%) said net income for the period ended Jan. 31 was $131.1 million, or 23 cents a share, compared to a loss of $147.7 million, or 27 cents a share in the same period a year earlier. Excluding special items, Nvidia said fourth-quarter earnings were 23 cents a share.

Revenue more than doubled to $982.5 million from $481.1 million.

Analysts polled by Thomson Reuters had expected Nvidia to post earnings of 20 cents a share, and $957.2 million in revenue.

"Nvidia's business continued to accelerate," Chief Executive Jen-Hsun Huang said in a prepared statement.

 

[nforce4max]

This is going to be a fail chip regardless of performance and Nvidia will agree to some extent being not what they had expected of their hard work. 40nm process is just not enough so a shrink is necessary to be successful at all in any price segment. I had read some ware but will look it up later that TSMC isn't doing the 32nm process but instead chose 28nm so that is Nvidia's best bet. Shrink the chip below that or equal to the G80 in order to get passed the yield problem. It is a shame that the first samples that we may see won't be full spec nor have intended clocks. With the GT200 out of production the only thing left is those 40nm low end cards and the G92 cards. For me I will stay with my 9800gt 1gb cards. For ATI I will wait till prices come down but will continue to add to my collection.

 

[hallowed_dragon]

In my opinion nVidia will be OK even if the Fermi generation is going to be a flop, but due to the delay it is released the problem will get bigger when the 6xxx series comes from ATI.
Maybe JHH will be replaced then...one can only hope .

 

[notty22]

Armchair engineers prophesizing about a product they know about as much as ATI's advertising monkeys do. Why do these companies employ engineers and teams of people in research + development when they could get it all from genius's on the internet.