A1200/30@40Mhz vs A4000/30@25Mhz

[UberFreak]

My adventures continue

After finally getting my A4000/30 up & running, I filled the HDD with demos & watched a few.

I have the same demos on my A1200 (1D4 board, GVP A1230+ EC030@40Mhz & 8MB RAM).
I noticed several strange things:

1) Some 3D effects (hard to categorize, but mostly textured objects) run noticeably FASTER on the A4000! 25 > 40 ?

2) On the A1200, in lots of demos, there are small "jitters" - movement that should be smooth isnt.
One very bad-looking example is the end-scroller in the demo Atome.
This problem does not appear on the A4000 at all!

Can this be related to timing issues on the 1200? I havent made the timing-fix mod since I was sure it only affected 040/060 accelerators.
Btw, both machines have an FPU, but I doubt the demos in question actually use it.

 

[Zetr0]

This is most likely the case that the two differing accelerators have different speeds in writting to chipmem.

Logic circuits have propagation delays, as such they may miss a read or write time to remain as synchnous as possible to the chipram / motherboard timing.

This can effect performance where you are mainly hammering Fast to Chip opperations (like in GFX Demo's) this however doens't mean the A1200 is slower in processing.

 

[UberFreak]

Interesting...
The "jitters" make watching certain demos on the 1200 not fun at all.

I'll attempt the timing-mod later today, to see if it helps.

 

[Justin]

so the 030 1200 is handling things differently to the 030 4000? even though the 1200 is faster?

 

[UberFreak]

@JuvUK:

As I understand from what Zetr0 wrote, this is accelerator-specific and depends on the design/implementation of the logic circuits on it.
As I dont have another 1200 accelerator to test, I cant check if this is indeed the case.

 

[Justin]

i thought the 1d4 board didn't need any timing fixes, wasn't that more for the 2b boards?

sorry scrap that, it is the 1d4 boards as well

http://www.ianstedman.co.uk/Amiga/amiga_hacks/A1200_Mobo_fix/a1200_mobo_fix.html

 

[Zetr0]

so the 030 1200 is handling things differently to the 030 4000? even though the 1200 is faster?

yes.... and no... lol

As far as I am away the Chip Architecture and FastRam architecture is only subtle different in the A4000. (ie. 32bit addressing space by default)

however that wont be the problem here, its quite likely the CPU cards when fetching and sending data back to the custom chips are opperating differently due to both thier logic *glue* and frequency.

If you imagine glue-logic like an old 1930's Telecom switchboard attempting to get a communication from out of the country to a localized number... how many switches it takes depends on a couple of factors.

1. the number of concurrent connections to find the given location
2. the speed of the data that being sent

now in the case of the latter, the data has to WAIT untill a write cycle can be synchronized between the motherboard and accelerator card.

Imagine this in the case of the A4000 CPU at 25 million times a second..... The motherboard is cycling at about 28 million times a second. so for the most part every write and read cycle should connect.

Imagine this with the A1200... the motherboard again is cycling at 28 million times a second, however the CPU is now cycling at 40 million times a second, as such it can only *catch* most (not all) of the write cycles times.

To improve write times, you can reduce the delay of the logic, and or in some cases with some expensive cards (on the A4000) re-sample the timing for better synchronization.

its a lot to get you head round for sure... but if you are partially deranged and some what lacking in most social skills then this should sit neatly in ones brain...

=D


This is why some cards are Faster talking to chip than others.

Noteably the Apollo 040/60 when compared to the Blizzard 040/060 (frequency alike) the Apollo is between 10 - 5% faster when talking to chipram.

This not to say its the better card, there are many factors features to a CPU card that one must weigh up.

 

[UberFreak]

@Zetr0:

Do you think the timing-fix should improve chip write synchronization?

 

[Zetr0]

@UberFreak

It is possible that it will improve performace, but more so for compatability.

Me personally I would do it without any hesitation, its quite simple modification, although i may make no difference to you graphics issues with the GVP card, it will improve the compatability with other cards like 040's.

However I suggest if you are not versed in what the modification is aboutor does for you, then take a moment to read up on Stedy's site =)

If you are unsure, then please ask for help, better to learn by asking than learn by breaking =)

 

[UberFreak]

OK, timing fix applied.
While I'm in there, I thought about also doing the PCMCIA reset-fix.
Since I've learned, the hard way, to always verify everything against the schematics, I tried to locate pin#5 of Gayle in the A1200 schematics, to see where its connected.
Strangely, its not in the schematics :huh:

Does this usually mean this pin is NC?

---------- Post added at 14:12 ---------- Previous post was at 14:09 ----------

Scratch that...

Gayle appears twice in the schematics, once with the cpu/alice/fpu & once with PCMCIA.
Quite confusing

 

[UberFreak]

Just closed the A1200.
Applied the timing-fix, PCMCIA reset fix & also fitted a disable switch for the GVP A1230+.

Ran a few problematic demos, no change, the problem is still there
Its strange since those demos were made for 030@50Mhz, according to their readme files, so I assume they should work fine.
I'll need to test my mobo with another accelerator sometime, although I doubt its worth it to get another accelerator just for this problem (unless its dirt cheap or free!).

Thanx for all the advice!

 

[Harrison]

Could different speed fast ram cause such issues?

 

[Kin Hell]

I've been away a few days & just caught up with this thread.

@ UberFreak

In a nut shell, what you're seeing is Faster Ram Access via the A1200 over the A4000. :nod:
We must not forget, when the A1200 & A4000 machines were released, the Big C= was fast approaching bankruptcy & several glitches within the OS & Hardware design we're never finialised.

The A1200 has a much faster access to any Ram expansion than the A4000. The A4000 actually has "wait states" set when accessing the ram on an A4KD, because it is so far away from the processor. :nod:

Stick an accelrator card in your A4000D/T with onboard ram (preferably with SCSI) & watch it fly.

Had the Big C= still been in power a few months/years longer, you can bet you wouldn't be seeing the cr4p you're seeing now.

Zetr0 is just brilliant with this sort of stuff in his description of things, but I thought I should try & post a "lay mans" description if nothing else.

 

[UberFreak]

@Kin:

So what you're saying is, the problem exists on the 1200 due to its faster RAM access?
Hmm... I wonder if this problem will be seen with other accelerators.

If you have an 1200 with a fast 030 card, please try to run the demo Atome by Skarla.
It has a plain (white text on empty black screen) end up-scroller, which looks really jerky on my 1200, but 100% smooth on the A4000.
Other parts are affected as well, to some degree, but its hard to spot without knowing what to look for.

 

[Kin Hell]

Hi UberFreak

I just watched that Demo on my Blizzard MKIV 256Mb 50Mhz baby & that end scroll Text is so silky smooth, it makes my 4Ghz Quad Core Duo with GTX295 Graphics card look like a bag of sh!t when trying to scroll.
Most folk still can not appreciate an Amiga can out-scroll Pee Cee's, period. :nod:

Try copying it to fast Ram & see if it's smoother for you. You might be seeing IDE hard drive lagg for the CPU loading. If you're using a CF card, the lack of Cache which all hard drives could be the culprit.

Kin