Charlie D Claim Watch

[Tempered81]

One cause of the heat is Nvidia increasing transistor density ~95% from gt200->gf100. ATI went from 1b to 2.15b transisitors and from 280mm2 to 334mm2 die area, increasing complexity 110% and size by 20% staying consistent with their transistor density, increasing ~87%. Nvidia went from 1.4b to 3.0b transistors and from 470mm2 to 530mm2 area, increasing complexity by 110% and size by ~11%.

Given
RV790 at 55nm, 1b tran, 280mm
Cypress at 40nm, 2.15b tran, 334mm
GT200b at 55nm 1.4b tran, 470mm
GF100 at 40nm, 3.0b tran, 530mm:
Ati squeezes 3,410,000 transistors into a square millimeter in a 55nm RV770 and 6,437,000 transistors into a square millimeter in a 40nm Cypress. Nvidia squeezes 2,978,000 transistors into a square millimeter in a 55nm GT200b and 5,661,000 transistors into a square millimeter in a 40nm GF100. Ati already had more efficient ICs with a higher transistor density, regardless of the performance in the end.

 

[BenSkywalker]

You know I have been approaching this from entirely the wrong angle. Let's say things like his shader counts and clockspeeds were 100% accurate, let's assume that is the case. Charlie's current claims-

There will be only one shipment of GTX 470s/480s ever

5870 is slow

The 58xx parts had unlaunchable yields

These are the things that are far more important in the long term. Everyone saying Charlie was 100% accurate should have no problem stating that all of these statements are true. These are not wiggle room statements, these are clearly black and white broad based claims that he has made. So, everyone saying Charlie is accurate agrees with those three points, right?

He did not claim the 480 would be 448... he thought the first part would be.

He said the top bin was 448, do people not understand what that means?

 

[1h4x4s3x]

Everyone saying Charlie was 100%

Show me just one person who said that? The only one who said Charlie was 100% wrong is you.
Mowr please.

 

[1h4x4s3x]

He said the top bin was 448, do people not understand what that means?

How about you read the article you linked to?
If you don't understand it I'm willing to give you a hand, just say please.
Hint: the picture tells more than 100 words.
:sneaky:

 

[Janooo]

He tells so many lies, I understand why you all may get confused.

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

... and how do you know that at the time of the article NV was not planning to go with the stated numbers?

 

[Daedalus685]

You know I have been approaching this from entirely the wrong angle. Let's say things like his shader counts and clockspeeds were 100% accurate, let's assume that is the case. Charlie's current claims-

There will be only one shipment of GTX 470s/480s ever

5870 is slow

The 58xx parts had unlaunchable yields

These are the things that are far more important in the long term. Everyone saying Charlie was 100% accurate should have no problem stating that all of these statements are true. These are not wiggle room statements, these are clearly black and white broad based claims that he has made. So, everyone saying Charlie is accurate agrees with those three points, right?

Oh for fucks sake....

No one is saying he is 100% accurate.. it is only in the crazy little Jedi world where things are either 100% or 0%....

Second, he never said anything like that about the 5870... Was he wrong about that comment for the 280? Yes.. does that have anything to do with the 5870? No.. Does being wrong about something discredit all other things they say? No...

No one has ever said he is right all the time, or that all of his rumours have merit, or that he isn't a biased jack ass a lot of the time... The fact is that many of his rumours were pretty accurate.. and even more contained a lot of useful tidbits.


I am a millionaire. I sleep with seven different porn stars at a time. The sky is blue because of Rayleigh scattering. The acceleration from gravity at sea level is 9.81m/s^2.

Since you agree with the last two that means that I must be rich and lucky eh?

 

[waffleironhead]

He tells so many lies, I understand why you all may get confused.

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

So the top bin on A3 on that date was a GTX470. Can you prove him wrong?

 

[Daedalus685]

He said the top bin was 448, do people not understand what that means?

/facepalm

You just linked and quoted the damn article!

"...600MHz for the half hot clock, and 1200MHz for the hot clock, and the initial top part will have 448 shaders."

He said INITIAL top part.. he said nothing about top bin, nothing about it being the 470, and nothing about there not being a better one shortly after.

For the love of god read the things you use as evidence!

 

[dank69]

...

There will be only one shipment of GTX 470s/480s ever
We will have to wait and see.

5870 is slow
As compared to what?

The 58xx parts had unlaunchable yields
Yes, they could not keep up with demand, resulting in raised prices.

...

So far he has been close enough for me...

 

[BenSkywalker]

He said INITIAL top part.. he said nothing about top bin,

the short answer is that the top bin as it stands now is about 600MHz for the half hot clock, and 1200MHz for the hot clock

For the love of god read the things you use as evidence!

Just posting that to help remind you that you should read things used as evidence

Second, he never said anything like that about the 5870...

Yes he did, I already quoted and linked it.

He claims that there will be one shipment ever on the 470/480 parts and then nVidia will have nothing competitive performance wise until 2011.

Really now, I'm trying to find the angle you all are taking, honestly I am. So far, I get the impression that people see him as nigh psychic because he figured out that a chip with an extra billion transistors is going to run hotter then a smaller one, and that's it. I also haven't seen people commenting on his assertion that the 58xx parts were unlaunchable given their yields.

Everyone think I'm being assinine splitting hairs over finer nuances like not exact clock speeds, shader cores or the other fine details- let's look at the big picture then. People saying Charlie was 'reasonably' accurate agree that the 58xx parts were unlaunchable and there will only ever be one shipment of 470/480 parts? These are very broad based claims, should be easy to agree or disagree with.

 

[Daedalus685]

Just posting that to help remind you that you should read things used as evidence

Yes he did, I already quoted and linked it.

He claims that there will be one shipment ever on the 470/480 parts and then nVidia will have nothing competitive performance wise until 2011.

Really now, I'm trying to find the angle you all are taking, honestly I am. So far, I get the impression that people see him as nigh psychic because he figured out that a chip with an extra billion transistors is going to run hotter then a smaller one, and that's it. I also haven't seen people commenting on his assertion that the 58xx parts were unlaunchable given their yields.

Everyone think I'm being assinine splitting hairs over finer nuances like not exact clock speeds, shader cores or the other fine details- let's look at the big picture then. People saying Charlie was 'reasonably' accurate agree that the 58xx parts were unlaunchable and there will only ever be one shipment of 470/480 parts? These are very broad based claims, should be easy to agree or disagree with.

So I'm a millionaire who sleeps with pron stars nightly am I? If there is no correlation with two statements one makes it is asinine to think that one has to agree with both.. It is total logical failure to assume that if he said two things then all those who agree with one must agree with the other...

I will give you the one on top bin.. seems he did say 'bin'... however, he never did mention the 480 as that part, and he was pretty clear on initial. Again we come back to the fact that finding one part of a statement that was a mistake does not discount the whole of the statement. Though we are talking about me, not Charles.


He never said the 5870 was unlaunchable.. I don't know how many damn times you have to be told that... He was referring to the 280... he was wrong as it clearly was launchable... but besides that.. so what if he thinks yields have to be 80% to be launchable.. many will say he is wrong but that has absolutely nothing to do with anything at hand... By the way, this is the third time I've commented on that... far stretch from you calling me out on ignoring that point.

What is his reasoning for claiming unlaunchable? Is it based on raw yields (thus you infer the 5870 is also unlaunchable according to him) or is he perhaps using the yields to represent cost per GPU. I may not agree that the 280 was unlaunchable (they launched it after all)... but if it was truly not profitable because of yields then I certainly could hear reasons about why it would not be a good idea to launch it if I want to make money.. or if I am a start up company. He never said 5870... you are being unreasonalbe and a fool to claim he did, there is more to it that you do not understand.. and besides that.. he can be wrong about what is and is not launchable and still useful other times.. I don't understand why that is difficult to get.

 

[BenSkywalker]

He never said the 5870 was unlaunchable..

He stated that 62.5% was barely launchable, the 58xx parts launched with 40% yields. I keep harping on these points as they are far bigger in terms of his credibility. He is claimed to be a respected industry insider in this thread(not saying you did, but it has been claimed). He stated that an acceptable yield level was someone over 60% for launching a part. We know the 58xx parts launched well below that and ended up putting ATi in a very nice position(being able to increase their SRP while still being suplly constrained).

Yes, he was clearly wrong on what an acceptable yield level is to launch a part, and this is the key point- everything he has talked about involving Fermi has been about the yields. He doesn't understand the most basic principles of it, and that is the main point.

So, do you think there will only ever be one shipment of Fermis? This has been a major theme with him, the driving theme he has used over and over again. If we are going to ignore the fine details, which I am doing at the moment, the bigger picture must be where his accuracy is held. Fermi can not be manufactured and nV will only produce one round of shipments then cancel it. Do you agree with that or not?

 

[T2k]

For some reason these kind of topics destined to descend into a pathetic nitpicking contest... I wonder why.

 

[The-Noid]

ATi 5870 could come in with <40&#37; yield because it wasn't price constrained. At the time it could have charged more and still gotten away with it. I think 5870 pricing was fair considering there was really no competition at the time. GTX 295 was out, but that was $600.

Overall, Charlie did very well. Shockingly well.

 

[Janooo]

He stated that 62.5% was barely launchable, the 58xx parts launched with 40% yields. I keep harping on these points as they are far bigger in terms of his credibility. He is claimed to be a respected industry insider in this thread(not saying you did, but it has been claimed). He stated that an acceptable yield level was someone over 60% for launching a part. We know the 58xx parts launched well below that and ended up putting ATi in a very nice position(being able to increase their SRP while still being suplly constrained).

Yes, he was clearly wrong on what an acceptable yield level is to launch a part, and this is the key point- everything he has talked about involving Fermi has been about the yields. He doesn't understand the most basic principles of it, and that is the main point.

So, do you think there will only ever be one shipment of Fermis? This has been a major theme with him, the driving theme he has used over and over again. If we are going to ignore the fine details, which I am doing at the moment, the bigger picture must be where his accuracy is held. Fermi can not be manufactured and nV will only produce one round of shipments then cancel it. Do you agree with that or not?

What one shipment are you talking about?

 

[T2k]

Overall, charlie did very well. Shockingly well.

qft

 

[Daedalus685]

He stated that 62.5&#37; was barely launchable, the 58xx parts launched with 40% yields. I keep harping on these points as they are far bigger in terms of his credibility. He is claimed to be a respected industry insider in this thread(not saying you did, but it has been claimed). He stated that an acceptable yield level was someone over 60% for launching a part. We know the 58xx parts launched well below that and ended up putting ATi in a very nice position(being able to increase their SRP while still being suplly constrained).

Yes, he was clearly wrong on what an acceptable yield level is to launch a part, and this is the key point- everything he has talked about involving Fermi has been about the yields. He doesn't understand the most basic principles of it, and that is the main point.

So, do you think there will only ever be one shipment of Fermis? This has been a major theme with him, the driving theme he has used over and over again. If we are going to ignore the fine details, which I am doing at the moment, the bigger picture must be where his accuracy is held. Fermi can not be manufactured and nV will only produce one round of shipments then cancel it. Do you agree with that or not?

Why don't you understand that he can be legitimately wrong about many things yet still provide useful information.

I have nothing further to say about the 5870 and not being launchable thing. He said the 280 was unlaunchable because it had 60% yields... That could mean any number of things, and while all of those things may be wrong in the end, it has NOTHING to do with the 5870.

He can legitimately think that the 5870 was launchable and the 280 was not, despite the 5870 having lower yields... Using price/GPU is just on eway he could validate this (not saying he did, or it is correct to do so.. just that it is valid).

There is no correlation between the two.. your comparison is not at all valid.


I don't know how many launches of Fermi there will be (Have not seen a single one yet..)... Now bear in mind he only said there would only likely be one launch of the 512 core part he thought there would be... Will you make a post claiming he is right if in June we get a 485 that has 512 cores but is released as a one time special edition? You have to get yourself free from the absolute BS. If you can't take anyone who is ever wrong seriously you may as well plug your ears for the rest of time...

 

[yh125d]

Oh for fucks sake....

No one is saying he is 100% accurate.. it is only in the crazy little Jedi world where things are either 100% or 0%....

Do (100% accurate) or do not (0%), there is no try (1-99%)

 

[Daedalus685]

Do (100% accurate) or do not (0%), there is no try (1-99%)

You just won the Internet sir.

 

[BenSkywalker]

What one shipment are you talking about?

The initial launch shipment. He stated how many parts were going to ship before the product was EOL, that is going to be covered by the first shipment to hit retail.

I'm honestly trying to find the accuracy that people are talking about does it really come down to-

Going to ship later then ATi
Going to be bigger then ATi
Going to be hotter then ATi?

Is that what everyone sees when they talk about the accuracy? I'm trying to grasp this, we have his biggest claim left on the table, and noone wants to say he's right so far- Fermi can't be mass produced. I'm ignoring the finer details, are the three points listed above why people consider him accurate?

Why don't you understand that he can be legitimately wrong about many things yet still provide useful information.

What information? That is what I'm asking. People say look at the OP, I go through the list and none of it is right, so I look to the broader picture that he has left, and the only things I come up with are the three general points above. All of those things were known before Charlie said anything.

 

[dank69]

The initial launch shipment. He stated how many parts were going to ship before the product was EOL, that is going to be covered by the first shipment to hit retail.

I'm honestly trying to find the accuracy that people are talking about does it really come down to-

Going to ship later then ATi
Going to be bigger then ATi
Going to be hotter then ATi?

Is that what everyone sees when they talk about the accuracy? I'm trying to grasp this, we have his biggest claim left on the table, and noone wants to say he's right so far- Fermi can't be mass produced. I'm ignoring the finer details, are the three points listed above why people consider him accurate?

The claim was that the chip was going to be so big/hot/power hungry that the yields would be so low that the chip would be "economically unmanufacturable." Funny how people leave out the "economically" when quoting him. Basically he said that Nvidia would not be able to turn a profit on this chip. We still have to wait and see but all signs point to yes.

 

[Daedalus685]

The initial launch shipment. He stated how many parts were going to ship before the product was EOL, that is going to be covered by the first shipment to hit retail.

I'm honestly trying to find the accuracy that people are talking about does it really come down to-

Going to ship later then ATi
Going to be bigger then ATi
Going to be hotter then ATi?

Is that what everyone sees when they talk about the accuracy? I'm trying to grasp this, we have his biggest claim left on the table, and noone wants to say he's right so far- Fermi can't be mass produced. I'm ignoring the finer details, are the three points listed above why people consider him accurate?

The accuracy we are referring to includes many things. Such as how the 512core part will likely not see the light of day. That the card will not blow the 5870 out of the water. That there are leakage problems, yield problems, etc. That the supply will be limited.

While he may say "ZOMG bloody Fucking FAilure Nvidia only going to release one truck of cards then EOL fail the entire Fermi", reasonable folks can glean from that such information as "Oh, I guess the supply will be pretty limited." We don't focus on the detail that there is exactly one truck of the things... as it largely doesn't matter. I do not think it will be easy for me to get a gtx480 even if I wanted one. Ever. I also thought it was nice to know just how power hungry the things will end up being.

You can't take him so seriously Ben..... You'll have a heart attack.. If him being wrong a few times means you can't trust a word he says then don't... but don't try to convince others to ignore him based on what amounts to a logical blasphemy.

 

[Janooo]

The initial launch shipment. He stated how many parts were going to ship before the product was EOL, that is going to be covered by the first shipment to hit retail.

...

Do you have a quote? Where does he say that it's only initial launch shipment and nothing else?
I don't believe 9,000 risk wafers is equal to one launch shipment.

 

[Daedalus685]

What information? That is what I'm asking. People say look at the OP, I go through the list and none of it is right, so I look to the broader picture that he has left, and the only things I come up with are the three general points above. All of those things were known before Charlie said anything.

If you are only capable of seeing right and wrong then there is no point in explaining anything to you... It would be like teachign a shark to fly...

There is much more to his articles than the biased drivel, as well as the point and shoot reduction in the OP... Hence my immediate dislike of trying to reduce it as such... He doesn't write with a point list, to create one just allows folks like you to say they are objectively wrong even though the points were subjectively created.

 

[The-Noid]

Do you have a quote? Where does he say that it's only initial launch shipment and nothing else?
I don't believe 9,000 risk wafers is equal to one launch shipment.

With how bad the yields supposedly are it may be

Let's just be honest here with what GTX480 is. It is a massive undertaking that hasn't panned out (yet). Nvidia probably will attempt refreshes at this mArch for the next 2-4 years and by then it should have the ability to create a much better running chip. Right now the part is exactly what Charlie said it was, HOT, UNMANUFACTURABLE, and HUNGRY FOR POWER.

This is not what Nvidia wanted Fermi to be, this is not what the AIB's wanted fermi to be, this is not what anyone on Team Green wanted fermi to be, and this is especially not want the market and Nvidia shareholders wanted Fermi to be. Right now it is a big, hot and expensive disaster.