Originally Posted by sdlvx
Charlie is saying that Intel is dropping Tick Tock for something different. Honestly, it's no surprise that they're doing this. Ever since the Tick Tock model was introduced, CPU performance increase has been abysmal.
Here's out everything looks since the end of 2008 when Nehalem was introduced. :
Nehalem -> Westmere (no IPC increase) -> Westmere -> Sandy Bridge (about 10% to 15% IPC increase) -> Sandy Bridge -> Ivy Bridge (0% to 5% IPC increase, performance gains only come from better working turbo which means nothing to overclockers. Also, Ivy Bridge took an overclocking penalty) -> Ivy Bridge -> Haswell (10% IPC increase).
So basically, since the end of 2008, we've seen generations of CPUs that offer pretty much 10% IPC increases every two years.
Intel definitely has realized this and it's realized that not only AMD, but ARM is growing in performance much more rapidly than Intel is.
And, everyone seems to forget this, but Intel is just refining the mobile Pentium design. It is a refine architecture that is around 20 years old. AMD has a brand new architecture.
My point is that Intel doesn't have a lot of room to squeeze out performance. You can notice this because Intel is working on power consumption, not performance now. Everyone assumes that Intel is just focusing on power consumption because that's where the market is growing, but they haven't stopped to realize that it may be that Intel simply can't milk much more out of their old mobile Pentium design and they're relying on processing nodes and power consumption improvements. Obviously, it's probably a combination of those two, but that never seems to get mentioned.
Contrast that with AMD, who released a brand new architecture that has a ridiculous amount of room for improvement, and it's easy to see that unless Intel changes things, AMD actually has a good chance of catching up.
I mean, look at Haswell, it's basically 10% IPC increase with no clock increase (from the leaks I've seen), and then Broadwell will be the same chip on 14nm, so probably same performance per clock with better working turbo. So, with Tick Tock, Intel will rely on no performance increase for the next ~2 years.
Meanwhile, AMD is going to release PD 2.0 which allows for 10% higher clocks (and possible IPC improvements), and then will release or start shipping Steamroller at the end of this year, offering 30% IPC increase.
Simply put, this is basic calculus. AMD's rate of positive change is far greater than Intel's. So, eventually, AMD would catch up to Intel (and exceed) Intel if these current trends continued. Which is probably why Intel is going to drop Tick Tock.
A 10% increase for the next 2 years can not compete with 10% and then another 30% in a single year. if AMD does manage to increase performance 40% per year continually, it will beat Intel unless Intel changes their rate of improvement or AMD slows down.
Intel has not been releasing good products compared to their past. They simply look good compared to what AMD is offering. Had Intel or AMD had performance increases on Intel's current level of 10% every two years in the 90s or early 00s, things would be absymal.
To put 10% every two years into perspective, lets look at Pentium 4 at 1.5ghz. Willamette was released in 2000. So
2002 we would have the equivalent of 1.65ghz single core Pentium 4
2004 we would have the equivalent of 1.815ghz single core Pentium 4
2006 we would have the equivalent of 1.99ghz single core Pentium 4
2008 we would have the equivalent of 2.19ghz single core Pentium 4
2010 we would have the equivalent of 2.4ghz Single core Pentium 4
2012 we would have the equivalent of 2.66ghz single core Pentium 4
2014 we would have the equivalent of 2.92ghz single core Pentium 4
Obviously, in the 00s, we had much better rate of performance increase than what I've listed. From 2000 to 2008 we basically went from 1.5ghz single core Pentium 4 netburst to Core i7 quad core.
AMD has not done much better, but they are increasing faster.