Originally Posted by MadeInBeats
The OoE efficient CPU posts are interesting and encouraging, but my question still remains: why did the Tekken producer feel the need to mention an issue with CPU performance? Why not just not mention it, and apply the supposed efficient tech architecture of the WiiU?
Well if that analysis is accurate, then ports are guaranteed to suffer because they're designed for a very different solution in mind, lack of SMT is a kick in the groin.
Out-of-Order code is very branch intensive (it's part of the reason Valve was so outspoken against the CPU-tech on the console side when this generation begun), but code meant for in-order-execution won't benefit that much from out-of-order (it'll benefit a little mostly from less cache trashing and branch prediction, but that's all) on the other hand not having 2-way execution will hurt it big time, because without out-of-order it was relying heavily on multithreading. Now, multithreading is basically like having a second virtual cpu inside the cpu in an attempt to max out the rest of the cpu juice with more than just 1-way data execution; without that, in a 2-way tied software configuration, you'll instead use the other CPU cores available; meaning they could be using 2 CPU's in some capacity for what a single X360 CPU could do before; the leftover overhead will be different, yes; but hardly ideal. This wasn't really an issue on the cpu field (Pentium 4 introduced Hyperthreading, then Pentium M, Core Duo and Core 2 Duo lacked it; only 2008 with Nehalem architecture (Core i3/i5/i7) it returned, but then again PC software never assumed you had to have a multithreaded CPU; not the case here when for a whole generation that's all they had.
On the efficiency side though, I doubt Nintendo went with the offshoot CELL/PPE/Xenos architecture solution when the stage pipeline was lenghtned in order to achieve bigger clock rates (note how in 2005 there was no such thing as a Power5 CPU clocked at 3.2 GHz, anything surpassing 2.5 in a dual core config needing extreme cooling configurations on G5 Power Macs
. They're probably left with a regular pipeline stage design. (I don't know exactly which, Pentium III had 10 stages, current i3/i5/i7 cpu's have 16 stages; PPC G5 had 16 stages as well; shouldn't exceed 17)
Anyway it should be quite a little more efficient; remember how initial Pentium 4 (netburst architecture) actually took a beating from Pentium III (and later AMD Athlon CPU's) to the point intel gave up on it and reverted to Pentium III pipeline (Pentium M that paved way to Core Duo architecture). The bigger pipeline stage the cpu has the worse performance per clock it'll have. One nice episode to express this was how P4 Northwood benchmarks took a beating on Prescott at the same clockrate, because due to the pipeline had been lenghtned again, making it hit the 31 stage number. Similarly, AMD Athlon xx00+ cpu series used to beat P4 while going at much lower speeds due to that.
I doubt there's something fundamentally wrong with the cpu they went with, but running code not meant for it better than the aforementioned platforms is certainly not apples to apples.
Kinda how it has been said that the MGS2 port on the Xbox suffered slowdowns because they were still running the rain particle system on the CPU. PS2's cpu was crap at cpu tasks, but all system's fillrate was there and thus it had more cpu fillrate than Xbox's Celeron could hope to achieve; doesn't mean the Xbox's cpu was worse when it came to cpu tasks. Anyway, happened before.
Originally Posted by MadeInBeats
Why not just not mention it, and apply the supposed efficient tech architecture of the WiiU?
Most directors are not tech heads, most of the time they can only convey what engineers told them in simpler words sans technical jargon.
And an engineer won't rewrite the whole code in order for it to run more efficiently when it comes to a port; a port is a work of translating something in order for it to run, if possible they'll left as many things untouched as possible.
I wouldn't be surprised if lot's of developers are actually running audio tasks on the cpu whilst having a dedicated SPU for instance; it's never as simple as moving it to the dedicated DSP and expect it to work.
Originally Posted by lwilliams3
Tekken, as well as most console games, are designed around the 360/PS3's hardware strengths and weaknesses. Wii U is designed a bit differently, as the CPU is OOE, clocked a little lower, maybe less threads but more "lean", a lot better at general-purpose tasks, but not as good with FLOPS. It will take a bit for devs to adjust their engines to that, but not relying as much on the CPU on much for FLOPs and physics will likely be a characteristic for the other next-gen consoles as well.