It just seems hard now. I also remember where getting 100 MB HDD for a personal computer was considered a hard task. It is just another wall to get past is all. In a few years of healthy competition it will seem like it wasn't hard to get around.
The process of using a 50 year technology will pass.. Transistors will be replaced.
(11-21-2017, 01:29 PM)JobeStroud Wrote: [ -> ]It just seems hard now. I also remember where getting 100 MB HDD for a personal computer was considered a hard task. It is just another wall to get past is all. In a few years of healthy competition it will seem like it wasn't hard to get around.
The process of using a 50 year technology will pass.. Transistors will be replaced.
Maybe. The main problems now are physics (limits clockspeeds) and the inherent lack of parallelism in most programs (how many instructions can you really execute in parallel when so many depend on others) though. I don't think competition is gonna make solving those any easier.
I mean, AMD spent 5 years (maybe even more) on a complete redesign, and it didn't exactly do much for single core performance overall. Ryzen is still behind Intel.. In fact, it's behind what Intel had in 2013 in some areas.
OK, i finally had the time to test my new i5 2500K @ 4.5 GHz with Passmark. Result: STP = 2430 . I guess that's good? At least i'm able to run most games (at least those games i own and i've tested so far) with 4x internal resolution with a stable framerate of 60 fps.
MUCH better than my Phenom II X6!
Finally the games are looking and running great.
(11-21-2017, 08:59 PM)dogen Wrote: [ -> ]Maybe. The main problems now are physics (limits clockspeeds) and the inherent lack of parallelism in most programs (how many instructions can you really execute in parallel when so many depend on others) though. I don't think competition is gonna make solving those any easier.
I mean, AMD spent 5 years (maybe even more) on a complete redesign, and it didn't exactly do much for single core performance overall. Ryzen is still behind Intel.. In fact, it's behind what Intel had in 2013 in some areas.
It is also a math issue. A perfect pipeline could generate 1 instruction by cycle. So you have an efficiency between 0% and 100%. If you're close of 0, it is easy to have a 2x factor (for example 10% -> 20%, give you a 100% increase). If you have already a good efficiency, let's say 80%, it would be very hard to improve it. Even you hit the perfect level you will only gain 25% (80% -> 100%).
So even Zen 10 and Intel i17 will have a similar IPC level.
(12-04-2017, 01:08 PM)gregory Wrote: [ -> ]It is also a math issue. A perfect pipeline could generate 1 instruction by cycle. So you have an efficiency between 0% and 100%. If you're close of 0, it is easy to have a 2x factor (for example 10% -> 20%, give you a 100% increase). If you have already a good efficiency, let's say 80%, it would be very hard to improve it. Even you hit the perfect level you will only gain 25% (80% -> 100%).
So even Zen 10 and Intel i17 will have a similar IPC level.
Yeah, this is what I'm expecting with the next Zen version.