- Jun 5, 2008
- 1,588
- 676
- 146
Rather than focus on the negative reception of the Bulldozer launch, I'd like to take a step back and acknowledge how amazing an architecture Sandy Bridge is. That being said, I wonder what makes it so amazing?
Thinking back to the evolution of the Core CPU line, I see it this way:
Conroe/Penryn:
Pro's: Massive IPC increase and still highly clockable compared to competition
Cons: You have to choose whether you want dual core or quad core performance (both from price, power consumption, and absolute performance in games)
Cons: Need massive amount of cache ($$) as a band-aid to off die memory controller
Nehalem/Lynnfield:
Pro's: Another big IPC jump, similary clocks to Core 2 but more overclocking headroom, moving memory controller on board was a good idea
Pro's: Finally an affordable quad core, and turbo mode that gives it single threaded performance of a dual
Sandybridge: Yet another big IPC jump, higher clocks than anything before, and much higher turbo boost speeds. While Lynnfield took advantage of turbo, Sandybridge really relies on it, esp. in mobile form. Finally practical to have a quad core laptop other than a desktop replacement.
Moving BCLK generator to CPU caused drama, but was likely because of the massive overclocking headroom Intel knew about. Unlocked CPUs make it so easy to overclock everyone on tech forums think they are 1337
-------------------------------
Feel free to let me know any other reasons you think I am missing!
I think if BD had launched back in line with Nehalem, it would've been a much more warm reception. Even if SB had brought it's IPC increase only and not the massive overclocking headroom. But SB is honestly the most fiercely competitive processor I have seen in a very long time.
I wonder things like: how did Intel get clocks to scale so high? Was that their intent or a side effect?
Why did Intel decide to make fatter cores rather than more cores? Was that solely because of hyper threading? Or better margins?
Exactly what changes to the microarchitecture are responsible for the IPC increase AND higher clockrate of SB? I feel like an architecture that brings both of these, on the same process node are very rare. I mean, 32nm Xeon quads and Lynnfield duals did not clock as well as their SB counterparts, and have less IPC. That's pretty amazing.
Thinking back to the evolution of the Core CPU line, I see it this way:
Conroe/Penryn:
Pro's: Massive IPC increase and still highly clockable compared to competition
Cons: You have to choose whether you want dual core or quad core performance (both from price, power consumption, and absolute performance in games)
Cons: Need massive amount of cache ($$) as a band-aid to off die memory controller
Nehalem/Lynnfield:
Pro's: Another big IPC jump, similary clocks to Core 2 but more overclocking headroom, moving memory controller on board was a good idea
Pro's: Finally an affordable quad core, and turbo mode that gives it single threaded performance of a dual
Sandybridge: Yet another big IPC jump, higher clocks than anything before, and much higher turbo boost speeds. While Lynnfield took advantage of turbo, Sandybridge really relies on it, esp. in mobile form. Finally practical to have a quad core laptop other than a desktop replacement.
Moving BCLK generator to CPU caused drama, but was likely because of the massive overclocking headroom Intel knew about. Unlocked CPUs make it so easy to overclock everyone on tech forums think they are 1337
-------------------------------
Feel free to let me know any other reasons you think I am missing!
I think if BD had launched back in line with Nehalem, it would've been a much more warm reception. Even if SB had brought it's IPC increase only and not the massive overclocking headroom. But SB is honestly the most fiercely competitive processor I have seen in a very long time.
I wonder things like: how did Intel get clocks to scale so high? Was that their intent or a side effect?
Why did Intel decide to make fatter cores rather than more cores? Was that solely because of hyper threading? Or better margins?
Exactly what changes to the microarchitecture are responsible for the IPC increase AND higher clockrate of SB? I feel like an architecture that brings both of these, on the same process node are very rare. I mean, 32nm Xeon quads and Lynnfield duals did not clock as well as their SB counterparts, and have less IPC. That's pretty amazing.