exxos's Atari, Amiga & retro forum

exxos wrote: 30 Apr 2020 11:45 So while this analogy does not increase ST-RAM speed per-se, the "available bandwidth" to RAM is somewhat increased because after the integer division we get to the next ram access a lot sooner.

I can't speak about Frontier (never used it), but people :-) would point out that division is something that you avoid like plague with this gen of hardware. You either try to get away with pre-calculated tables, shift aproximations or multiplication (using the reciprocal) before you consider using divisions.

Would probably be an interesting case study in an emulator to do an opcode analysis over a larger amount of apps at runtime and my bets would be on divisions are an edge case in the big picture. As long as you don't get the surrounding operations much faster, the effective speedup doesn't scale well with the frequency increase.

Best regards,

pixelpusher wrote: 30 Apr 2020 16:56 the effective speedup doesn't scale well with the frequency increase.

Indeed. I document benchmarks of various speeds several months ago. Basically after 32MHz you really hit diminishing returns without speeding up something else.

pixelpusher wrote: 30 Apr 2020 16:56 Would probably be an interesting case study in an emulator to do an opcode analysis over a larger amount of apps at runtime and my bets would be on divisions are an edge case in the big picture. As long as you don't get the surrounding operations much faster, the effective speedup doesn't scale well with the frequency increase.

That's why I tend to use EEMBC CoreMark® to assess the speed of Atari systems and accelerators. Imho, CoreMark contains multiple algorithms that you would also find in real software and, therefore, gives a much better indication of total speed than, e.g., Gembench's "Integer Division" number.

Source code of my port to Atari: https://github.com/czietz/coremark
Compiled version and some results in this thread: https://forum.atari-home.de/index.php/t ... 024.0.html

For example, you can see that -- without cache -- doubling the CPU clock to 16 MHz in the MegaSTE only gives about 10%(!) speed increase. That's because RAM access is not accelerated. Only with cache you're able to see a big boost.

EDIT: Just to mention, CoreMark is meant to assess CPU and RAM speed. By design, it does not call any OS functions during the benchmark. Thus, to see the effects of faster ROM access (e.g., on VDI drawing speed) by Exxos' boosters you still need to use Gembench.

No one will ever know...


I've tested several CPU's and GAL's in my test board and all seem OK. So that booster will be the first one heading to my store next week...

This one has been running for about 3 hours in total now.. So I will start on a second one. If the second one works, I may take some pre-orders for this so people can reserve one.

Hello,
First post here ! I have a question about those booster. As you remove the CPU socket, is it still possible to add an ide board ? Or do you have to choose between one or the other ?
Thanks !

Reylan wrote: 03 May 2020 22:48 Hello,
First post here ! I have a question about those booster. As you remove the CPU socket, is it still possible to add an ide board ? Or do you have to choose between one or the other ?
Thanks !

I had one of the IDE boards from Alan Hourihane in mine, and it wouldn't work boosted. Also a bit of a nightmare because of the stack of stuff plugged in.

Reylan wrote: 03 May 2020 22:48 First post here ! I have a question about those booster. As you remove the CPU socket, is it still possible to add an ide board ? Or do you have to choose between one or the other ?

I don't know of any compatible IDE solutions for the STE.. There is very few to start with.

exxos wrote: 03 May 2020 23:21

Reylan wrote: 03 May 2020 22:48 First post here ! I have a question about those booster. As you remove the CPU socket, is it still possible to add an ide board ? Or do you have to choose between one or the other ?

I don't know of any compatible IDE solutions for the STE.. There is very few to start with.

@exxos Agreed, the only one I could find was Alan's.