4MB Upgrade & 16MHz Booster progress

[exxos]

4MB.jpg

This is a sneak preview onto my new 4MB upgrade kit. My last one would only fit one style of motherboard which limited it for a lot of people. Plus it needed a lot of soldering. This new kit will fit almost all revisions that have a socketed MMU. The Install is very much a like the Marpet kit, only there is no simm board now. The DRAM is held on 2 small chips on the PCB which plugs in over the MMU. The old DRAM (be it 512K or 1MB) is not used as the DRAM on the kit replaces all the RAM to give 4MB total. This kit is pretty much "plug and go". The old DRAM will have to be removed or disabled. Removing it (can be cut out) is ideal as it takes a huge load off the PSU. I have also added in a jumper link on the board to isolate MAD10 so anyone who wants to switch between 1MB and 4MB is free to add a switch if they desire.

The 16MHz mod is the basic booster one which gives overall about 30% increase in machine speed. Its designed to solder on top of the CPU. There is only a few connections to be made. Though it is possible to wire it to the CPU and locate the board elsewhere if there is other upgrades already over the CPU. I've had thinner PCB's made as height is at a premium (as always). The CPU booster should fit over the CPU and still fit under the metal shield. Though if not, it can be routed away next to the CPU somewhere anyway.

I'm still waiting for the cables to come for the 4MB kit currently. I've been told there's huge shipping delays with the China post (more than normal) So it could be a few weeks before the rest of the stuff arrives. I will post here progress as it happens. I normally get really busy with work a month before xmas, so its probable they won't be on sale until after xmas. Stay tuned!

[qq1975b]

[GadgetUK164]

Wow, that's looking sweet!!! What an awesome project this has turned into! =D

[artik-wroc]

I have a similar solution
http://atari.wroclaw.pl/hardram.html#myram

[exxos]

I did notice that one a while ago, but when I did the MMU layout, there is 3 sides of the MMU which needed to be linked into, but on your design, it only seems to use 2 sides of the MMU ? I can't see where it would pick up MAD 7,8,9 without another row of pins on the MMU ?

[artik-wroc]

http://atari.wroclaw.pl/pliki/stram.JPG

[exxos]

ahh

[exxos]

For the larger 16mhz kit with fast rom, Here are the scope waveforms of the CPU & ROM running at 16mhz..

16mhzrom.png

We can see the CPU enter 16mhz when /AS is HI, or ROM_CS is LO. Of course when /AS & ROM_CS are both LO, this still results in 16mhz mode without glitching back to 8mhz.

16mhzrom2.png

A little while later in the cycles, we can see the CPU switch from 8mhz to 16mhz there it remains at 16mhz for basically ages!

I will re-test 16mhz rom and 8mhz rom to see if there is any difference. Though how its looking at the moment, 16mhz ROM access isn't giving any boost in speed at all according to GemBench4. As GB4 isn't even showing 200% ROM access, I can only assume it was never tested at higher speeds.

I could do with someone chipping in who knows more about TOS calls, who could possibly write a small app to benchmark various TOS functions. I would assume there would be some routines in TOS which do not need to access RAM or some long routines which take many CPU cycles which should run faster at 16mhz. I guess a program to just access a address in ROM might be enough to see if there is a speed boost at all without actually calling any functions.

[GadgetUK164]

Nice progress! I wonder if we can get PP to take a look here, he might be able to shed some light.

[exxos]

Nice progress! I wonder if we can get PP to take a look here, he might be able to shed some light.

Possible, though I just assume its a RAM bottleneck slowing things down. I guess if GEM apps were run in Alt-ram and the CPU could access alt-ram at 16mhz, as well as TOS at 16mhz, then I would assume it would give a huge boost in GEM app speeds. I think AlanH was doing some Alt-ram boards a while back, but I will have to ask him if they are likely to be capable of running at 16mhz.

I still want to try faster than 16mhz speeds, though big "d'oh" as when I got the other pcbs made, I forgot to include the clock booster boards. Not really worth putting them into production on their own either

[AtariZoll]

As I see, TOS ROM is selected by own logic, instead slow Glue. But Glue still generates DTACK when TOS ROM is accessed - what may cause problems. Override of it is using TOS 2.06 on different address space.
I will code some simple ROM access speed SW - based on using movem - then RAM speed should have minimal influence.

[exxos]

As I see, TOS ROM is selected by own logic, instead slow Glue. But Glue still generates DTACK when TOS ROM is accessed - what may cause problems. Override of it is using TOS 2.06 on different address space.
I will code some simple ROM access speed SW - based on using movem - then RAM speed should have minimal influence.

TOS206 isn't using GLUE DTACK, it is generated on the card. I think rodolphe said it was 62ns. I did try faster speeds but it did not work. Rodolphe is going to see if he can speed up ROM access any soon.

It would be really useful if you can do simple ROM speed testing software. CPU & ROM are running at 16mhz. Of course CPU also run 16mhz when /AS is HI so this will have some effect on software running.

Gadgetguy has 16mhz booster (with normal TOS access 8mhz) so we will have to benchmark his machine to get the base line figures. Then try on my ST with 16mhz ROM and see if figures improve.

[AtariZoll]

Here is simple test prg:

ROMSPT.ZIP

At 8 MHz it shows 212 . Should show about 120 when TOS ROM is accessed with 16 MHz CPU speed. 85% of testing time is pure ROM read . For TOS 2.06 - accessing area at $E00200 . On Mega STE at 16 MHz shows 104, but it caches RAM and ROM too, so it's all with full speed.

[exxos]

ok great

TOS206 on STFM shows 139... We just need gadget guy to try results as he has 16mhz cpu and 8mhz TOS..

[AtariZoll]

So, ROM really runs at 16 MHz. At least most of time.
With this test speed gain on Gagdet's machine will be minimal, max some 20% - as said most time is ROM read.

[exxos]

So, ROM really runs at 16 MHz. At least most of time.
With this test speed gain on Gagdet's machine will be minimal, max some 20% - as said most time is ROM read.

Well that is good news if ROM running 16mhz speeds. I think its 17% for CPU only 16mhz.

Though in gembench4, nothing shows up as running faster in TOS. Mostly ROM speeds, GEM windows etc, all show no speed improvement. So do you have any idea what routines would actually benefit from 16mhz TOS ?

I am starting to think GemBench4 isn't showing the proper speed measurements as the ROM speed in there is still 100%. Clearly the ROM speed is faster. So if that speed test is incorrect, then its possible all the GEM related speed tests are not showing correct figures either.

I wonder if there is any other benchmark programs out there which may be better than gembench in relation to TOS functions ?

[AtariZoll]

I used some SW long time ago, mostly with Gemulator. But can not remember name at moment.
There are some ROM routines which actually run in RAM from diverse reasons - short code. But then it should run faster too, as RAM is accelerated too, although not so much. And Line-A goes via blitter in many cases, so should ensure that it is off.

[exxos]

QINDEX22.zip

Need gadgetguy to try his machine also to see if its the CPU speed causing the faster results or TOS access speed.

Results are (with 1.04 setting)

q22.jpg

I half think GEM runs faster, though it was when the CPU was boosted to 16mhz anyway, but im sure GEM is running faster with 16mhz ROM. Its hard to tell.

[GadgetUK164]

Gadget.jpg

Here's that app running on mine:-

I tried PP's util there but I just get a flashing cursor and then it returns to GEM? Could it be due to TOS 1.04?

I did all tests on that other benchmark prog (seperately), excluding the RPM test as that was hanging.

EDIT: I attribute those higher readings I got to the Blitter, and the 64Kb DMA is probably the Ultra Satan.

[GadgetUK164]

Here is simple test prg:

ROMSPT.ZIP

At 8 MHz it shows 212 . Should show about 120 when TOS ROM is accessed with 16 MHz CPU speed. 85% of testing time is pure ROM read . For TOS 2.06 - accessing area at $E00200 . On Mega STE at 16 MHz shows 104, but it caches RAM and ROM too, so it's all with full speed.

Doesn't show anything on mine =/ I get a black flashing cursor for a couple of seconds and then it just goes back to GEM? Presumably I just double click the .TOS file here and it should display something?

[GadgetUK164]

Gadget1.jpg

Same thing with Blitter switched off:-

[exxos]

Here's that app running on mine:-

I tried PP's util there but I just get a flashing cursor and then it returns to GEM? Could it be due to TOS 1.04?

I did all tests on that other benchmark prog (seperately), excluding the RPM test as that was hanging.

EDIT: I attribute those higher readings I got to the Blitter, and the 64Kb DMA is probably the Ultra Satan.

Awesome, yeah, I forgot you had blitter, you need to turn that off and do the tests... the DMA speeds and such are not important.

[exxos]

ahh, you beat me to it on no blitter tests. There is a slight & difference, but that is probably down to TOS104 vs TOS206 so some small error variation has to be included. So how its looking, 16mhz ROM isn't giving any speed boosts.. again.. Though with PP's util, it does show around 85% speed boost. Very strange indeed.

[qq1975b]

maybe a silly question...but is it useful yo compare it against an Atari Mega Ste at 16MHZ?

[exxos]

maybe a silly question...but is it useful yo compare it against an Atari Mega Ste at 16MHZ?

AtariZoll already tried his program on MSTE, it was slightly faster than my setup, though I think he said there was cache involved.