4 ms·

Seems like this was a scheduler issue in Windows, again. In Linux they show a decent performance uplift. > When taking the geometric mean of those nearly 400 r

Seems like this was a scheduler issue in Windows, again. In Linux they show a decent performance uplift. > When taking the geometric mean of those nearly 400 raw benchmark results, it sums up the greatness of Zen 5 with the Ryzen 5 9600X and Ryzen 7 9700X processors. The Ryzen 7 9700X delivered 1.195x the performance of the Core i5 14600K competition or 1.15x the performance of the prior generation Ryzen 7 7700X. The Ryzen 5 9600X came in at 1.35x the performance of the Core i5 14500 and 1.25x the performance of the Ryzen 5 7600X. Or if still on Zen 3 for comparison, the Ryzen 5 9600X was 1.82x the performance of the Ryzen 5 5600X. ... > The raw performance results alone were impressive for this big Linux desktop CPU comparison but it's all the more mesmerizing when accounting for the CPU power use. On average across the nearly 400 benchmarks the Ryzen 5 9600X and Ryzen 7 9700X were consuming 73 Watts on average and a peak of 101~103 Watts. The Ryzen 5 7600X meanwhile had a 92 Watt average and a 149 Watt peak while the Ryzen 7 7700X had a 99 Watt average and 140 Watt peak. The Core i5 14600K with being a power hungry Raptor Lake had a 127 Watt average and a 236 Watt peak. The power efficiency of these Zen 5 processors are phenomenal! https://www.phoronix.com/review/ryzen-9600x-9700x/16 https://www.phoronix.com/review/ryzen-9600x-9700x/16

zamadatix 2y ago
The Windows scheduler/admin issue is the difference of a few percentage points and it also applies to Zen 4. That it is such a big thing this generation is precisely why it's a lame release. E.g. look at the Phoronix numbers you linked. More than "it's Linux" they benchmark a bit heavier in the workstation/server tasks which is particularly good for focusing on what this generation improves (things like AVX 512 performance). Despite this: 5600X -> 7600X = 45% increase 7600X -> 9600X = 25% increase The last generation was nearly twice, and not just in hyper specific workloads. If anything levels out about this generation it'll probably be when AMD gets around to releasing the 3D versions early next year. They shoot themselves in the foot from a media/review perspective on that each release because there are always CPUs from the previous generation that smash certain cache heavy workloads better than the new shiny thing at release and then the 3D version comes out with less general discussion and coverage months later.
- eightysixfour 2y ago
  The 5600x -> 7600x generation went backwards, dramatically, in power consumption though. The way that I look at, in two generations, we got an 82.9% increase in performance with essentially no power consumption change. That's incredible. There are also substantial gains in areas that are not AVX accelerated - looking at Python performance, database performance, etc.
  - zamadatix 2y ago
    The 7600X did average 19 W higher in Phoronix's benchmark but that's certainly not the explanation for the 45% performance gain and benchmarks matching the 9x00X parts to the wattage levels of the 7x00X parts don't show perf/watt improvements, though it is nice to see saner default limits rather than eking that last 5% of perf instead. 82.9% in two generations/4 years is decent but few are disappointed with the performance gains of getting off Zen 3, they're disappointed with the performance gains of Zen 5 itself.
    - eightysixfour 2y ago
      The 7600x used 28% more energy than the 5600x to get those big gains. That’s not a small amount - it kept me from upgrading my SFF. My guess is that if you power limited the 7600x, the generational gains would look more linear.

zamadatix 2y ago

The Windows scheduler/admin issue is the difference of a few percentage points and it also applies to Zen 4. That it is such a big thing this generation is precisely why it's a lame release. E.g. look at the Phoronix numbers you linked. More than "it's Linux" they benchmark a bit heavier in the workstation/server tasks which is particularly good for focusing on what this generation improves (things like AVX 512 performance). Despite this: 5600X -> 7600X = 45% increase 7600X -> 9600X = 25% increase The last generation was nearly twice, and not just in hyper specific workloads. If anything levels out about this generation it'll probably be when AMD gets around to releasing the 3D versions early next year. They shoot themselves in the foot from a media/review perspective on that each release because there are always CPUs from the previous generation that smash certain cache heavy workloads better than the new shiny thing at release and then the 3D version comes out with less general discussion and coverage months later.

eightysixfour 2y ago
The 5600x -> 7600x generation went backwards, dramatically, in power consumption though. The way that I look at, in two generations, we got an 82.9% increase in performance with essentially no power consumption change. That's incredible. There are also substantial gains in areas that are not AVX accelerated - looking at Python performance, database performance, etc.
- zamadatix 2y ago
  The 7600X did average 19 W higher in Phoronix's benchmark but that's certainly not the explanation for the 45% performance gain and benchmarks matching the 9x00X parts to the wattage levels of the 7x00X parts don't show perf/watt improvements, though it is nice to see saner default limits rather than eking that last 5% of perf instead. 82.9% in two generations/4 years is decent but few are disappointed with the performance gains of getting off Zen 3, they're disappointed with the performance gains of Zen 5 itself.
  - eightysixfour 2y ago
    The 7600x used 28% more energy than the 5600x to get those big gains. That’s not a small amount - it kept me from upgrading my SFF. My guess is that if you power limited the 7600x, the generational gains would look more linear.