NVIDIA Fermi: Architecture discussion

trinibwoy said:
Yeah but what's up with that crack about AIB's growing up? Are they pissing and moaning about not having cards to sell so Nvidia is gonna teach them a lesson? Or are they holding Fermi close to their chest to avoid the typical AIB leaks before launch? The latter would make sense if Fermi performance is a big surprise - especially in the underwhelming direction.
Click to expand...

I don't think it's in NVIDIA's interest to piss on its partners, especially the exclusive ones. I'm sure they are in direct contact and the partners know for a long time when they'll receive the first Fermi parts.
As for NVIDIA holding Fermi close to their chest, to avoid leaks, they've been doing it ever since they showed off the architecture itself. The fact that there are no benchmark leaks of even A2 chips, shows how secretive they want to be until they are ready to show it for themselves.
 
Sontin said:
And you base your theory on which facts? The tesla specification? Will they use the A2 Chips for that business?
Click to expand...

I don't think so. Tesla parts go through a different certification process than the GeForces, so clocks wise, they are much more conservative.
 
silent_guy said:
I believe I gave 6... And they're actually plausible.


That's one more thing: the 2% yield number. It makes for a spectacular story, but it doesn't say anything about how much dies can be sold as working, lower performance parts. The moment you add redundancy to the mix, the yield story changes dramatically. It also becomes much harder to compare two different chips based just only area.
Click to expand...

I never claimed that was a result that was generalizable. I was quite specific in saying it was a result for a specific batch of wafers. I haven't commented specifically on GF100 yields since then because other than anecdotal evidence, I don't know any numbers.

<sarcasm> What is this redundancy thing? </sarcasm>

-Charlie
 
trinibwoy said:
Does anyone know what Mr. Bennett is talking about?


http://www.hardforum.com/showpost.php?p=1035036711&postcount=19

Charlie also has an Nvidia article up and it's full of optimism. Must be the holiday cheer kicking in :D

http://www.semiaccurate.com/2009/12/10/fermi-a3-silicon-oven/
Click to expand...

Yeah, NV builds all their high end boards, and sells them to customers as a unit. If you have ever seen their price lists, they say, GPU X Kit - $Y, GPU X Board - $Z. One is a GPU + a few components and RAM, the other is a fully working board. Some parts, usually the high end versions, are exclusively sold as boards.

The lower the part in the line, the more likely they are to let an OEM/AIB do their own thing from day 1. Most high end parts are sold as boards only for a few months, then opened up. The idea is that if volume doesn't support a differentiated design, you don't handicap the little guys, or force people not to make a certain class of board. ATI does much the same thing.

This time around they are going to lessen the reliance on their boards, hopefully to get partners to differentiate a bit. Partners have been promised silicon in late Feb to start their own designs, so if it launches on Feb 1, you will see differentiated boards about 4-6 weeks later, say April 1.

As for the nice article, they had some good news, I wrote it up. I actually don't hate them, they just won't stop lying and playing stupid little power games, so it is impossible to get 'good' news officially.

-Charlie
 
FrameBuffer said:
And NO, I would not say Charlie is comparing Fermi v. Cypress based upon clock frequencies alone.. he is stating that when Fermi info was coming forth a few months back that the target clock rate was approx. 750 mhz and that supposedly current sillycon is yeilding 500Mhz.. and that at the projected 750 Mhz pro nV source were claiming that G100 would outpace 5870 by 40%. It's not THAT outlandish to come to the conclusion that at 66% projected (500 supposed / 750 projected) that performance would drop equally), unless you are claiming that a 50% increase in clock rate wont make a difference in regard to market placement and that there would be no such penalty imposed by a difference in clock rates.. really ?!
Click to expand...

All true, but then Charlie's citing nv pr drones, who told since last spring that fermi would beat cypress by 40%. I cannot help but wonder how exact the guesstimates on nvidias side where with regard to cypress.

Plus: What kind of clock's currently at 500 in fermi and (to what extent) does it affect (the) other clocks in there.
 
Yes, I figured that, it's just that in the direct link to Tom's Hardware given on the previous page, I didn't see the slides from NVIDIA (at least at a quick glance), only graphs from Tom's. On the Nvnews forum thread, I saw the slides from NVDIA. A graph from Tom's may not look too convincing to anyone, but a slide from NVIDIA is something else entirely, I think ;)

These numbers are not too surprising to me (if true). NVIDIA has stated for some time now that they anticipate the new card will be the fastest card on the market when it launches, and they surely knew about the 5970 when making those statements.
 
You must log in or register to reply here.

Similar threads

hoom
AMD: Zen 4, Speculation, Rumours and Discussion
8 9 10
Replies
199
Views
39K
MfA
M
R
nVidia Denver discussion
2 3
Replies
56
Views
31K
A1xLLcqAgt0qc2RyMz0y
A
J
GPGPU running on FPGA
Replies
16
Views
10K
silent_guy
S
W
Xbox One Architecture Panel *Main Console HW Points* Transcript
3 4 5
Replies
89
Views
34K
Deleted member 11852
D
T
Is Quadro's SP count more Marketing than performance?
Replies
5
Views
5K
trinibwoy
T
Back
Top