FYI: Tyan S2885 with more than 4GB: New BIOS
FYI: Since some people reported problems with more than 4 GB and AGP on a Tyan S2885: The new BIOS at ftp://ftp.tyan.com/Chipset_AMD/K8/2885BIOS/2885001a.zip is supposed to fix these issues (I haven't tested it myself yet). Andreas -- Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj SuSE Linux AG, Maxfeldstr. 5, 90409 Nürnberg, Germany GPG fingerprint = 93A3 365E CE47 B889 DF7F FED1 389A 563C C272 A126
FYI: Since some people reported problems with more than 4 GB and AGP on a Tyan S2885: The new BIOS at ftp://ftp.tyan.com/Chipset_AMD/K8/2885BIOS/2885001a.zip is supposed to fix these issues (I haven't tested it myself yet).
I just tried this out. Conclusions: The Missing Memory is addressed (set a "Software Memory Hole" in BIOS setup) AGP performance is now identical regardless of how much memory you have, but fast write performance is still not good (4x is 280 MB/s, 8x is 333 MB/s). Problems: Using the Software Memory Hole to get your full RAM available quickly leads to system crashes as soon as you set write-combining on the video framebuffer (so presumably X will crash it with a nonzero memory hole, but I haven't tried this). Things to do: Mess with memory hole/AGP aperture to find a stable configuration for write-combining the framebuffer. Compare AGP performance to Iwill 8131/8151 dual-Opteron board when it arrives -- bad b/w might just be a fastwrites problem with the 8151? Write a benchmark that does framebuffer writes as AGP DMA texture transfers and compare results with fastwrites. Dump the bridge configuration registers and figure out just how things are getting set up, why all the weird MTRR entries are there, and if there's a plausible reason for the still-unexpectedly slow (but at least uniform) AGP fastwrite performance I'll poke around with the memory hole later and see if I can figure out the reason for the crashes; these feel just like the crashes I was seeing before when changing the weirdo MTRR entries that the 1.01 BIOS set up. Interestingly, the default MTRR (without a memory hole) is even crazier now: reg00: base=0xc0000000 (3072MB), size= 512MB: write-combining, count=1 reg01: base=0x00000000 ( 0MB), size=2048MB: write-back, count=1 reg02: base=0x80000000 (2048MB), size= 512MB: write-back, count=1 reg03: base=0x100000000 (4096MB), size=2048MB: write-back, count=1 reg04: base=0xa0000000 (2560MB), size= 128MB: write-back, count=1 reg05: base=0xa8000000 (2688MB), size= 64MB: write-back, count=1 reg06: base=0xac000000 (2752MB), size= 32MB: write-back, count=1 And with a 2GB memory hole: reg00: base=0xc0000000 (3072MB), size= 512MB: write-combining, count=1 reg01: base=0x00000000 ( 0MB), size=2048MB: write-back, count=1 -mcq
Hi! That's a very good bios. I have not discovered any new problems yet. The graphics system using an NVIDIA 5600 card got a speed-up from 10fps to 50fps (720x576 pixel) when using plain x11 and 20fps to >100fps when using sdl. AGP is now automatically detected and used (NVIDIA_GLX and NVIDIA_kernel). 6GB are detected and used. Thank you very much for the info! Marcus On Monday 15 December 2003 20:53, Andreas Jaeger wrote:
FYI: Since some people reported problems with more than 4 GB and AGP on a Tyan S2885: The new BIOS at ftp://ftp.tyan.com/Chipset_AMD/K8/2885BIOS/2885001a.zip is supposed to fix these issues (I haven't tested it myself yet).
Andreas
-- Marcus Barkowsky Telecommunications Laboratory Multimedia Communications and Signal Processing University of Erlangen-Nuremberg Cauerstrasse 7 91058 Erlangen Germany Tel: +49 9131 85 27103 Fax: +49 9131 85 28849 www.nt.e-technik.uni-erlangen.de
participants (3)
-
Andreas Jaeger
-
John McCorquodale
-
Marcus Barkowsky