• 1(current)
  • 2
  • 3
  • 4
  • 5
  • 17
AMD Fusion based hardware for a HTPC?
#1
Fusion products (Ontario & Zacate to start) are soon starting to ship to OEM's. How ready is XBMC for this new architecture? I am especially interested in Linux...

Is XvBA and / or vaapi implementation in XBMC ready for this new hardware? Will ALSA drivers handle sound?

And what about the drivers for the GPU? Catalyst drivers? Or the open source drivers are easier with XBMC? Luckily, normally Ati is a lot more open than NVidia regaring GPU specifications so I won't foresee any troubles related to Fusion GPU drivers.

Thanks for them who provide inside information. (I know most knowledge in this forum is based around VDPAU and NVidia but hopefully this thread will expand readers views on hardware acceleration related to AMD/Ati and XBMC)

btw: It seems that Fusion has an ION-like tdp and performance-wise blows away CORE i5!

Info:
http://forums.guru3d.com/showthread.php?p=3707221
http://www.avsforum.com/avs-vb/showthrea...p=19190748 (with some specs)
http://www.anandtech.com/show/3920/amd-b...an-core-i5 (don't forget the follow-up article)

Thnx in advance!
Reply
#2
ati graphics and linux still don't play nice.. for an HTPC I wouldn't get anything other than nvidia graphics.
Reply
#3
PatrickVogeli Wrote:ati graphics and linux still don't play nice.. for an HTPC I wouldn't get anything other than nvidia graphics.

This all has to do with drivers and hardware acceleration in XBMC. If mentioned techniques are in place, you won't hear a thing about ION and NVidia.
Reply
#4
Robotica Wrote:Luckily, normally Ati is a lot more open than NVidia regaring GPU specifications so I won't foresee any troubles related to Fusion GPU drivers.
I was a bit taken aback when I read that! Neither company releases much technical info regarding their GPU's for fear of giving away their proprietary advantage. However, Nvidia has been much (much, much, much, much) better at supporting Linux than ATI. Running an ATI card is a nightmare on Windows, doing it on Linux brings new meaning to the word "masochism". I'd thoroughly enjoy being wrong about these new chips, but there's been no evidence in the past of ATI being anything other than second rate (at best) on Linux hosts.
Reply
#5
ati have never cared much about the linux community. What makes you think the suddenly do now? Just dont get your hopes up
Reply
#6
teaguecl Wrote:I'd thoroughly enjoy being wrong about these new chips, but there's been no evidence in the past of ATI being anything other than second rate (at best) on Linux hosts.

That isn't completely true. AMD and Nvidia take different paths with Linux.

Nvidia puts out ONLY their closed source driver for Xorg (Linux's display system). But this closed source driver is full featured: Nvidia had composite in their driver before anyone else, OpenCl in their driver before anyone else, and VDPAU working a year before VAAPI mattered.

Nvidia is bad for Open Source fans because they don't open up their specs or driver code, but they are awesome for HTPC fans because quite simply when their feature work, then work better than anywhere else.

AMD (post ATI purchase) puts out (a pretty crappy) closed source driver, and then releases all the specs of their cards so that open source developers can make proper open source drivers.

This opening of the specifications on their cards is the where the OP gets the claim that "ATI is more open."


But with all that said, I personally will NEVER buy an AMD GPU. Why?

Because time has shown that the old Open Source adage of "open up the specs and the community will write the drivers" has turned out to not apply to GPUs as it does simpler things like NICs. Despite AMD's specs being in the open, more than a year after the specs were opened AMD's open source drivers still don't have all the features that AMD's closed source drivers have, let alone Nvidia's drivers.

I don't want to take anything away from the AMD team that makes the driver- they have done amazing work! But making a graphics driver is hard, and its difficult to dedicate enough time unless you are getting paid to do it. Nvidia pays a pretty large and talented team to create their closed source driver, and for that reason it remains the best Linux GPU driver (by features) in existence.

I will add that in my years of being a Xorg nerd, I have always been impressed with the Nvidia developers. Look up the contributions the VDPAU developer has made in the community- they make Nvidia seem like a genius factory!

In summary, even though AMD is more open for HTPC use Nvidia is currently the most practical option and it looks like it will stay the most practical option in the near future.

Reply
#7
Robotica Wrote:And what about the drivers for the GPU? Catalyst drivers? Or the open source drivers are easier with XBMC? Luckily, normally Ati is a lot more open than NVidia regaring GPU specifications so I won't foresee any troubles related to Fusion GPU drivers.

Xorg GPU acceleration outside of VDPAU is a nightmare.

VAAPI is nothing more than a shim for other various backends (such as VDPAU) so its no magic bullet. What matter is what happens underneath.

And underneath AMD's closed source driver still completely sucks compared to NVidias. Those who try to use VAAPI still run into tons of rendering errors, now almost a year after VAAPI has been in the open.

Is it getting better? Yes.
Will one day it work like VDPAU? Hopefully.
Does it today? No way.

I don't suspect that AMD's open driver effort will help much- the specs have been out for more than a year and just recently have decent drivers for 3D been released by the community. Since it took so long to master 3D, I'm pretty sure it will be a few more years until the open driver has VAAPI support with near VDPAU performance.

You are talking like the open drivers are like a light at the end of a tunnel, when really the only thing that has been proven in the time since AMD released the specs is that Nvidia might have been right that GPU drivers are too hard for the community to produce at a high level.

Honestly I think these new AMD nettops will accelerate the community focus towards the Windows editions. I am sure most will come with Windows 7, which means easy XBMC support without fighting. Add in the fact that the Windows version (eventually) will have a monopoly on HD bitstreaming support and I don't think AMD's poor Linux support will be a problem.

Of couse obviously you want to run Linux, as do I. We need to stick to Nvidia hardware for the time being, unless your purpose for buying is to help development.

Its like composite with Xorg. Nvidia had Xorg composite support in 2004. The community drivers had decent composite support by around 2007/2008. I imagine the community will have an equal or larger lag for GPU decoding support.

Reply
#8
Robotica Wrote:This all has to do with drivers and hardware acceleration in XBMC. If mentioned techniques are in place, you won't hear a thing about ION and NVidia.

if they are in place.. which I highly doubt. I might be wrong, though.
Reply
#9
Thanks for your valuable answer poofyhairguy.. Lot's of inside info. I was looking for.....

Hopefully, this Fusion platform will speed up development off those open source drivers since hardware specs look awfully good.... Either way, I don't expect problems in playing 50 Gb Blu rays by either CPU or GPU so this platform has my attention for now.
Reply
#10
Just checked some development highlights regarding this subject:

Quote:The hopes with VA-API are to ultimately replace X-Video Motion Compensation as being the predominate video API for Linux systems. Unlike XvMC that is limited to MPEG-2 and IDCT / Motion Compensation, VA-API is designed to support other standards such as MPEG-4 and VC-1 along with accelerating IDCT, Motion Compensation, VLC, bitstream processing, spatial-temporal deinterlacing, inverse quantization, and other functions.

1> vaapi decoding in XBMC relies on FFmpeg;
2> vaapi is implemented in FFMPEG/Mplayer in december 2008; The libVa library provides the VA-API implementation (also in Ubuntu Mobile, Intel's Moblin)
3> vaapi is activly developed: http://www.splitted-desktop.com/~gbeauch...yer-vaapi/
4> there is some XvBA development named xvba-video, which is an XvBA back-end for VA-API.
5> VAAPI/XvBA playback in XBMC is much worse then in Mplayer: Looking at XBMC forum support or dev trunk that isn't strange.
6> FFMPEG isn't very active on VAAPI/XvBA

Source:
http://www.phoronix.com/scan.php?page=ar...aapi&num=1

My conclusion:
I think AMD hardware acceleration can be a lot better if devs are willing.... Besides, AMD attitude towards Linux is likely to change since releasing the Fusion platform (in answer to ION), will create a lot more (not just gaming) Linux demand for hardware acc. Just look at the HTPC possibilities....
Reply
#11
First of all Robotica I want to say thank you for a high level Xorg conversation. I love Xorg stuff. Let kick it up a notch.

My personal opinion is that Intel is gumming up the works, and Nvidia keeps setting a faster pace.

You are right that VAAPI support has existed in FFMPEG for a while, and that there are some people in the Mplayer community working on it. It has also been worked on by the XBMC devs, I know you have seen the big thread. Me too. And from reading that thread it seems it isn't quite there yet.

From my understanding the basic holdup with VAAPI is that its not stable enough yet. I mean the actual VAAPI API is stable, there is no problem there. What instead is the problem is that it has been hard to make it so that the many different types of hardware covered work well.

And I really blame Intel for that.

It was their GPU group that could never make a decent GPU so that they had to license that PowerVR tech. The first video platform to make the best use of VAAPI was the GMA500 from Intel powered by the PowerVR, and it was full of proprietary tech that made the driver very closed off. Which it had to be, because Intel's GPU arm failed at its mission in life.

That was a big problem because up until that point (and still really) Intel was doing a LOT of heavy Xorg lifting. I mean whoever employes Mr. Keith Packard owns the majority share of Xorg as far as I am concerned. And for a while the synergy was great- until the big AMD source dump open source GPU hardware meant Intel.

But then they release this closed off monstrosity that has the features we need behind a gilded gate. And so suddenly VAAPI ran into the NVidia problem- the only drivers that worked for it for a while were closed!

The Nvidia problem of course is that if you have closed drivers then its impossible to figure out if a Xorg problem is in Xorg or if its in the driver.

Nvidia gets around this problem by hiring Xorg wizards to fix all the problems: When VDPAU was released Nvidia didn't just wait for the community to work it in, they released a patched Mplayer with it. That is how Nvidia rolls.

But this problem was new for the open source side of the Xorg driver community. And it wasn't the only problem.

There was a lot of infighting over whether XvMC should be extended or a new thing rewritten. Also the open source drivers have pretty much redone memory management in the last few years (plus all the composite work) so even without VAAPI there is tons of more basic things to work on.

In that environment, Nvidia didn't stand still. VDPAU based color correction, sharpening, and de-interlacing was added. In newer Nvidia GPUs, divx upscaling was added. They kept cranking out hit after HTPC hit to a very receptive audience. We can't get enough ION!

The good news for VAAPI is that:

-The infighting has stopped and its been full steam ahead by all camps (even Via amazingly). XvMC is dead.

-Thanks to some wonderful work by some amazing folks in this community and others, the Broadcom card exists as a complete open source option.

-And as you stated the large popularity of AMD fusion devices in the future will drive the development.

-Intel Ironlake has real VAAPI support now

-We saw the same slow and steady process for composite on Xorg (while Nvidia had it nailed early) and today a non-Nvidia option is actually easier when it comes to composite support (since it works right after install). The result was excellent.


With that said, Nvidia VDPAU delivers more today that it seems that VAAPI will deliver for a while, which makes it a tempting option for practicalists. Also you can bet that Nvidia will keep rolling out the good stuff (Maybe HD Audio bitstreaming in the 4xx series?) which makes them a safe bet for the future....

Reply
#12
Poofyhairguy... Your answers saves me nights of researching. On the other hand it will cost me nights since it's interesting. ;-)

I'll go reading for a while...

ps. From a individual user perspective I can imagine you advise to stick with ION (Intel) but from a open source community perspective I can not understand that attitude: why don't open source communities like XBMC, ffmpeg, Xorg, etc. aren't sticking their neck out to make Intel less dominant and thus have lower prices for all consumer within a few years...

But on topic: Hardware acceleration for AMD Fusion.
Reply
#13
Robotica Wrote:why don't open source communities like XBMC, ffmpeg, Xorg, etc. aren't sticking their neck out to make Intel less dominant and thus have lower prices for all consumer within a few years

The XBMC community did stick its neck out (especially a few) and that is why open source Broadcom Crystal HD support exists.

I think the reason it hasn't shown up with the open source AMD driver is simply because those who are developing that have barely gotten the basic 3D down- x264 decoding is icing on the GPU cake. They do great work and will probably get there one day, but for now getting an accelerator to work is easier than getting a whole GPU to work.

I honestly don't think Fusion will be the next big step for HTPCs. I think set-top boxes with decoder cards will be. Think Popcorn Hour + XBMC. That is the future because Apple has already shown the price point on that stuff can be sub-$100. At those entry prices, the community will explode.

Reply
#14
...Assuming that you want a
*passive (at least silent) mITX-motherboard (thus SoC) and*
  • a appealing case where everything is removed except for
  • backside: hdmi 1.3a, SPDIF, gigabit-ethernet and 2 usb3.0 ports
  • Frontside usb3.0, eSATA interface, minijack
  • and add dual channel 1000+Mhz DDR3 memory,
  • 4 gb CF card
  • a 60W power supply like PicoPSU....
  • Optional CI-slot (DVB) + PCIe x16 slot (PVR)
  • RF/IR chip (remote)

Functionwize I would love to:
  • Playback Flash 10.1
  • Hardware acc. for most important codecs
  • HD Audio bitstreaming (or similar)

Then possibility's are very limited:
1> Netgen ION
2> ATOM + CrystalHD (or a cheap NVidia G2xx card).

So instead of waiting for AMD Fusion I will investigate on focus an those options. Is there good info. on option 2?
Reply
#15
Honestly you can get everything you want with a first gen ION box. Your main problem in your list is that Linux based Flash is not accelerated completely yet.

ION + Windows 7 gets around that because of the superior Flash 10.1 for Windows has proper GPU decoding.

My ION box is pretty much exactly what you want as far as a build. I started with this mobo:

http://www.newegg.com/Product/Product.as...6813500036

I get HD Audio support though PCM, I get full x264 acceleration even with huge files. With an SSD inside the interface is faster than on my quad core.

ION does it all but Flash, and that isn't ION's fault- its Adobe's. Flash remains one of the worst programed platforms out there, and non-Windows versions are really second class citizens.

So a few options:

1. Forget about Flash (its what I did)

2. Get something with an i3 to overpower Flash with CPU

3. Run Windows 7

4. See how the new Broadcom Flash support shapes out, maybe get Atom + Broadcom. This might be the best option.....

Reply
  • 1(current)
  • 2
  • 3
  • 4
  • 5
  • 17

Logout Mark Read Team Forum Stats Members Help
AMD Fusion based hardware for a HTPC?2