Dolphin, the GameCube and Wii emulator - Forums

Hey,

I'm somewhat saddened that no one has posted anything from a bulldozer/piledriver cpu yet, so I decided to fix that! (at least partially.)

Hardware configuration used:

CPU: AMD FX-8350 - Stock 4.0 Ghz (~Turbo 4.2 Ghz)
GPU: AMD 6950 1GB - Cat 13.1
RAM: 16GB DDR3 at 1600Mhz
OS: Windows 8 Pro 64bit
Build: Dolphin 3.0 x64 (The one Starscream uploaded in the first post.)

NTSC-U Wind Waker
Save 1 (56|111|185)
Save 2 (45|90|150)
Save 3 (70|141|235)


That's what you can expect from Piledriver!

If anyone wants to get a nice page made with all of the stats that we have so far, feel free. We can even start to discuss how everything should look and what the layout should be if you like. Unfortunately, I do not have the time or the will to make this "page" myself. This doesn't mean that the benchmark is over or that people should stop submitting results, but I do hope someone considers gathering all of the information and making something nice out of it eventually.

It's official, Vishera is faster than a Sandy Bridge i3 in Dolphin.

Or at least Vishera with 8MB of L3. (FX-4300 only has 4MB of L3 compared to the others). I would be particularly interested in how well an FX-6300 does considering it also has 8MB of L3 cache but costs considerably less than the FX-8350, not to mention has a lower TDP at 95w.

I wish we could have an official LLE benchmark
I saw FX-8150 could run Super Mario Galaxy full speed all the time with HLE ...But those CPU only get 30-50 FPS with LLE
On the other hand , the slower clock speed Sandy Bridge i5 2400 could run SMG full speed with LLE
Afaik , Piledriver is 10% faster than Bulldozer -> FX-8350 is a bit slower than i5 2400 in Dolphin ?

Raydonn Wrote:I'm somewhat saddened that no one has posted anything from a bulldozer/piledriver cpu yet, so I decided to fix that! (at least partially.)

....we already had two people post piledriver results.

We haven't had anyone post vishera results yet, which is what I suspect you meant to say.

Nintendo Maniac 64 Wrote:It's official, Vishera is faster than a Sandy Bridge i3 in Dolphin.

Barely. But yes. I'm moving it into the the very fast category (category 2). We still don't have any bulldozer results (which I suspect belongs in category 3).

Nintendo Maniac 64 Wrote:Or at least Vishera with 8MB of L3. (FX-4300 only has 4MB of L3 compared to the others). I would be particularly interested in how well an FX-6300 does considering it also has 8MB of L3 cache but costs considerably less than the FX-8350, not to mention has a lower TDP at 95w.

You forget that the FX-6300 has a much lower clock rate than the other two. Its main selling point is that it offers good multithreaded performance for a similar cost to the FX-4300 at the expense of poorer single threaded performance (which as you know dolphin desperately needs). It will certainly perform the worst in dolphin out of the three.

IKarlCPFC Wrote:This is using a clean install of Dolphin 3.0 with no settings changed.

Follow the OP!

Use the included build and saves and stand still.

IlovetheGamecube Wrote:It looks like its running better because of the older version of Dolphin.

Our benchmark is using 3.0, and so is he.

IKarlCPFC Wrote:Are there any settings that I can change (from default) that might maximise FPS?

You're not supposed to. That's the whole point of a benchmark. Everyone needs to be using the same settings so that their results are comparable.

IlovetheGamecube Wrote:Maybe you could shrink the window so it 320 by 480? You're using HLE audio right?

His cpu is clearly the bottleneck so a lower resolution won't help him. Also he said he's using the default settings so that means he's using HLE.

IKarlCPFC Wrote:Yeah I am. I'll try shrinking the window.
Tried making the window smaller. Not sure of exact window size but,

Shrinking the window won't necessarily lower the IR depending on what you have the IR set to. Speaking of which. What do you have the IR set to?

IlovetheGamecube Wrote:Have you set Dolphin as a high priority on task manager? That might help. I found it interesting how the gamespeed was higher but the FPS was the same. Did you find it playable as it was?

This is a benchmark. Stop trying to find gimmicky little ways to boost his performance. All you're doing is throwing the accuracy of our results out the window.

Besides this should have little to no effect in dolphin. Changing task manager priority belongs with "guys I got a 1% speed boost from cleaning my registry" logic in the bin of pointless advice.

admin89 Wrote:I wish we could have an official LLE benchmark
I saw FX-8150 could run Super Mario Galaxy full speed all the time with HLE ...But those CPU only get 30-50 FPS with LLE
On the other hand , the slower clock speed Sandy Bridge i5 2400 could run SMG full speed with LLE
Afaik , Piledriver is 10% faster than Bulldozer -> FX-8350 is a bit slower than i5 2400 in Dolphin ?

Honestly it wouldn't matter. If it's faster here it will be faster under that situation too. This benchmark is actually better because it's so lightweight on the GPU that it pretty much guarantees a cpu bottleneck even with the worst IGPs.

(02-18-2013, 05:41 PM)NaturalViolence Wrote: [ -> ]You forget that the FX-6300 has a much lower clock rate than the other two. Its main selling point is that it offers good multithreaded performance for a similar cost to the FX-4300 at the expense of poorer single threaded performance (which as you know dolphin desperately needs). It will certainly perform the worst in dolphin out of the three.

Oh I didn't forget it has a lower clock. Even in single-threaded workloads the FX-6300 performs considerably closer to the FX-8350 than to the FX-4300 - heck it even beats an FX-8320, most likely due to a slightly higher turbo clock.

You can tell the following benches were lightly-threaded since the Ivy i3 ranked higher than the FX-8350.

SOURCE: http://www.xbitlabs.com/articles/cpu/display/fx-8350-8320-6300-4300_7.html

Here are the results so far!
If you notice any mistake I've made, please let me know.

Link to the spreadsheet on Google Drive.

@naturalviolence I was trying to further help his speed like he was asking and I asked him to do this benchmark in his thread so we could help him further. I guess I posted in the wrong place?

Yeah I would have kept the performance stuff in the original thread.

Anyways I notice a few odd data points on that graph. Take a look at the 3570K results in save 1 for example.
73 @ 3.7GHz
74 @ 4.4GHz
90 @ 4.5GHz
97 @ 4.8GHz
103 @ 5.0GHz

Something is obviously wrong with the second result. It's way off from the rest. Also for my results (3.7GHz) did you use my latest post where I used the GTX 660 or the older post where I used the HD 4000?

Also I think we can safely say that something is wrong with the system of the guy you used the A10-5700 since his results are wwwaaayyy off. I'll look over it later but for now those ones stand out to me.

Nintendo Maniac 64 Wrote:heck it even beats an FX-8320, most likely due to a slightly higher turbo clock.

Yeah that has to be why. The larger L3 cache shouldn't make a difference. Bulldozer/piledriver modules have ridiculously high L2 hit rates.

Thanks to Garteal for taking the time to make that graph. I'll add it to the first post at some point.