exxos's Atari, Amiga & retro forum

Steve wrote: 22 Feb 2021 19:56 @Badwolf Already have, the second Gembench picture (these were all with Emutos)
@stephen_usher Is that with a blitter installed? Also yeah... mine is 200% slower lol, surely that must be a bug in the latest firmware.

No, no blitter.

By the way, I'm not sure why you have a heatsink on the 68030 as it hardly gets warm.

Steve wrote: 22 Feb 2021 19:56 @Badwolf Already have, the second Gembench picture (these were all with Emutos)
@stephen_usher Is that with a blitter installed? Also yeah... mine is 200% slower lol, surely that must be a bug in the latest firmware.

Ah, brilliant. Many thanks. I'll use that as my standard reference, then.

Some better (my underlying architecture is 16MHz rather than 8) others a bit down. I covet that memory speed, though! That's probably the difference. I wonder how much effect burst mode support has?

Cheers,

BW.

stephen_usher wrote: 22 Feb 2021 20:52

Steve wrote: 22 Feb 2021 19:56 @Badwolf Already have, the second Gembench picture (these were all with Emutos)
@stephen_usher Is that with a blitter installed? Also yeah... mine is 200% slower lol, surely that must be a bug in the latest firmware.

No, no blitter.

By the way, I'm not sure why you have a heatsink on the 68030 as it hardly gets warm.

Mine does get warm and will crash without it.

stephen_usher wrote: 22 Feb 2021 20:50 True for the GEM part as the AES is faster in TOS 2.06 but not the raw CPU.

As you can see it was 880% stock STFM in that benchmark and after the firmware update it dropped to 606%.

Those figures look wonky, @stephen_usher .

The 'CPU' section of the statistics considers Integer Division, Float Maths (not applicable in either case), RAM, ROM and ALTRAM tests.

The only difference in the timings of these sections between your screenie and Steve's is a ~1% drop off in RAM Access (31.645 seconds vs 31.290). Tiny.

However, your screenshot evaluates AltRAM at 0% relative and Steves's evaluates it at 847% for *exactly the same figure* of 8.580. I think we can deduce from this that the TOS 2.06 STE @ 8MHz reference archetype contains an AltRAM stat and the TOS 1.04 reference archetype you're comparing against doesn't.

Quite how that affects the headline CPU percentage I don't know, but it clearly breaks it somehow as the difference doesn't exist in the underlying timings.

I think it's best to take that headline figure with a pinch of salt anyway, but you also look to be on different GB6 versions -- perhaps different averaging methods in use too?

BW.

Badwolf wrote: 23 Feb 2021 10:36 Those figures look wonky, @stephen_usher .

Yep. @stephen_usher your using a really old build of GB6. The tests themselves could may well be different under the hood. Alt-ram benchmarks were only added in much later builds. So 0% is being taken into account on the overall scores as BW says.

https://www.exxosforum.co.uk/atari/last/GB6/index.htm

Also @Steve is using a blitter benchmark file. So the blitter running and being taken over by the CPU can affect the scores as well. You really have to be careful what your comparing against.

... to compare with a 25 years old hardware design

PAK68/3-030 - 50Mhz with 64MB Fastram

... to compare with a 25 years old hardware design

PAK68/3-020 - 40Mhz with 64MB Fastram

exxos wrote: 23 Feb 2021 11:27 You really have to be careful what your comparing against.

Completely agree @exxos. I have seen this a lot on many forums, that the reference (test) file being used is often different so results are not comparable, like comparing apples to oranges.

It would be good to have a standard to test everything against, i.e. what version GemBench, what version TOS and BLITTER.

So far on this thread I have seen systems tested against TOS2.06 and TOS1.04 with BLITTER enabled or disabled etc. The results would be very different if everyone were to use the same reference system (test file)

Icky wrote: 23 Feb 2021 15:00 So far on this thread I have seen systems tested against TOS2.06 and TOS1.04 with BLITTER enabled or disabled etc. The results would be very different if everyone were to use the same reference system (test file)

Yeah, I think with GB6 having so many ways of doing things, people are mixing it all up. Technically "stock" should be TOS104 no blitter, or TOS104 with blitter.. but results can differ because of TOS version... I am tending to use the STE with TOS206 with blitter on as the main reference file. But it really depends on what the user wants to compare against vs what machine. I think some think "load GB6 and run it - job done" , but GB6 doesn't work like that.. thats just a bad habit from GB4 days which didn't have any proper references to start with.

Its all one reason why I put the GB6 build number in the title bar some time ago. As earlier builds can differ from later builds.

I've always tested against 'DEFAULT' assuming others probably would/should do the same. I think there are only two logical test methods: current system vs stock same system & current system vs gembench default.