bluestar2k11 :
I didn't say anything about it not taking a lot of power to do so, I just said it has been done on x86 hardware, so it can't be as hard as their implying, esp since these guys have the know how of everything in both systems, so designing a emulator to run previous games can't possibly be as hard as it is for the emulation community to make a wii one work well, who don't have all that information.
Though even if they got it working, how playable it would be (stares at halo 2's 360 emulator) it likely wouldn't be very good, unless they really put some work into it.
The first thing I want to clear up is that the people who develop emulators are those who completely understand the ISAs they're working with and the architectures of the machines in question. I would argue that in many ways, they're much more capable than the people that MSFT will ultimately task into writing a emulator if it's ever done. The lack of documentation to them is probably less of an issue than it sounds like it is.
Now, the question isn't as simple as simply translating the code from one ISA to another. Like I've said before, a simple functionally correct translator can be written by any competent CS student in college. The real trick is to make it fast enough to make it playable. If the final result is like 3fps they may as well not invest the effort (and they certainly won't announce it).
Why is performance so hard? Well, how each processor is architected will significantly impact the final result. [strike]For starters the Xenon uarch (powerPC) was only 3 cores. Jaguar (x86) is 8, so a direct translate will already eliminate 5 of the cores. They will always be idle because the game was never designed to scale to 8 cores.[/strike]
edit - alextheblue has corrected me:
alextheblue :
Actually, Xenon is an SMT chip, so you're only partially correct. It has three cores, but six threads. So a lot of games actually ARE capable of scaling easily to at least 6 cores, leaving two free to assist in emulating other components. Also the modern-day out-of-order Jaguar cores are pretty darn competent compared to the older in-order PPC cores used in Xenon. More so than you give them credit for...
My response to that is (for those of you who care about architecture):
I see that Xenon actually has 2 VMX128 engines per core. This makes Xenon significantly higher performing with SMT (Simultaneous multithreading). Based on Wikipedia (I'll fix the Wiki in a minute), I thought it was just 1 VMX128 per core which makes the second thread essentially worthless so I didn't bother mentioning SMT. This puts the Xenon core+SMT vs Jaguar 2 cores at par in terms of SIMD engine count (lets ignore which one is more efficient).
My point still stands, just not as strongly, A direct translation would end up using only 6 of the 8 cores the xbone has. There's more details below for anyone who cares. While you're down there, go give him a thumbs up for his catch
/edit
The next question is, why can't the translator expand that to 8 cores? Well, lets just say even using a high level language, multi-threading is difficult at best (there reaches a point where the overhead of adding more cores actually hurts performance more than it helps), doing at the assembly level would be insane. It's the primary reason why most things are still single threaded, the cost of multithreading isn't worth the investment. I would argue that it would be easier to take the source code and compile the game for x86 and release the game on xbox live as a download than it is to get the translator to do the equivalent change correctly and more importantly consistently. This is a translation inefficiency (compared to compiling the game to be native to the ISA).
Another hazard in this translation is features in one ISA that the other ISA doesn't support. Any feature that PowerPC has that isn't supported in x86 will have to be circumvented via a set of instructions. Now instead of one instruction doing something, it takes 3-4 instructions to the same function. This makes the code longer for the emulated architecture to do the same thing. This is also a translation inefficiency.
How each core compares between the two architectures also have a huge impact. Look at Shin-san's post, he started going into a little bit of detail of PowerPC's capability. I will, however, do a bit of handwaiving here because I don't have the time to look up the actual details on the 2 architectures. Jaguar was written for the mobile space, it's inherently a small simple core (which is why MSFT can stamp 8 of them on 1 die). PowerPC was designed in the day to be a much more powerful core (closer to AMD's Piledriver core). Essentially, we're comparing today's netbook CPU with the top of the line PowerPC a few years ago. I question that a Jaguar core is that much faster vs a Xenon in a core vs core comparison.
If Microsoft pulls this off, I would be very impressed. If they do it via the cloud, they could throw significantly more processing power behind it.
P.S. So why does emulation work for the gamecube (PowerPC "Gekko" at ~500Mhz) and Wii (PowerPC "Broadway" at ~750Mhz) etc? Well, the opposite is true for those consoles. They are significantly less powerful compared to the machine you're running the emulator on. Look at those specs, it should be pretty apparent why they are possible but a 360 may not be (Xenon at 3.2GHz). Even then, I probably still wouldn't want run a GC or Wii emulator on Pentium/Celeron which are faster than the Jaguars found in the xbone...
I'm a high level Visual Studio C# developer and doing multi-threading is not easy to keep track of, so I can only imagine trying to emulate at the low level these emulators work, OUCHY no thanks.
Facebook
Twitter
Instagram