16-Dec-2018 04:08 GMT.
UNDER CONSTRUCTION
Anonymous, there are 43 items in your selection
[News] The long time awaited Petunia test programs are downloadable now!ANN.lu
Posted on 14-May-2002 08:40 GMT by Chip43 comments
View flat
View list
Download the tests programs here, and compare your results with the petunia's one. What do you think? Write your results here and don't forget: Petunia will be much more faster in a context switch free enviroment (like AmigaOS4).
The long time awaited Petunia test programs are downloadable now! : Comment 1 of 43ANN.lu
Posted by hmm on 14-May-2002 10:49 GMT
how does it work then currently, if not the WHOLE system is emulated (eliminating the need for context switches)? Just curious ..
The long time awaited Petunia test programs are downloadable now! : Comment 2 of 43ANN.lu
Posted by Alkis Tsapanidis on 14-May-2002 10:51 GMT
In reply to Comment 1 (hmm):
It currently runs under WarpOS
The long time awaited Petunia test programs are downloadable now! : Comment 3 of 43ANN.lu
Posted by Eric Chernoff on 14-May-2002 12:18 GMT
I know this can be counted as flamebait, but FYI:
I ran these test progs on my Amithlon 1.4gz system, and I got the following results:
c2ptest - 1 second
chunkycopy - <1 second
mandel - 1 second
demoeffect - 399.6 fps
Voxeleffect wasn't included in the LHA archive.
The long time awaited Petunia test programs are downloadable now! : Comment 4 of 43ANN.lu
Posted by Graham on 14-May-2002 12:29 GMT
In reply to Comment 3 (Eric Chernoff):
Interesting...
Probably on par with a 600MHz G3 based AmigaOne when it comes out then, if Petunia is able to emulate a 68060 at around 300MHz speeds.
The long time awaited Petunia test programs are downloadable now! : Comment 5 of 43ANN.lu
Posted by Don Cox on 14-May-2002 12:47 GMT
In reply to Comment 3 (Eric Chernoff):
My results:
Demoeffect - 448.87 fps
Mandel <1 sec (probably around 3/4 sec)
ChunkyCopy - <1 sec
c2ptest - 3 secs
Presumably the last one depends partly on the graphics card used.
The long time awaited Petunia test programs are downloadable now! : Comment 6 of 43ANN.lu
Posted by Ole-Egil Hvitmyren on 14-May-2002 13:09 GMT
In reply to Comment 4 (Graham):
Very much so. A factor 4 in speed from 2.6 times higher clock, no WarpUP context switches _and_ the addition of a full speed L2 cache is not impossible at all. _That'll_ be fun, when our 600MHz G3 outperforms the Amithlon beasts on m68k emulation. Of course, then the x86-fanatics will just go out and spend their dough on faster x86'es to prove their point, instead of buying Amiga's... I fear this is a fight we cannot win ;)
The long time awaited Petunia test programs are downloadable now! : Comment 7 of 43ANN.lu
Posted by twicer on 14-May-2002 13:23 GMT
In reply to Comment 5 (Don Cox):
what's your setup like?
The long time awaited Petunia test programs are downloadable now! : Comment 8 of 43ANN.lu
Posted by Eric Chernoff on 14-May-2002 13:24 GMT
In reply to Comment 4 (Graham):
Well, there's no way I *only* have a 300Mhz 68040 on this Amithlon machine. It's much more like a 1Gz 68040.
I say this based on watching it compile *nix programs and executing filters in ArtEffect.
I say this as someone who *really* wants an AmigaOne (mainly OS4), but there's no way that the AmigaOne is going to be faster than the Amithlon setup I have here.
The long time awaited Petunia test programs are downloadable now! : Comment 9 of 43ANN.lu
Posted by Robert on 14-May-2002 13:39 GMT
So is this program ready to go?
I mean: Is this part of OS4 complete?
Robert
The long time awaited Petunia test programs are downloadable now! : Comment 10 of 43ANN.lu
Posted by Don Cox on 14-May-2002 13:41 GMT
In reply to Comment 7 (twicer):
AthlonXP 17, 512 Megs RAM, 40Gig drive, SCSI CD writer. It replaces an
A2000, I'm still using the A4000.
ISA Catweasel.
CD writing with MasterISO is much better than on the A2000.
No other OS on this machine. There is a WinNT machine on the networtk,
but it doesn't get much use.
The long time awaited Petunia test programs are downloadable now! : Comment 11 of 43ANN.lu
Posted by Ole-Egil Hvitmyren on 14-May-2002 13:41 GMT
In reply to Comment 8 (Eric Chernoff):
Except a 300MHz 68040 isn't even close to a 300MHz 68060, which is what we were talking about. Stop comparing apples and oranges! :)
I only have a 40MHz 68040 myself, my guess would be that's half the CPU a 50MHz 68060 is...
If you compare the numbers from the 233MHz 604 with the 1.4GHz Amithlon setup someone mentioned, you'll see that it's a 5-fold increase in speed and a 5-fold increase in performance. Now add the fact that the A1 has a 600MHz 750 (possibly effectively a 3 times faster PPC), it does not have the context switch problems he get's under WarpOS, and it has an L2 cache. I think that 1.4GHz Amithlon is gonna get asswhooped by the A1. Do you?
I also run Amithlon from time to time, but have had too many projects hanging over my head to have as much fun with it as I wanted. I agree it feels like a pretty fast Amiga, but it's like 500 channels and nothing on. I ended up cracking RC5 :)
I had no end of problems with sound and net as well...
The long time awaited Petunia test programs are downloadable now! : Comment 12 of 43ANN.lu
Posted by Ole-Egil Hvitmyren on 14-May-2002 13:42 GMT
In reply to Comment 9 (Robert):
It's still lacking FPU and MMU, according to the webpage. Who needs FPU anyway, eh? ;)
The long time awaited Petunia test programs are downloadable now! : Comment 13 of 43ANN.lu
Posted by Don Cox on 14-May-2002 13:54 GMT
In reply to Comment 11 (Ole-Egil Hvitmyren):
"500 channels and nothing on"
Why haven't you installed your Amiga software? There are loads of
programs that run well - Pagestream, Samplitude, Audio Evolution,
Image Engineer, Transwrite, DOpus4, SoundFX, Digital Almanac, World
Construction Set, Imagine, MasterISO, ADPro, TVPaint, Personal Paint,
etc etc etdc. Even Voyager works better.
Network working fine here - I used the card from an old Linux machine.
The PCI-128 isn't a good sound card, but it does work reliably here.
I access Internet via the A4000 which is a gateway (MiamiDx). Works
well, but I may set up aLuinux firewall-router.
The big frustration is not being able to use the AGA paint programs.
We need a good UAE with full AGA emulation.
The long time awaited Petunia test programs are downloadable now! : Comment 14 of 43ANN.lu
Posted by Anonymous on 14-May-2002 14:01 GMT
In reply to Comment 2 (Alkis Tsapanidis):
sorry, but how do you want to do this? You just can't emulate an 68K without replacing it completly if you run it in the same environment .. just think of such basic things as memory allocations!! AFAIK the PPC has its own functions and normal Exec functions can't (they can but shouldn't) be used due to alignment issues, cache flushing and such .. how do you keep memory-contents consistent? how do you want to seperate between a single program and the library functions it calls and why should they then run on PPC instead of the builtin 68K? all these points have to be addressed and make it, in my eyes, impossible to create such a thing as you're suggesting. Please give answer.
The long time awaited Petunia test programs are downloadable now! : Comment 15 of 43ANN.lu
Posted by Ole-Egil Hvitmyren on 14-May-2002 14:30 GMT
In reply to Comment 14 (Anonymous):
Uh? How is emulating a 68k any different to any other program running under WarpOS? Really? When a program under emulation calls OS routines running on the real m68k, you get a emulated-m68k-on-ppc-to-real-m68k switch, which takes a far bit longer than just skipping out to another program on the same CPU. I don't see any reason why this wouldn't work. Do you? Really? Care to explain your view?
The long time awaited Petunia test programs are downloadable now! : Comment 16 of 43ANN.lu
Posted by Ole-Egil Hvitmyren on 14-May-2002 14:39 GMT
In reply to Comment 13 (Don Cox):
Well, I never got _anywhere_ with sound (double clicking on ahi-prefs crashed the emulator, same with clicking on "test sound" in sound-prefs). onboard ac97 sound. The same thing happens when I tried to use the NE2000 PCI board I had... Which is now serving life in my A1200 instead. I'm sure there are ways to get things to work, but when real-life suddenly sneaks up on me and someone demands results by the end of May 2002, I tend to drop those things that give me splinters of wood planted in my forehead first. It might just be lazyness, bad luck, incompetence or whatever, but that's how not-so-far I got when playing with Amithlon.
And back to the list you presented. If you had no sound and net, would the list be that impressive? ;)
Come over here and fix my Amithlon for me, and I'll gladly use it. But I don't think I have enough time for now to fix it. And it's not even my PC, so when I move in 3 weeks time it'll be all for nothing. Too bad, really.
The long time awaited Petunia test programs are downloadable now! : Comment 17 of 43ANN.lu
Posted by Bernd Meyer on 14-May-2002 14:47 GMT
In reply to Comment 14 (Anonymous):
As far as I can guess (and you can imagine I am following this closely :), Petunia under WarpOS currently acts as a kind of shell or program loader. Once the 68k program is loaded, V68k execution starts. Whenever a library call is made through JSR xxx(A6), a context switch to the real 68k is executed, the corresponding function is called by the real 68k, then the results are transferred back to the V68k and V68k execution continues after the JSR instruction.
On the plus side, this allows running 68k code fragments (or small benchmark programs) in an emulator that isn't completely finished yet --- only the small parts of 68k code explicitly loaded by Petunia get JIT-compiled; The OS and the whole environment are still running on the real 68k. So without having a fully debugged JIT compiler that fully covers the whole 68k instruction set, you can still run things, look for weird effects, and debug. That's invaluable, believe me (in UAE-JIT, the equivalent is falling back on the UAE interpretative emulation).
On the flip side, this means that every library call will incur a PPC->68k->PPC context switch. Obviously, that isn't too bad, seeing as the whole project is aimed for a PPC-only system, anyway, and thus this is only a stopgap measure.
Worse, though, it means there are only two ways of entering V68k mode... Either by starting the 68k program in the emulator shell, or by returning from a real-68k library call into the V68k environment. What appears to be missing so far is a way for hooks, callbacks, or function pointers to work. If the V68k passes on a function pointer to an exec library function (for example to be called when an interrupt occurs), and exec (running on the real 68k) later calls that function, there seems to be no mechanism for that call to end up inside the V68k environment. Which, considering the kind of problem it could create, is something I am sure Mr. Rajnai has worked out a way to handle quite some time ago. It just appears he hasn't implemented it yet (and once again, from painful experience with respect to 68k->x86->68k transitions, I can attest that this sort of thing is hard to get right).
Mind you, all this is my speculation based solely on the information on the project web page, and my experience with JIT 68k emulation.
On a related note, I would like to recommend caution with respect to the selection of benchmark programs that are being used for the JIT compilers both in MorphOS and Petunia. In particular, the "demoeffect" program seems to be very popular.... For a good reason. Its main loop (which does all the perspective-mapping) consists of a block of 32 unbroken instructions, with only one of them using flags, which were created in the immediately preceding instruction. This is almost as good as the RC5 client (MorphOS's other public benchmark). Also, both programs only use full 32 bit arithmetic in their main loops (with the exception of one 16by16-to-32 multiply in demoeffect), which is what JIT compilers have few problems with.
However, such main loops are rare. Extremely rare. The size of the average executed code segment between two control instructions (such as branches, JSR etc) is somewhere between 3 and 6 instructions for typical programs. Control flow instructions are hard to handle in a JIT compiler. As are not-full-sized arithmetic instructions. So while the published results are certainly very interesting, I would advise against over-estimating their representativity with respect to real Amiga applications.
The long time awaited Petunia test programs are downloadable now! : Comment 18 of 43ANN.lu
Posted by Bernd Meyer on 14-May-2002 14:58 GMT
An interesting aside: To have a fair comparison, I went and set up Amithlon on a contemporary to the CSPPC. It's a PPro-200 board, using 72 pin EDO SIMMs and a Matrox Millennium for gfx (board manufactured 1/97, gfx card 26/2/97, with its add-on ram hailing from August 96). I bought these components sometime in '98, because they were dirt cheap. But this is the sort of machine the CSPPC would have competed with when it was first released. Results:
Mandel: 3.26s
chunky: 6.38s
c2p: 8.28s
demo: 65.5fps
voxel: 43-53fps
The long time awaited Petunia test programs are downloadable now! : Comment 19 of 43ANN.lu
Posted by Bernd Meyer on 14-May-2002 15:03 GMT
BTW --- you can use "timecmd" from aminet to time the execution of c2ptest, chunkycopy and mandel. Gives more precise results than a stopwatch....
The long time awaited Petunia test programs are downloadable now! : Comment 20 of 43ANN.lu
Posted by Kronos on 14-May-2002 15:19 GMT
In reply to Comment 17 (Bernd Meyer):
>...is almost as good as the RC5 client (MorphOS's other public benchmark).
In April the c't (german IT-magazine) tested Amithlon
and it reached about 50% of the x86-native performance
on RC5. It was running on an outdated Athlon800 with a
GFX-Card in VESA-mode (ATI Rage Pro), and the Win-Client
was probraly compiled with a much better compiler than
GCC-68k.
So anyone who expect 060@300mhz performance on a G3@600mhz
in real world tasks should recalculate his predictions.
The long time awaited Petunia test programs are downloadable now! : Comment 21 of 43ANN.lu
Posted by Bernd Meyer on 14-May-2002 15:39 GMT
In reply to Comment 20 (Kronos):
To be pedantic --- neither the 68k nor the x86 RC5 client cores are compiled from C code; Both are (collections of) lovingly handcrafted assembler code written specifically to match the capabilities of the respective CPUs. It's a heck of a lot of work to get those last few percents (been there, done that --- the linux/Alpha client has my code in it :), but both 68k and x86 camps have very good coders doing the job.
Of course, JIT compilers were never the target of the 68k coders, and by some coincidence, a slightly older client (with dedicated 68040 core) will do significantly better than the current (combined 040 and 060 core) one.
However, as I mentioned before, RC5 calculations are really not particularly interesting benchmarks. The main loops are very very long, and use only fairly-easily JIT-compiled instructions, with practically no need whatsoever for flags generation.
Which means JIT-compiling RC5 cores avoids all the interesting and difficult bottlenecks, and any halfway decent JIT compiler will do just fine on it. The Bytemark test suite ("bytemark68kPPC.lha" on aminet) or a quick demo-run of doom will test those aspects much much better....
The long time awaited Petunia test programs are downloadable now! : Comment 22 of 43ANN.lu
Posted by Ben Hermans/Hyperion on 14-May-2002 15:51 GMT
In reply to Comment 20 (Kronos):
What people do seem to forget is this:
As Bernie Meyer explained, Petunia is running in a very "performance hostile" way right now.
The following issues slow down Petunia considerably:
- the context-switches (which involve a cache-flush of each CPU)
- the absence of L2 cache
- the fact that the 68K when active steals a considerable amount of cycles from the PPC
If you factor in all these issues, I'm confident that a G3@600 Mhz may be able to reach 300 Mhz/060 speed.
It will however be very difficult to measure this in real life in an unbiased way as large parts of the OS will also be PPC native such as Intuition, layers, input.device, the RTG system etc.
68K programs will run with these native modules without being aware of it whilst on Amithlon you would still be emulating those OS functions.
In the end, benchmarks are all artificial to some extent or another.
The long time awaited Petunia test programs are downloadable now! : Comment 23 of 43ANN.lu
Posted by Ole-Egil Hvitmyren on 14-May-2002 15:52 GMT
In reply to Comment 17 (Bernd Meyer):
You trying to shut up a good argument with facts or something? :)
We don't want that! We want flamewars!!! Uhm, no we don't. Who said that? *peering around his office"
Hmm.
The long time awaited Petunia test programs are downloadable now! : Comment 24 of 43ANN.lu
Posted by Kronos on 14-May-2002 15:59 GMT
In reply to Comment 22 (Ben Hermans/Hyperion):
>68K programs will run with these native modules without being aware
Just like any SW running on Amithlon with powerfb.elf ;)
(GFX-Routines are probraly where you can expect the most gains)
Emulation-speed isn't linear to host-speed so I wouldn't make
any predictions before I did some tests on the actual HW.
The long time awaited Petunia test programs are downloadable now! : Comment 25 of 43ANN.lu
Posted by Bernd Meyer on 14-May-2002 16:31 GMT
In reply to Comment 22 (Ben Hermans/Hyperion):
Of course, it should be mentioned that of the programs used to benchmark petunia, only one (Voxelspace) does any significant amount of JSR xxx(A6) calls (and thus context switches).
So while context switches are certainly going to hurt Petunia if they happen, they don't seem to happen in the time-critical loops of 4 out of the 5 benchmarks.
Of course, 68k cycle stealing still goes on, and the lack of 2nd level cache is probably hurting Petunia quite badly. On the other hand, when the G3-600 in the AOne *does* miss its cache, the resulting latency, in terms of clock cycles, is probably longer than on the 604e-200.
So all in all, speculation as to the performance of Petunia on an AOne with a G3-600 under OS 4.0 on typical Amiga applications seems kinda pointless as long as the AOne has yet to materialize, Petunia hasn't been integrated into OS 4.0, and no benchmarks for real world applications are known. It's kinda like looking at my 1975 Beetle (yay!) and, based on that, speculating on the milage and top speed of next year's Porsche model.....
The long time awaited Petunia test programs are downloadable now! : Comment 26 of 43ANN.lu
Posted by Don Cox on 14-May-2002 16:53 GMT
In reply to Comment 25 (Bernd Meyer):
Still, we can be reasonably sure that Petunia running on a 600MHz CPU
will not give the effect of a 68060 running at more than 600MHz.
The long time awaited Petunia test programs are downloadable now! : Comment 27 of 43ANN.lu
Posted by darklite on 14-May-2002 16:54 GMT
In reply to Comment 11 (Ole-Egil Hvitmyren):
>If you compare the numbers from the 233MHz 604 with the 1.4GHz Amithlon setup
>someone mentioned, you'll see that it's a 5-fold increase in speed and a 5-
>fold increase in performance. Now add the fact that the A1 has a 600MHz 750
>(possibly effectively a 3 times faster PPC), it does not have the context
>switch problems he get's under WarpOS, and it has an L2 cache. I think that
>1.4GHz Amithlon is gonna get asswhooped by the A1. Do you?
I don't. Why not lets stop making these silly guesses and wait (sigh) till we can actually run some benchmarks under OS4 running on the A1?
The long time awaited Petunia test programs are downloadable now! : Comment 28 of 43ANN.lu
Posted by Ben Hermans/Hyperion on 14-May-2002 17:51 GMT
In reply to Comment 25 (Bernd Meyer):
Yes, it's all extrapolation ofcourse and certainly nothing better than guestimates.
In fact, our emulation concept for OS 4 extends far beyond just Petunia which is why I was bold enough to make the claim of a 300 Mhz 050 on a 600 Mhz G3 in the first place.
But like you said, time will tell.
I certainly don't advocate starting a "benchmark war" before the AmigaOne and OS 4 have actually materialised.
The long time awaited Petunia test programs are downloadable now! : Comment 29 of 43ANN.lu
Posted by cOrpse on 14-May-2002 18:13 GMT
In reply to Comment 27 (darklite):
Whats wrong with guesses , and its pretty easy to guess that this emu is gonna be SMOK'IN , its a beast on the current crap ppc cards and cripled by its twin cpu nature , so i reckon we can expect alot more emu performance per mhz when we benchmark a current ppc and an AmigaONE :). The current guess kinda rely on the performance just be multiplied by the mhz and not the other factors ( such as memory ;) ).
Or are people just getting pissed cuz a 600mhz aOne is gonna make uae (x86 of course) look like a speccy emu hehe.
/me can't wait !! , didn't need to get an 060 blizzardppc after all :D
/me also thanks god he got the highest ppc speed available ( for the blizzard nehows )
ROCK'IN !.
The long time awaited Petunia test programs are downloadable now! : Comment 30 of 43ANN.lu
Posted by cOrpse on 14-May-2002 18:18 GMT
In reply to Comment 28 (Ben Hermans/Hyperion):
It does show that even a shoddy ( current ppc cards ) ppcs can keep up and give the goods where us amiga users need it , and thats all that matters :D
The long time awaited Petunia test programs are downloadable now! : Comment 31 of 43ANN.lu
Posted by Alkemyst on 14-May-2002 19:42 GMT
most benchmark programs if not all are not ment to measure emulation as how they go about getting the info is ment for real HW.
a emulated 600Mhz Amiga will never be as fast as a real 600Mhz amiga.
take sysinfo, useing this on emu is totally silly as it does not measure sound or gfx.
& even if it did it would be fauls.
so with sysifno you get a 600Mhz 040, but if thats all that you wanted emulate then fine but bring sound & gfx into it then the x86 or ppc would not be able use all its power to give you 600Mhz 040 anymore, as your emulating a whole amiga not just the cpu.
but then some one uses another bench prog for gfx & goes oh it shows AGA gfx emu at 600Mhz gfx AGA chip, wich is wrong again as the x86 or ppc used all it had just to emu the gfx & the same would go for paula sound emu
in real world test you should be testing cpu,gfx & sound emu all at the same time & not one after the other.
keeping it simple you you could end up with 200Mhz 040 200Mhz AGA gfx 200Mhz paula emu, but we know that it could never panout like that.
The long time awaited Petunia test programs are downloadable now! : Comment 32 of 43ANN.lu
Posted by Graham on 14-May-2002 19:50 GMT
In reply to Comment 31 (Alkemyst):
It is made all the more simple by the fact that there will be no effort to emulate any of the custom chips.<p>
Also, to be honest, a real 600MHz 68060 would not have been 10x faster than a 60MHz 68060, because you have to take the underlying platform into account, and that doesn't scale as well. But when we talk about 300MHz 68060 equivalent speeds, we are ignoring that to all extents and purposes. Did that make sense?
Simply put, 68k Imagine should run around 6 times faster under Petunia on a 600MHz G3 on the AmigaOne platform than a 50MHz 68060 based Amiga system. In theory. Luckily we will be able to test the theory when the AmigaOne+OS4 comes out. The results will be very interesting. Of course there will be variances - some apps might only show 3x improvement whereas others might show 10x improvement - it all depends on the app and the resources it uses...
The long time awaited Petunia test programs are downloadable now! : Comment 33 of 43ANN.lu
Posted by Alkemyst on 14-May-2002 20:07 GMT
In reply to Comment 32 (Graham):
>>It is made all the more simple by the fact that there will be no effort to emulate any of the custom chips.<p>
yes i was going to finsh off with something like that, but could not be bothered.
as my spelling & writing grammar & puntuaition is bad & its a pain as alot of the time i can say thing as i realy want to,
so i dont bother posting at all most times. :)
The long time awaited Petunia test programs are downloadable now! : Comment 34 of 43ANN.lu
Posted by Alkemyst on 14-May-2002 20:19 GMT
In reply to Comment 32 (Graham):
Also, to be honest, a real 600MHz 68060 would not have been 10x faster than a 60MHz 68060, because you have to take the underlying platform into account
yes true busses & stuff.
i have a 060 at 80Mhz & on most tasks i get good speeds over a 060/50 or 060/66 but on a few tasks i get
very little noticeable difference
The long time awaited Petunia test programs are downloadable now! : Comment 35 of 43ANN.lu
Posted by Arild Kvam on 14-May-2002 20:50 GMT
In reply to Comment 34 (Alkemyst):
How did You get the Fast-mem to run at 80Mhz? Any sort of HW trickery?
I have a original 50Mhz 060 overclocked to 75Mhz, as is the 16MB EDO (60NS)
But i know from the Amiga's respone that the EDO memory is pushed to it's
absolute limit.
The long time awaited Petunia test programs are downloadable now! : Comment 36 of 43ANN.lu
Posted by Don Cox on 14-May-2002 20:50 GMT
In reply to Comment 32 (Graham):
The underlying platform is also much faster than in a classic Amiga.
Your guess of around 6x the speed of a 50MHz 060 is probably not far
off. I get around 10x on this Athlon 17 (a CPU of rather uncertain
speed). The faster RAM access on a moderm motherboard must help.
The long time awaited Petunia test programs are downloadable now! : Comment 37 of 43ANN.lu
Posted by Alkemyst on 14-May-2002 21:13 GMT
In reply to Comment 35 (Arild Kvam):
PowerTower A1200,060/80Mhz,Heatsink&Fan,66MBRam,PowerFlyerGold,50xCDRomdrive,250Zip,2.1GB&34GB HD,internal Scandoubler & FF,19"Monitor,Mediator,Voodoo3-3000,PaceSolo 56k ,PortJnr2,ZEKeys-XS,SMON ,Os3.9
Apollo 060/66MHz FPU & MMU working just fine clocked to 80Mhz 60ns EDO & very nice indeed.
simple crystal change
The long time awaited Petunia test programs are downloadable now! : Comment 38 of 43ANN.lu
Posted by Álmos Rajnai on 15-May-2002 04:51 GMT
In reply to Comment 17 (Bernd Meyer):
Mr. Bernd Meyer were right in his long explanation. (Of course, he did the job before... :) Demoeffect test can be optimized quite well, like the other codes, which have an unbreaked main loop (like: image/sound processors, some of the games, demos). But most applications spending its time with calling the system functions, therefore the optimalization there is not as effective as it is in huge loops.
But... Calling the system will speed up a lot, because of AOS4 is rewritten for PowerPC, and not only emulated.
It is pointless to get far with speculations about the speed of the final release in AOS4. There are quite a lot of aspects, which will have (a good or bad) influence on the troughput. I am afraid, nobody (including me, I mean it) can forecast the final speed. All I can promise to you all, that I do my best to get it working pretty good.
Why I haven't published benchmarks about more complex programs? Because only a very few is working without problems, there are a sort of issues: backcalling, interrupts, badly written multitasking between several tasks, etc. I don't want to solve problems, which are residing to only the actual dual processor architecture. Later all of them will disappear together with the 68k processor.
I published all the benchmarks to inform you about the magnitude of the speed of the ACTUAL form of Petunia, and not about the final speed. I tought it was clearly written on the page, but it seems it was not clear enough.
The long time awaited Petunia test programs are downloadable now! : Comment 39 of 43ANN.lu
Posted by Ole-Egil Hvitmyren on 15-May-2002 05:53 GMT
In reply to Comment 38 (Álmos Rajnai):
There, there. Be a nice little coder and finish Petunia. We big bad trolls won't listen to reason anyways, so it's best not trying to tell us :)
The long time awaited Petunia test programs are downloadable now! : Comment 40 of 43ANN.lu
Posted by Lennart Fridén on 15-May-2002 09:07 GMT
@ Bernd Meyer
Very interesting, thanks for sharing.
The long time awaited Petunia test programs are downloadable now! : Comment 41 of 43ANN.lu
Posted by Anonymous on 15-May-2002 16:42 GMT
In reply to Comment 26 (Don Cox):
>Still, we can be reasonably sure that Petunia running on a 600MHz CPU
>will not give the effect of a 68060 running at more than 600MHz.
How can you be so sure? The Audiolabs DSPtest shows Amithlon running at
some 1900MHz (sic!) 060 speed - http://www.audiolabs.it/dspspeed/index.html
The long time awaited Petunia test programs are downloadable now! : Comment 42 of 43ANN.lu
Posted by Anonymous on 15-May-2002 16:46 GMT
In reply to Comment 29 (cOrpse):
>its a beast on the current crap ppc cards and cripled by its twin cpu nature
The PPC cards are neither crap nor crippled. Try MOS with JIT and you will see.
Hell, even APUS Linux puts the 68k to sleep afaik, so you should have no problems
with the 68k stealing cycles.
The long time awaited Petunia test programs are downloadable now! : Comment 43 of 43ANN.lu
Posted by priest on 17-May-2002 15:40 GMT
More benchmark options for the future:
http://www.illuvatar.demon.co.uk/lambda/info/LInfo.html
Anonymous, there are 43 items in your selection
Back to Top