Dolphin, the GameCube and Wii emulator - Forums

Nintendo Maniac 64 Wrote:But IPC can differ largely based on what is used by the program. In particular, Cinebench is supposedly very floating-point heavy.

Nevertheless, this could theoretically mean that we could use the single-threaded results in Cinebench R11.5 as a "cheat sheet" for Dolphin performance.

Kinda sorta. I doubt they match up perfectly since the workload is not exactly the same. Which is why I would prefer not doing that or at least taking it with a grain of salt. Keep in mind that dolphins IPC will also vary depending on settings, OS, other hardware, which game you're running, revision, etc. But mainly settings and game.

Since historically x86 architectures have generally been designed as incremental improvements (minor to moderate revisions) over their predecessors rather than being completely redesigned from scratch (bulldozer is one of the few examples of the latter ever being done) they tend to improve IPC in all applications compared to their predecessors (though by varying amounts based on the workload being done). However while not always true it is generally true that most of the noticeable improvements tend to effect nearly all applications fairly equally. You will notice for example that nehalem's IPC is almost always around 20% higher than core 2 quad give or take a few percent. Bulldozer and piledriver on the other hand are all over the place depending on the workload for obvious reasons.

For the record dolphin is very floating point heavy (at least compared to most applications) too. And so are games in general.

(02-11-2013, 11:08 AM)NaturalViolence Wrote: [ -> ]Kinda sorta. I doubt they match up perfectly since the workload is not exactly the same.

Uhh... it already does match up pretty much perfectly. I haven't benched my Brisbane in Cinebench at 3.1GHz, but it gets a single-threaded score of 0.24 @ 1GHz and 0.47 @ 2GHz, therefore you can extrapolate that 3.1GHz should give a score from 0.712 to 0.744 - darned near EXACTLY the same IPC score you gave my Brisbane.

Ok here is the 1100t at 2.4 ghz

OS: Windows 7 home premium x64
CPU: Phenom II x6 1100t (2.4 Ghz, turbo disabled)
GPU: Radeon HD 5570 ddr3 version
RAM: 8GB Crucial Ballistix 1600
Dolphin: 64bit From this thread only options changed where the ones Starscream changed and auto adjust window to make pics not as big.
GAME: NTSC


Save 1

Save 2

Save 3 (drops to 42 once the boat comes on screen)


I ran the tests again with two cores disabled but it didn't make a difference.

@Nintendo Maniac 64

Having never actually run cinebench before is that your score or your IPC that you're talking about? It sounds like you're just talking about a score. I normalized all of the results to 4GHz to compare for IPC.

Thanks for taking the time to do that, Xalphenos.

(02-11-2013, 11:24 AM)NaturalViolence Wrote: [ -> ]@Nintendo Maniac 64

Having never actually run cinebench before is that your score or your IPC that you're talking about? It sounds like you're just talking about a score. I normalized all of the results to 4GHz to compare for IPC.

Oh, yeah, that was my Cinebench score... so seeing how you normalized to 4GHz I guess it really doesn't apply then.

...but still, what the crap! The point values you listed are still darned near the same as the Cinebench scores for those CPUs at ~3.3GHz!

Core i5 3570k - Cinebench @ 1.60:
http://pcpartpicker.com/benchmarks/cinebench/render-single/?detail=1&rid=117

Core i5 2500k - Cinebench @ ~1.45:
http://pcpartpicker.com/benchmarks/cinebench/render-single/?detail=1&rid=72

Phenom II x6 1100T - Cinebench @ 1.11:
http://www.legitreviews.com/article/1481/5/

Seriously, the results of ~3.3GHz CPUs line up PERFECTLY with your 4GHz IPC calculation.


EDIT: HOWEVER, Llano seems to fail a bit with this calculation. An A8-3850 at a stock 2.9GHz scores ~0.87. If you extrapolate that to 3.3GHz you only end up with ~1.00 Cinebench points rather than your calculated 1.08.

Maybe I know what I'm talking about then? Seriously though this doesn't surprise me at all. Nehalem is generally 20% faster clock for clock than core 2 quad. Sandy Bridge is generally 20% faster clock for clock than Nahelem. Ivy Bridge is generally 10% faster clock for clock than core 2 quad. And so on. There are few applications where those aren't true. All of my calculations in all threads up to this point have been based on these assumptions due to lack of real data. Although it's still very incomplete this thread is the first real usable data that we have on this stuff.

Oh, I thought your IPC calculations were based purely off of the Dolphin results and not other benchmarks.

were my scores on pat with simular processors?

---

par* srry for double post. *ps3 doesnt like edit mode on these forums*

Why are you browsing the web on a ps3?