-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
20nm gpu's ?
- Thread starter Wilko_Dog
- Start date
nm = nanometer
In short, in means smaller tech, more efficient, more 'chips' or tech occupying the same space on a PCB (Printed Circuit Board) - so in theory in means you would have a more powerful device (In this case a GPU) which is smaller, potentially cheaper (Doubt that) and consumes less power whilst being much more powerful.
In short, in means smaller tech, more efficient, more 'chips' or tech occupying the same space on a PCB (Printed Circuit Board) - so in theory in means you would have a more powerful device (In this case a GPU) which is smaller, potentially cheaper (Doubt that) and consumes less power whilst being much more powerful.
nm = nanometer
In short, in means smaller tech, more efficient, more 'chips' or tech occupying the same space on a PCB (Printed Circuit Board) - so in theory in means you would have a more powerful device (In this case a GPU) which is smaller, potentially cheaper (Doubt that) and consumes less power whilst being much more powerful.
Ah I see !
cheers ...
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Generally speaking each new process node brings a shrink in how small they can make the individual transistors on a piece of silicon.
It went 90-80-65-55-40-32-28nm Some are half nodes some full nodes.
Effectively a full node drop means roughly twice the transistor density. So if you could fit 1billion transistors in a specific size of chip, say 300mm^2, then you could fit 2billion transistors in the same amount of space on the next full node down.
So currently chips are around the 4-5billion transistor 450-550mm^2 and 250W range on 28nm. 20nm should let them fit say 8-10billion transistors in the same 450-550mm^2 size and 250-300W power bracket... in theory.
That is about the usual change in the past, transistor count has near enough doubled every 2-3 years and that is where the real performance improvements come from. unfortunately it looks like 20nm has some serious issues, the industry is basically coming up against theoretical limits and each new shrink is getting exponentially more difficult.
20nm is missing the real power saving but has the transistor density so it's looking more like a 550mm^2 chip on 20nm would also be pushing 350-400W which it's doubtful AMD or Nvidia will want to do. So it's likely the truly high end next generation chips may have to wait an extra 12-18 months for 16nm(it's a marketing name more than anything, it's 20nm but with new transistors which finally bring the more normal power drop).
Effectively once you get a 450-550mm^2 chip that is about the maximum size of chip you can make and be financially profitable. So once you have a chip launch at those sizes, this gen would be Titan and 290x, you can't really get more performance until you get a new process.
Effectively when a new process node works well offering 2x transistor density and 50% power reduction per transistor(this is pretty normal for every node drop in the past decade) then you can look to get 70-80% more performance from a new process.
A new architecture on the same process(ie if they made a new 28nm chip of the same size/transistor count) is unlikely to be more than 10-15% faster. So process node is WAY more important for new levels of performance than architecture, but combining both is best of both worlds.
It went 90-80-65-55-40-32-28nm Some are half nodes some full nodes.
Effectively a full node drop means roughly twice the transistor density. So if you could fit 1billion transistors in a specific size of chip, say 300mm^2, then you could fit 2billion transistors in the same amount of space on the next full node down.
So currently chips are around the 4-5billion transistor 450-550mm^2 and 250W range on 28nm. 20nm should let them fit say 8-10billion transistors in the same 450-550mm^2 size and 250-300W power bracket... in theory.
That is about the usual change in the past, transistor count has near enough doubled every 2-3 years and that is where the real performance improvements come from. unfortunately it looks like 20nm has some serious issues, the industry is basically coming up against theoretical limits and each new shrink is getting exponentially more difficult.
20nm is missing the real power saving but has the transistor density so it's looking more like a 550mm^2 chip on 20nm would also be pushing 350-400W which it's doubtful AMD or Nvidia will want to do. So it's likely the truly high end next generation chips may have to wait an extra 12-18 months for 16nm(it's a marketing name more than anything, it's 20nm but with new transistors which finally bring the more normal power drop).
Effectively once you get a 450-550mm^2 chip that is about the maximum size of chip you can make and be financially profitable. So once you have a chip launch at those sizes, this gen would be Titan and 290x, you can't really get more performance until you get a new process.
Effectively when a new process node works well offering 2x transistor density and 50% power reduction per transistor(this is pretty normal for every node drop in the past decade) then you can look to get 70-80% more performance from a new process.
A new architecture on the same process(ie if they made a new 28nm chip of the same size/transistor count) is unlikely to be more than 10-15% faster. So process node is WAY more important for new levels of performance than architecture, but combining both is best of both worlds.