How are they going to connect the 2, interposed like amd or like the status quo gddr5? Or will it be on die?
How are they going to connect the 2, interposed like amd or like the status quo gddr5? Or will it be on die?
I think what is the more pertinent (and quite possibly important) question is will more than just the flagship die use HBM moving forward...... I certainly hope so. It would bode really well for laptops and notebooks if the low end Pascal chip got a 2.25 or 2.5x perf/w boost over GM107. That would be today's r9 290 performance at 65 watts.
Im sure Nvidia will also use interposer. They will just license it.
Im sure Nvidia will also use interposer. They will just license it.
The point is that the CPU won't be tightly integrated on the interposer. Only Intel and AMD can do that.
HBM2 may end up as HMC.
HBM1 will be limited to 1GB stacks until some time in very late 2016 if not early 2017 when HBM2 finally arrives. While you can buy 2 and 4GB HMC stacks today.
The Pascal picture comes to mind with its PCB implementation like HMC would use.
HBM2 may end up as HMC.
HBM1 will be limited to 1GB stacks until some time in very late 2016 if not early 2017 when HBM2 finally arrives. While you can buy 2 and 4GB HMC stacks today.
The Pascal picture comes to mind with its PCB implementation like HMC would use.
Turn the clock 15 months forward and HBM is looking...troublesome. Much lower densities, lower speeds and delays. Its not going to be easy for manufactors if we have to wait for a 4GB+ HBM card till late 2016/early 2017.
Dont take bait.You mean like the way people love HBM because it was developed by AMD?
HBM2 may end up as HMC.
HBM1 will be limited to 1GB stacks until some time in very late 2016 if not early 2017 when HBM2 finally arrives. While you can buy 2 and 4GB HMC stacks today.
The Pascal picture comes to mind with its PCB implementation like HMC would use.
This is just a render, not a photo of the gpu and memory. JHH already talked about how they're implementing stacked RAM on Pascal at the Nvidia GPU conference and he specifically stated that they are not going through the PCB but through an interposer (though he used the term wafer). HBM with an interposer has a large advantage over HMC for many applications by using an interposer.
A render?
Turn the clock 15 months forward and HBM is looking...troublesome. Much lower densities, lower speeds and delays. Its not going to be easy for manufactors if we have to wait for a 4GB+ HBM card till late 2016/early 2017.
HBM troublesome in 15 months from now? Wow, add another fail prediction to your prediction track record considering AMD will already have HBM this summer.
90%+ of all future predictions about GPU tech you post are wrong, especially when it relates to AMD or any tech related to them that will be adopted by other firms in the future since it is the industry standard. Most of this forum can't even remember the last time you predicted anything correctly. You even managed to get all the tech in current consoles wrong. Recently you made statements how Pascal won't be a big leap over Maxwell due to high prices. Prices always fall over time which means your prediction on the Pascal perfoemance leap is also wrong.
Pascal = HBM2
http://nl.hardware.info/nieuws/43807/nvidia-gebruikt-hbm2-en-finfet-voor-pascal-gpus
I think that one is gonna haunt him for the rest of his life. XDYou even managed to get all the tech in current consoles wrong.
To be fair the Shintai, I believe that the console manufacturers would have been wiser to use better tech. Even if he was wrong, his sentiment was correct. As it stands, I believe the lack of power in these machines is helping pc gaming grow. So, I suppose there is always a bright side. I really don't see how they can possibly hope to drag these machines out for 10 years. Then again, they just might....