(02-26-2014, 06:08 AM)degasus Wrote: [ -> ]Maybe haswell just improved on executing our shitty jit code.
PCSX2 performs more or less the same way - so they are probably same as shitty ;-)
drhycodan Wrote:Well it just proves Haswell is much faster in single threaded performance and IPC over Ivy Bridge. Maybe no other app has tapped into Haswell's true potential because they are mostly multi-threaded based? Why do you think the hexa-core i7 extremes always come out on top? Surely, it's not about their single threaded performance.
That doesn't make any sense. The performance per core should not change between singlethreaded and multithreaded applications. Haswell is usually around 5-15% faster clock for clock than ivy bridge in either case. In dolphin and PCSX2 however it's 30% faster. i7 extremes only come out on top in highly multithreaded applications. They perform the same as regular sandy/ivy bridge cpus in applications that use 4 or less cores (most apps) including dolphin.
degasus Wrote:Maybe haswell just improved on executing our shitty jit code.
Well obviously, but why? Which specific optimizations are doing this? 30% is a really big gain.
(03-04-2014, 10:26 AM)NaturalViolence Wrote: [ -> ]drhycodan Wrote:Well it just proves Haswell is much faster in single threaded performance and IPC over Ivy Bridge. Maybe no other app has tapped into Haswell's true potential because they are mostly multi-threaded based? Why do you think the hexa-core i7 extremes always come out on top? Surely, it's not about their single threaded performance.
That doesn't make any sense. The performance per core should not change between singlethreaded and multithreaded applications. Haswell is usually around 5-15% faster clock for clock than ivy bridge in either case. In dolphin and PCSX2 however it's 30% faster. i7 extremes only come out on top in highly multithreaded applications. They perform the same as regular sandy/ivy bridge cpus in applications that use 4 or less cores (most apps) including dolphin.
degasus Wrote:Maybe haswell just improved on executing our shitty jit code.
Well obviously, but why? Which specific optimizations are doing this? 30% is a really big gain.
My guess seeing the impact in ipc faster cache has in the dolphin benchmark for people who overclocked their cpu cache and the graph below:
http://media.bestofmicro.com/K/0/386928/original/cache.jpg
is that perhaps is due to faster cache memory.
Hmm. I wonder if delroth knows what kind of cache hit rates dolphin encounters.
Actually, I don't see how it's faster in PCSX2. The MGS3 intro still slow down to 60-70% speed, no improvement at all over my 3930K 6 core Sandy Bridge. This is with software rendering of course because pcsx2's hardware renderers are shit. I don't understand why don't they implement all the post processing effects and all the other missing effects in the hardware renderer, maybe then pcsx2 would be playable.
(03-09-2014, 07:25 AM)drhycodan Wrote: [ -> ]Actually, I don't see how it's faster in PCSX2. The MGS3 intro still slow down to 60-70% speed, no improvement at all over my 3930K 6 core Sandy Bridge. This is with software rendering of course because pcsx2's hardware renderers are shit. I don't understand why don't they implement all the post processing effects and all the other missing effects in the hardware renderer, maybe then pcsx2 would be playable.
4770K stock w/ turbo boost + N580 3GB Lightning Extreme and I'm getting full speed in MGS3 intro.
Were the builds and settings identical? The official pcsx2 benchmark shows a 30% increase just like dolphin.
Official build 1.2.0 no speedhacks + hardware mode.
I was asking drhycodan. Sorry if that wasn't clear.
i think it would be because it works different on many cpu's and the results can be completely different