Newsgroups: comp.sys.ibm.pc.hardware
Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!think.com!snorkelwacker.mit.edu!ira.uka.de!fauern!NewsServ!!roell
From: roell@informatik.tu-muenchen.de (Thomas Roell)
Subject: Re: Mentor & Trident VGA cards - which one is the best ?
In-Reply-To: roell@informatik.tu-muenchen.de's message of Wed, 10 Apr 1991 10: 09:31 GMT
Message-ID: <1991Apr11.095521.26017@Informatik.TU-Muenchen.DE>
Sender: news@Informatik.TU-Muenchen.DE
Organization: Inst. fuer Informatik, Technische Univ. Muenchen, Germany
References: <1991Apr3.071713.6290@fel.tno.nl> <3725@d75.UUCP>
	<PJV.91Apr4121841@fogo.fct.unl.pt>
	<1991Apr6.192215.29729@leland.Stanford.EDU>
	<1991Apr9.004129.20991@odin.corp.sgi.com>
	<1991Apr10.100931.27295@Informatik.TU-Muenchen.DE>
Date: Thu, 11 Apr 1991 09:55:21 GMT
Lines: 46

>Yes but the problem is ONLY with scrolling. Perhaps with some chaching tricks
>the overhead could be reduced. For example with the ET4000 you have one
>read operation followed by one write operation. This means the processor has
>to wait for getting videomemory access. The ET4000 reads one line (32 bits)
>at once and buffers it. But cause of the following write this buffer is once
>again flushed. But the write will be buffered in a write-buffer internally.
>If the TVGA8900 has the same architecture the sequenece of reads and writes
>will make full use of such internal buffers. And if you read only small blocks
>of data that fit into one line of the 386 cache there might be an additional
>improvement. Concluding I assume that the overhead of the TVGA8900 of the
>ET4000 in scrolling would be about 40 to 60%. All other operations shounld
>have the same speed.

Ok, I think this was a little bit to abstract, so lets fill in some numbers.
Assume we do a simple bitblit, only 32bit words, all correctly aligned, no
shift necessary. Under this assumptions a ET4000 makes about 1.7 MBytes/sec.
This was measured on a 33MHz 386. Thus one movsl I used for blitting too about
78 cpu cycles. Thus the raw VGA accesses took about 74 cycles.
Now to the TVGA8900. I assume that a read and a following write are as fast as
for the ET4000, 74 cycles. But here we have first to buffer the stuff:

	74	VGA read/write
	 4      movsl for buffering
	 5      storing the buffered data into memory
	 4	movsl for bltting from memroy into VGAs.
	 0      reading data from cache
        --
        87      cycles

This means that the TVGA8900 is at least 12% slower. At an average I assume
about 20% slower performance than the ET4000 is they have the same memory
troughput.

But if you really need speed, forget about VGA. Buy a 8514. Benchmarks:

	VGA 640x480x256          6600 xstones	ET4000
        VGA 1024x768x256         4800 xstones
       8514 1024x768x256	26300 xstones	WD9510

- Thomas

--
_______________________________________________________________________________
E-Mail (domain):	 roell@lan.informatik.tu-muenchen.de
UUCP (if above fails):   roell@tumult.{uucp | informatik.tu-muenchen.de}
famous last words: "diskspace - the final frontier..."
