Intrinsic said:
A few things.
- this 7nm process we are all talking about isn't even gauranteed to be available by 2018/2019. Thats just powerpoint talk right now and barring any surprises they don't always come out exactly when expected. Theer is always some sort of hiccup or delay. Like take the 14nm process for instance, believe it or not that has been available since around 2013 when the PS4/XB1 launched. Albeit only from intel and in their highest of the high end processors. It took it another 3 years to become mainstream (i.e: AMD support). Right now, no one has even gone out of risk production phase for the 7nm chiips. Thats assuming any of them has even started.
|
Keep in mind that Intels 14nm is NOT the same as TSMC's/Global Foundries/Samsung/IBM and so on's 14nm.
They are using the "nm" as a marketing gimmick and isn't really representative of geometry sizes.
DonFerrari said:
shikamaru317 said:
Well, I was mainly basing 7-8tflops on the fact that the 6 tflop XB1 X isn't hitting Native 4K on every game, though that may be more down to the Jaguar CPU than the GPU. I also seem to recall Mark Cerny saying that 7-8 tflops is the requirement Sony would need to hit native 4k consistently.
You're right about flops not being a good measure, but it's the most convenient way to talk about a GPU's power.
|
Also, if we have ceteris paribus for all else on the architeture, looking at how much more flops would be necessary on a "scalar approximation" isn't really that off.
|
I actually had to google "ceteris paribus" as it's a term I wasn't familiar with. In English it's a statement often associated as "everything else being equal".
However even then flops isn't really an accurate representation of anything. - You could have double the flops but only see 20% of the gain in gaming performance as your are bottlenecked elsewhere in the GPU pipeline.
