Comment by yardie | Hacker News Reader

Comment by yardie | Hacker News Reader

Parent Full thread

PPC era saw macs that were significantly slower than their intel/amd counterparts.

This is subjectively correct. The G4 and G5 had Altivec coprocessing units which were the precursors to SIMD and GPGPU. A few apps could take advantage of it to do video processing and vector calculations but most cross platform apps(like Office or Photoshop) just didn't use it at all, putting the G5 at a severe disadvantage. Also, Intel was, sort of, cheating with their NetBurst architecture. Deeper and deeper pipelines allowed them to get higher clock speeds without adding much actual performance. Ops just sat in the queue longer, but their predictive caching algorithm was very advanced (I'll give them that). Once they got as far as they could go, clock wise, with the P4 they simply started packing more processors on the die (core duo). In fact, their saving grace came from their competitor AMD with the x86_64 extension set.

PPC had the ability to go much faster (as seen in the 3GHz X360, PS3) they just couldn't get the die smaller nor the power down. IBM had the performance advantage with their Power7 CPUs but weren't really interested in the consumer CPU market and left Apple holding the bag. Apple wanted smaller, low power CPUs for their laptops and found it in Intel. Windows compatibility (at least through VM) was icing on the cake.

Some corrections.

NetBurst was not 'cheating' by using a long pipeline to get higher clock speeds. NetBurst was purposely designed to get higher clock speeds therefore better performance. Intel thought it would get to 4 ghz, 5 ghz, 6 ghz, and beyond on NetBurst which would theoretically be better than what we have today if it had worked. Obviously it didn't.

The Core Duo was not based on NetBurst (i.e. Pentium 4). It was based on the Pentium M which was an improved Pentium 3. The Core Duo and everything that came after it are based on the old P6 architecture that started with the Pentium Pro.

Intel's saving grace was going back to the P6 and abandoning NetBurst. Intel had adopted x86_64 on NetBurst and it was worthless.

PowerPC can get smaller dies and lower power consumption just fine. In fact most PowerPC chips today are smaller and lower power than the G4 ever was. Motorola makes them all day long for automaters, defense contractors, and network device makers. The problem was Motorola's other customers didn't care about performance. Nobody wanted to run Photoshop on guided missiles.

The PPC 970 (i.e. G5) was powerful, but IBM made CPUs for big iron. When a CPU ran hotter than expected it was the mechanical engineer's problem to fix in the enclosure. IBM just didn't need to figure out how to make chips with tight operational constraints. IBM didn't have the expertise to get the PPC 970 to work in a Power Book and IBM didn't have any incentive moving forward to build that expertise.

It just so happened that Intel was in the middle of its NetBurst epiphany. NetBurst was bad. Centrino was great. The Core Duo was the future. Now Intel was talking Apple's language.

PS. IBM's POWER7 just came out in 2010. The PPC 970 that Apple called the G5 was a low power version of the POWER4.

Just like G4 and G5 had Altivec, Intel had MMX and SSE. Around the same time too, so it was not something magic for PPC.

As a owner of G3 and G4 machines, I can tell you that they were underperfoming. 7450 was slower than equivalent Pentium M (that's what was in PowerBooks and windows laptops at the time). Going back further, 466 MHz G3 couldn't keep decoding divx movie at SD resolution. 400 MHz P2 or 433 MHz celeron had no problems.

10 years ago I compared the IO performance of a 450Mhz G4 running Yellowdog Linux to a PIII 700 Mhz server running RedHat, and the G4 literally blew the PC out of water IO-wise (disk, RAM and network performance), particularly running MySQL benchmarks, though slower in raw computing speed. So it depends upon your applications I guess, and your OS.

Reply on news.ycombinator.com