home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.ibm.pc.hardware
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!usc!cs.utexas.edu!qt.cs.utexas.edu!yale.edu!ira.uka.de!ira.uka.de!math.fu-berlin.de!informatik.tu-muenchen.de!roell
- From: roell@informatik.tu-muenchen.de (Thomas Roell)
- Subject: Re: [IBM] XGA-2 Specs.
- In-Reply-To: kaul@bcrvm1.vnet.ibm.com's message of Thu, 12 Nov 92 23:18:56 GMT
- References: <1992Nov11.082729.15062@hippo.ru.ac.za> <1992Nov12.231856.19708@watson.ibm.com>
- Sender: news@Informatik.TU-Muenchen.DE (USENET Newssystem)
- Organization: Inst. fuer Informatik, Technische Univ. Muenchen, Germany
- Date: Mon, 16 Nov 1992 08:21:17 GMT
- Message-ID: <1992Nov16.082117.15908@Informatik.TU-Muenchen.DE>
- Lines: 92
-
- > The "official" specs are"
- > Max video PEL rate: 90MHz
- > Max screen resolution: 1280x1024
- > Max palette colors: 16.7 million (at 1024x768 - 256 colors simultaneously)
- > Max palette greys: 256
- > Max direct colors: 65,536 -- at 640x480 (at up to 75Hz refresh)
- > -- at 800x600 (at up to 60Hz refresh)
- > Video Memory: 1 MB VRAM (which is why you only get 256
- > colors at 1024x768)
- > Adapter Type: 16/32-bit bus master
-
- I think I should followup on this ... The 16/32 bit busmastering won't
- help you in a ISA system, since normally the whole address space of
- the ISA bus (16MB) is filled with your normal DRAM ... From what I
- have seen this feature is just a way to force clone manufactures to
- waste more gates on a compatible feature that hardly anybody would use
- (think about busmastering in a virtual memory system).
-
- > > Is it really as fast as
- > > the ATIs of this world and how does it compare to fast framebuffer
- > > cards like ET4000s -- I just want to get my appetite whetted.
- >
- > The XGA doesn't do well on the WINMARK benchmark (pulling only about a
- >> 10), but that's because of the weighting of that benchmark. On
- > practical applications and programs it's much faster. And this is
- > from a guy who has an ET4000 on his home 486/33 comparing to his work
- > 386/25. I believe we have a study on several Windows programs
- > comparing the timings, and the XGA is very, very strong on the mix of
- > operations shown by real programs. This is especially noticable on
- > OS/2 text in a window.
-
- Hmmm ... I agree with you that MS-Windows benchmarks turn out to be
- real hype. They won't tell you to much. Hence here are two number the
- should give you some idea of the speed. I have measured solid fills of
- 400x400 rectangles. No pattern, just COPY mode, no clipping, no
- planemasking. The second test was a aligned bitblit from screen to
- screen, using no pattern, no clipping, no planemask, just COPY mode.
- These two tests show basically the rough performance of the board.
- What the driver writes make out of these for your application is
- something different:
-
- ET4000 XGA-2 481(8514/A clone)
-
- Fill: 4 34 72 MegaPixels/second
- BitBlt: 2 14 31 MegaPixels/second
-
- Just for reference, S3 boards with 80ns VRAMs are at the same speed as
- the XGA-2, and the Ultra Pro is just a little bit slower than the 481
- in this case. (Sorry for this 481 here, but I don't have the exact ATI
- numbers online)
-
- > As far as future enhancements, there will be some. That's about all I
- > can say :-)
-
- Ok, here is my whishlist (from a X-Server developer, not MS-Windows ;-)):
-
- - reintroduction of some sort of command queue. Right now you
- have to wait for the XGA to finish it's operation before you
- can write to the XGA registers to set up a new one. That
- means you cannot have the same concurrency as for the 8514/A
- where you just could stuff everything in the command queue,
- and let the 8514/A work pretty much interleaved with the
- driver.
-
- - some sort of HW pattern registers. Right now if you do a
- patterned fill, the pattern is in VRAM. Hence you have
- normal bitblit cycles to do the fill (i.e. RMW). If you had
- pattern registers, you wouldn't have to relead the pattern
- from VRAM, and if the rasteroperation would be a simple copy
- mode, then normal write cycles would be all that's necessary
-
- - extend the XGA internally at least to use 16 bit
- corrdinates. If you draw for exaple lines in a drive that
- allows you 16bit coordnates, you cannot simply use HW
- clipping, since the XGA has only 12bit coordinates (and
- wraparound for huge coordinates would occure). Hence you
- have to do at least SW preclipping against the device coordinate
- space, which does cost unecessary time.
-
- - there is no way to copy a single plane of pixels of a n-bit deep
- image to another plane in the same image. This was possible
- in the good old 8514/A (using the read-mask).
-
- - 1280x1024 non-interlaced at 256 colors (read: support for 2MB
- VRAM)
-
- - Thomas
- --
- -------------------------------------------------------------------------------
- Das Reh springt hoch, e-mail: roell@sgcs.com
- das Reh springt weit, #include <sys/pizza.h>
- was soll es tun, es hat ja Zeit ...
-