How fast is MorphOS?

[Dietmar]

Believe me, MOS on Pegasos feels a lot faster than UAE on any PC, gigahertz CPU or not. Pure emulators just don't come close.

Below are the compilation times of a program (GoldED) compiled with SAS/C under UAE, UAE with JIT and MOS without JIT. Compilation is a reasonable test for raw performance. Integer performance obviously. The compiler is run with precompiled headers to take HD speed out of the equation.

UAE 0.8.17r3 is used, which is not the latest version, for obscure personal reasone: it works better with a tablet. The PC@1.4 GHz has 512 MB DDR-RAM, 40 GB Maxtor HD and 160 GB RAID; Maxtor used for test. The Pegasos@600MHz has 256 MB SD-RAM and a 40 GB Barracuda IDE.

UAE /w JIT: 1: 59 min
UAE no JIT: 11:52 min
MorphOS 1.3 no JIT: 4:19 min

Given those figures, the PC with UAE+JIT is about 2.5 times faster than MorphOS. However, the PC's CPU is also clocked 2.3 times faster. A crude guesstimate would be that a Pegasos/MOS1.3 and UAE/JIT are running emulated code equally fast on CPUs with the same clock frequency.

Now the interesting part: JIT massively contributes to the speed of UAE. JIT makes UAE compile seven times faster. If JIT for Pegasos (which will start shipping with MOS1.4?) has an effect anything like that, it is quite possible that a Pegasos at 600MHz will, with respect to emulated code, meet or exceed the performance of very fast PCs.

[Dietmar]

To put numbers above a bit into perspective, it took 3.5 hours to compile the four different versions (68000-68040) on an A3000/25

[Dietmar]

>I think that GoldED, like any other event-driven program, spends most of its time waiting for user input rather than in OS-functions.

Doesn't help much. Event-driven applications may idle along most of the time but when user input arrives, it still has to be processed quickly. In the case of GoldED, scrolling is a demanding scenario: while a user scrolls through a C source code, each word is compared against 16000 known words before it is displayed. If spellchecking is on, it is additionally compared against megabytes of dictionary content. There are other CPU-banging functions on top of that. A fast computer is not wasted on it. A G3@600 can run it reasonably fast but it's not fast enough to avoid getting friendly with gcc in the long term (to have room for more CPU-banging features

[Dietmar]

>Since the 600Mhz G3 is half the speed of UAEJIT then the G4 1GHz will be the same speed as UAEJIT?

Question makes no sense The G3 can not be "half speed of UAE JIT" because any speed difference obviously depends on the PC you throw at it, not just UAE's JIT: A G3-based Pegasos without JIT may very well be much faster than UAE/JIT. And vice versa. Faster or slower is very much a matter of Ghz. Also, please keep in mind that we are talking about how fast classic m68k code is executed, ie. "old software". This thread is not about the speed of the G3 but about the speed of the emulator. That said, some comments on speed:

I've previously posted some of my own timed tests. As far as I can tell from those, both emulators reach about the same speed on "similiar" hardware (that's of course a very loose, crude, questionable term). Compilation was used as benchmark because it primarily tests the emulation rather than the OS. The result is surprising because MorphOS does not yet have JIT. Given the effect JIT had on UAE - it made my UAE benchmarks seven times faster - a G3 with JIT should be a very fast platform for running classic software.

>When will MorphOS get JIT?

JIT has been announced for MorpHOS 1.4, making it a matter of days, I guess. As to the original question, I don't know what effect G4 will have on the emulator. 400 MHz isn't a massive difference and the CPU is only one of many components.

[Dietmar]

> I must say that what you wrote down surprised me: a 600 MHz CPU not yet fast enough for an editor...?

There are editors and there are editors, it's all a matter of features. Old Amiga editors obviously were designed for the hardware at that time. They run fine with 8 MHz. Hardware gets better, features get better, CPU takes some hits. Compared to good developer-quality PC editors like Source Insight, GoldED is harmless in that regard: those editors compile code in the background (incremental compilation), analyze and tag source codes, build class trees and do a lot of other amazing things.

Btw, I didn't write that a G3/600 is not yet fast enough but that it runs GoldED (ie. emulated software) reasonably fast and that more feature would require a gcc port, ie. a native version.

[Dietmar]

>Sorry, what I meant was that Morphos without JIT is equal in speed to UAE with JIT on a MHz for Mhz basis

Obviously not, MorphOS is faster because the OS runs directly on the CPU and does not need an emulator. UAE needs an emulator to run the OS. Big difference. But I guess/hope what you really mean is how the speed of emulators compare, ie. the speed at which "old" software is executed? If you mean the emulator in MorphOS, yes, in my experience, with my own PC, a crude MHz comparison vs a typical PCs (of the last year or two) seems to be a reasonable rule of thumb. While waiting for JIT, that is.

[Dietmar]

Below are the compilation times of a program (GoldED) compiled with SAS/C under UAE, UAE with JIT and MOS without JIT. Compilation is a reasonable test for raw performance. Integer performance obviously. The compiler is run with precompiled headers to take HD speed out of the equation.

[...] The PC@1.4 GHz has 512 MB DDR-RAM, 40 GB Maxtor HD and 160 GB RAID; Maxtor used for test. The Pegasos@600MHz has 256 MB SD-RAM and a 40 GB Barracuda IDE.

UAE /w JIT: 1: 59 min
UAE no JIT: 11:52 min
MorphOS 1.3 no JIT: 4:19 min

With MorphOS 1.4 released today, I'll use the opportunity to update those figures:

UAE /w JIT: 1: 59 min
UAE no JIT: 11:52 min
MorphOS 1.3 no JIT: 4:19 min
MorphOS 1.4 /w JIT: 1:05 min

MorphOS 1.4 is almost four times faster than the previous version in this test (which is a test about running CPU-intensive classic software, not about PPC software). It outperforms my 1.4 GHz UAE PC by a factor of two.