Announcement - new QL Interface

Nagging hardware related question? Post here!
User avatar
Dave
SandySuperQDave
Posts: 2765
Joined: Sat Jan 22, 2011 6:52 am
Location: Austin, TX
Contact:

Re: Announcement - new QL Interface

Post by Dave »

It will just be useful to know how much difference RAM shadowing following Nasta's plan can help a QL. I've done some experiments but following a different plan to Nasta's.


martyn_hill
Aurora
Posts: 909
Joined: Sat Oct 25, 2014 9:53 am

Re: Announcement - new QL Interface

Post by martyn_hill »

Hi again Dave

So, a pretty crude, repeated scroll test performed identically in SuperBasic on my (non shadowed) Issue 7 QL and the shadow-RAM Issue 5 QL gave about 10% improvement, thus:

Issue 7 (Minerva + TkII 2.23) 6447 'ticks' == 128 secs
Issue 5 (MGUK + TkII 2.23) 5765 'ticks' == 115 secs

Both QLs have zero-WS SRAM above 256k, and apart from the ROM and MB versions, are otherwise identical. If both ROMs were MG, I'd expect even higher relative performance

I used the DIYTK 20ms Timers and a tight, one line FP FOR loop scrolling a 512x256 window, thus:
T_START 1 : FOR x=1 TO 1000 : SCROLL 10 : END FOR x : T_STOP 1

SCROLLing seemed like a good but simple benchmark, due to the need to both read and write the video memory.

Better performance results would come from a pure read test, but that wouldn't be especially representative of typical video memory accesses...

M.


User avatar
Dave
SandySuperQDave
Posts: 2765
Joined: Sat Jan 22, 2011 6:52 am
Location: Austin, TX
Contact:

Re: Announcement - new QL Interface

Post by Dave »

Nice. The OS difference might really skew it since I find Minerva to often be 30-50% faster than JS. So the difference may be far more than shown by this. Minerva incorporates most of speedscreen, so screen handling is much more efficient.


martyn_hill
Aurora
Posts: 909
Joined: Sat Oct 25, 2014 9:53 am

Re: Announcement - new QL Interface

Post by martyn_hill »

Yes, agreed - I'll swap over the ROM on the Issue 5 with a Minerva +TkII tomorrow to validate that point.

(I used the MG ROM on this QL simply because I found the Network receiving by Minerva + TkII problematic in my setup and the principal reason I have two QLs is to explore QNet in preparation for a QNet/USB adapter that I'm currently developing for my QPC/laptop setup - I'll be writing-up that project in the next week or so...)

:-)


User avatar
Dave
SandySuperQDave
Posts: 2765
Joined: Sat Jan 22, 2011 6:52 am
Location: Austin, TX
Contact:

Re: Announcement - new QL Interface

Post by Dave »

I'm investigating having an FTDI USB driver IC on the hobby card I'm working on. But then, I'm also looking at the Wiznet hardware stack, separately, to provide ethernet.


martyn_hill
Aurora
Posts: 909
Joined: Sat Oct 25, 2014 9:53 am

Re: Announcement - new QL Interface

Post by martyn_hill »

Yeah - I read your thread - looks very interesting!

The QNet/USB Bridge adapter I'm working was prompted instead from a need to develop QL code more easily on my QPC/laptop and then transfer to the BBQL, without messing about with swapping the SD-Card in and out each time, or using my null-modem cable and the serial ports.

I've also always been fascinated by the QL Network - simple but elegant - especially with TkII FSERVE. The thin, 2-wire connection is also more convenient for what I had in mind.

The IPNet driver recently developed is another truly great piece of work - between emulators - but I wanted a means of communicating between emulator and BBQL using native QL protocols and available ports. Anyway, more to follow on that - I fear I may be sabotaging your original thread - sorry!

Must go to bed now.

M.


Nasta
Gold Card
Posts: 443
Joined: Sun Feb 12, 2012 2:02 am
Location: Zapresic, Croatia

Re: Announcement - new QL Interface

Post by Nasta »

Also note that the shadowed/non shadowed results depend on the speed of access to the various OS tables and SuperBasic structures which may (on a normal system) reside in screen 1, which is still slow on writes. That being said, with the scroll test, and regular 68008CPU, it will be a very minor difference.
On the other hand, the shadowed/non shadowed difference increases as the clock speed of the CPU or the add-on memory bus width increases, as now reading and writing do not have a 1:2 duration relationship, rather the reading keeps getting faster to the point where the time needed to perform a read all but vanishes in comparison with writing. However, on more advanced CPUs such as the 68020, internal parallelism may decrease the difference in some cases. Internal operations run in parallel with external bus cycles, so if complex addressing modes are used, for example, to access data being moved, the complex address calculation may actually be the bottleneck, which now gets executed internally to the CPU while the external slow access is being performed.


martyn_hill
Aurora
Posts: 909
Joined: Sat Oct 25, 2014 9:53 am

Re: Announcement - new QL Interface

Post by martyn_hill »

So, just to close the loop on the 'Scroll test' results between shadowed and non-shadowed RAM QLs, with Minerva 1.98 fitted to the Issue5 BBQL with shadow-RAM, we do get slightly better performance again, thus:

Issue 7 (Minerva + TkII 2.23) 6447 'ticks' == 128 secs - as before
Issue 5 (Minerva + TkII 2.23) 5523 'ticks' == 110 secs - an improvement of 5 secs over MG ROM - about 14% speed-up over the non-shadowed RAM unit (was 10% with MG ROM.)

I've since written a short m/c routine to further test the ram speed - I'll post here once I've run it on both units.

M.


User avatar
Dave
SandySuperQDave
Posts: 2765
Joined: Sat Jan 22, 2011 6:52 am
Location: Austin, TX
Contact:

Re: Announcement - new QL Interface

Post by Dave »

That's a 16% speed difference. Impressive.


martyn_hill
Aurora
Posts: 909
Joined: Sat Oct 25, 2014 9:53 am

Re: Announcement - new QL Interface

Post by martyn_hill »

I was never sure, when comparing difference, whether you divide the difference by the original result (128s) or the improved result (110s)... :D


Post Reply