Nvidia GeForce GTX 460(GF104 GPU) supports full audio bitstreaming - Page 56

Well, all of these may be silent one's (Sparkle definitely is) but I would still chose the third option (Gigabyte 450 card). Why? Simple, because it blows the air out of the case, not circulate it inside like the other two. Most people chose small (at least smaller than midi towers) cases for HTPC and going with passive like Sparkle 450 would require quite a nice case solution with case ventilators also. Just my opinion.

Indeed, passive at that performance level requires quite good case ventilation, which in turn then defeats the purpose of not having a fan on the card, instead you have one or two in the case..

I picked the Gigabyte Windforce for that exact reason, the MSI Cyclon is equally silent, but it doesn't blow out the hot air.

Thank you sir, completed building the HTPC with 440 and silverstone MC 10E case. Would share the experience as soon as I tweak my XBMC to behave like I want to ......

The new checkactivate.dll did the trick, but only partially.. Now, ArcSoft appears in the DXVA Checker list for VC-1 interlaced clips, but only in the format as in the attached screenshot.

I am able to benchmark the VC1 VLD mode under EVR (got 132 fps with GT 430 and all driver post processing enabled), but the support is not explicit DXVA2 EVR like it is for the MPC Decoder / MS DTV-DVD Decoder for the H264 clip in the other attachment.


Now, I also understand how you were able to benchmark LAV CUVID with the DXVA Checker ( I found that you could still benchmark even if the decoder comes out as Unsupported -- So, it just looks like you need it to appear in the list )

So, the new question is: What decides whether DXVA2 / EVR appears against a particular mode for a particular decoder?

As long as its showing a decode device (and low CPU usage) during benchmarking, its using DXVA. I don't know how DXVAChecker determines those values in the table.

Yes, I confirmed that And the GPU as well as VPU load was pretty high during the benchmarking (just like any other DXVA2 / EVR combo test).

I hope renethx has some insight I believe he is well versed in Japanese, and maybe there is some interesting stuff in the DXVAChecker documentation (which was in Japanese the last time I saw)

Edit

Unfortunately I have no insight. Readme.txt includes only routine usage of the program. For me end results are everything, whatever DXVAChecker says:

- Low CPU usage -> low power (may not be true any longer with a newer Intel processor, however)
- Not excessive high GPU/VE/video memory load
- Good PQ.

BTW where does EVR-CP get this information?

Which decoder is used can of course be gathered from the DirectShow filter graph.

For the DXVA detection, it has a Hook into DXVA which will allow it to detect if some decoder requests a DXVA decoding device. It can even detect if LAV CUVID is using the DXVA processing mode.

I found this glitch happens only in dual display (NVIDIA+NVIDIA or NVIDIA+Intel) under 7. Playback with GT 430 is fine in a single display under 7/Vista or in dual display (NVIDIA+NVIDIA only) under Vista. GPU load is very high, however. Compare (at the playback of La Traviata BD 1080i60 AVC):

GT 430 DDR3 1GB (with Pentium G840)

GTS 450 GDDR5 1GB (with Pentium G840)

GPU load and Memory Controller Load are greatly reduced.

I tested several 1080i60 files (e.g. Cheese Slices) and saw the same tendency. GTS 450 such as GIGABYTE GV-N450OC-1GI is highly recommended if you want to use LAV CUVID Decoder + madVR.

BTW full screen mode is windowed to take a screenshot. If exclusive is selected, the CPU usage is ~20% in either card.

Thanks!

maybe someone from here can help. I have a gtx460 running an extended display. my main display is a 20 inch monitor but my extended display is a 46" samsung going through a marantz receiver. I am having problems bitstreaiming to the receiver. any ideas? i can get DTS and Dolby, but not DTSMA and DD

I ran into a problem with using custom resolutions. I had a CR defined for 23.976 which worked great until I tried to watch a 3D movie. After enabling 3D mode, Nvidia would not go into frame-packing mode unless I selected 720 FP (and I assume 1080/60, but my projector doesn't support that).

I had to delete the CR to get 24FP to work. I haven't tried defined a custom 3D resolution, maybe that would fix it.

Hopefully, I have created a decent summary of the currently important topics in this particular forum here

http://www.anandtech.com/show/4380/d...-gpus-shootout

That is hands down the best article I've ever read on gpus information and relevant software for htpc use. My only complaint was that they should have used the gts450 and gtx460. The gt430 had less stream processors than the lowest stream processor count on the amd card. I think that really should have been brought to the attention of the readers especially when addressing "videophiles", "all the great recent open source software" like cuvid an madvr. Oh, one other thing they incorrectly stated is their is no bitstreaming support for linux. Other than that it was spot on when addressing the 23.976 and other relevant issues, and all other post processing benchmarks. Wonderful article to read for anybody trying to figure out to go amd or nvidia for htpc use or for anybody wanting to gain knowledge about gpus for htpc use.

Good summary of the current state of the art. The GT430 is best
compromise for the small form factor video card if the material
does not exceed 50MBit/s (BD spec).

The coverage of the VP4 decoder in the 520 was also interesting in
that it can handle a substantially higher bitrate, useful for 3D

This has helped me decide on a 560Ti for the SVP FI box I just finished
building. It is using a GTX460 borrowed from another machine.

The Gigabyte 450 Windforce card has gotten some praise in this thread. I was considering purchasing one for my HTPC, but I noticed that the newegg reviews for the Gigabyte 450 and 460 cards have recently gotten a number of complaints about cards working for a few weeks or months and then failing with freezes and artifacts.

Has anyone using one of these cards with a HTPC had any similar problems?

Was the version of MadVR the later ones like 0.62 and above ?

The GT430 does not have enough SPs to do the scaling over a 4K desktop.
Looks like you have tapped out the GPU.
Locking MadVR to 1 display might be a workaround. Another option
would be to use a less computationally expensive scaling algorithm
like Mitchell-Netravali or Catmull-Rom for the luma rather than the default
3 Tap Lancsoz

The older versions of MadVR did chroma scaling to the full screen res,
the newer ones only double up, much less load on the shaders.

I wished someone made a half height GT440 GDDR5 that can fit in my ISK310

Its kinda interesting that the only half-height 440s they produce are DDR3 models - the different ram can't make a difference on PCB size, or can it? Maybe its about cooling the RAM..

GDDR5 takes more drive current so GPU runs hotter.

ATI had a similar issue with the GDDR5 5570 needing to be full height for
the bigger cooler.

BTW the YV12 fix is much appreciated.

jakmal pointed out that the new Video Processor in GT 520 can handle more bit rates than GT 430 in his article. Maybe GT 440 has a newer VP? I haven't compared. Update: Never mind, my comment is not true. From the same page:

Feature Set D could be called PureVideo HD Gen5 or Video Processor 5 (VP5).

I have a OC 768mb galaxy 460. Great card, quiet, removable fan, top connector for tight fit jobs.

The 440 and 450 cards are both based on the GF106 so I expect the
VP to be identical. 440 uses DDR3 and is lower clocked
http://www.hwcompare.com/7482/geforc...ce-gt-450-oem/

The really interesting point is that the 440 has the lowest power consumption
of the 4xx series at 56W, lower than the 430 at 60W. The 450 is about 2X @ 106W

http://www.hwcompare.com/7242/geforc...-gt-440-1-5gb/
On the 430 in your example the DDR3 memory channel is only 51% loaded
there is still enough headroom.
(Taking a closer look here, I think the need for a GDDR5 440 is moot)

The video load in your example is nearly identical for the 430 and 450
as the VP block is the same across the entire GF10x family,
except the GF100 (maybe?)
The performance difference is down to the number of SPs/TMUs and
how fast they are clocked at.

In practice, the 430 driving a 1080p display has sufficient SPs for MadVR + LAV.

Going beyond that to multi monitor setups, the picture changes and there
is a need for more SPs/TMUs

The 520 is poorly suited for this application as jakmal's numbers show. The 530 OEM
may change things although I doubt we will see a half height 540 anytime soon.
The product line is already fragmented.

The faster VP on the 5xx will have to wait for the relevant content to
catch up

All 4xx and all 5xx cards, with the one exception of the 520, have exactly the same video processor, and its not dependant on the performance of the card itself.

Only the 520 has the faster processor for now, and it truely is too slow for any serious usage, according to NVIDIAs own statements, it cannot even deinterlace properly.
The 5xx OEM cards are rebranded 4xx cards, so they all have the old VP as well, and i would rather get a real 4xx card then rebranded OEM cards of the last generation (its not generally available anyway)

Somehow i really wish Intels IVB will fix the 24p issue and be fast enough to handle madVR scaling, it would make things so much easier in HTPC land.
A SNB HD3000 iGPU takes around ~25ms to scale a image with madVR using 3-taps algorithms (Lanczos), which is good enough for 24 and 25 fps movies, but on 60p its too slow (16.6ms frame time).
Rendering a frame without scaling (1080p movie on 1080p screen) takes around 10ms (just upsampling chroma)

A promised 30-40% increase in GPU performance could make it all possible, now lets just hope they also fix the 24p issue with the next PCH version.

Nevcairiel,HDGT

Thanks for the clarification on the decoder feature set C/VP4 vs D/VP5 on the 5xx

Given the rebranding games on the 5xx, I am thinking that Kepler
would be the next real step, that is if AMD does not get its house
in order by the time Kepler shows up.

If AMD fixes the driver software and improves on their version of CUVID
Nvidia could find itself in trouble again on more than 1 front.

Intel has been the most aggressive and innovative but the results on SNB
leaves much to be desired when using MadVR +LAV. IVB is a wait and
see proposition for me.

Greetings Rene.

Just to clarify:

1. The GT 520 has VP5?
2. The GT520 can handle Ginza Cat ad Birds using DXVA?

Many thanks and best regards.

It is not officially called VP5, but just VDPAU Feature Set D.


Yes, it can. However, post processing is not good. Also, all the previous GT 4xx cards could handle 1080p60 (AVCHD / broadcast) clips without issues.

SNB can be made to work properly with madVR, kind of.
Intel implemented their own Lanczos scaling optimized directly in hardware, so if i were to write a LAV CUVID like filter for Intel, which uses Intels own Lanczos 4-tap scaler to scale the image to the desired display resolution (usually 1080p), in theory the GPU would be fast enough to do the remaining madVR operations, even on 60p content. I'm not sure if the quality would be the same in comparison to pure madVR, but it probably at least wouldn't be far off.

Only the silly 24p issue left.
Also, this is purely theoretical, days need to be longer, etc.

Interesting... Is that exposed through the VPP function in the MSDK?