..:: PCSX2 Forums ::..

I suspect that everyone will jump to the obvious conclusion on this one but give the whole post a read first

1) I have Fedora 24 running a git build of PCSX2.
2) The build of software has not changed between old hardware and new.
3) The old rig was an i7 920 @ 3.6GHz with 6 gig tri channel ram and a GTX 770
4) The latest build is an i7 6700k @ 4.5GHz with 16GB dual channel ram (3000Mhz) and a GTX 770

The "old rig" used to really struggle with some games such as MGS / SSX / Jak and Daxter. In order to make them playable I had to mess around with the Speed hacks to adjust the EE cyclerate and VU cycle stealing settings and with SSX especially there was no "good" speed hack configuration.

The "new rig" doesnt even break a sweat with most of the games I have tried so far. I have turned all of the speedhacks off and everything runs at full speed. SSX especially is like a chalk and cheese comparison.

So my question is, is this really the result of a ~1GHz increase in CPU speed or does PCSX2 take advantage of some new instructions in the 6700k that the i7 920 did not have? I always associated performance with clock speed but this really does seem like there is more at play here and I would love to know what the real game changer is.

Thanks.

PCSX2 relies heavily on single thread performance. And your new CPU has much more oomph per thread/core. It would definitely explain the difference.

You can check out the difference here: https://www.cpubenchmark.net/singleThread.html

(11-09-2016, 10:57 PM)FlatOut Wrote: [ -> ]PCSX2 relies heavily on single thread performance. And your new CPU has much more oomph per thread/core. It would definitely explain the difference.

You can check out the difference here: https://www.cpubenchmark.net/singleThread.html

Thanks for the link.

I never really considered single thread performance as an issue as I somewhat naively tied performance to clock speed. I am familiar with the advantages of parallel processing in a multicore environment but never considered that single thread performance would have improved so much over the past 5 years.

If I am reading correctly, single threaded performance has seen most of its gains using a form of parallelization down at the instruction level resulting in double the performance between the 920 and 6700k. This fully explains the situation.

Thanks for info. Hope it provides some useful information to others in the same situation.

I had a similar jump when I upgraded from my old i7-875k to an i5-4570. The improvements in the chip combined with some of the progress gregory has made has really improved a lot of my games, and I just bought a new laptop that i have yet to try anything on It is amazing though how far CPUs have come in the past decade, and right now I am amazed that my new laptop runs basically everything while topping out around 60*C

6700K has an issue running GSdx in software mode, though. And as of late, the appveyor vs2013 build does not help with the software mode, either.

FWIW, some instructions (gather) are more than 2 times faster on skylake vs haswell. Gather instruction is used to sample texture data (SW renderer AVX2).

Quote:Haswell : 20 uops, 9 cycles of latency
broadwell: 10 uops, 6 cycles of latency
skylake: 4 uops, 4 cycles of latency.

(11-25-2016, 01:44 PM)gregory Wrote: [ -> ]FWIW, some instructions (gather) are more than 2 times faster on skylake vs haswell. Gather instruction is used to sample texture data (SW renderer AVX2).

Does it have much impact on games using haswell on sw ?

So far, few code uses this new instruction. So impact is rather small. It is at least use once to convert texture in the SW renderer. Skylake will be faster here but some people complain on issue with Skylake/Windows (I'm not sure it impacts Linux too). Hopefully it will be used more often in the (distant) future.

This is a good example of how someone learned how important single thread performance is. I see A LOT of posts where people don't understand why the games run so slow when they think they have a powerful pc. I wish there was a way they cold be better educated on it, so we don't see so many posts where we have to explain that there cpu isn't as powerful as they thought it was. At least where pcsx2 is concerned