PDA

View Full Version : Radeon/Mediator as FastMem (HowTo)

Ratte
07-11-2009, 06:09 AM
This guide is for ZIII-Amigas (A3000/4000) only !!!

http://www.a1k.org/forum/showthread.php?goto=newpost&t=17108 (org. german-guide)

Even before I bought a Mediator for my A4k (Mediator4000Di) I thought about the idea to add some of the gfxmemory from a radeon to the amiga memorypool.
Last night I had some time to try it out.
(If there is allready a guide or tool out there ... i didnt found it.)

My current setup is a A4k (Blizzard4030/50Mhz) with 2 mb chipmem & 16MB fastmem (onboard) and a Mediator4000Di with a 64MB Radeon 9250.

How to ...
First open/create the file "sys: prefs/env-archive/mediator/radeonmem" and replace/insert the comment "15 MB" direct at the begining of the file.
Save the setting and reboot to activate it.

Now your radeon as "only" 16MB graphicsmem (doublesize of a cvppc).
1 MB is reserved for PCI-communication ( thats the reason why we used "15 MB" ).
All other memory is unused (64MB Radeon = 48MB unused / 256MB Radeon= 240MB unused)

Next step:
Start the tool "CheckBoards" from the Picasso96-archiv (its inside of the "debug"-directory).
CheckBoards should find a gfx-board named Radeon with ? 3dfx Chipset (driver-bug) ? and a memorysize of 15712256 bytes if you dont made any mistakes.
If you own more than one pci-card, keep the "MemoryBase" of the gfxboard in mind!

With the Mediator-Software you received a tool named "pciinfo" in your c:-directory.
If you call "pciinfo" youŽll get a full list of all installed pci-cards including the used memoryareas.
Compare your "MemoryBase" from CheckBoards with the MemoryAreas from pciinfo.
If you found the correct pci-card keep also the end of its memoryarea in mind.
My 64MB Radeon is (for example) at $5000.0000 - $53ff.ffff.
Our current setting is using only the first 16MB for gfx ... $5100.0000 - $53ff.ffff are unused and free!!!

Last step is to add the memory to the memorypool:
Download the "addmem" tool from aminet.
And add the following line to the end of your "user-startup" inside of the sys:s-directory:
addmem 51000000 to 53ffffff pri 10 name pci-memory
The memoryareas could be differ from my example, it depends on your setup.

A "avail flush" could be nice behind the addmem-command,but it is not really needed.
Reboot and ...
... Done.

Have fun with your cheap memexpansion.
(btw. os4-research cant be done, as long as iam waiting for my repaired csppc from a german reseller)

regards
Ratte / a1k.org
cv643d
07-11-2009, 07:56 AM
Nice, so you could add 240 MB fast mem for free ? :)

Is the mem as fast as Z3 memory?
aggro_mix
07-11-2009, 10:43 AM
So Radeon 9250 works "out of the box" in Mediator 3000/4000?
AmigaMance
07-11-2009, 11:01 AM
I wish this could be done with CyberGraphX too.
Piru
07-11-2009, 01:03 PM
Is the mem as fast as Z3 memory?
Much slower, especially when reading. VMEM is not meat to be read.
Ratte
07-11-2009, 02:50 PM
Much slower, especially when reading. VMEM is not meat to be read.

Yes, you are right.
Today I found some time to make some tests.
I started "Analyser" and made MemTests for over an hour ... no errors.
On a1k.org we had an older discussion regarding a (new) ZIII-memoryexpansion and we had collected some benchmarkresults from different memorys and expansions.

The Radeon/Mediator-Combo seems to be comparable to the DBK3128.
The user ville9/a1k.org send us results from his DKB3128.
reading 7.7 mb/s & writing 4.6 mb/s
The Radeon/Mediator results are:
reading 5.1 mb/s & writing 7.7 mb/s
A FastlaneZIII is faster ...
reading 15.5 mb/s & writing 8.6 mb/s

In normal use it is not different to any other memory.
But it is not comparable to memory on cpu-cards!

On the other side, this memory is really cheap ...
cv643d
07-11-2009, 03:52 PM
Yes, who would say no thanks to 240 MB slow mem? :)
HammerD
07-11-2009, 05:19 PM
No, you need a true 5V card. Best to buy it from Elbox's online store to ensure the card will work.
Akiko
07-11-2009, 06:07 PM
This would be absolutely invaluable for CyberstormPPC OS4 classic users! :-)

Hopefully the next update from Hyperion will see native Radeon 9250 drivers.
Ratte
07-12-2009, 12:04 AM
The delivered Radeon9250 64MB bundled with the Mediator (bought 1-2 month ago) is a normal 3,3V Radeon.
Elbox modified the Card with a powerconverter ...
... on the same way it can be done with a 256mb version from ebay. :)

btw. the radeon.card driver for p96 is for 9200 and 9250 radeons.

pictures from the bundled radeon-card with powerconverter
http://www.a1k.org/forum/showpost.php?p=264387&postcount=31
(click on the pictures for fullsize)

ragards
Ratte/a1k.org
Gulliver
08-30-2009, 08:35 AM
Why will it only work on ZIII? Could it also work on an accelerated (030/040/060) Amiga 1200 with a mediator?
Piru
08-30-2009, 09:08 AM
Why will it only work on ZIII?
ZIII can be addressed directly.
Could it also work on an accelerated (030/040/060) Amiga 1200 with a mediator?
No
Kronos
08-30-2009, 09:13 AM
The A1200-Mediator only has an 8MB-Zorro-window to the A1200...

One could do some heavy MMU-trick, think about virtual-memory that isn't paged to disk but to VMEM, but I doubt you'll find anybody interested in writing such a hack.
mongo
08-30-2009, 09:23 AM
The A1200-Mediator only has an 8MB-Zorro-window to the A1200...

One could do some heavy MMU-trick, think about virtual-memory that isn't paged to disk but to VMEM, but I doubt you'll find anybody interested in writing such a hack.

If someone wants to buy me a Mediator for my A1200, I'd be happy to write it.
Piru
08-30-2009, 10:00 AM
If someone wants to buy me a Mediator for my A1200, I'd be happy to write it.
You're aware of the fact that many applications allocate memory as assume it doesn't get paged away? And that MEMF_PUBLIC doesn't work as a "swappability" flag?
mongo
08-30-2009, 10:51 AM
You're aware of the fact that many applications allocate memory as assume it doesn't get paged away? And that MEMF_PUBLIC doesn't work as a "swappability" flag?

The application would never know if the memory was paged or not.
Piru
08-30-2009, 12:05 PM
The application would never know if the memory was paged or not.
It isn't that easy. Even the most advanced virtual memory system ever written for amigaos doesn't work right for all applications. You need to have blacklists of misbehaving applications (or whitelists of well behaving ones).

Just look into http://aminet.net/package/util/misc/VMM_src and you'll soon figure out it is hardly trivial, nor can it be made 100% transparent.
mongo
08-30-2009, 12:19 PM
It isn't that easy. Even the most advanced virtual memory system ever written for amigaos doesn't work right for all applications. You need to have blacklists of misbehaving applications (or whitelists of well behaving ones).

Just look into http://aminet.net/package/util/misc/VMM_src and you'll soon figure out it is hardly trivial, nor can it be made 100% transparent.

Huge difference between paging memory from disk and paging memory from bank switched RAM.
Kronos
08-30-2009, 12:27 PM
Huge difference between paging memory from disk and paging memory from bank switched RAM.


Other than it should be faster actually not .......

One could send out memory-pages per serial cable to another computer, would still make no differnce (expect for speed).
mongo
08-30-2009, 12:34 PM
Other than it should be faster actually not .......

One could send out memory-pages per serial cable to another computer, would still make no differnce (expect for speed).

No!

To page memory from disk requires that the OS be running and in a state that allows you to do disk I/O. To page from bank switched RAM, you don't need the OS at all. This is why you don't have the same problems as with paging from disk.
Piru
08-30-2009, 12:46 PM
Huge difference between paging memory from disk and paging memory from bank switched RAM.
That part is true. It removes the trouble of accessing the memory while in Forbid/Disable. That, however, is just one of the problems.

Additionally the performance would be really bad. Much slower than chip memory.
clusteruk
10-08-2009, 02:19 AM
Has anyone notice C:Iprefs crashing on bootup with this patch.
clusteruk
10-09-2009, 04:35 PM
Ignore last post, I believe it has nothing to do with the memory adding. My OS3.9 is screwed so I am going to reinstall it.
Karlos
10-09-2009, 06:12 PM
One could send out memory-pages per serial cable to another computer, would still make no differnce (expect for speed).

There is probably an OpenVMS fan somewhere that has done that :lol:
johnim
11-25-2011, 05:26 PM
any new on how to get 2 radeons working together on a a1200 mediator