Discussion in 'News' started by btarunr, Aug 31, 2010.
It was already scaling well on GF100. Check the review charts
being an ati fan this gpu looks spanking! this would defiantly be my gpu of choice.
who the hell is "colorful"???
900 MHz core clock....pci-e extensions...original cooler design...DO WANT!
I will buy this, but this company is not listed at newegg, where to buy?
really? ..hmmm they used to be
it says the card is not released yet, and when it is it will be to the asian market...
perhaps the company itself decided to only sell/ship to the asian market now.
That card looks absolutely mental! I'm liking the heatsink design.
I really want two of those.
I seriously doubt nV would be revising GF104 past A1. There's no apparent problems to fix.
I wouldn't call 10% OC (700MHz -> 770MHz) anywhere near good scaling. I'm guessing the L2 cache, which GF104 lacks, is holding Fermi back.
I believe he is talking of SLi scaling, and you of OC scaling.
so I guess it doesnt take a second core revision to unlock all the SP's on die?
bad fucking ass
No, it's not needed.
Seing how good GF104 is in comparison, if Nvidia has another chip in their sleeves that's going to be a "GF104"-like GPU with 3 clusters instead of 2. IMO that would definately compete and even beat a Southern Islands all while being smaller than GF100 so they would be in a better situation than now (GF100 has 3.1 billion transistors, GF104 has 1.9, so 1.9 + 50% will always be less than GF100). But IMO they could even aim a little bit higher and take a "minor" risk and release a real monster. That mosnter would be the aforementioned 3 cluster "GF104-like" chip but instead of 48 SP (3x16) per SM, it would have 4x16= 64 SP.
Architecturally it makes all the sense: GF104 is superscalar with 2 warps schedulers, but only 3 SIMD execution units. That doesn't make a lot of sense, since in one clock cycle it can issue two warps, but on the next only one can be issued, hence the move to 4 SIMD units is unnavoidable. Now, regarding the die size of such a chip, if you compare GF104 and GF100 it's clear that adding the extra 16 SPs into the SM did not increase transistor budget a lot, considering they also doubled SFUs and TMUs on GF104. GF104 is way more than 75% of a GF100, but has 60% the transistor count and die area. IMO Nvidia might be in a position right now to release that chip consisting of 3.2-3.4 billion transistors and around 560 mm^2 die area at 40nm (not the biggest chip Nvidia has ever done). It would be a risk, yeah, again yeah, but a risk that may very well pay off, we are talking about a chip not much bigger than GF100 but with 768 SPs! But that's where the Fermi architecture was created to go and beyond that. Next on 28nm, who knows, it could be 4 clusters with maybe 96 SPs (6x16) per SM and 6 SMs per cluster for a total of 2304 SPs!! (Impossible? R600 started off with 320, Cypress has 1600, SI will have how many?)
Nevermind just a little bit of speculation...
As of GF104: It's taking them too long to unlock them tbh. I've been wondering lately, if they are not into price fixing practices again (Ati and Nvidia). AMD doesn't lower prices, Nvidia does not want to be aggresive with GF104, it makes me wonder, really.
The excuse with GF104 has been "improving yields" (although those excuses come from the rumor mill), but no one with a brain believes that. OC potential is so vast that there's never been a valid excuse for not releasing the card at 750-800 Mhz stock. And if clocks are so good there's no way they cannot get 25%-40% of the dies with full 384 SPs and 800 Mhz capable, which is what they would need to create an SKU capable of beating GTX470/HD5870.
your long posts are always an interesting read Benetanegia! and I agree with most of what you're saing too. the changes made to Fermi to create GF104 should well and could well carry onto their next high end monster GPU. bring on 768-sp's and 64 ROPS. Nvidia have generally done well for themselves (performance wise) making monster single GPU's than resorting to using two lesser one's in a single card-dual GPU solution.
Nvidia's high end user base seems to, by-and-large, not care much about the heat and power consumption, which is the main put down of the AMD fanboys.
and a big +1 on a fully enabled (384-sp) 800mhz+ GF104 to ceat the GTX470 and most of the time a 5870 too IMO.
EDIT, check out this speculation thread I made
GF104 refresh, GTX475 speculation
I must admit if i get a 460 it really looks like this could be the one i would get, i love its features and it would be nice to have a card that i leave the stock cooler on for once.
Wow that is one crazy looking card, hmmm i bet it would perform very well!
Dang that card means business and certainly no need to OC the stock over clocks Almost makes my cards heatsink look tiny! Little bit envious to all that get this.
but OnBoard, if you got it you'd have nothing to do to it, wouldn't that take away all the fun of your own modding and overclocking which we all know you love so much
I sure know I love reading about it here
Hypothetically, how would this compare to a 5850? About on par?
this should beat a stock 5850 quite easily, overclock the 5850 however and it would overtake quite quickly again.
Don stop your swearing! My niece is watching
She loves the gpu also!
Seeing this, iCame, almost.
I email them to ask if this card was going to be seen in the U.S. and this was their responce.
Now we got offices in germany, and korea,,,,but no one in USA.
Maybe you could suggest me some dealer there, i could introduce them our nice card.
then you could get one or two pcs from them
how do you think about it?
How they could not know of any retailers here I don't know?
But, I like most Americans tend to think we are the epicenter of all things.
I would love to get my hands on a couple of those though.
She sounds open to testing out the market here.
Maybe someone should ask for a card to review.
Separate names with a comma.