PAK030 40MHz & STorm ST issues

[Darklord]

Hmm, I've got some puzzling behavior here (I know, it's a recurring theme with me).

There is something odd going on with the Pak board and the STorm-ST board. It's to do with
Gembench results. Look at these pictures and see what I mean.

The first is with the Pak board, NO STorm-ST, NO NVDI:

STacy-Pak-GB01.JPG (247 KiB) Viewed 1073 times

The second is with the Pak board, the STorm-ST, NO NVDI:

Stacy-Pak-GB02.JPG (200.03 KiB) Viewed 1073 times

The third is with the Pak board, NO STorm-ST, NVDI:

STacy-Pak-GB03.JPG (194.49 KiB) Viewed 1073 times

The fourth one is with the Pak board, STorm-ST, NVDI

STacy-Pak-GB04.JPG (187.91 KiB) Viewed 1073 times

It appears that the STorm-ST board being on results in a lower CPU score and
if NVDI is present, the CPU goes even lower?

Anyone else seen anything like this or have any idea what it's doing?

I sure don't...

[Steve]

That's really weird because integer division is pretty much the same across the board. Although I do wish you'd have done CPU only tests, rather than display & cpu. As it's really difficult to dissect the differences in what Gembench is using to calculate the CPU score.

[exxos]

I suggest you repeat your tests without the caches enabled.

[Darklord]

That's really weird because integer division is pretty much the same across the board. Although I do wish you'd have done CPU only tests, rather than display & cpu. As it's really difficult to dissect the differences in what Gembench is using to calculate the CPU score.

I can do that. I'll report back later today with the results. Thanks.

[Darklord]

I suggest you repeat your tests without the caches enabled.

Okay, I think I can do that. Now, just to be clear because I'm
horrible at taking instructions , are we talking about
using the J2 jumper on the Pak board to disable the CPU
cache before running the tests or just clicking off "Cache"
in the drop-down menu under TOS v3. 06?

Also, the Pak board has an SLC (secondary level cache). Does
it need to be disabled as well?

Thanks much.

[exxos]

You turn the caches off / on in the GB6 menus.. I don't know however it has any effect on the L2 cache stuff though. It's mostly designed for internal CPU caches.

[Darklord]

Okay, here were go.

First picture is the STacy, with the Pak, with the STorm-ST board active, and both GB caches turned off:

STacy-Pak-STorm_ST-NO Caches.JPG (187.17 KiB) Viewed 946 times

And the second picture is the STacy, with the Pak, with the STorm-ST INactive, and both GB caches turned off:

STacy-Pak-NO STorm_ST-NO Caches.JPG (184.82 KiB) Viewed 946 times

[agranlund]

It looks like the aggregate cpu score goes down when your (slower) alt-ram is enabled.
Not sure if that makes sense or not, maybe it does.

All other things being equal, if you have 200% ram speed from st-ram (assuming due to pak cache?) but only ~117% when using non-cached altram on the storm board then perhaps it makes sense that the aggregate overall score ends up somewhere in between.
Chances are some programs will run from alt-ram, and thus a bit slower.

The individual test results look sane and roughly equal for storm vs no-storm, it’s mostly that alt-ram is taken into consideration on the overall average when storm is enabled.
NVDI probably loads into altram when you have storm enabled and may or may not become a small bit slower than your (probably Pak L2 boosted) ST-RAM.

[Darklord]

Okay, all of this was kinda triggering a memory so I went back to Derkom's summer STacy message thread and
found some interesting stuff on Page 11, about halfway down or so:

viewtopic.php?f=29&t=3009&start=100

Capture.JPG (107.01 KiB) Viewed 894 times

[Badwolf]

It looks like the aggregate cpu score goes down when your (slower) alt-ram is enabled.
Not sure if that makes sense or not, maybe it does.

Concur. The headline CPU figure has a statistical dilemma on its hands: how to deal with a ratio that simply wasn't there in the first place.

The interesting scores for me are the display ones. I particularly like VDI Enquire as an indicator of grunt, personally.

With NVDI, as Anders suggests, it probably loads into AltRAM and so relative speeds go down, so that expected.

But they appear *faster* with the Storm-ST board (without NVDI) than without despite GB6 running from ST-RAM. What gives there?

I believe blocks of Amiga RAM have the concept of a priority. I don't think we do in the ST world. AltRAM is AltRAM and if a program can run from it, it generally will. Even if it's slower than ST-RAM.

BW