Printable View
Wasn't even their article to begin with, they translated it from some site "hardspell". They weren't privy to secret knowledge directly from Nvidia. Check out the inquirer link I put up, sez all there, even though theINQ is just at bad spinning stuff, Fuad usually can admit a mistake.Quote:
Originally Posted by DilTech
Ok, I'm going to throw some obvious things out there, which presumably everybody already knows.
3D redenring is all aout parallelism. Its also about using as many shortcuts and clever tricks to do as little work possible and still get the end effect. A G70 (7800) isn't faster than NV40 (6800) only because of higher clock speed or even increase of number of shader units.
Undoubtedly G80 will be very fast. And it will surely be faster than current G71. However, it is not necessary that it have higher clockspeed or higher transistor count to achieve this.
By extensive use of compression, HyperZ and other clever tricks to increase efficiency, Radeon 9700pro was able to provide uprecedented Antialiasing and Anisotropic Filtering perfromance. nVidia could have doubled GF4 clockspeed and not matched the results. Even if they doubled the number of execution units and doubled clockspeeds, their design would still be crippled by its existing inefficient algorithm.
I hope you're still with me on all this.
That is why nVidia's next generation part will be assigned next generation name... because its not just increased clockspeed or increase of execution units which will deliver its performance, but fundamental improvements to the architecture.
(PS: Please stop that stupid 1/3 transistors for DX10 balloney.. if its being designed for DX10, then its being designed for DX10. Making separate sections on the chip for each DX version would be just plain stupid. If the whole design is developed for DX10, then ALL the architecture and logic (and thus transistors) is made with DX10 in mind.
for your amusement,
NV40 info, from vr-zone feb 14, 2003:
"0.09 Micron Process
300-350 Million Transistors
750-800 MHz Core
16MB Embedded DRAM (134M trans.)
1.4 GHz 256-512MB DDR-II Memory
8 Pixel Rendering Pipelines (4 texels each)
16 Vertex Shader Engines
204.8 GB/sec Bandwidth (eDRAM)
44.8 GB/sec Bandwidth (DDR-II)
25.6 GigaTexels per Second
3 Billion Vertices per Second
DirectX 10 (or extended 9.1)
Release H2 2004
Source: Quake "
edit: if you are having trouble figuring out where is the amusement, I'll give you a hint.. each and every single thing is wrong.. it even doesn't match 2006 G71 specs.
First things first...the last thread was pulled because of flaming. As a matter of fact, your last post was the one that pushed it into abyss of dead threads. n.b., watch what you say, you're getting close to an official warningQuote:
Originally Posted by phi|os
As for why VR-Zone pulled their link...only they know. We all have our theories...:)
The interesting thing about these specs is the outside-the box extremeness combined with complete plausibility and logic. DX10 is going to require a serious step-up in transistor count (no, it's not a separate section...but to do the same amount of work, it needs many, many more transistors) and power usage, ATi and nV have both made public apologies for this and said that this generation they went for all-out performance disregarding power usage. Future generations would focus on getting that/better performance at lower power usage figures.
The daughter-die in Xenon was one of the greatest ideas for image quality, no doubt we'll start seeing more and more of that throughout the entire market. A multi-die solution also enables physics to be done 'on-card'. The dual memory controllers also hint at this. Dual-dies will become common-place in the market over time. For this upcoming gen, the 700m transistor count probably is split over two dies (at 90nm or 80nm the die is just too big), and the dual memory controllers also hint at this. Again, doesn't mean it's true, or even '50%' true, but it means it's plausible and the the pieces fit together.
There's nothing in this design that's out of reach or particularly extraordinary, it's just completely untraditional. Again, it doesn't mean that this is true, but in no way does it make it false either.
Quote:
Originally Posted by ***Deimos***
Now notice, that article didn't get removed. This one did. Goes along with my point that fake news never gets pulled from VR-Zone, only news that gets the company they're talking about's attention gets pulled.
This tidbit, obviously, did. Infact, even hardspell has pulled it, which is VERRRRY rare as it is. I can't think of a time they've ever pulled....well.... anything.
This is the reason I'm digging away at this info, there's some reason sites are being made to pull it down after all.
from i heard from one of the NV brand.....
g80 will take about 200~250W
and cooler made by CM
A aquagate modified to the pcb dimensons perhaps?
Vapor, I see you like multi-chip solution.
A little walk down memory lane. multi-chip solution was the death knell of 3Dfx. Maybe a fluke huh? Well, how well did Rage Fury Maxx do? Do you really think nVidia is going to lay the entire future of the whole company on a new untried unproven scheme?
Besides, I dont understand who so many of you have a fixation that physics should be done seperately on another processor (or worse yet, on a seperate chip). WHY?
7900/X1900 have already demonstrated ability to accelerate physics immensely and much more elegantly because of general pupose processing vs more rigid function in Ageia.
Likewise, why do people insist that many more transistors are required for DX10. What specifically creates this demand? DX7 meant adding a large group of fixed function circuitry. DX8 expanded that to include general integer processing of pixel and vertex values. DX9 made it floating point to enale higher precision and range. DX10 isn't complicating things.. its simplying them. As far as I know it still uses FP32. There will still be bump maps, specular maps, normal maps. Textures and MIP maps wills till be there. As will vertex meshes and keyframe animation. From what I understand DX10 simplifies the whole shader programming aspect. Instead of using separate similar code/hardware for vertex shaders and for pixel shaders... instead combine into general pupose unit which can do either one or both. It simplifies things for developers, and allows them more freedom to do more things.
turtle, I very highly doubt nVidia would make a design that would necessate water cooling unit. Perhaps maybe a special OC version from BFG or something, but surely not standard. Not practical. Not economical. Too complicated and error prone installation for user.
guess2098, if one G80 indeed puts out anywhere remote close to 250W, I'm prepared to eat poo for breakfast, lunch and dinner. Thats a higher power conumption/ heat output, than a WHOLE Athlon64/Core2Duo computer system - including CPU, memory, motherboard, hard drive, 7900GTX video card and accessories.
The X-box 360 GPU is actually split between 2 dies, and it works fine.Quote:
Originally Posted by ***Deimos***
So no, it's not untired, or unproven.
@***Deimos***
3dfx died cause of many reasons and not just multi chip designs.
and btw high-end cards arent the thing that makes most of the cash for companies like ATi and Nvidia its simply the cardds they use for marketing.
So its unlikely that if G80 is multi chip theyl also use that for the mid-end and low-end market.
If the use multi-chip because physics on seperate chip as some here propose, why would they integrate it for mid-low end and not for high end?Quote:
Originally Posted by Starscream
Mark my words, if they become reliant on multi-chip solution, they will fail in the marketplace.
u know... my friend had meeting with Nvidia last week...
he is ..... hummmm someone relat to you video card XD
maybe he iswrong :D
Quote:
Originally Posted by ***Deimos***
you know you forget one really importaint new card on your list called 7950gx2
oooh lets look at the specs
500mhzx2 frequency (would be called 1000mhz in marketing dept)
512mbx2 ram (again 1gb is whta they call it)
24x2 pipelines or 48 pipes for the marketing department
cmon man can you not see how this works.
looks like a dual core card with the gpu totaling 1500mhz somehow it sounds like it may be lopsided with one core being more powerfull and one less
im thinking like a new tech hybrid 7900gtx paired with a 7600gt to do all the antaliasing work or something like this or maybe something much newer and smarter.
hey the 7950 card makes use of 556million transistors therefore that is in practice a 556m transistor card not far off of the 700 they claim
all im saying is think of what these specs could mean instead of disgarding them so fast
Yes, finally someone thought of it! I was thinking it tooQuote:
looks like a dual core card with the gpu totaling 1500mhz somehow it sounds like it may be lopsided with one core being more powerfull and one less
im thinking like a new tech hybrid 7900gtx paired with a 7600gt to do all the antaliasing work or something like this or maybe something much newer and smarter.
you make a very incitefull point. That was also what I first thought.Quote:
Originally Posted by brandinb
250W is for two GPUs/cards, single is 125W
750M transistors is two GPU's/cards, single is 375M
1500Mhz is two GPU's/cards, single is 750Mhz (perhaps at 65nm, but not right away first generation).
1GB is two GPU's/cards, single is 512MB.
it all sort of makes sense then. Well, at very least its at least within the realm of feasability.
Still not sure what to make of the 384bit memory interface rumours.
-------------
in 2004, nVidia luanched 6800s. Secret weapon was SLI. Unfortunately, now they're screwed. When they launched 7800GTX in 2005, they had to make it SLI to increase performance past 6800SLI. Similarly with 7900GTX. nVidia made it even more complicated with quad SLI. Now,they've set a very high bar of performance to beat. Naturally everyone will compare G80 not only to a single 7900GTX, but also to 7900GTX and even quad SLI!!
So, are we doomed to have SLI/CF video cards, and motherboards to the end of eternity? Its getting kind of ridiculous. Afterall, what's next.. 8 video cards?
Objections with SLI/preferance for one chip solution:
- efficiency is only good in select few games, otherwise poor.
- redudancy of data, resources, hardware, processing.
- not elegant.
- special motherboard with special chipset required.
- big power consumption, and heat output. Cooling problem because two very hot cards right next to each other.
- although appears you have, say 1GB of memory, you really have access to only 512MB.. the other is just a copy.
- when its time to upgrade, you have twice as many obsolete cards to sell.
- you have double DVI/VGA ports, purevideo, 2D.. but most of this stuff goes unused.
x1800xtx was going to have 32 real pp's i remember lots of link given here to prove that and lots of the people here stated lots of logical statements that it will surely have 32pps today we didnt even see 32 pps together
we believe what we want to believe
I thought that too, until I saw this:Quote:
Originally Posted by Starscream
http://www.theinq.net/default.aspx?article=34503
http://www.theinq.net/images/articles/jpr.gif
*back on topic*
The idea that it is a 2 GPU single card (a la 7950 GX2) would validate those specs. and nvidia should have a working quad sli driver out by then, maybe they are even holding off till then?
Judging by the info we have...the two dies will not be identical (from: mem controllers). Therefore, they won't necessarily be SLI'd together....just natively 'together' (i.e., one won't work without the other and really don't need any software to get them talking/functioning).
The talk of drivers is intriguing....DX10 and a outta-left-field tech-structure both debuting at the same time has gotta be a strain on the driver dept. At least it'll leave tremendous room for improvement....
Of course, nVidia might just put it all on one die and laugh at our speculation.....or be kicking themselves as to why they didn't do a split-die to get better yields :p:
Quote:
Originally Posted by Vapor
I bet we will all be wrong!
I bet anyone a cookie (that famous cookie bet) those specs will be alot diffrent when the final product is up for sale :p:
Quote:
Originally Posted by ***Deimos***
Who said mid- and low-end have physix integrated?
if the g80 is a 2 chip solution and the 2nd chip is like guessed a physix chip this is simply (like i said before) a way to make the perf gap between high-end and mid-end bigger.
So that people that spend the extra cash for a high-end card get more for their money.
2 single chips 1 for grafix and 1 for physix should be better then 1 chip for both (if done properly).
The idea of a seperate physics chip on card may have some extra credence:
http://www.reghardware.co.uk/2006/09..._nvidia_beans/
The same die may be found on add in cards as well.
Although i'd imagine physics on gpu would be more likely than an nvidia produced specialised physics chip.
Something about that just seems wrong. I bet when they include 6600's in the performance and ethusiast. Also, how does one determine the value of an integrated GPU, since it is part of the chipset?Quote:
Originally Posted by Stuperman
Quote:
Originally Posted by onewingedangel
ya saw that new Asus board a few hours ago.
but before we all start to expect news from nvidia on their physix plans we better wait till more nvidia partners put 3 16x PCI-e ports on their boards.
As it can just as well be an ASUS marketing thing.
@j-mag
i agree that thing looks a wierd.
As i always thought that the profit margin on high-end cards was low due to the expensive ram and bigger cores etc.
But if that pie chart is true they either make a crapload of profit on high-end or the chart info is wrong.
That doesn't look like a 'profit' chart. Looks more like revenue. The overhead markup could very well be lower on high-end cards.
Ryan