Dolphin, the GameCube and Wii emulator - Forums

He only want to see how different models of cpus and architectures, performs on dolphin, in this way you can make a more organized list to have a better reference of performance.
If you can overclock your cpu its good for you, but its not the point of the test.

CPU: i7 4800MQ @ stock - 2.7GHz/3.7GHz Turbo Boost
OS: Win 8.1 x64
Time: 9 minutes, 3 seconds
Screenshot: http://i.imgur.com/nJOrUGC.jpg

CPU: Intel i7 CPU 920 @ 2.67GHz (2.93 GHz Turbo speed)
OS: Lubuntu 13.10 x64
Time: 16 minutes, 9 seconds
Screenshot: http://i.imgur.com/xCikd6Y.png


CPU: AMD A8-6600k @ 3.90GHz (4.20GHz Turbo speed)
OS: Lubuntu 13.10 x64
Time: 14 minutes, 9 seconds
Screenshot: http://i.imgur.com/tAA0K42.png

Edit: Listed the turbo speed for the A8-6600k & i7 920

(01-22-2014, 03:03 AM)AnyOldName3 Wrote: [ -> ]It is a bit stupid. If 90%+ of the userbase of a chip bought it specifically to overclock, because once it was overclocked it would be the fastest thing available, then it doesn't support the "the point is just to know how well a given CPU performs" idea. In Dolphin, it's a fact that users are going to find the 4670K faster than the 4670, because when they spend the extra money on the K, they're spending that money because they want to overclock.

There is a point to the idea that "overclocking adds so many dimensions to the problem that the results become meaningless", but most of the time, a stable overclock of a specific model of chip at a specific frequency will give pretty consistent performance between individual chips, and as the frequency is included in the results, it's not like people will muddle a 0.2 GHz OC with a 1 GHz OC.

Finally "people trying to optimize for a benchmark" isn't a huge issue, as if we're locked down to a specific revision, and a specific test which should be entirely CPU bound, the only things optimisation will bring about are overclocks and new CPU purchases, both of which should cause a direct increase in emulation speed in game, and are therefore relevant to Dolphin users (or prospective users) trying to decide what CPU to purchase or gauge their own performance before going to the hassle of ripping their game library.

I've probably said something dumb here which needs picking apart (and therefore should be), but in its current form, I don't see Delroth's reasoning making much sense.

Anyhow, I've got several older computers which I never got around to benching last time, but will bother to now the process is simpler.

OC data is not interesting. On this specific benchmark, clock rate correlates extremely well to performance increase on the same architecture. I could interpolate an i3 and an i7 result and give you a pretty good approximation of the time that you will get with your OC'd clock rate.

Then people started doing weirder OC: OCing memory (when I explicitly said I'm interested in CPU performance here - cool, you made your point, but that doesn't help anyone figure out how differeut microarchitectures perform on Dolphin), OCing cache, partial OCs, OCs with turbo, ... There are way too many variables to keep track of for results that are not interesting to compare microarchitectures (and again, time*clock rate correlates very well to microarchitecture if you remove outliers).

If you want to maintain results that take into account OC, feel free. You can just make a copy of my current results list and maintain it on your own. No hard feelings about that, I personally find it completely useless but do whatever you want. I personally don't have the time nor the interest to think about how to integrate OC results (I already don't have time to add more stock clock results into that spreadsheet... yay for full time work + moving + buying furnitures + planning FOSDEM trip + doing dolphin development).

(01-22-2014, 03:03 AM)AnyOldName3 Wrote: [ -> ]

(01-22-2014, 02:47 AM)Anti-Ultimate Wrote: [ -> ]

(01-22-2014, 01:47 AM)Gvaz Wrote: [ -> ]

(01-21-2014, 11:38 PM)Anti-Ultimate Wrote: [ -> ]No overclocked cpu's anymore. Delroth should really put that into the first post. :/

Why?

For example the 3570k is an extremely popular cpu because you pay slightly more and you get a free, stable 400+mhz out of it right off the bat. More if you can do it.
Going forward in the future it's likely that even more cpus will be "unlocked" with the ability to have a stable clock at those frequencies. And what difference does it make to this benchmark really?

According to delroth, OC is too complicated with Cache Ratio, BLCK changes and so on. He says he doesn't want to publish data taken from a CPU that is overclocked like ass only for this benchmark and will catch fire else. (He said it like this).

That makes a little more sense than how I read it originally. Surely the ideal solution would be to request only stable overclocks, and just hope that no-one lies. If people were going to lie about that, they could also have 'shopped the numbers in the screenshot. If someone cares about winning enough to burn their CPU out, then there are cheaper and easier ways to generate useless results.

It's not a question of lying or not lying, it's a question of people seeing ordered numbers in a spreadsheet and thinking it correlates with e-penis size. I've lost count of the number of people I've seen saying "oh, I'll push my OC again so that I get first place on the list!". That is not helping anyone in my opinion.

In a separate message, because it might be interesting to people not interested in the drama. There is a second sheet in the results that shows "time * clock speed" for each of the results (including OC). The correlation by microarchitecture is very strong: for example:




Interesting factoid: the time*clock for a real Wii is 766908. That means Haswell is so efficient that although Dolphin's JIT is pretty terrible, it still averages only 0.43 PPC instructions/clock. That's VERY good.

(01-23-2014, 05:52 AM)delroth Wrote: [ -> ]

Code:

Sandy Bridge    i5-2500k    4700    0:09:20    2632000
Sandy Bridge    i5-2500k    4500    0:09:50    2655000
Sandy Bridge    i5-2500k    4500    0:09:57    2686500
Sandy Bridge    i5-2500k    4200    0:10:34    2662800
Sandy Bridge    i7-2600k    4000    0:11:04    2656000
Sandy Bridge    i7-2600k    3400    0:12:28    2543200
Sandy Bridge    i5-2500k    3400    0:13:17    2709800

I take joy in knowing that my e-penis was bigger than any of the other Sandy Bridge users.

In seriousness, I can repeat the test with the stock clock if it would still be helpful at this point.

CPU: i7 4702MQ @ 2.2ghz (3.2/3.1/2.9ghz TB)
OS: Win 8.1 x64
Time: 11 minutes 1 second
Screenshot: http://i.imgur.com/gOULqGE.jpg?1?9546

CPU: i3-4000m @2.4ghz
OS: Windows 8 x64
Time: 14 minutes 44 seconds
Screenshot: http://imgur.com/fWsUhDb
I did much better then I thought I would.

CPU: I5 2500k @ 4.2ghz
OS: Windows 7 x64
Time: 10 minutes 28 seconds
Screenshot: http://i.imgur.com/nCM6JsL.png

Here's my result redone with the overclock disabled / bios defaulted:

CPU: i5-2500k @3.3ghz, all settings stock
OS: Windows 8.1 Pro x64
Time: 12 min 56 sec
Screenshot: http://farm3.staticflickr.com/2849/12113696283_c8bb8fa3cd_o.png