exxos's Atari, Amiga & retro forum

Darklord wrote: 23 Jul 2024 20:02 And the second picture is the STacy, with the Pak, with the STorm-ST INactive, and both GB caches turned off:

Those results look perfectly normal to me.

Don't forget the results in the statistics window use "geometrical averages" over all the test results in the main window. So the main CPU score will be lower as its "weighted" by the slower alt-ram speeds.

Or maybe you are thinking all the scores in the main window simply multiplied together as you expect the CPU score to be higher maybe ?

exxos wrote: 24 Jul 2024 19:54
Those results look perfectly normal to me.

Don't forget the results in the statistics window use "geometrical averages" over all the test results in the main window. So the main CPU score will be lower as its "weighted" by the slower alt-ram speeds.

Or maybe you are thinking all the scores in the main window simply multiplied together as you expect the CPU score to be higher maybe ?

I wasn't sure what to make of it, to be honest. :)

So my takeaway is that the CPU really isn't slowing down when ALT-RAM is in use, it's that GB is reporting the
slower ALT-RAM speed and averaging it into the total score. Is that right?

After going back and reading the section in Derkom's message thread, where he ran into the same thing (except that
he was going for the additional 8 megs when he switched down to 8mhz/compatibility mode), things became a little
more clear.

I hadn't thought about ALT-RAM being slower than regular ST-RAM. I still get the terminology mixed up.

This is correct, right?

ST-RAM = 512k -> 4 megs

ALT-RAM = >4 -> 12 megs

FAST-RAM (or TT-RAM) = above ST/ALT-RAM ranges and faster

Is that right?

Thanks much, and thanks for the explanations everyone.

Darklord wrote: 24 Jul 2024 21:32 This is correct, right?

ST-RAM = 512k -> 4 megs

ALT-RAM = >4 -> 12 megs

FAST-RAM (or TT-RAM) = above ST/ALT-RAM ranges and faster

People have very entrenched positions over this!

A shorthand may be that ST-RAM is RAM which normal, on-board, video (ST shifter, TT shifter, Videl) can be output from. It can also transfer data to and from ACSI/SCSI and the floppy.

AltRAM is anything else.

TT-RAM exits in the space above 24 bits (> 16MB), but is still just AltRAM to the OS. I've had conversations where my interlocutor was of the opinion TF536 and DFB1 don't have TT-RAM as they're not TTs. TTs can DMA into TT-RAM. Nothing else can. It's a position as defendable as any other.

So if you were being precise you might say that a TF536 has AltRAM in the TT-RAM space.

Now between 4MB and 15MB there is an address space that is sometimes completely available for AltRAM (base STs with TOS < 1.4), sometimes only 10MB is available (STEs), sometimes only 9MB is available (MegaSTs, TTs) and sometimes it's available for either AltRAM or ST-RAM too (Falcon).

That certainly can't be called TT-RAM, but, according to the OS, if it's declared as AltRAM, is no different.

Your mileage may vary.

Which is why I mentioned priorities. You may potentially have TT-RAM, AltRAM and ST-RAM all on the same (TT) system. You'd really like the fastest one used first, but you don't get much of a say in that. Generally the OS assumes AltRAM is preferable so if a program is allowed to use it, it will. You can set program flags to say 'only use ST-RAM', but not 'please choose the fastest'.


BW

Darklord wrote: 24 Jul 2024 21:32 So my takeaway is that the CPU really isn't slowing down when ALT-RAM is in use, it's that GB is reporting the
slower ALT-RAM speed and averaging it into the total score. Is that right?

Correct.

I hadn't thought about ALT-RAM being slower than regular ST-RAM. I still get the terminology mixed up.

I'm not sure on STram Vs alt-ram.. as there are normally faster CPUs accessing alt-ram. So altram is generally faster and can even be 32bit.

I'm not sure there was a official benchmark file for GB6 with altram as I think I might have just copied STram speed to the altram speed tests.. so assuming your alt-ram is 8mhz then it would mean it's slightly faster then STram.

BUT, as you got 18 seconds for STram and almost 200%... Then 36seconds would be 100%. But your altram shows 64seconds. So I'm slightly puzzled why it's showing 113% as it takes allmost twice as long as STram.

I guess really we need a system with just a basic altram running 8mhz CPU and a new benchmark file creating...Maybe @Badwolf can run a new benchmark with TOS206 8MHZ so we can have a proper benchmark of what 8mhz altram is actually supposed to be.. assuming there's no wait states on access.

EDIT:
Or maybe I copied ROM speed as it hangs off the CPU bus like altram does... But that be wrong as ROM is read only anyway.. so I dunno.. we need a better altram test with sram really..

exxos wrote: 24 Jul 2024 22:27 I'm not sure there was a official benchmark file for GB6 with altram as I think I might have just copied STram speed to the altram speed tests..

Looks like it in the newer versions. I think that's perfectly reasonable as at 8MHz the CPU can't access RAM faster anyway.

Unfortunately my AltRAM code is a bit nobbled at the moment speed-wise so it only gets to 3MB/s writing at 8MHz. It'll max reading, but writing would mess up the timings.

BW

"People have very entrenched positions over this!"

Right. I'm not surprised by any of this - it's the world of Atari, after all! :lol:

Oh, don't think I mentioned it and I don't think it makes a difference,
but this was all with the 68030 clocked at 40mhz...I never changed
the speeds during any of the tests.

Thanks again to everyone for the detailed explanations - appreciated!

:)

I ran Gembench overnight, with the STorm installed and everything running at 40mhz:


Best I'm capable of getting, I reckon. I still have the same problem I've had since day 1
with this board - I can't run full speed (50mhz) without problems. I can disable the FPU
on the Pak board (J3) and it will run at 50mhz and seem very solid. Just not with it
enabled. It will boot up, but sometimes gives a 2 bomb error that continues on to the
desktop. From there, FPU Test 6-7 bombs completely. Gembench will run until it accesses
the FPU, then also crashes.

At 40mhz, both Gembench and FPU Test complete without errors.

I tried different 68882's - I have 3 of them, all rated at 50mhz. Here they are:


All 3 will pass Gembench at 40mhz. Only #2 will pass with no errors with FPU Test.

At 40mhz, with #2, it seems rock solid though. That was the setup when I ran GB
overnight without issue.

<shrugs> I guess I can live with a 40mhz 68030/40mhz 68882, TOS v3.06 STacy. :)

(it just irks me that I can't eeek out that last 10mhz!!!) :x

If you look at all the FPU problems I had with the DFB1X... You really need perfectly stable circuit to run the FPU at 50MHz. But also with a lot of stuff being reprinted, it is really questionable if you *really* have a 50MHz FPU. Unfortunately I have no way to test the PGA versions. I could certainly buy genuine chip, but it would be insanely expensive but I doubt the FPU alone is the only problem.

... if such an FPU is reprinted use Acetone to remove the printing. Original prints cannot be removed with Acetone.