MSTE vs STE TOS206

Benchmark screenshots for various boosters & machines.
User avatar
exxos
Site Admin
Site Admin
Posts: 8244
Joined: Wed Aug 16, 2017 11:19 pm
Location: UK
Contact:

Re: MSTE vs STE TOS206

Post by exxos » Tue Jul 31, 2018 1:03 pm

Petari wrote:
Tue Jul 31, 2018 12:03 pm
ROM access should be exact 100% at 16 MHz. But RAM access is only 101%, lower than ROM access - and that's not realistic. Of course, real RAM access ratio depends from running SW.

ROM should be higher speeds for 16MHz.. CPU will run some instructions faster, so code runs faster, so ROM speed will show faster. Maybe not because of faster ROM itself, as still the same really, but results should still show higher for 16MHz ROM access.. It shows 100%.

When cache is on, results are 198%.. because code runs in cache, so to be expected.

When CPU runs 16MHz like on my V1 booster, RAM, ROM still technically run at same speed , but loops are much faster, so still show real world gains, and of course shows 117% ROM speed.

Actual faster ROM access gives 163% minimum in case of my V2.2 booster. So MSTE should equal that easily.. but it does not.. its still 100%!

So the MSTE should show minimum of 17% faster ROM, but it does not.. ROM is still running at 8MHz speeds there.. Similar with other tests, like whole system running 8MHz, other than when CPU is not accessing the bus, which is 16MHz speed for int-div test. That is how V2.2 booster works, but MSTE still is much behind in speeds.

Speed wise over all, its just a STE with cache.. it is nothing close to proper 16MHz machine, just 4% faster for 16MHz speeds.. clearly something very wrong with design somewhere.

Maybe someone can scope CPU clock with 16MHz enabled and see if really 16MHz or 8/16MHz switching ?
4MB STFM 1.44 FD- VELOCE+ 020 STE - 4MB STE 32MHz - STFM 16MHz - STM - MEGA ST - Falcon 030 CT60 - Atari 2600 - Atari 7800 - Gigafile - SD Floppy Emulator - PeST - HxC - CosmosEx - Ultrasatan - various clutter

https://www.exxoshost.co.uk/atari/ All my hardware guides - mods - games - STOS
https://www.exxoshost.co.uk/atari/last/storenew/ - All my hardware mods for sale - Please help support by making a purchase.

Petari
Trusted Guru
Trusted Guru
Posts: 550
Joined: Tue Nov 28, 2017 1:32 pm

Re: MSTE vs STE TOS206

Post by Petari » Tue Jul 31, 2018 4:43 pm

First to say that subjective impression after running plenty of SW at 16 MHz on Mega STE is that speed is better about 60 % in average . And running speed of usual SW is what really matters. It is much better than simple solutions with only faster CPU, no cache, where CPU works at 16 MHz only with internal operations - but they still have speed gain of some 20 % . So, I really don't see how you got that 4 % . Speed is for sure 16 MHz except when accessing RAM or ROM and cache miss. No need for scope. It switches of course all time. Here comes the proof:
16MBlitOnTOSinRAM.png
16MBlitOnTOSinRAM.png (55.27 KiB) Viewed 1365 times
RAM access and ROM access are both 200% . Not mistake. Only that I made test with TOS 1.04 running from RAM on Mega STE . And Gembench is not smart enough to see that it is not ROM - it just sees TOS start address, and that's in RAM. Graphic operations are over 200% because it is with blitter on, and compared with blitter off system.
16MBlitOffTOSinRAM.png
16MBlitOffTOSinRAM.png (73.38 KiB) Viewed 1365 times
Blitter off . RAM and ROM speed same - of course. Now graphic operation speed is real. And that's approx same as speed gain I mentioned at start. CPU runs most of time at 16 MHz, because cache hit ratio is good.

Explanation of what ROM and RAM access speed really means: it is not time (speed) of complete test code execution, but it should be (if testing SW is correct) just memory access speed, regardless from code execution speed - so regardless from CPU speed, and depending only from memory access speed.
It looks like - simplest case:
Opcode fetch: 2 cycles
Memory access: 2 cycles
But 2 cycles for opcode fetch must be not same time as for memory access - for instance because ROM access may be slower. So, to get valid result must know real speed of opcode fetch - what works in short loop, so will be in cache for sure after some 1mSec for sure.
Then can calculate real memory access speed:
Total time is: opcode fetch time + mem access time > Speed of whole code is : cycle count/total time . But must compensate it with factor:
total time/mem access time . That will give real mem access speed.
And yes, with 16 MHz only when accessing cache it is 200% with Gembench 3 - simply because it tests not larger memory block - what is indeed flaw.
On the other side, it has no sense to test with very large memory block, just because real SW operates not with it normally.
16MBlitOffTOSinRAMGB6.png
16MBlitOffTOSinRAMGB6.png (65.11 KiB) Viewed 1365 times
With GB6 it is little different. But that RAM and ROM speed difference is not real. Especially when both are actually RAM . ROM speed is not real, and I'm sure that reason to be almost 200% is same - small memory block tested.

Btw. it detects not Mega STE cache, but there is 68020/30 cache. I think that there is still more Mega STE around that ST(E)s with 68020/30 .
And because Mega STE cache is 16 KB, while in 2 32-bitters it is only some 512 bytes - guess which is more efficient. Really should be more careful before making some negative conclusions. Maybe to run some games first :D
There is 2 kind of people: one thinking about moving to Mars after here becomes too bad, the others thinking about how to keep this planet habitable.

User avatar
exxos
Site Admin
Site Admin
Posts: 8244
Joined: Wed Aug 16, 2017 11:19 pm
Location: UK
Contact:

Re: MSTE vs STE TOS206

Post by exxos » Tue Jul 31, 2018 5:50 pm

Can you try GB6 with reference machine STE20608.. blitter on.. see if you get the same results as my machine... ?

My comments are not negative, just showing what my machine does... that is not option its fact. If your new tests are different than mine then maybe may machine is faulty ?

Also GB3 totally different tests, can't compare to GB6.. RAM & ROM routines totally re-written in GB6. GB3 is far from accurate and results pretty meaningless.

I don't understand comments about saying ROM is actually RAM ?! Routines run in RAM yes, but ROM speed is direct reads from ROM into CPU registers, so little access to RAM other than read loop. If we had 16MHz RAM & ROM then results would be 200%.. but if we only have 8MHz RAM, overall results are slower for ROM access speed. We can actually see in V2.2 booster ROM is 168% not 200%, because we lose 32% because of 8MHz RAM access. So results are as they should be and can't be any more accurate than that.

GB6 does not detect external caches only internal CPU caches. You need to use CPX to change external caches.
4MB STFM 1.44 FD- VELOCE+ 020 STE - 4MB STE 32MHz - STFM 16MHz - STM - MEGA ST - Falcon 030 CT60 - Atari 2600 - Atari 7800 - Gigafile - SD Floppy Emulator - PeST - HxC - CosmosEx - Ultrasatan - various clutter

https://www.exxoshost.co.uk/atari/ All my hardware guides - mods - games - STOS
https://www.exxoshost.co.uk/atari/last/storenew/ - All my hardware mods for sale - Please help support by making a purchase.

Petari
Trusted Guru
Trusted Guru
Posts: 550
Joined: Tue Nov 28, 2017 1:32 pm

Re: MSTE vs STE TOS206

Post by Petari » Tue Jul 31, 2018 5:57 pm

I said clearly that I made tests with TOS 1.04 running in RAM - special RAMTOS version to say so . So, it accessing ROM not at all in tests above. GemBench SW (3 and 6) is just not able to detect it.
I will now do more tests. And I'm sure that your Mega STE is OK. Just need some time to realize how great machine is it :lol: That was little exaggeration. of course, but yes, I think that it is fine for year 1991.
There is 2 kind of people: one thinking about moving to Mars after here becomes too bad, the others thinking about how to keep this planet habitable.

User avatar
exxos
Site Admin
Site Admin
Posts: 8244
Joined: Wed Aug 16, 2017 11:19 pm
Location: UK
Contact:

Re: MSTE vs STE TOS206

Post by exxos » Tue Jul 31, 2018 6:10 pm

Well should not run ROM in RAM... I don't remember RAM test, maybe does RW tests and ROM is read only, so RAM speed would be slower in that case and faster for read only in ROM. GB3 maybe does read only for RAM and ROM so may explain same result there. So result you got in GB6 seems to be as expected.

You need to repeat my test in very first image and see if you get the same.
4MB STFM 1.44 FD- VELOCE+ 020 STE - 4MB STE 32MHz - STFM 16MHz - STM - MEGA ST - Falcon 030 CT60 - Atari 2600 - Atari 7800 - Gigafile - SD Floppy Emulator - PeST - HxC - CosmosEx - Ultrasatan - various clutter

https://www.exxoshost.co.uk/atari/ All my hardware guides - mods - games - STOS
https://www.exxoshost.co.uk/atari/last/storenew/ - All my hardware mods for sale - Please help support by making a purchase.

Petari
Trusted Guru
Trusted Guru
Posts: 550
Joined: Tue Nov 28, 2017 1:32 pm

Re: MSTE vs STE TOS206

Post by Petari » Tue Jul 31, 2018 7:15 pm

The problem with your tests is that blitter is enabled. It works not faster at all when CPU is at 16 MHz - and that's the reason why speed gain is so low. Tests need to be done with blitter off. And we know that blitter is not much used in diverse SW.
T206ROM_16MHz_GB3_BlitOff.png
T206ROM_16MHz_GB3_BlitOff.png (58.13 KiB) Viewed 1346 times
T206ROM_16MHz_GB6_BlitOff.png
T206ROM_16MHz_GB6_BlitOff.png (72.49 KiB) Viewed 1346 times
T206ROM_16MHz_QI22_BlitOff.png
T206ROM_16MHz_QI22_BlitOff.png (73.21 KiB) Viewed 1346 times
Above are tests with TOS 2.06 in ROM, CPU at 16 MHz, blitter off . And both GemBench v. give 200% for ROM speed, what is simply total wrong. Why same 200% as for TOS in RAM ? Because Mega STE cache obviously caches ROM access too . What is good. But tests should access larger memory area, what will lower cache hit %, and that will be close to realistic work. Actually, old Quindex gave here most realistic RAM access speed. GB3 RAM access 200% is nonsense - probably small RAM area for test.

So, you need to do your tests with blitter off .

I will make some simple rutines for real ROM and RAM access speed tests. This with Mega STE cache is hard problem - because speed depends a lot of accessed RAM size.
There is 2 kind of people: one thinking about moving to Mars after here becomes too bad, the others thinking about how to keep this planet habitable.

Petari
Trusted Guru
Trusted Guru
Posts: 550
Joined: Tue Nov 28, 2017 1:32 pm

Re: MSTE vs STE TOS206

Post by Petari » Tue Jul 31, 2018 7:20 pm

exxos wrote:
Tue Jul 31, 2018 10:43 am
dhedberg wrote:
Tue Jul 31, 2018 8:04 am
Why does the CPU only show 124% with cache OFF? Shouldn't it be close to 200% considering the frequency is 16MHz? What more test is taken into consideration besides integer divison?
int-div is the only logical result... My V1 booster has ...
Image
Thats a 8/16MHz switch.. the MSTE doesn't even get close to that :roll:
So either the MSTE has a very poor 8/16MHz switch, or CPU is running constant 16MHz with very poor state machine.. :shrug:
Seems to be its not really a 16MHz machine.. Its basically a STE with a cache giving ST-RAM a boost.. and thats pretty much it aside from 16MHz int-div test..
That's what I talked about - if no cache, only CPU is faster - then speed gain is low - about 20% . In case of Mega STE using 16 MHz without cache simply has no sense - and nobody using that setting. And RAM test with that old GemBench gave good result. While ROM test is not OK. It should be same - about 100% .
As said, in case of division, there is minimal RAM access, so speed is close to 200% . And I must say that this are not unknown things. I only was not sure about caching TOS ROM too.
There is 2 kind of people: one thinking about moving to Mars after here becomes too bad, the others thinking about how to keep this planet habitable.

User avatar
exxos
Site Admin
Site Admin
Posts: 8244
Joined: Wed Aug 16, 2017 11:19 pm
Location: UK
Contact:

Re: MSTE vs STE TOS206

Post by exxos » Tue Jul 31, 2018 7:42 pm

OK I will try with blitter off.. but in case of caches, they not give reliable results, best to turn caches off to get raw MHz power of machine.
4MB STFM 1.44 FD- VELOCE+ 020 STE - 4MB STE 32MHz - STFM 16MHz - STM - MEGA ST - Falcon 030 CT60 - Atari 2600 - Atari 7800 - Gigafile - SD Floppy Emulator - PeST - HxC - CosmosEx - Ultrasatan - various clutter

https://www.exxoshost.co.uk/atari/ All my hardware guides - mods - games - STOS
https://www.exxoshost.co.uk/atari/last/storenew/ - All my hardware mods for sale - Please help support by making a purchase.

Petari
Trusted Guru
Trusted Guru
Posts: 550
Joined: Tue Nov 28, 2017 1:32 pm

Re: MSTE vs STE TOS206

Post by Petari » Tue Jul 31, 2018 8:01 pm

exxos wrote:
Tue Jul 31, 2018 7:42 pm
OK I will try with blitter off.. but in case of caches, they not give reliable results, best to turn caches off to get raw MHz power of machine.
That's very old fashion thinking. Do it with your PC, and will see how much slower it will be . RAM speed was always lower than CPU speed since 199x .
There is 2 kind of people: one thinking about moving to Mars after here becomes too bad, the others thinking about how to keep this planet habitable.

User avatar
exxos
Site Admin
Site Admin
Posts: 8244
Joined: Wed Aug 16, 2017 11:19 pm
Location: UK
Contact:

Re: MSTE vs STE TOS206

Post by exxos » Tue Jul 31, 2018 8:10 pm

Petari wrote:
Tue Jul 31, 2018 8:01 pm
That's very old fashion thinking. Do it with your PC, and will see how much slower it will be . RAM speed was always lower than CPU speed since 199x .
PC have much more advanced cache's.. no relevance to simple caches on the ST. Internal CPU caches like on 020 will give much higher speeds than external cache on ST for example. True 020 more efficient core, but internal caches will run faster internally with same 16MHz... You can't compare L1 and L2 caches in terms of speed, totally different technology. I am not saying the caches are bad or useless, obviously works, whole point is that the MSTE is not running the CPU at 16MHz! It is not a 16MHz machine. Maybe 16MHz to cache, but nothing much else.
4MB STFM 1.44 FD- VELOCE+ 020 STE - 4MB STE 32MHz - STFM 16MHz - STM - MEGA ST - Falcon 030 CT60 - Atari 2600 - Atari 7800 - Gigafile - SD Floppy Emulator - PeST - HxC - CosmosEx - Ultrasatan - various clutter

https://www.exxoshost.co.uk/atari/ All my hardware guides - mods - games - STOS
https://www.exxoshost.co.uk/atari/last/storenew/ - All my hardware mods for sale - Please help support by making a purchase.

Post Reply

Return to “BENCHMARKS”