The Ryzen PCU's do much worse than Intel recent gen ix units in the Dolphin 5.0 CPU benchmarks. I keep hearing that it is not a fair benchmark and that Ryzen does much better on more recent versions of Dolphin than the one the benchmark build is based on.
Is this accurate?
Is there a better comparison out there?
Does anybody know how well Ryzen processors do in other emulators, in general?
Ryzen performs about as well as any given haswell processor at the same clock speed for emulators.
If you're curious about an actual score, just download the latest development build, get a haswell or newer Intel system and a Ryzen system, benchmark the two using the luabench benchmark. Note clock speed differences will greatly influence the score. Do not submit it to the spreadsheet for 5.0 stable benchmark results.
And yes, that is accurate. We don't know why changing compiler versions improved perf on Ryzen when the most time spent on CPU is in the JIT, where we emit our own instructions anyways, but, hey.
Thanks for the tip. This gives me a good sense of performance. Sounds like we can expect SkyLake and newer to outperform Ryzen on a per-clock, per-core basis since those newer Intel processors outperform Haswell. It comes down to intended use, price, etc.
I like to think about it this way -
If you're lucky enough to get a Ryzen chip that can run at 4 ghz, you can run just about everything in Dolphin, except maybe Rogue Squadron after Dynamic BATs landed. Because the G3258 pretty much can do that with no voltage bump.
At that point, it doesn't matter which Ryzen you get, 3/5/7, they all have the same core. They're just binned differently and that will change how lucky you are with how high you can get the clock speed. But I think even Ryzen 5 clocks at 3.7 ghz or something? More than enough for the majority of games people care about in Dolphin.
Yeah, I've been setting mentally on i3-8100 for my next upgrade when memory prices come back down (someday). I want at least 4 physical cores for gaming and possibly for Cemu and RPCS3 which apparently can benefit from them. I also want the best possible per cycle performance and clock speed combination at a low price. I could be tempted to chase up an i5-7600K to get better clocks since they can be had for about $190 on Microcenter and sometimes even less during sales.
Unless a benchmark is purposely modifying things to give a product (or company) an edge [which would take effort and be pointless for a Dolphin benchmark to do (I doubt any company is sponsoring Dolphin)], then no benchmark is really unfair. All it does is measure performance under certain conditions. If conditions change and performance levels change as a part of that then you could say a benchmark is out of date. The 5.0 stable is out of date compared to recent developer versions (every stable anything will be out of date as soon as the next dev version after is released). But if you run 5.0 stable for whatever reason (some people only want to run stable versions), then the data is accurate.
As far as other emulators goes
PCSX2 it performs about on par with Haswell, but no improvements have been made that specifically effect it either way. PCSX2 is really just single threaded performance dependent no matter the processor.
PPSSPP any x86 processor that is not either really old or a low power (think atom line) will do fine on most everything
Other newer system emulators start depending on a mix of CPU and GPU performance also they change so much week to week that there is really not a good answer.
Actually Dolphin is one of Ryzen's worst case scenarios, possibly because of how latency sensitive emulation is can be (and we know that latency is a bit of a weak point on the 1000-series Ryzen currently). Alternatively, this may be because Haswell and newer on the Intel side of things are weirdly fast at emulation (20-30% faster than Ivy Bridge) compared to most other workloads (5-10% faster than Ivy Bridge).
Take a look at the
Dolphin 5.0 benchmark - Ryzen actually ends up consistantly performing worse than Haswell on a per-GHz basis, and
this persists even when using much newer builds of Dolphin (though to a lesser degree).
However, the
Ryzen 2200G and 2400G both launch in literally 5 days and should have some improvements to the latency of the memory subsystem, so it's very possible that they may end up with performance in Dolphin that exceeds the current 1000-series Ryzen CPUs.
(02-08-2018, 06:45 AM)Nintendo Maniac 64 Wrote: [ -> ]Actually Dolphin is one of Ryzen's worst case scenarios, possibly because of how latency sensitive emulation is can be (and we know that latency is a bit of a weak point on the 1000-series Ryzen currently). Alternatively, this may be because Haswell and newer on the Intel side of things are weirdly fast at emulation (20-30% faster than Ivy Bridge) compared to most other workloads (5-10% faster than Ivy Bridge).
Take a look at the Dolphin 5.0 benchmark - Ryzen actually ends up consistantly performing worse than Haswell on a per-GHz basis, and this persists even when using much newer builds of Dolphin (though to a lesser degree).
However, the Ryzen 2200G and 2400G both launch in literally 5 days and should have some improvements to the latency of the memory subsystem, so it's very possible that they may end up with performance in Dolphin that exceeds the current 1000-series Ryzen CPUs.
I think the point here is that the "5.0" benchmarks aren't quite representative of the latest -development release performance, as the performance on ryzen has increased in proportion to the intel equivalents (apparently due to an update of the compiler?)
That's one of the problems with "Benchmarks" - they never quite match with real-world performance, and people often misunderstand what they're saying (here the benchmark list is saying that it's slower in dolphin 5.0 performance. What it is not saying is they are slower in the latest dev release, or generalized to "all dolphin releases ever". Those statements may be true, but that benchmark list isn't evidence for either.)
Thanks all for the insight. I'll stick with Intel for emulation purposes, for the moment.