The Legend of Zelda: The Wind waker CPU Benchmark

[Starscream]

You can overclock, just be sure to list your CPU speed and make sure that you didn't change any of the settings in the build provided.

[Nintendo Maniac 64]

Uhh... I know I can overclock seeing how your own Llano is OC'd.... I was asking about the "Auto adjust Window Size" setting in Dolphin. In the build provided it was unchecked, but the window size was actually larger than my display resolution (1152x864), so I had to check that checkbox.

[Axxer]

It took me a while, but I finally ran the benchmark on my laptop again and fixed the resolution stuff. I don't think anything changed at all, but this is still more accurate.

[Starscream]

I'm going to let this thread sit around a while more, I'm in no hurry to do anything with the results yet. Unless someone else wants to compile all the data, that'd be fine.

[NaturalViolence]

My desktop is up and running again but I still have a lot of stuff to install and dolphin is near the bottom of the list. I'm at work/school all day sunday and monday so the earliest I will have my results in by is tuesday.

[Starscream]

There's no rush. I plan on giving as many people as I can a chance to be a part of this benchmark, so that means it'll still be here for a while.

[rpglord]

Here are my results:

Cpu: i5 2400
Gpu: geforce 460 gtx
OS: win 7 x64
Dolphin : x64 downloaded from first post

[Xalphenos]

I just got interested in dolphin again, and this benchmark doesn't seem to be finished so I'll throw my results in there.

OS: Windows 7 home premium x64
CPU: Phenom II x6 1100t (Stock)
GPU: Gigabyte GTX 660 2GB GDDR5
RAM: OCZ Gold 8GB DDR2
Dolphin: 64bit From this thread only options changed where the ones Starscream changed.
GAME: NTSC










If I can get some of my stuff out of storage I have a few more systems I could test on as well. I also plan on trying to OC this 1100t and see if I can get some significant gains from it.

[Chriztr]

Testet with one old desktop and one old laptop.

Desktop:
OS: Win 7 Ultimate x64
CPU: AMD Athlon 64 x2 Dual Core 5000+ 2,60 GHz (stock)
GPU: ATI Radeon HD 4600 Series ( 4670 )
Ram: Crosiar 2gb stick and Kingsone 1 GB stick = 3GB, both DDR2.
Dolphin setup: x64 from the thread. Just as Starscream wanted
Game region: NTSC

Screens:




Laptop:
OS: Win 8 Enterprise x64
CPU: Intel Core 2Duo P8600 2,40 GHz (stock)
GPU: Mobile Intel® 4 Series Express Chipset Family (Microsoft Corporation - WDDM 1.1)
Ram: 2 GB
Dolphin setup: Still the x64 from this thread
Game region: NTSC

Screens:

[NaturalViolence]

It's FINALLY done.

OS: Win 8 Professional x64
CPU: Intel Core i5 3570K (stock)
GPU: Intel HD 4000 (stock)
Ram: 2 x 4GB (8GB) DDR3 1600 MHz
Dolphin build: 64 bit
Game region: NTSC

All images resized to half width/height with a lanczos filter so that they wouldn't be obnoxiously large.











Quick reference:
213%
186%
275%

The memory is probably still running at 1333MHz (which will bottleneck the IGP). I'm going to check on that right now.

Edit: Here we go:




As you can see the memory is still running at 1333 MHz effective (666 MHz real). The cores are running at 3.6 GHz (max turbo is 3.8GHz) with the first core fluctuating between 3.6 and 3.7GHz. (probably due to shifting load on/off the core as the threads jump around). The IGP is running at its normal 650 MHz clock rate not its 1150 MHz turbo clock which means the processor is detecting low load on the IGP and turboing the cpu cores instead of the IGP. I'm going to run some more thorough tested to see how high I have to push it before it detects high IGP load.

Edit 2: With lock threads to cores all 4 cores are fluctuating between 3.6 and 3.7GHz instead of just the first one.

Edit 3: I just realized I had my web browser with a bazillion tabs open in the background. I'm going to retest.

Edit 4: Apparently not. I get the exact same speeds without the browser open. I tested Lock threads to cores and it reduced my performance in spot 1 from 213% to 203%. Windows 8s scheduler is apparently pretty good.

Edit 5: Retested with 2x IR:
175% (-38%)
160% (-26%)
177% (-98%)

I have to get the memory running at its proper 1600MHz speed. HD 4000 benchmarks all show that 1600MHz ram provides a 20% speedup in IGP performance over 1333MHz ram. The IGP does turbo to 1150MHz when running at 2x IR.

Edit 6: The cpu cores are turboing even when the IGP is turboing. I activated the XMP profile for the new ram and now it's running at 1600MHz. here are the results:
1x:
230% (+17%)
191% (+5%)
298% (+23%)

2x:
191% (+16%) (-39%)
175% (+15%) (-16%)
195% (+18%) (-103%)

First parenthesis is the increase over DDR3 1333MHz. Second parenthesis shows the decrease at 2x using the same speed ram. It's interesting to note that the faster ram improved performance across the board. Up to an 8% performance increase at 1x and up to 10% at 2x. Another interesting thing to note is that the first two spots take a small performance hit from moving from 1x to 2x but the third spot takes a 33% performance hit from moving to 2x IR. You would think then that the faster ram would make little difference in the first two scenarios and a tremendous difference in the third, effectively reducing the IGP bottleneck in the third spot. But it doesn't, it boosts performance at 1x and 2x by the same amount and leaves the performance hit untouched. I can safely say that the IGP is not a bottleneck at 1x IR. But at 2x IR it definitely is in many situations even with the faster ram. The load is below 50% at 1x IR which means that it should be able to handle 1.5x IR without any bottleneck. Which is what I would expect considering the HD 3000 can run graphically demanding games at 1x IR without a bottleneck and the HD 4000 is supposed to be 50-66% faster than the HD 3000.

Since haswells IGP should eliminate memory bottlenecks using edram and provide an IGP that is twice as fast as the HD 4000 it should be safe to assume that haswells IGP should be capable of 2x IR without a bottleneck. Which means we're not that far away from 1080p GC/Wii emulation on IGPs, probably 1-2 more years for AMD and 3 more years for Intel.

I know I should update my original results now that I have my ram running at its proper speed but I'm going to keep it there for reference.

Edit 7: Yet another update. For curiosity I ran the benchmark with the other two backends and 3x IR to see how this IGP would handle them.

d3d9
1x:
230%
191%
298%

2x:
191% (-39%)
175% (-16%)
195% (-103%)

3x:
96% (-95%)
110% (-65%)
96% (-99%)

Performance is cut in half in spots 1 and 3. Spot 2 saw a 33% reduction in framerate. Overall the game is almost fullspeed but not quite.

openGL
Efb copy to texture gave me a blue screen (spot 1) or white screen (spot 2) with the HUD elements on top. Efb copy to ram showed the image but it only filled the bottom left 1/4 of the window. I blame the build.

d3d11
1x:
136% (-94%)
148% (-43%)
283% (-15%)

As we can see in spots 1 and 2 d3d11 is very significantly slower than d3d9. But in spot 3 it is only slightly slower.

2x:
130% (-61%) (-6%)
146% (-29%) (-2%)
240% (+45%) (-43%)

The first parenthesis shows the difference from d3d9, the second parenthesis shows the difference from the next IR down. What we see here is almost no drop at all in framerate going from 1x IR to 2x IR in spots 1 and 2, and a slight drop in spot 3. Compared to d3d9 at the same settings spots 1 and 2 are moderately slower and spot 3 is moderately faster! The IGP is not a significant bottleneck at 2x IR on d3d11.

3x:
118% (+22%) (-12%)
139% (+29%) (-17%)
122% (+26%) (-118%)

Here we see that at 3x IR d3d11 is now faster in all three spots. Spots 1 and 2 showed almost no drop in framerate compared to 2x IR while spot 3 once again acted as the outlier and showed a 50% reduction in performance.

This means that d3d11 scales better to higher resolutions on my setup. Even when the average framerate is better on d3d9 the minimum framerate seems to usually be better on d3d11, making it an overall better choice. With d3d11 I can run the game at 3x IR and it still stays well above fullspeed as I'm running around the 3 spots. It seems like the d3d11 backend is lighter on the IGP but heavier on the cpu (or maybe memory?). The would explain the big drop in performance at 1x but the better scaling as you climb to higher IR, what do you think neobrain?