Originally Posted by budgetgamer120
Do better than what I did. So far you are making stuff up.
Making stuff up?
It is actually a fact that the Xenon processor uses three, dual-issue, in order, cores vs. two, much wider, out-of-order, three issue cores in an Athlon 64 X2. It's a fact that it has 165 million transistors vs. 233 million in an Athlon 64 X2. It's a fact that they were made on similar processes by similarly (at the time) capable companies. It's also a fact that the console CPUs were low power parts.
If you think a low power part with 65% of the transistor count is going to outperform a larger part of a contemporaneous architecture on a similar process with four times the power consumption, as a general purpose CPU, you're insane.
Saying the XBox 360 Xenon is faster than an Athlon 64 X2 is like saying the Pentium N3530 is faster than an i3- because it's got more cores.
The Xenon was 2005's Jaguar equivalent. Multiple, slow, low-power cores to do what needs to be done in a console while fitting in power, cooling, and budgetary needs...not some fantastical, physics defying super-chip. Impressive, in many ways, to be sure, but it was never going to be a faster general purpose CPU than the leading desktop x86 parts of the time, which lacked such constrains.
Some architectural details:
Some benchmarks of the PPEs in the Xenon and the Cell BE:
In the end you have a part whose cores are weaker than a Pentium 4 core and that will never come anywhere near it's theoretical performance figures except in perfectly optimized vector code...which does not exist.
Using those GFLOPs figures is even worse than trying to use LINPACK performance to compare processor performance on PC. My 5820K is four times as fast in LINPACK as my X5670 at the exact same clock speeds, and twice as fast as my 4930K. There is essentially no real-world application that will show anywhere near that differential. Theoretical GFLOPS figures are even more useless, as they aren't derived from any actual benchmark, synthetic or not...they take the clock speed and retire rate of the biggest VMX instructions the part can do and just multiply them together...not even the most synthetic benchmarks will reach those figures.
Originally Posted by AmericanLoco
Really curious how you were playing Quake a year before the N64 came out, when they both launched within one day of each other in 1996
Demo came out in February 1996. So I was off a bit, but this was still well before the N64 launch. I am recalling stuff I did more than twenty years ago here, so you'll have to forgive me being ~6 months off on some of it.
Originally Posted by AmericanLoco
Also super curious how you were playing Quake 1 @ 1024x768 in 1995 on a Pentium 75 with a Tseng Labs card. Considering Tseng Labs only made 2D accelerators, and Quake 1 will barely manage 20 FPS in software mode at 320x200 on a Pentium 75, nevermind 1024x768.
A good 2D accelerator made a big
difference in Quake (which had no hardware acceleration on launch) and many other games.
Those benchmark results seem unusually low to me, though I never ran the timedemos (im not sure they were in the demo), I used 'timerefresh' at the difficulty/chapter select stage, which was pretty demanding. At stock a Pentium 75 at 320x240 would pull in more than 30 fps with even a mediocre 2D accelerator.
At 100MHz + with a good 2D accelerator, I recall the game was smooth at 640x480 and mostly playable at 1024x768.
Edit: Take a look at the second part of the benchmarks you posted. Those are mostly 2D accelerators and in the case of the 3D ones, the 3D component wasn utilized...the CPU used is shown in parenthesis.
Those figures look plausible and I may have been using a faster CPU than my OCed Pentium 75 by late 1996. My next part was a Pentium 200MMX that I ran at 266MHz, but I'm not sure if that was 96 or 97.
Edit2: Fount my receipt folder...it wasn't until early 1997 that I purchased the Pentium 200MMX, so I still had the 100MHz P5 part when N64 dropped.
Edit3: Compared 'timerrefresh' at the start of the game to the timedemos and the time demos are indeed slower, but probably a better representation of some of the most demanding areas. I was probably using 512*384 or 640*480 to get Quake playable on my late 1995 setup. Regardless, this was not the fastest PC available at the N64 launch, and still ran Quake better than the N64 version that didn't show up until 1998, which had to be downgraded relative to the original PC version of two years prior because of lack of RAM and storage.Edited by Blameless - 4/11/17 at 12:13am