Bettering Radeon Gallium3D Performance With PCI-E 2.0

[AnonymousCoward]

galium3d is supposed to do the same -share code among OSes- but i haven't seen anyone jumping on it and releasing an FOSS driver for all OSes

What systems would that be? For Windows and Macos it'll presumably take a lot of work to get them working at all, and then the proprietary drivers will still be better, and all those small niche systems like Haiku and so on will probably be hard-pressed to come up with the required manpower.

[liam]

But I imagine looking at e.g. the power saving code could still tell a very good programmer why radeon on the "low" profile still uses more power than fglrx.

I mean, why doesn't AMD simply have somebody going over the code of catalyst, deleting everything patented or "secret" and release the nonfunctional rest? Could still be helpful for low level stuff like power saving and communicating properly with the hardware...

Do you have any idea who long that would take? I think Bridgeman has said catalyst includes something like 10s of millions of lines of code (IIRC). Going through that, line by line, and having legal agree with you, would take many years.

[smitty3268]

sure you are a nice dev but economic isn't your strength.

In economic you deal with the future not with the past or with the presence.

this means the grow factor is more important than the market-share.

right now linux do have 3,5%(2%andorid) market share (source: wikipedia.org) and a grow rate of 40% every 6 month (source: Net Applications ).
and market analyst found out that nearly 80% of all companys wana grow there opensource usage to lower costs. (source: linux foundation market survey)

just do the mathematics: calculate grow rate 40% every 6 month on the market share 1,5%

this numbers prove : Microsoft is allready death

AMD doesn't have any hardware in phones/tablets, so you can basically take Android out of those numbers, and end up with very little left over.

[smitty3268]

the point is: amd want the andorid part to.

Sure, but they're probably still at least a couple years out from getting that hardware. It's completely fair to say the market for software drivers isn't there yet since they don't have anything.

I think their last CEO was fired basically for the decision to sell of their mobile graphics division to Quallcom. It will take them some time to recover from that decision.

[ryszardzonk]

So what are the plans for using pcie_gen2=1 by default? Couldn't the incompatible motherboards be blacklisted instead of turning it off for everyone?

my thoughts exactly. Sam goes for Intel RC6 and others just like they went with ASPM

[Drago]

AMD looses some very important point here. Now with not optimal FOSS driver, and since APU initiative, they loose not only GPU sale, but CPU and chipset sales.
For people that don't care, there is always some Linux geek whom they consult(or just the salesman), and he will tell them "Yes do choose intel, they are flawless"
I like AMD, but if Trinity fails, I would go intel way.
On the topic...can't this PCI-E 2.0 setting being enabled at least for the APUs(Brazos,Llano)? They have internal PCI-E link, so no motherboard involved.

[Drago]

sure you can activate this option with an Llano.

I meant by default.

[HokTar]

sure you can activate this option with an Llano.
and the Trinity opensource announcements come tomorrow or the day after tomorrow.

i think they will grow the opensource team and focus on openCL and video acceleration.

but in fakt the biggest impact will not be in the "trinity" generation because its a VLIW based gpu architecture like the hd6970.

the successor of trinity will be the one with the greatest open-source impact.
because then its a hd7970 based gpu.

Wait!
This time you actually make sense! For the following reason:
AMD will switch to FSA (Fusion Systems Architecture), which they renamed yesterday or today to HSA (Heterogeneuos Systems Architecture). This is their new intruction set that is open, meaning that anyone can support it. It's better than Intel's, because they let all supporters to have a voice in it's developement.
Here comes the rational for the renaming: Fusion is already accociated with AMD which is bad for other backers, e.g. ARM. So they simply changed it -- the hardware will come in a few years anyways, so no problem.

Now, here's the big news: HSA is an ISA that is an extension of OpenCL.
So your OpenCL code will run pretty efficiently on these systems.
The catch is that the architectures for the CPU and GPU part will be fully integrated, will have the same address space, etc. Hence fglrx would be needed for your CPU as well!

Hordes of Linux users may not give a crap about fglrx today but everybody will be pissed if they can't even boot a live CD without a blob!

Can you see now why they invest in OpenCL and why would they further increase that?

+1: HPC industry wouldn't like problematic binary blobs, especially for CPUs. Don't forget that the market share of Linux is about 90% there.

Edit:
+2: Have you noticed that there's _no_ patent issues whatsoever with OpenCL?

/me hoping for the rise of Mesa in the following years!

[agd5f]

On the topic...can't this PCI-E 2.0 setting being enabled at least for the APUs(Brazos,Llano)? They have internal PCI-E link, so no motherboard involved.

This option is irrelevant on APUs since they are not really on the PCIE bus per se. It only applies to discrete cards.

[bridgman]

On the topic...can't this PCI-E 2.0 setting being enabled at least for the APUs(Brazos,Llano)? They have internal PCI-E link, so no motherboard involved.

I don't think the GPU goes through an internal PCIE link so wouldn't expect any performance increase from enabling the option on APUs :

http://realworldtech.com/page.cfm?ArticleID=RWT062711124854&p=2

EDIT - Oops