$ dd if=amdcal-1.01.1_beta.x86_64.run of=amdcal.tar.gz bs=1 skip=16384
$ mkdir amdcal
$ cd amdcal
$ tar xzvf ../amdcal.tar.gz
$ sudo alien amdcal-1.01.1_beta-1.x86_64.rpm
$ sudo dpkg -i amdcal_1.01.1_beta-2_amd64.deb
$ dd if=amdbrook-1.01.0_beta.x86_64.run of=amdbrook.tar.gz bs=1 skip=16384
$ mkdir amdbrook
$ same as above
Read *.run and you can find the line of dd.
Sorry for my poor English.
I apologize for the incorrect note for Vista and Server 2003 support. I've informed engineering to remove those notes.
All 6xx cards are unofficially supported but we normally test on 3870s and FireStream 9170s (RV670 cards). Only the RV670 has double precision support and scatter support.
We are considering adding additional support as I have seen quite a few other flavors of Linuxes tried on this forum. We'll see what we can do for you in future releases.
Originally posted by: rahulgarg
I have just installed the new beta on win32. From the install notes, faq etc bundled with the CAL sdk, it appears the following are supported :
a) Vista and Server 2003 appear to be supported?
b) For cards, radeon 3xxx is mentioned and not 38xx as earlier. So is radeon 3650 supported for example?
c) Linux support : any plans of supporting debian and derivatives? I am currently running ubuntu and the installer has dependence on rpm. I will see if I can somehow do an install using alien and will post results in a few days.
I also would like to install this on Ubuntu 64. I am right now, unfortunately, having to run this on Windows XP.
I have positive tests with brook+ running on a 3870 graphics card. For future applications, there are 3 limitations : on-board memory, limited size of doubles and double versus single precision. The documentation says that the FireStream9170 is the solution to these problems and I am in the process of purchasing one. However I wonder how brook+ will handle that : is there a specific version without the limitation to 8200x8200 and double precision variables? this would mean that I missed a recent release. Could you please point me to the appropriate URL?
The FireStream 9170 has 2GB of memory so that should take care of your on-board memory problem.
I think the 8200x8200 is currently something somewhat tied to hardware limitations (the GPU itself). Also, what is the double precision variables limitation you are referring to (I'm slightly backlogged on emails and forum postings at the moment...)
The most recent release is v1.1-beta. See the forum link at the top of the post listings for the link to v1.1-beta.
At some point I experienced a limitation of the stream sizes. This appeared to be related to the fact that I used an out of date release of the sdk. Changing to a more recent made a dramatic improvement, with the same hardware. My count says that 8200*8200*4 is 268 megabytes, well below the 500 megabytes which I believe the 3870 video card has on board. With 2 gigabytes, it should be possible to allocate 2D streams of 5000x5000 double precision words (sqrt[2.e9)/8). We need less than that, but I wanted to make sure that it would be possible.
Looking at your announcement of 1.1-beta, I see various new interesting features, but double precision variables are not mentioned. I know that the best way to check that would be to install it and look into the documentation, but somehow I did not yet find the time to do it. Since 1.0-beta is working fine for me at the moment, I am a bit reluctant to change anything!