Overclocking an 8bit?

[danwinslow]

Ok, could someone explain to me why an 8 bit couldn't be overclocked in an even multiple of its speed now? My dim understanding is that the 1.79 mhz rate was chosen to sync with the tv signal and vblank, etc. But tv issues aside, are there technical reasons we could not run the whole atari at say twice that speed?

[analmux]

I think the Chips can't handle the speed and will soon burn down. Another thing is: on the computer board there are all kinds of signals that have their unique synchronized delay w.r.t. the main clokc (especially for the DRAM: /ras and /cas signals and others). Maybe this gets f**ked up too, when overclocking the system.

 

 

.....

I still wonder how they did the 65816 implementation (running at 14MHz)

[Bryan]

Ok, could someone explain to me why an 8 bit couldn't be overclocked in an even multiple of its speed now? My dim understanding is that the 1.79 mhz rate was chosen to sync with the tv signal and vblank, etc. But tv issues aside, are there technical reasons we could not run the whole atari at say twice that speed?

862093[/snapback]

 

Everything in the A8 runs off the same clock. If you change the CPU clock, then suddenly you're not generating properly timed TV sync or color signals, all the sound would change in pitch, your SIO timing would be off, etc... Basically, any change to the clock renders the system unusable (plus, you wouldn't get the existing chips to run much faster anyway).

 

Sooo.... a CPU accelerator must be designed to run internally at a faster speed, while externally communicating with the rest of the system at 1.7x MHz. This is why such upgrades are never simple.

 

-Bry

[danwinslow]

Well, I expected the effects you mentioned. The TV issue seems manageable given that the new rate would be, say exactly twice the current rate...you could just ignore every other vblank or something along those lines, or maybe just switch to a upscanner for VGA output or something. The sound issues and game timing are also to be expected...kind of like the old turbo-mode issues on PC's. I expect an OS patch to change delay loops and so forth could handle the OS issues, SIO, etc. But providing we are talking about new software only, and not concerned with maintaining compatibility with previously existing games, is there any deep technical issue why we couldn't double the clock rate?

 

*edit*

Hmm which chips, Bry? The CPU is rated easily high enough, IIRC. Memeory chips we could probably handle. The custom chips might be an issue, though.

Edited May 26, 2005 by danwinslow

[Bryan]

I doubt you could get the chips to run at twice the current rate. The Sally 6502C is most likely a 2MHz core, which would not tolerate a 3.5Mhz clock. The stock RAM certainly would not work at this speed, and I doubt the custom chips have that much timing leeway either. The whole thing would also get very hot (some chips in an A8 get hot at the regular speed).

 

Assuming it did work, you'd have a screen with a 31Khz horizontal refresh, a 100/120Hz vertical, and a screen that's only ~200 lines tall. Ignore one of the verticals, and you'd get 2 stacked screens with a big black bar between them.

 

-Bry

[danwinslow]

I don't think the screen issues are that insurmountable. Maybe just running antic itself on every other cycle would help, or switching away from TV output completely to something else via some kind of converter. Then again, I don't really know wth I am talking about as I am strictly a software geek. It still seems like the video and sound and SIO issues could be handled...especially if game compatibility were not an issue...the memory might be able to be replaced too. The CPU maybe could be a 68c15 or whatever that 16 bit chip is, but run in 6502 mode only. The 65c02 sally was modded to have some kind of dma halt for antic IIRC, so maybe that would be a stopper, although draco seems to have dealt with it.

Edited May 26, 2005 by danwinslow

[Bryan]

I still think the simpler route is to design a CPU upgrade that leaves the rest of the system at the current speed.

 

BTW... You can get a 65C02 at speeds up to 14Mhz, so you don't have to use a 65816.

 

-Bry

[danwinslow]

Really...hmm wow. But wouldn't the speed gains from a processor only overclock be sorta small? Maybe with some kind of on chip fast RAM cache.....hmmm....

[Bryan]

You would need to have RAM (and possibly ROM) on the upgrade. You would make all accesses to RAM happen at the accelerated speed, and all access to hardware happen at the slow speed. When Antic does DMA, it would need to have access to the RAM on the upgrade as well (that is, unless you integrate some of the RAM already in the system as slow RAM).

 

-Bry

Edited May 26, 2005 by Bryan

[danwinslow]

Some interesting linkage :

 

http://retrobits.net/altcpu.html

http://www.overclocked.org/emu2600.htm

[DamageX]

OK what if... one were to swap out the DRAM for SRAM and then feed the CPU clock signals that were separate from the rest of the system (and running at some multiple of the speed). If you had for instance the CPU running at 3.58MHz and everything else still running at 1.79MHz would there be any difficulties there? Would the CPU have to slow down to read/write registers on the other chips perhaps?

 

ANTIC's DMA would use up that many more CPU cycles this way but maybe it would be easier than using a cache or separate RAM for the faster CPU.

[Bryan]

OK what if... one were to swap out the DRAM for SRAM and then feed the CPU  clock signals that were separate from the rest of the system (and running at some multiple of the speed). If you had for instance the CPU running at 3.58MHz and everything else still running at 1.79MHz would there be any difficulties there? Would the CPU have to slow down to read/write registers on the other chips perhaps?

 

ANTIC's DMA would use up that many more CPU cycles this way but maybe it would be easier than using a cache or separate RAM for the faster CPU.

863140[/snapback]

 

This is sort of what upgrades do, but you cannot combine different bus access speeds on the computer's bus (all the timing and decoding logic would get really confused). What you still need to do is keep the fast and slow stuff separate, and only connect them when the CPU is running slow.

 

I looked at the data for the custom IC's and there's no way you could feed them a doubled clock. Atari's engineers were very fond of asynchronous logic, meaning much of the logic doesn't run on the system clock, but rather uses delay elements to get the signal where it needs to be on the right cycle (sort of the analog approach to digital).

 

This works...if you time everything just right, and it makes the design much smaller than using clocked flip-flops everywhere. But if you vary much from the specified timing, things will go very wrong.

 

-Bry

[danwinslow]

Hrmmm...sounds like the hardware equivalent to hard-coding stuff in a program. Well, that explains the lack of overclocked Atari 8bits.

[Bryan]

Looking at the data some more, it appears that clocking is used more often than I thought. The schematics just hide it inside simple logic symbols. I wish some better copies had survived.

 

-Bry

[Heaven/TQA]

maybe the polish guys good explain what is discussed at atariarea.histeria.pl as Draco recompiled and patched the OS to handle the 65816....

[Heaven/TQA]

good = could

[danwinslow]

Yes, I've seen draco's project, but there is very little information in english, just kind of a high level description.