3 ms·

Does anyone know how good the PowerPC chips from that era actually were compared to the competing chips from Intel? I remember reading a lot of marketing claim

Does anyone know how good the PowerPC chips from that era actually were compared to the competing chips from Intel? I remember reading a lot of marketing claims, but back in those days there wasn’t so many review sites that would test them against each other with a multitude of precise benchmarking suites. Nor were there so many pieces of software which ran on both platforms.

aidenn0 3y ago
TL;DR: Clock-for-clock the 7400 was arguably faster than the Katmai, the Coppermine was arguably faster than the 7400, and the 7450 was arguably faster than Coppermine. TFA is specifically discussing FLOPS. The PPC 7400 (first G4) had an AltiVec unit that was considerably more powerful than the Katmai (first P3) SSE engine. The 7400 altivec was 128-bits wide and had separate permute and arithmetic engines; this means it could retire up to 2 4x32 SIMD floating point operations per cycle (if you scheduled permutes . The SSE engine on the Katmai was only 64-bits wide, so it took 2 uops to do a 4x32 add and 2uops to do a 4x32 multiply. However, it could do an add and a multiply simultaneously, so if you were alternating adds and multiplies, you could retire one 4x32SIMD instruction per cycle. Coppermine (2nd generation P3, released fall of 1999) used a 128-bit unit, but I'm not sure if it could still retire adds and multiplies in parallel (I think not). For general (i.e. not just FPU) performance: Coppermine and 7450 introduced on-die L2 cache to the P3 and G4 series respectively. This probably had a bigger performance impact on a broader base of code than other changes made in and around the time, as the FSBs were not keeping up with the CPU clocks.
- akamaka 3y ago
  Thanks for the detailed answer!
  - aidenn0 3y ago
    Something I forgot to mention is that Intel obtained and maintained a process advantage over Motorola around this time; The early G4 PowerPCs were on Motorola's 200nm process, but had to be downclocked to get the yields up. Intel used their existing 240nm process for the Katmai P3, which allowed them to release a 600MHz stepping of the Katmai the same month that Motorola failed to launch a 500MHz stepping of the 7400. Of course Intel famously then pivoted to the NetBurst architecture, for the P4, which turned out to be an evolutionary dead-end thanks to increased leakage at 90nm. The idea was to tradeoff reduced IPC in exchange for higher clocks, but they were never able to get the ALUs running over 7.6GHz, and they had to reduce IPC so much to get there that any advantages of the high clock speed were lost. Nevertheless the fastest P4s on the 130nm process were significantly faster than the fastest P3s on the 130nm process.
watersb 3y ago
A great series of deep dives from the 2000s by Jon Stokes: https://arstechnica.com/features/2004/08/ppc-1/ https://arstechnica.com/features/2004/08/ppc-1/ Comparison between the Pentium 4 and the Power PC G4e: https://archive.arstechnica.com/cpu/01q4/p4andg4e2/m-p4andg4e2-1.html https://archive.arstechnica.com/cpu/01q4/p4andg4e2/m-p4andg4... Stokes had so much material on these deep dives, he collected it into a book, "Inside the Machine": https://nostarch.com/insidemachine.htm https://nostarch.com/insidemachine.htm Chapter 7 is the PowerPC vs Pentium 4 analysis. Curiously, I was going over it again just last night.

aidenn0 3y ago

TL;DR: Clock-for-clock the 7400 was arguably faster than the Katmai, the Coppermine was arguably faster than the 7400, and the 7450 was arguably faster than Coppermine. TFA is specifically discussing FLOPS. The PPC 7400 (first G4) had an AltiVec unit that was considerably more powerful than the Katmai (first P3) SSE engine. The 7400 altivec was 128-bits wide and had separate permute and arithmetic engines; this means it could retire up to 2 4x32 SIMD floating point operations per cycle (if you scheduled permutes . The SSE engine on the Katmai was only 64-bits wide, so it took 2 uops to do a 4x32 add and 2uops to do a 4x32 multiply. However, it could do an add and a multiply simultaneously, so if you were alternating adds and multiplies, you could retire one 4x32SIMD instruction per cycle. Coppermine (2nd generation P3, released fall of 1999) used a 128-bit unit, but I'm not sure if it could still retire adds and multiplies in parallel (I think not). For general (i.e. not just FPU) performance: Coppermine and 7450 introduced on-die L2 cache to the P3 and G4 series respectively. This probably had a bigger performance impact on a broader base of code than other changes made in and around the time, as the FSBs were not keeping up with the CPU clocks.

akamaka 3y ago
Thanks for the detailed answer!
- aidenn0 3y ago
  Something I forgot to mention is that Intel obtained and maintained a process advantage over Motorola around this time; The early G4 PowerPCs were on Motorola's 200nm process, but had to be downclocked to get the yields up. Intel used their existing 240nm process for the Katmai P3, which allowed them to release a 600MHz stepping of the Katmai the same month that Motorola failed to launch a 500MHz stepping of the 7400. Of course Intel famously then pivoted to the NetBurst architecture, for the P4, which turned out to be an evolutionary dead-end thanks to increased leakage at 90nm. The idea was to tradeoff reduced IPC in exchange for higher clocks, but they were never able to get the ALUs running over 7.6GHz, and they had to reduce IPC so much to get there that any advantages of the high clock speed were lost. Nevertheless the fastest P4s on the 130nm process were significantly faster than the fastest P3s on the 130nm process.

watersb 3y ago

A great series of deep dives from the 2000s by Jon Stokes: https://arstechnica.com/features/2004/08/ppc-1/ https://arstechnica.com/features/2004/08/ppc-1/ Comparison between the Pentium 4 and the Power PC G4e: https://archive.arstechnica.com/cpu/01q4/p4andg4e2/m-p4andg4e2-1.html https://archive.arstechnica.com/cpu/01q4/p4andg4e2/m-p4andg4... Stokes had so much material on these deep dives, he collected it into a book, "Inside the Machine": https://nostarch.com/insidemachine.htm https://nostarch.com/insidemachine.htm Chapter 7 is the PowerPC vs Pentium 4 analysis. Curiously, I was going over it again just last night.