Nvidia GT300 core: Speculation

[Kowan]

Sometimes the biggest "punishment" is to simply ignore someone if you think he doesn't fit your imagery or you strongly disagree with him. It doesn't cost anything and you don't add to his "value" (publicity doesn't come necessarily with quality).

That's what I do. Imagine a Paparazzi without a camera...

 

[neliz]

He has shown he not only reports false information, he also makes stuff up. He is also a liar. The stuff he does I think is criminal. You can not make stuff up about a company that can have damaging effects on their price per share. His lies about a 2% yield on the new chips saw a selloff from their 52wk high the day before at 16.5 or so. The stock is now down about 15% since his article.

He should be investigated. You cannot do stuff like that.

Charlie doesn't make stuff up, he posts what is told to him. Much like anyone else here.
nvidia's "our yields are super, didn't you see the half a million 40nm products we've shipped this year?" statement is based on the same "facts" as Charlies 1.67% number.

you're talking about corporate spinners here who's interest it is to present the unknowing with their version of the truth. This are the people that say GT300 has been taped-out in the beginning of this year and there's no problem with the yields.

All in all, I wonder why there are still people that think that nv's 40nm products are flawless. Then why has no board partner any to show?

Certainly Sir, have you verified these messages above or just Rumors ?

It's the same people that said that GT215 is coming in December for a couple of months. So far that hasn't showen up either.

 

[Florin]

nvidia's "our yields are super, didn't you see the half a million 40nm products we've shipped this year?" statement is based on the same "facts" as Charlies 1.67% number.

What, Nvidia gets their numbers from mistranslated forum posts now?

You guys

 

[neliz]

More crystal ball doom forcasts? You must be Charlie's insider.

Oh, Charlie's insider is much more reliable than I am.

So far his accuracy is 33% on the "back to school"
Fuad is a firm believer of a launch after Win7.. so that makes Charlie 67% correct.
[strike]My source[/strike]"An unnamed source high up in management" tells me GT300 will lead the Cypress refresh by about 2 Months. (that should make Cyphresh DX11 done right-er according to fudo logic, briding Q1 and Q2 in 2010.)

 

[Razor1]

Oh, Charlie's insider is much more reliable than I am.

So far his accuracy is 33% on the "back to school"
Fuad is a firm believer of a launch after Win7.. so that makes Charlie 67% correct.
[strike]My source[/strike]"An unnamed source high up in management" tells me GT300 will lead the Cypress refresh by about 2 Months. (that should make Cyphresh DX11 done right-er according to fudo logic, briding Q1 and Q2 in 2010.)

That's a hell of a lot better batting average dan Degustators "we'll have pictures soon (around Computex)" back in May.

neliz, 33% on back to school isn't that good, if he didn't know anything about the dx 10.1 cards, what is he going to know about the g300, and no he said there will be no win 7 parts, all dx 10.1 cards are win 7 parts that is the minimum requirements for it, thats why I knew he didn't know sh*t the moment he posted that, he just comes out and says things to get attention.

 

[Ailuros]

Ladies I'm sorry I'm going to spoil your happy little tea party but in case any of you hasn't noticed this isn't material for a place like B3D. If you have something to contribute on the topic then by all means let's hear it, but that juvenile pissing contest of X vs. Y leaves me fairly in the cold.

Or else feel free to carry on in private or in SA's fora even better; that is if you're even allowed to have an different opinion than the administrator and his entourage there....oooops my God it's contagious aaaaaarghhhhh......

 

[Scali]

But I can't see them outright lying about insanely good performance and then not delivering, I think they'd see their own fanbase turn against them so who knows.

Why not? It worked for AMD with the Barcelona

 

[Chris123234]

He has shown he not only reports false information, he also makes stuff up. He is also a liar. The stuff he does I think is criminal. You can not make stuff up about a company that can have damaging effects on their price per share. His lies about a 2% yield on the new chips saw a selloff from their 52wk high the day before at 16.5 or so. The stock is now down about 15% since his article.

He should be investigated. You cannot do stuff like that.

Not his fault stock market people act stupid at times in attempts to be as greedy as possible. I have no respect for the virtual legal mandate of stock price increases anyways.

 

[ShaidarHaran]

He has shown he not only reports false information, he also makes stuff up. He is also a liar. The stuff he does I think is criminal. You can not make stuff up about a company that can have damaging effects on their price per share. His lies about a 2% yield on the new chips saw a selloff from their 52wk high the day before at 16.5 or so. The stock is now down about 15% since his article.

He should be investigated. You cannot do stuff like that.

Are you actually contending that Charlie Demerjian is responsible for the actions of investors?

Anyone that takes investment advice from a site named "semi-accurate.com" deserves whatever they get.

 

[AlexV]

Gentlemen, a minor jump in signal about the GT300 and a major reduction in noise about Charlie Demerijan and feelings for him would be very much appreciated. Thank you.

 

[Ailuros]

My hero! ooops

 

[Arty]

Where's that popcorn smilie when you need it hm?

Arty,

Not that I know anything but on a pure layman's speculative basis:

Assume they recognized early enough that the high end part is going to delay their first X11 introduction even more, let's say that they might have had enough time to pull forwards a slightly castrated part. That would be of course if all the above is for real and we're talking about a 384bit only buswidth (and not a bunch of disabled units).

So you are talking about bringing forward a G302 or something like that? Assuming G300 is 30-40% faster than Cypress, the castrated part might not give them the advantage they might need. This is the only problem I see with this theory.

You have reached a blank page in my mind here.

If Nvidia is truly launching in November (best case scenario), then they need a significant performance advantage over Cypress to make it look like a bad deal. Or they should be ready to offer this product at a lower price but ATI would also follow up with a price cut, so I'm sure if launching a product 10-20% faster 3 months late is going to work. They have to have a significant lead, I think trinibwoy was also suggesting something like this in one of these threads.

Doesn't sound impossible; if I combine the hypothetical dynamic warp idea in my mind with the possibility of SPs being also capable of DP it could work.

Sounds good, but can it be a major selling point?

Assuming (1) is true sounds doable.

See above, only plausible if this castrated part allows them a significant performance lead or launch at a lower price point.

A large pile of useless marketing yadda yadda which are typical for each new product's launch

Sounds typical.

If there were anyway they could, I'm pretty sure they would. I don't think they're exactly holding back their launch right now for strategic purposes.

Yeah, that makes perfect sense.

I'm pretty sure ATi could respond with an X2 of their own.

But will it be enough?

What I fear is a paper launch of a card that looks to have >50% performance increase over 5870 where they claim to have working hardware that a very few handful have reportedly seen but can't discuss, in other words a FUD campaign from hell.

But I can't see them outright lying about insanely good performance and then not delivering, I think they'd see their own fanbase turn against them so who knows.

That would be the absolute worst case scenario and they need to screw up further than they actually have until now.

What, Nvidia gets their numbers from mistranslated forum posts now?

You guys

That number was backed up by Kyle so psssh.

 

[fellix]

GPU TechCon schedule

 

[no-X]

We should get some info much sooner. There are some NDA slides flying around, which will be public very soon :smile:

 

[Ailuros]

So you are talking about bringing forward a G302 or something like that? Assuming G300 is 30-40% faster than Cypress, the castrated part might not give them the advantage they might need. This is the only problem I see with this theory.

The rumor in that former link was speaking of a 512SP/384bit chip with the 512bit chip following later on. If true that would suggest a high end chip with castrated ROP partitions and not a performance GPU.

Albeit frankly yes ROPs might take up quite some die area but I can't figure out why theoretically 6 out of 8 partitions would make such a big difference to yields.

If Nvidia is truly launching in November (best case scenario), then they need a significant performance advantage over Cypress to make it look like a bad deal. Or they should be ready to offer this product at a lower price but ATI would also follow up with a price cut, so I'm sure if launching a product 10-20% faster 3 months late is going to work. They have to have a significant lead, I think trinibwoy was also suggesting something like this in one of these threads.

Of course would they have to have a significant lead; I just don't know or pretend to know (which is quite a popular trend lately during silly season *cough*) what the case exactly is. Now let's give the former rumor and the above thought a nice logical chain to follow: it sounds like a relatively bandwidth starved solution to me which could be good enough to surposs the competition's performance GPU by a good enough but not the best enough margin.

Sounds good, but can it be a major selling point?

The maximum peak double precision throughput of a 5870 is at 544GFLOPs/s according to AMD itself.

See above, only plausible if this castrated part allows them a significant performance lead or launch at a lower price point.

The scenario isn't even plausible from the get go in my mind. It could be some el weirdo test/dev sample like you can find from about any IHV before each launch, it could be a chip that is meant for a dual chip SKU, it could be some false translation, or Lord knows what else. Shall I remind you how specs for Cypress fluctuated until they finally came to the light of day? It nearly sounded like a rubber chip that you can stretch into all 4 directions.

So the die area D12U is supposed to consume varies between 452 and 576mm2.

Sounds typical.

Like the one from AMD for the Evergreen launch or an even older one:

http://users.otenet.gr/~ailuros/evergr.pdf

 

[digitalwanderer]

GPU TechCon schedule

California? I think I'll just listen in on this one.

Odds that they'll mention the GT300, high! Odds that they will announce an actual launch date, well that seems to be the million dollar question right now.

 

[digitalwanderer]

We should get some info much sooner. There are some NDA slides flying around, which will be public very soon :smile:

Really? Any juicy info on 'em or are they just fluff PR ones?

 

[SiliconAbyss]

Really? Any juicy info on 'em or are they just fluff PR ones?

They're fluffy AND juicy.

 

[MfA]

The maximum peak double precision throughput of a 5870 is at 544GFLOPs/s according to AMD itself.

Which a 512 SP chip is unlikely to crack without a 2 clock FMA throughput ... that's expensive, a little over half the multiplier hardware would be idle in normal operation.

 

[no-X]

Really? Any juicy info on 'em or are they just fluff PR ones?

I'm quite curious, what excuse will Jen Hsun use this time. If GT200 was "too much CUDA", how will he denominate its successor,

Seems they aimed much effort at nonexistent markets... Anyway, slides are one thing, reality another