Dolphin, the GameCube and Wii emulator - Forums

(03-17-2014, 05:13 AM)QuentinX5 Wrote: [ -> ]The "high-end" retailer one out currently, the FX-8350, for instance, has a "processing power" of 33.6 GHz, because 8 cores * 4.0 GHz

No it doesn't. Even games that use 4-cores (the few, I don't think any game uses over 4 at this point) don't use it like that at all. Dolphin is a single-core application and only uses 3 cores to emulate the GC's 3 microprocessors. I contemplated getting that (or similar) AMD cpu a while ago and people here were pretty insisting that it was far inferior to a high-end i3/5/7. It's cheaper for a reason, why would 8 cores @ 4.0 be cheaper than 4 cores @ 3.4?

(03-17-2014, 05:58 AM)KazumiZorah Wrote: [ -> ]It's cheaper for a reason, why would 8 cores @ 4.0 be cheaper than 4 cores @ 3.4?

The short answer to this is "marketing" and "research". AMD presents an excellent price/value in some of their newer product lines for general computing. Their IPC is poor as a result of inferior R&D compared to Intel. Some of it is also the fact that it is Intel and someone has to pay for all of that great marketing material. Right now you can get an AMD CPU that is about 1/2 as performant (generally speaking) as what Intel offers. It will come at about 1/3 of the price of that Intel chip, so in a sense, that makes it a great bargain.

In practical terms, IPC is much more important for Dolphin than raw clock speed. If you're doing any kind of intensive work like video encoding, the i5-4670 tears the doors off anything AMD offers. I will also say anecdotally that it is a hell of a lot faster than any AMD CPU I've ever used for general computing. If you have the money to spend and want your computer to go fast, AMD is just not your ticket right now for CPUs.

QuentinX5, do you even know what IPC is? It is how many instructions a core can handle per cycle. AMD's IPCs are very, very low right now, while Intel's are rather high. To get the total peak IPC of a processor, you do not simply multiply the clock rate by the amount of cores. You must multiply the clock rate by the IPC, then multiply THAT by the number of cores. And then, you need to factor in architecture differences.

Currently, Intel not only has higher IPCs than AMD (AMD's peak IPC on most chips right now being 2-4), but (correct me if this is wrong) also does some things using less instructions.

So, no, AMD doesn't have the most processing power right now.

Even if your equation were right, they don't, because Intel has a 12-logical-core beast with a clock rate that I can't remember which is higher than a stock 8350. Maybe a tiny bit slower than a 9590, but a 9590 still has 8 cores.

(03-17-2014, 05:18 AM)rokclimb15 Wrote: [ -> ]I'm not particularly interested in wading into the details of your reply, but that whole thing about the processing power of various CPUs is just way off base.

However, I would like to explain a little about what is going on in your screenshot above. You're running at about 50% realtime speed. Normally, that's a CPU bound function, but you are arbitrarily constrained by GPU in this situation because you're using 1. a Dolphin build after tev_fixes_new merge 2. OpenGL backend and 3. an Nvidia card. If you use these three things together, you'll get about a 50% reduction in FPS.

I recommend switching to D3D in that revision and re-performing your experiment with the screenshot and re-post. Or, switch to some revision before 1198 (I think). The above experiment does not reflect the capabilities of your hardware accurately.

It's not always a 50% reduction in FPS though (unless somehow I should actually be getting 60 FPS in certain areas). I can wander around just fine in the beginning area (Orion Village?) and be at 30 FPS, but entering the forest ahead of it launches me down to the 15 FPS.

Wait, I remember reading somewhere in the tutorials about D3D not being that great for NIVIDA graphics... Oh! It was on the setup guide for best recommended settings:
" Backend - OpenGL is the fastest for Nvidia graphics cards, and Direct3D is usually the fastest for AMD graphics cards. Note: OpenGL is the only backend available on Linux and Mac OSX. "
So why isn't this the case here? My guess behind it was that OpenGL was developed more for NVIDIA and D3D developed more for AMD. I mean, I guess it wouldn't necessary apply to every scenario, but why would changing between the who cause a speedup?

*NOTE* I wasn't using D3D because it was giving me some issues, but using it now I get about 25 FPS in the areas where I was seeing some intense lag and slowdown.
Picture below:

(03-17-2014, 05:19 AM)Anti-Ultimate Wrote: [ -> ]

(03-17-2014, 05:13 AM)QuentinX5 Wrote: [ -> ]The "high-end" retailer one out currently, the FX-8350, for instance, has a "processing power" of 33.6 GHz, because 8 cores * 4.0 GHz. However, a CPU comes down way more to just it's "processing power" (I'm using the word processing power for lack of a better term. )

Please tell me you are joking. Are you seriously working in a computer store? That sentence above there made me cringe really hard. You can't just multiply the frequency * number of cores, that's not how it works.

You seem to of taken my quotes about how processors work to be 100% accurate. Please refer to the quote below where Kinkinkijkin explains it without the use of quotation marks.

(03-17-2014, 05:58 AM)KazumiZorah Wrote: [ -> ]

(03-17-2014, 05:13 AM)QuentinX5 Wrote: [ -> ]The "high-end" retailer one out currently, the FX-8350, for instance, has a "processing power" of 33.6 GHz, because 8 cores * 4.0 GHz

No it doesn't. Even games that use 4-cores (the few, I don't think any game uses over 4 at this point) don't use it like that at all. Dolphin is a single-core application and only uses 3 cores to emulate the GC's 3 microprocessors. I contemplated getting that (or similar) AMD cpu a while ago and people here were pretty insisting that it was far inferior to a high-end i3/5/7. It's cheaper for a reason, why would 8 cores @ 4.0 be cheaper than 4 cores @ 3.4?

Correct, and once again, as I said, I used the term "processing power" for lack of a better term. The point of my post is in no way to say AMD is better, because throwing more processors onto a CPU in no way makes a computer, or even a CPU for that matter, work harder or faster.

(03-17-2014, 06:28 AM)kinkinkijkin Wrote: [ -> ]QuentinX5, do you even know what IPC is? It is how many instructions a core can handle per cycle. AMD's IPCs are very, very low right now, while Intel's are rather high. To get the total peak IPC of a processor, you do not simply multiply the clock rate by the amount of cores. You must multiply the clock rate by the IPC, then multiply THAT by the number of cores. And then, you need to factor in architecture differences.

Currently, Intel not only has higher IPCs than AMD (AMD's peak IPC on most chips right now being 2-4), but (correct me if this is wrong) also does some things using less instructions.

So, no, AMD doesn't have the most processing power right now.

Even if your equation were right, they don't, because Intel has a 12-logical-core beast with a clock rate that I can't remember which is higher than a stock 8350. Maybe a tiny bit slower than a 9590, but a 9590 still has 8 cores.

Thank you for explaining it in a way in which quotations weren't needed. As I stated above, I used the term "processing power" for lack of a better term. In marketing and the world we live in, people look at that and say "HOLY COW! A 8-CORE PROCESSOR THAT RUNS AT 4.2 GHz AND IS $100 CHEAPER THAN INTEL! AMD MUST HAVE THEIR STUFF TOGETHER!" and go run off and build an AMD computer and have at it. Which, don't get me wrong, there are "some" (NOTE THE USE OF A QUOTATION MARK) advantages to having more processors in a CPU. But, as you said, at the moment, Intel's ability to have higher IPC's per core gives them a HUGE advantage over AMD, considering that (rough estimate of a number coming up here) probably 95% of programs people use access nowhere near 8 cores.

The guide you read was completely accurate until about 10 commits ago. It hasn't been amended yet to stipulate that from a performance perspective D3D is probably the best backend for Nvidia cards right now. There were a few notable bugs in D3D rendering, but many have been fixed by the very commit that caused the slowdown. So, absent a particular bug or situation, I'd suggest D3D for an Nvidia setup unless you have a Titan card (you don't).

You might not see 50% everywhere, but in motion areas or areas with lots of textures, you will. IMHO that makes the game unplayable.

I couldn't say at this point whether your CPU or GPU is constraining your gameplay. To find out, reduce your IR and see if framerate goes up. If it does, it is a GPU bottleneck (possibly still tev_fixes related, D3D still takes about a 5% hit). If it doesn't, it's a CPU constraint. Although many people will recommend disabling TurboBoost on laptops to prevent overheating, you might want to turn it on and check your dynamic clock rate with a turbo boost monitor. That extra 400 MHz comes in handy.

the only way the fx 8350 could rock over an intel (maybe over i5) due to the 4 extra cores would be the ability to process data faster in 3d rendering, vector calculations, video editing, sculpting.... but that as well depending on the app itself (if it is optimized for single threaded power or multi threaded power).
intel has better single threaded processing power while amd owns better multi-threaded power (at their respective price).
a xeon will be a nice cpu to have but errm at least if you gonna make a bit use of it's reason of being made.. cad, render, calculation, particle, simulation or whatsoever. with a xeon the rig might turn into a small personal server and media/gaming rig xD
for only normal use, gaming and emulation gaming I wouldn't bother with something expensiver than an i5.
and yes it is possible to play and tweak with the data of a mainboard. but mod it? that might turn creepy if there is even a 0.1 margin of mod error or mistake... I mean you don't need the mb anymore O_O ?
I know many cheaper mainboards that own a cmos flashback button o.O the mb you own is a mb created for high overclocking potential. just a suggestion but you could sell the 4771 and get an unlocked cpu even i5 (if you are not into rendering and "hard" editing). oc is not meant only for games but for productive softwares as well. yet the question is if it is really needed.
you can have a cpu with 100 cores clocked at 5ghz. that won't get you a processing power of 500ghz O_O you would need super duper nitro cooling for that x.x
all the cores will work together in harmony at 5ghz (if the apps pushes all the cores to work). this theory gets blown away by the titan supercomputer. but that's a supercomputer not a rig ...

I'm going to try to not pull myself into this, but I'd just like to point out that modern iGPs can actually be faster than low-end discrete GPUs and older high-end GPUs. I know it's a synthetic test, but according to PassMark the HD 6530D iGP in the A6-3670K is actually faster than the GT 610:

HD 6530D (iGP) - 520 points: http://videocardbenchmark.net/gpu.php?gpu=Radeon+HD+6530D&id=109
GT 610 - 444 points: http://videocardbenchmark.net/gpu.php?gpu=GeForce+610&id=2453

However this doesn't mention the RAM speed, which can have quite an effect on the 6530D's performance.