From: anton@mips.complang.tuwien.ac.at   
      
   Michael S  writes:   
   >On Tue, 30 Dec 2025 11:13:37 GMT   
   >anton@mips.complang.tuwien.ac.at (Anton Ertl) wrote:   
   >   
   >> Michael S  writes:   
   >> >Do you happen to have benchmarks that compare performance of Alpha   
   >> >EV5 vs in-order Cortex-A ?   
   >>   
   >> LaTeX benchmark results (lower is waster)   
   >>   
   >> Alpha:   
   >> - 21164 600 MHz CPU, 2M L3-Cache, Redhat-Linux (a5)               8.1   
   >> ARM A32/T32:   
   >> - Raspberry Pi 3, Cortex A53 1.2GHz Raspbian 8                    5.46   
   >> ARM A64:   
   >> - Rockpro64 (1416MHz Cortex A53) Debian 9 (Stretch)               3.24   
   >> - Odroid N2 (1896MHz Cortex A53) Ubuntu 18.04   
   >> 2.488   
   >> - Odroid C2 (1536MHz Cortex A53) Ubuntu 16.04                     2.32   
   >> - Rock 5B (1805MHz A55) Debian 11 (texlive-latex-recommended)   
   >> 2.105   
   >>   
   >> A problem with the LaTeX benchmark is that it performance is   
   >> significantly influenced by the LaTeX installation (newer versions   
   >> need more instructions, and having more packages needs more   
   >> instructions).  But it's the only benchmark results I have.   
   >>   
   >> - anton   
   >   
   >Thank you.   
   >Two 64-bit A53 results about the same as EV5 clock for clock and one   
   >result is significantly better.   
   >So, either wide in-order is indeed not bright idea or 21164 suffers   
   >because of inferioriority of Alpha ISA relatively to ARM64.   
      
   Hard to tell from these results.  In addition to the problems   
   mentioned above there are also differences in cache configuration to   
   consider.  And the A55 does quite a bit better in IPC than the A53,   
   although it superficially has the same resources.   
      
   The benchmark may be worse for the Alpha than many other benchmarks   
   because LaTeX probably uses many byte accesses, but the used binary   
   does not use BWX instructions (the 21164A has them, but there has been   
   no Redhat for Alpha with BWX).  This can be seen as an inferiority of   
   the Alpha ISA relative to ARM A64.   
      
   I think all these problems make these results too fuzzy to draw any   
   conclusions about the question of interest.  You would have to run a   
   larger set of more well-defined benchmarks to answer this question.   
      
   But I think that one can say that the performance-per-clock is roughly   
   similar between the 21164 and the Cortex-A53.   
      
   >BTW, Odroid C2 score appears suspiciously good. Could it be a turbo   
   >clock frequency was much higher than reported?   
      
   No.  The cycles and GHz number comes out of perf.  And the Odroid C2   
   does not reach higher clocks without overclocking, and does not reach   
   the promised 2GHz even with overclocking, as the manufacturer had to   
   admit.   
      
   - anton   
   --   
   'Anyone trying for "industrial quality" ISA should avoid undefined behavior.'   
     Mitch Alsup,    
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)