SymbOS MSX multitasking operating system - help needed!

Page 376/392
369 | 370 | 371 | 372 | 373 | 374 | 375 | | 377 | 378 | 379 | 380 | 381

By Manuel

Ascended (15287)

Manuel's picture

31-12-2015, 11:41

Any idea why on a normal turboR (so without GFX9000) the graphics score is higher than on MSX2?

By giuseve

Paladin (732)

giuseve's picture

31-12-2015, 18:16

Waiting for a full release with all those new stuff !!!

By edoz

Prophet (2127)

edoz's picture

31-12-2015, 21:39

Manuel wrote:

Any idea why on a normal turboR (so without GFX9000) the graphics score is higher than on MSX2?

I have no idea why this is. First i was thinking it was because of the emulator but real machines (gives a bit other results) a standard MSX has a better score if it is about filling things.. Could it be that the Turbo-r is less effective then a standard MSX2 if it is about VDP or communication with the VDP?

Trubo-r

MSX:

The move speed is somehow the same.. The bitmap is much better on turbo but it seems the fill speed is better on a normal MSX2.

By edoz

Prophet (2127)

edoz's picture

31-12-2015, 22:05

By Manuel

Ascended (15287)

Manuel's picture

01-01-2016, 01:01

Ah, I didn't see that higher numbers are slower than lower numbers. Anyway, I would expect that the turboR is perhaps a little bit slower on the gfx side than the MSX2, but it should definitely not be faster. If it would be faster, you're not measuring correctly (not measuring the gfx speed, but the CPU (partly) instead). The VDP's are of course of the same speed in both machines, so simply speaking, they should have the same speed. But there may be a bit more I/O delay to the VDP on the turboR causing some slow downs.

So, make sure your tests are correct: make the majority of the time of the test spend on waiting for the VDP, or at least only measure the time you are waiting for the VDP. I understand you can only start and stop measuring when executing some graphics primitive of the OS. It's possible that some of these make a lot use of the CPU and not so much of the VDP. Those instructions may run a lot faster on turboR of course.

Bear in mind: proper benchmarking is not easy! Think carefully when creating such tests to be sure you're actually measuring what you want to measure.

By Prodatron

Paragon (1788)

Prodatron's picture

01-01-2016, 14:51

BTW, for the CPU benchmarks I used these instructions:
- simple (fast 4cycle 1byte instructions with no additional memory access): nop/add a/ld a,b/cpl
- complex (1byte instructions with the maximum memory access): ex (sp),hl
- 16bit (1byte instruction with no additional memory access but internal "16bit" operation): add hl,hl
Especially the first and second have been choosen for demonstrating the different wait-state behaviour between the MSX and the other platforms (CPC, Enterprise, PCW). The third have been mainly choosen to demonstrate the internal 16bit capability of the R800, but it also shows again the different wait-states.

Manuel wrote:

If it would be faster, you're not measuring correctly (not measuring the gfx speed, but the CPU (partly) instead).

Yes, the intention is to measure the whole process of graphic operations, not to test the VDP isolated. The Amstrads and the Enterprise don't have a VDP. IMHO including the CPU part makes sense, if you really want to know the total amount of time, which is required on all the different computer platforms.

Manuel wrote:

Bear in mind: proper benchmarking is not easy! Think carefully when creating such tests to be sure you're actually measuring what you want to measure.

At the beginning the progress bar has been updated during a single test multiple times, which was probably not a good idea to have clean results Smile

By edoz

Prophet (2127)

edoz's picture

01-01-2016, 16:01

The turbo-r is faster in the bitmap test, that make sense but i also wondering why the first test is better on the MSX2. (Fill speed) Is there something special about it?

By Manuel

Ascended (15287)

Manuel's picture

01-01-2016, 16:04

Why does it make sense that the turboR is faster in the bitmap test? Does that test involve a lot of CPU operations?
And the fill speed, does that involve mostly VDP operations?

By edoz

Prophet (2127)

edoz's picture

01-01-2016, 16:12

Because bitmaps in SymbOS are in normal memory (not VDP memory) So there is more CPU needed in that case. Bitmaps are in normal memory because MSX is the only machine that has a VDP. Indeed fill speed is mostly VDP. So i wonder why it seems the turbo-r is slower here. At least in this test.

By Manuel

Ascended (15287)

Manuel's picture

01-01-2016, 19:39

My guess is that it's due to the aforementioned VDP I/O delay on the turboR. But it's a guess... I don't know the details at all and thus I might be completely wrong.

Page 376/392
369 | 370 | 371 | 372 | 373 | 374 | 375 | | 377 | 378 | 379 | 380 | 381
My MSX profile