End of GPU wars?

DO be really frank, I do not really care how AMD gets there.
If I can have an elegant solution to play 1440P, CPU+GPU or APU is fine by me...

I'd settle for an APU that does 1080P on my HD TV, which is what I have until they figure out who wins the HDR format war. Next TV will be 4K HDR.
 
and now imagine it costing 999 to 1499 per card plus a cpu :hmm: :(


lots more people will go with APU's


True, APU's made at 7nm will be a lot more powerful in their own right......Even if the die space is shared with the CPU, a 40 billion transistor budget allows to pack the core of the fastest gaming card on the market right now with it's 12 billion transistors, and still has tons left over for the CPU portion and both run at much higher clock speeds...


Add that both the GPU and CPU are in the same die, the communication speeds and extremely low latency put PCI-e express to shame, and then wrap that up into some nice HBM also made at 7nm, so memory capacity shouldn't be a problem either, as the current version of HBM 2.0 already allows up to 32GB when using 4 stacks a using a fab process no where near that small.


One gets to the point of nearly stating......"Dedicated video cards?....Why?"
 
GPU Wars are far from over :)


Well, there will eventually be 5K screens on the market and later on it's 8K's turn ( basically 4x more demanding than 4K for a given Fps ), along with even more demanding games to run to keep future hardware busy, but beyond that the GPU war at least at the high end of the scale is truly done, simply because 8k resolutions equals the same at that as an IMAX screen, and that same resolution equals the eyesight capabilities of someone that has perfect 20 / 20 vision according to optometrists.


There's physically no point in going beyond 8K unless you have the eye sight capabilities of these guys:


owls.jpg



Or:


golden-eagles-picture.jpg
 
Well how small can you create on silicone? After 7nm? What will be the next big leap in graphics software? What comes after silicone?
 
If I recall correctly, they think they can get down to about 4 nm, but after that point it's questionable whether the laws of physics will allow for further reduction in size. Realistically that's probably at least a decade out though, and possibly longer considering the increased time/difficulty involved in rolling out new process nodes.
 
If I recall correctly, they think they can get down to about 4 nm, but after that point it's questionable whether the laws of physics will allow for further reduction in size. Realistically that's probably at least a decade out though, and possibly longer considering the increased time/difficulty involved in rolling out new process nodes.

AMD Zen Plus CPUs, Vega 20 & Navi GPUs To Be Built On 7nm FinFET – Globalfoundries Announces Risk Production For 2018

So you think its going to take 8 more years after 2018 to get down to 4nm?
 
AMD Zen Plus CPUs, Vega 20 & Navi GPUs To Be Built On 7nm FinFET – Globalfoundries Announces Risk Production For 2018

So you think its going to take 8 more years after 2018 to get down to 4nm?

I doubt very much we will see 7nm in 2018 even 2020 - delays, problems etc. Maybe for ram, ssds etc. but for complex GPU's/CPU's I don't think so.
 
AMD Zen Plus CPUs, Vega 20 & Navi GPUs To Be Built On 7nm FinFET – Globalfoundries Announces Risk Production For 2018

So you think its going to take 8 more years after 2018 to get down to 4nm?

It wouldn't really surprise me. Intel's 10nm was pushed back to mid 2017. Intel's 7 nm would have to be out around 2020, then 4 nm in 2023. That's with no delays and no intermediate node between 7 and 4 nm (5.5 nm?). I think GF and TSMC having 7 nm in 2018 is very optimistic. Most likely those get pushed back. Also the naming conventions are not the same across different companies. So who knows if GF's 7 nm is really equivalent to Intel's?

The wild card is the last process shrinks have proven to be harder and harder to pull off each time. If 4 nm is really pushing the physical limits then there could be a lot of challenges to get there. So, who knows, it could come earlier, but based off recent history I don't think a decade is that far out of line.
 
Last edited:
It wouldn't really surprise me. Intel's 10nm was pushed back to mid 2017. Intel's 7 nm would have to be out around 2020, then 4 nm in 2023. That's with no delays and no intermediate node between 7 and 4 nm (5.5 nm?). I think GF and TSMC having 7 nm in 2018 is very optimistic. Most like those get pushed back. Also the naming conventions are not the same across different companies. So who knows if GF's 7 nm is really equivalent to Intel's?

The wild card is the last process shrinks have proven to be harder and harder to pull off each time. If 4 nm is really pushing the physical limits then there could be a lot of challenges to get there. So, who knows, it could come earlier, but based off recent history I don't think a decade is that far out of line.


The main problem is the gate portion within the transistor itself as the process gets smaller and smaller, and at 4nm the thickness of the gate isn't much more than a few atoms at most, so regardless if the logic circuit requites said gate to be closed and not allow current to pass thru from one side to the other of the transistor, it still does anyhow since it's so thin.....It becomes useless as quantum effects kick in like quantum tunneling in this case.


Other materials or completely new technologies are needed to take it to the next step beyond that and allow those smaller structures to be viable, and as it's been mentioned above, even Intel and it spending billions in R&D every year is finding it harder and harder to pull it off with each step getting longer.....It won't surprise me one bit if the 7nm step takes 5~6 years after the 10nm one is released, and even more than that for the 4nm step to happen.
 
Yeah others 7 is more like Intel a 10nm expect awhile to get real 7nm production

Sent from my VS990 using Tapatalk
 
Intel has probably the best scientists/engineers for process node tech and the bucks to pour into it for equipment, testing equipment etc. They are having a rather tough time with 10nm as it is.
 
Back
Top