PC still playing Amiga catchup

[Karlos]

Why not try it out on a REAL PC with standard hardware rather than specialized graphics cards.

First of all, it is a "real PC". The graphics card is no more out of place in this machine than any of the custom chips in my "real Amigas". This simulation runs on any PC that has a G80 or above based graphics card.

Secondly, there's nothing "specialised" whatsoever about the card. All current generation graphics cards tend to have fully programmable parallel arithmetic units.

Thirdly, I have run the same simulation entirely on the CPU. A naive single threaded implementation in the same "real PC" is about 250x slower than the GPU version on this machine. I can boost it significantly by optimizing it for four core execution and even further by using specific SSE3 vector operations. However, that naive version would still be 400x faster than my humble 040 could manage in a perfect world where memory bandwidth was infinite and all operations took one cycle. In the actual real world, it couldn't run it at all, there's not enough memory available to even hold the state information for this simulation.

Criticising the use of the GPU is also shooting your own argument for the Amiga in the foot. So far, you've extolled the virtue of using the custom chips for "realtime" performance, such as polling the joyport. If our experiment were simply filling flatshaded polygons on a 68000 amiga, would you be advocating the use of the CPU and not some "specialised graphics processor" ? Of course not.

Incidentally, do you really think a 2.6GHz CPU with 12MB of cache, even running the least optimized code in existence, isn't capable of polling a piece of hardware at 1kHz without missing a single iteration?

[Karlos]

Depends on the task. If your real-time task only involves modifying some registers, you can use the Copper and it's guaranteed with accuracy of 558ns (no +/- latency bullcrap).

Also, the argument was about "real-time OS", not what constitutes a real-time task in general. AmigaOS and Linux are in no way, shape or form "real-time" OS. You should probably look up the definition of what constitutes a real-time OS if you think differently.

A real-time OS guarantees that an event (an interrupt ot whatever) shall be dealt within a specified minimum time limit. A failure to do so is considered a complete failure of the OS.

No commonly used desktop OS makes this guarantee. You'll only see it in embedded hardware and mission critical systems.

[Karlos]

All I have to do is show one example, where Amiga wins out and your claims above are FALSE and the topic of this thread is valid-- PC still playing catchup.

Then please do. I'm genuinely intrigued as to what this one example could be. And once you deliver it, I suppose you could look at it the other way around. I would suggest the amiga has been playing catchup to the PC for some time now. Or at least it would be, if it was even in the race still. Which it isn't and it hasn't been since the last hardware amiga rolled off the production line.

There's no greater demonstration of how far behind the amiga is in hardware terms than a quick glance at the hardware upgrades that are available for it. Most of the heavily expanded classic amigas out there are using some form of (now totally obsolete) commodity PC hardware to upgrade their native kit. I have two A1200 towers, one has a BVision, the other has a voodoo 3000. Both have nice displays that are significantly better than AGA could manage for productivity work, yet both are using utterly ancient PC hardware to deliver them. Apart from retrogaming and demos, I can't use AGA at all. Which is fine, because I don't ask anything more from it.

You are probably sitting there, p!ssed at me, thinking I'm some PC fanboy and that I hate the Amiga etc. The fact is, you couldn't be more wrong. I still own two towered 1200's, both pretty heavily expanded, two desktop 1200s, modestly expanded, a presently dead (but hoping to revive) A600 and a "next generation" Amiga, namely the A1XE. I get plenty of enjoyment out of them but I had to accept reality. Aside from some of the really cool FPGA stuff, the classic line is never coming back. The "next gen" line is interesting but I don't think there's any niche it can fulfil, outside of entertaining we few enthusiasts. And at the end of the day, that's what the Amiga is today. An enthusiasts platform. It's yestyear's classic car; beautifully designed and engineered, a joy to see still out on the road, but easily outclassed at a raw technical level by modern machines.

[Karlos]

I'll give you THREE:

1. Boot up time
2. Shut down time
3. Application launch time  (and don't give me the spiel about how much bigger modern apps are because they do so much more so ofcourse they' take longer to load, thats all negated by the fact that you're running those apps on hardware specs that are many many factors faster clock speed and higher capacity ram and bussess, than the hardware used to launch Amiga apps.)

In what sense is (1) and (2) remotely relevant unless you spend all your time rebooting? Bootup times on most machines are limited by having to wait for things. Network connections, hardware initialisation etc. All systems wait at the same speed. My current PC boots into linux in about 25 seconds, during which time it starts several server processes. It's true that my A1 can A1200 can boot faster from a warm reset, but then, I never have to reboot the linux box unless I consciously turn it off or intend to upgrade the kernel. My A1200 waits for 10 seconds just to see if anything is attached to the BPPC SCSI. It never boots in less than 20 seconds, even warm. There are additional delays whilst AmiTCP fires up and talks to the router.

3) Complete and utter rubbish, I am afraid. I can launch applications that take up more memory on this PC than I have fitted in all my amigas put together and they open faster than most applications on my actual amiga. For example, Firefox 3 is reportedly using 350MB of memory at the moment. It took about 2 seconds to start.

What's more, like 1 and 2, it's a contrived measure of user experience. I don't know about you but I tend to start applications once and then carry on using them for as long as needed. I don't sit there opening and closing them repeatedly and marvelling at the speed.

My ZXSpectrum boots from cold in about 1 second. It's obviously the best of the lot, by your reckoning.

[Karlos]

Application (game) launch times generally weren't very good on the ZX series though... (cassette tapes).

I had the expansion interface that had the rom cartridge slot. They launched pretty damn quick.

To this day I have no idea what happened to that expansion.

[Karlos]

well you must be the only person on the planet that sees no value in fast booting and shut down

Wow, really? Sh!tfire, I must be a total freak if I think waiting 20 odd seconds either way is fast enough for something I do at most once a day. My work machine takes a whopping minute to start on a slightly less powerful PC. However, it's only been rebooted about 6 times in 3 years, 3 of those times were due to office moves.

I get on the net on my A1200 68060 far more quickly than my PC. I have the google home page up faster on ibrowse than i do with Firefox

No way! Rock on. Guess what? I don't actually have to connect. I just start a browser (this is also true on my A1200 and A1, both of which are connected to the same router). I just opened firefox and it was at my homepage instantly. I really can't shave any time off.

I think maybe lynx is slightly faster.

BTW you only needed ONE, you've got three..

Three what? Totally limp wristed non-arguments I've heard dozens of times that have never gotten any more meaningful no matter how many times repeated? If you ask any sane person if they have to choose between slightly faster boot times or the actual need to reboot in the first place, I wonder which they'd choose?

[Karlos]

i can assure you its not 20 seconds, its no less than 2 minutes, often 2 and half. Well you ARE a freak if you boot up once per year, excluding office moves.

Assure me? It's my system. Would you like me to time it for you? It isn't as if I don't have any better use for a computer than rebooting it.

[Karlos]

BS.  it takes time to start Firefox, and that time is longer than Ibrowse.  you're not at you home page "instantly".

Yes, I am a dirty liar:

karlos@Megaburken-II:~/Desktop$ time firefox

real   0m0.138s
user   0m0.032s
sys   0m0.012s

That's very nearly 0.14 seconds OMGZ!

[Karlos]

No, I really thought the amiga had fixed time slicing.

A nice read:rtos

Well, AmigaOS generally tasks run for at least an exec quantum of time before being preemptively switched out. However, as Trev says, all kinds of other things can happen in unpredictable ways.

AmigaOS is most assuredly not a RTOS, however it is so damned efficient that for the most part it behaves as if it were one. Right up until it hits heavy CPU load.

[Karlos]

@Karlos

http://www.techpowerup.com/95445/ASUS_Designes_Own_Monster_Dual-GTX_285_4_GB_Graphics_Card.html

Holy crap :laughing:

I'm waiting to see what the G300 series does, though. Let's face it, fast as they are, there are lots of improvements that could be made to the arithmetic units. Double precision performance is still dire (compared to single, still faster than your CPU).

[Karlos]

Thanks for clarifying.  You are just using some souped up graphics card.  Nothing to do with PC running 100s of processes.  And you forgot to answer the rest of the message.

And you still violate the law of conservation even with quad core since there's still some overhead involved.

Wrong. There were > 100 CPU processes, not GPU processes. There was precisely one CPU process for that simulation. "ps aux wwwf | less" is hardly going to report on GPU processes, now is it? :roflmao:

Four of those CPU processes are capable of being in the run state concurrently provided they aren't making atomic operations on the same area of memory. Which is highly unlikely because this is a virtual memory system. Most of the processes have no idea the others exist, let alone have the opportunity to lock something they own.

Just for you, I've dumped the current process list. Too big to post here, so here it is: process.txt

I make that 171 processes at the moment. You should see it when it is actually busy.

-edit-

Oh, and as for the rest of your objection:

I doubt even more that you can have EUAE running on top with PCI bus busy as it is. Why only look at the processor speed-- there's other things Amiga can do besides compare processor speeds. Why don't you try reading the joystick at 1Khz while doing all those things? Why not time things to cycle accuracy while your running your PCI transfers? I can't even time a 500Khz event on the latest PC without synchronization going off if I have a WIFI card plugged in (not even surfing the web).

My god, you are living so far in the past it's hilarious. EUAE runs just fine, even on top of that CUDA simulation.

If you can't time a 500khz event properly on your PC without sync going off it's probably because you are either

1) Using a crap OS (which given the rest of your assertion I'm assuming is windows)
2) Using a crap PC. Harder to verify, but not impossible.

As for your claim the machine is somehow too maxed out with the simulation to run EUAE. try reading the specifications for PCI Express 2.0 16-lane and then come back when you've understood it. Total bandwidth for a single 16 lane device is ~8000MB/s. The simulation was only shifting around 400MB/s. Plenty of room left to shove EUAE's framebuffer into the primary X display.

I'm using a decent OS on good hardware. It may have escaped your attention back there in the 1980's, but present day POSIX compliant systems now require nanosecond accuracy for timing. The Linux kernel uses a busy loop to calibrate the maximum execution speed of your CPU at boot time. With this, you can, if you need it, have a busy wait that is accurate to a few machine cycles, though it will cost you load.

Oh by the way, PCs do slow down to a crawl because so much viruses/spyware comes into the PC easily because no one knows exactly which file is meant for what and whether it's in its original state or tampered

Well, again, you are failing to differentiate the OS from the machine. For the nth time, Windows != PC. Use a decent OS, you don't get these problems.

[Karlos]

Let's not forget two threads in the same process attempting to access data in the same cache line, which one core will implicitly lock. Pad those data structures. NUMA systems, e.g. multi-socket AMD hosts, could see implicit locks on the memory bus as well.

Of course, you can't totally avoid situations where you are going cause cores to wait on each other for access to some exclusive resource. Still, I wouldn't trade my Q9450 for a single core processor

[Karlos]

I already did-- you have a very shallow understanding of PC architecture so you did not even see it.  Reading joysticks on PC vs. reading joysticks on Amiga is a clear cut example where Amiga wins hands down.  Do you need exact numbers?  I have done the timing tests on PCs going from 90Mhz up to 2.8Ghz.  I have many more examples where Amiga wins over PCs, but if you don't understand this example others will be very difficult for you.

You still didn't clarify how you were doing this timing, or on which OS. Therefore your assertion cannot at this stage be solely blamed on the hardware. For example, if you disable speedstep in your application and wrote your delay loop in assembler, you can pretty much get it down to bus cycle level of accuracy.

If you were relying on OS level sleep()/usleep() type functions you aren't going to get accuracy any more than you get from a DoIO() on the timer.device, probably worse, in fact.

>There's no greater demonstration of how far behind the amiga is in hardware terms than a quick glance at the hardware upgrades that are available for it...

I have a PC hooked up to my Amiga so I guess my Amiga is better than your PC since I can consider the PC as a new hardware add-on.

If it makes you happy to believe that, carry on.

>You are probably sitting there, p!ssed at me, thinking I'm some PC fanboy and that I hate the Amiga etc. The fact is, you couldn't be more wrong...

No, I'm not pissed off at anyone.  You are just wrong in stating that PC has surpassed Amiga in all respects and that's all I'm pointing out since some blind followers seem to be taking your views.  In fact, Amiga just surpassed PC in another catagory without even doing further research/development.  Parallel ports are no longer put on new PCs so Amiga has another thing that PCs don't have.  The ability to do parallel port based software projects.

Well, you got me there. I don't have a parallel port. I'm really up the creek without a paddle now. I wonder whether or not USB would make a reasonable alternative for control applications? I have about 12 of those to play with.

FWIW, I don't have a joystick port either. Tragic, really.

[Karlos]

This is your shallow understanding of PC architecture.  I/O is not caches.  I/O has it's own clock way slower than processor clock.  In fact, even the memory doesn't even run at the processor clock and I/O to hardware is much slower than memory.  

You don't say...

Only specialized AGP type buses can access memory mapped areas faster than I/O transfers but even in those cases their control ports for I/O are still slower than regular memory access.

Good grief. AGP is obsolete, PCI Express 2 is the standard now and has been for some time. Devices talk to each other via high speed point to point serial communication, multiple devices can be using the bus at the same time. It's all a far cry from PCI and AGP.

[Karlos]

Your reply is irrelevant to the point.  Standard joystick gameport still uses slow I/O regardless of whether you have PCI/AGP/PCI Express.  Just because you put a new bus in doesn't make all the devices in the system automatically run at that rate.

There is no fscking "standard joystick gameport" on current PC hardware. Along with the parallel port, it has vanished into the hazy world of yesteryear. Where you live :p  In case you didn't notice, it has all moved to USB in recent years. Now, USB is very slow compared to any internal bus, yes. However that makes absolutely no difference on a system where the communication with the USB device is arbitrated by hardware. My USB device sends a packet, the controller handles it, puts it into memory somewhere and issues an interrupt. Magic. It might not be ideal for your specific polling needs but I think you'll find its ideal for most peripherals.

And no, AGP is not obsolete as far as people out there having AGP.

There are people out there with 8-bit home computers still. It doesn't mean they aren't obsolete. A given technology essentially becomes obsolete when it ceases to be manufactured or improved upon.AGP was better than PCI for graphics specific applications, PCIe is better than both, is fully generic and has subsequently rendered them obsolete.

I'm fully understanding your argument that the Amiga's native hardware is ideal for your purpose of polling the joystick port at a precise interval and that you couldn't, for whatever reason, duplicate this on a PC. It's quite amusing that you assert this as your premier example of the Amiga is still in some way streets ahead of the PC, yet complained that my particle simulation was "too specific" a metric.

[Karlos]

@Linde

I don't think he's entirely serious there. This is a good old "which platform is better" war, like we all used to enjoy back in the late 80's early 90's.

Anybody that takes anything in this thread too seriously, really needs professional help