I was reading the article below about the K1 64bit/Denver.
http://classic.slashdot.org/story/14/08/12/0158245
According to the article the upcoming dual-core k1 device will use something named "In-order execution".
The comments in the article clearly shows that many are not happy with nvidias approach.
My question is:
Is this something that will have a good or bad effect on Dolphin? Or maybe it dosn't have any effect at all on Dolphin. Maybe sonicadvance1 could help us get a better picture of this
In-order execution means not having
out of order execution. The original Atom, Pentium 1 (not Pro), and Cortex A8 (I think?) are examples of CPUs without OOE.
(09-23-2014, 08:58 AM)Fiora Wrote: [ -> ]In-order execution means not having out of order execution. The original Atom, Pentium 1 (not Pro), and Cortex A8 (I think?) are examples of CPUs without OOE.
Any idea on how that will impact performance in Dolphin, or no way to know at present?
I have no idea personally; I'd be kinda worried since compile-to-in-order-VLIW didn't fare so well last time (Itanium), but maybe it'll be better implemented this time around?
(09-23-2014, 10:58 AM)Fiora Wrote: [ -> ]I have no idea personally; I'd be kinda worried since compile-to-in-order-VLIW didn't fare so well last time (Itanium), but maybe it'll be better implemented this time around?
I thought as much, seeing some expert commentary from SoC development experts it seems there can be definite payoffs to using in-order Vs OOO, particularly to battery life which is a very important consideration in mobile applications, however there are some very significant hurdles also. Will be interesting to see how it pans out, I'm hoping well, as I'd really like to see a solid focus on optimising per-core performance in mobile, rather than the largely fruitless and marketing-driven emphasis on 'MWOAR CORES!' we currently have.
That all depends how they will implement it. I believe there's a reason why they've chosen in-order instead of OOOE. They call Denver their super core, so I guess it could be something big if done correctly. Anyway this was surprise and it'll be a big thing, either big win or big fail though.
Hopefully its a big win, im just afraid that it will have a negative impact on dolphin, not because i know anything about this in-order thing, but more because amny of the people commenting the article dosnt like this approach by nvidia.
Well, if we cant get an answer on what impact this will have on dolphin we will just have to wait and see, ill get a k1 denver device here in sweden as soon as they are released here. Ill make sure to have some videos and results uploaded
(09-23-2014, 11:27 PM)zxcvbad Wrote: [ -> ]That all depends how they will implement it. I believe there's a reason why they've chosen in-order instead of OOOE. They call Denver their super core, so I guess it could be something big if done correctly. Anyway this was surprise and it'll be a big thing, either big win or big fail though.
The reason is that OOE design eats up silicon.
It means true Cortex-A57 chip will be faster.
I think the silicon saving makes sense. Everything about the denver variant suggests nvidia wants to max out the amount of transistors they can pump into the most efficient setup. I suppose that means dual core, in order (apple style).
The A7 Cyclone (from what I heard) supposedly has a ~192 uop reorder buffer, which is about as far as you can get from in-order.