But not in the form expected (Tesla), 7.1 Billion transistors, Looks like the rumours about G104 being midrange rebadged were right.
http://www.nvidia.com/content/PDF/ke...ay_2012_LR.pdf
But not in the form expected (Tesla), 7.1 Billion transistors, Looks like the rumours about G104 being midrange rebadged were right.
http://www.nvidia.com/content/PDF/ke...ay_2012_LR.pdf
The most interesting thing to me is the comparison to the 680
e.g.
GK104 = 3.54 billion transistors
GK110 = 7.10 billion transistors
It sounds like it might offer double the performance!![]()
You do realise ORNL is only going to get GK110 based Kepler K20 cards in early 2013.
The latest rumours indicate that August would be the earliest a GK110 based consumer card might be out. OTH,that would make the GTX680 be "top dog" for only 4,perhaps, 5 months which would be less than the GTX480.
Knowing what Nvidia did with the G80 and G92,they will release what they have and not hold back. Remember how the AMD 2000 and HD3000 series performed against them(poorly). If the GK110 was ready by the time of the GTX680 launch,Nvidia would release retail cards or at the very least show demos of games running on it months before. If people are buying £800 GTX690 cards or GTX680 SLI or HD7970 Crossfire setups,there will be those who would spend £600 to £700 even on a GK110 based card.
Even if it is a second generation Kepler GPU,then you cannot say whether is twice as fast,50% faster or 150% faster.
For instance a GTX580 is not 50% faster than a GTX560TI despite having 50% more transistors,and there is a reason why this is. There is also the aspect of added redundancy for bigger chips and from what I gather it will be the biggest chip Nvidia has designed.
Last edited by CAT-THE-FIFTH; 16-05-2012 at 06:09 PM.
I disagree. A company that plans years in advance does not need to show its hand just because it can.
All businesses are about making money. The people buying high-end / bleeding edge are only a small percentage of customers.
These same customers will likely by the latest and greatest. If you had a product that outmatches your rivals and another that blows them away, you would keep the better one a trade secret until the market is right.
Why? Well you can sell the lesser card at a more profitable price point. If this pays for the R&D the secret card will gain better profit margins in the future. If they showed it now then sales of the GK104 would be affected. It like shooting yourself in the foot. Not a good idea.
Also...Why do you always attack peoples comments? There is just no need.
You don't recoup R and D if you do not sell what you have developed and the last 6 years of how Nvidia has acted,indicates they release what they can make in sellable quantities.They do not hold back and it is only some conspiracy that they have.
The G80 and G92 are excellent examples of this - AMD was not hugely competitive with the HD2000 series and even the HD3870 was not faster than the 8800GT. Nvidia had no reason to launch an 8800 Ultra,let alone an 8800GTS 512MB(which I owned BTW).
I also updated both Kepler rumours threads on Hexus and OcUK for months,as I have been following Kepler for quite a while - the same goes with AMD GCN and Intel MIC.
If it had been production ready in February or March,people at places ORNL would be playing with them at that time with prototypes and we would hear about it. We are hearing about Intel Knights Corner already from researchers and that is not being released to fulll production until the end of the year AFAIK.
That is thing though - with Intel entering the fray and AMD improving compute ability with GCN,it makes no sense for Nvidia to hold back. If anything,it is in there own long term interest to marginalise the competition.
I think it was not large scale production ready and August seems to be the earliest date ATM. People forget one thing though - fab costs. Nvidia is not happy that the current 28NM wafers are very expensive and if anything they are probably doing what AMD did with the HD4000 and HD5000 series. Make a smaller chip which in this case negates the massive rise in wafer costs and in theory should make it easier to make. IIRC,the agreement with TSMC for the 40NM process was completed chips since Nvidia whinged at them,but now it is full wafers AFAIK including the non functional chips(I could be wrong).
It makes a lot of sense.
Attacking what?? That I don't agree with what you said??
You should see some of the other members on here during some of the more heated debates!! That is when the mods step in and waive their magic wands!!
The forum is actually quite peaceful ATM.
I only questioned whether it is valid to expect a doubling of performance for a doubling of the number of transistors.
The higher end Kepler cards have compute considerations too which adds to the transistor budget and internal re-organisations towards compute too. The GTX580 is not 50% faster than a GTX560TI as a result even with 50% more transistors.
Now it could be a further refined second generation Kepler architecture,but even with that consideration I would be very impressed if it is twice as fast as GTX680,ie, faster than a GTX690. That would also make the GK114 very impressive.
Last edited by CAT-THE-FIFTH; 17-05-2012 at 12:11 AM.
Just checked again,it seems between September to December is when ORNL will be starting to get their K20 cards:
http://www.olcf.ornl.gov/computing-resources/titan/
It seems the upgrade will done by early 2013.
I shall bring the Moose with his finest antler wear.
Last edited by CAT-THE-FIFTH; 16-05-2012 at 07:52 PM.
Platinum (17-05-2012)
I am still not sure that the GK110 is or ever was intended for graphics cards, the rather compute light GK104 excels at graphics and the 110 with it's far more complex compute units might not end up that much faster in graphics anyway, one thing that people don't seem to realise is that nVidia don't just make graphics these days, the Tesla range is a huge money maker for them, and designing a chip just for that purpose is definitely something they would consider, and if they got to sell a few of those chips that don't make the grade as a tesla part as graphics cards, then it's a nice bonus.
*̡͌l̡*̡̡ ̴̡ı̴̴̡ ̡̡͡|̲̲̲͡͡͡ ̲▫̲͡ ̲̲̲͡͡π̲̲͡͡ ̲̲͡▫̲̲͡͡ ̲|̡̡̡ ̡ ̴̡ı̴̡̡ *̡͌l̡*
Originally Posted by Winston Churchill
There are currently 1 users browsing this thread. (0 members and 1 guests)