Nvidia GeForce GTX 460(GF104 GPU) supports full audio bitstreaming - Page 55

Those are crappy OEM cards, re-badged 4xx cards.
OEMs like those, because they can put a higher number on the box, and customers that buy pre-built PCs usually don't notice.

I want a real 540, or a 650 while we're at it. :P

As GT 430 supports HDMI 1.4, GT 430 = GT 530 OEM. It's surely underpowered even for video playback.

Model	Core	Processor	Memory Bandwidth	CUDA cores	TDP	Length	Street Price
GeForce GT 440 GDDR5	810MHz	1620MHz	51.2GB/s / 128bit	96	65W	5.7 in	$70
GeForce GT 440 OEM	594MHz	1189MHz	43.2GB/s / 192bit	144	56W	5.7 in
GeForce GT 545	720MHz	1440MHz	43.2GB/s / 192bit	144		5.7 in	$150
GeForce GT 545 OEM	870MHz	1740MHz	63.9GB/s / 128bit	144	105W	8.25 in
GeForce GTS 450	783MHz	1566MHz	57.7GB/s / 128bit	192	106W	8.25 in	$100
GeForce GTX 550 Ti	900MHz	1800MHz	98.5GB/s / 192bit	192	116W	8.25 in	$125

550 Ti is not terribly expensive, it's a good choice for HTPC IMO. If the retail version of GT 545 is released for less than $100, then it's also good (8.25 in is too long in my taste, however).

renethx, Why do you say that the GT 430 is underpowered for video playback? Any specific issue that you have seen with it? Yes, it is able to get only ~41 fps with LAV CUVID + madVR for the 1080p60 basketball clip, but that is with artificial scaling not really necessary for the 1080p clip.

Yup, I am talking about CUVID + madVR. Besides the 1080p60 clip (BTW do you have an idea what's the minimum GeForce card?), I saw lots of dropped frames at playing back 1080i60 AVC/VC-1 BDs (documentaries, music conerts) with GT 430 DDR3. Upgrading it to GT 440 GDDR5 fixed it perfectly. (The difference is core/processor clock and memory bandwidth.)

Upsampling 4:2:0 to 4:2:2 is always necessary (why). madVR does it nicely.

That's a pretty niche use case. Have you seen similar issues with DXVA + EVR?

No, never. If one limits himself to the DXVA/EVR framework, GT 430 is a perfect choice.

The main reason I bother to get 550 is the Cuda cores. I would like to see more because now it's exactly the same as the 450's (192). And here comes the 460 with it's nice number of cores which is 336. Sure, the 460 is quite another level but when you compare these 3 cards you actually see that the 460 is rocking in everywhere. It's faster than 550 for about 10-20% (may be less or more), power consumption at idle is quite equal on all cards and probably is the same while playing videos, noise is surely depended on cooling solution whether it's a referrence or not but I haven't seen or read (heard?) about noisy GTX460, no matter it was referrence or not but I have read a review about quite noisy GTS450. The GTX460 is just perfect by price/power/noise/energy consumptions ratio. Also, what really benefits or improvements (cool technologies or something like that, sorry for my poor english) do you get from Geforce 500 series cards over 400 series cards?

I also wonder if there is a real difference between 192 and 336 cores in real life tasks, on particularly 2D -> 3D? I really want to know that because now I am stucked in here. I have Pentium E5500 and currently no discrete card. I really want to know if I am forced to change CPU for converting tasks or I can just get a good video card which will be able to handle it?

I have a suitable PSU for GTX460, it's Corsair VX550. CPU uprade involves MB upgrade also and it costs quite a lot if you choose let's say i3 and I would still need a discrete card to handle MadVR which requires GTS450 for comfortable use.

There is quite a mess in my head with these HTPC solutions and it actually kills me. I am actually a poor guy who counts on every penny even so I want to get as good quality and features as possible. Please help me (-:

Yes, I agree upsampling is necessary. However, the madVR benchmark I have been using does some artificial scaling. Let me fix that offline with madshi, but I think people should be happy just getting 1080p60 to decode with the driver's post processing for now [ I am talking about passively cooled GPUs here , of course one can always go for the ones with higher power and fans and what not and get what one wants ]. It has been a long time since I put in one of the 450 / 460 into my HTPC testbed, will have to see if that solves the issues with the basketball clip.


1080i60 (VC-1) doesn't get DXVA + EVR on any of the ATI / NV systems. However, they play back flawlessly in the commercial BR software players.

That said, I did use LAV CUVID on the GT 430 and GT 520 for 1080i60 VC-1 and also H264 and there were no issues with dropped frames (no madVR, but EVR-CP). I do agree that anything HD (i/p) and near 60 fps is a no-go for madVR + LAV CUVID in the GT 430 and GT 520. renethx, do you have a sample with dropped frames that I could try out? Just want to see how different they are from the test clips that I am using...

Of course it does, you just have to use a VC-1 interlaced compatible DXVA codec, like ArcSoft or Cyberlink.



For a pure HTPC, those extra cores don't do anything. Full deinterlacing and scaling don't even make a GTS 450 sweat at normal playback speeds. Once you start benchmarking, you'll hit the performance limit at around 100 fps deinterlacing performance.

There is no advantage in getting a 460. The 450 is a scaled down 460, there are models that are equally silent (mine is perfectly silent, i dont hear a thing), and use less power then a 460.
For pure decoding and post-processing, a 450 is plenty fast, and any more is overkill. Only if you ever plan to game, the upgrade may make sense. I have no experience with a 440, but the GDDR5 version should probably also be fast enough.

I still want a 520 with some more 3D performance, at least at the level of a 440.

But, that doesn't use EVR, but VMR (at least according to DXVA Checker). Any way to get frame rate stats for the ArcSoft / Cyberlink VC-1 decoders? VMR benchmarking doesn't go above the display refresh rate.

I just registered the ArcSoft decoder, opened a vc-1i m2ts file, and benchmarked it with DXVAChecker against EVR, with display off. Avg 144 fps.

Seems to work just fine.

@Uka5tic

Noise: This entirely depends on the cooling solution each card has. For example, GIGABYTE GTS 450 with WindForce dual fan is close to silence (for me). MSI's Twin Frozr II is another nice choice (450 Twin Frozr II does not exists, however). Twin/dual/triple fan is the current trend of silent cooling solutions.

Right now every 2D -> 3D conversion technology (CyberLink, ArcSoft, TriDef) relies only on CPU. Core i3-2100(T) or Athlon II X3 does it nicely.

madVR uses stream processors, but does not require that many processors. GT 430 96 sp is enough.

BDs I tested are

- La Traviata (by Lorin Maazel, Teatro alla Scala) 1080i60 AVC
- Nature's Journey 1080i60 VC-1

These are copy-righted material, so I shouldn't upload a clip. You can obtain them from amazon.com.

Thanks for your answer, it really showed that I was wrong and missdirected on some things. Actually I thought that PowerDVD can use the advantage of the Cuda cores. Probably I will wait for upcoming Liano and see what happens, the prices of the CPU's obviously will go lower and maybe I could afford some nice 4 cores CPU. I will be busy during the summer time and most probably won't even have time to watch movies so I can stay still for a while.

Wait, you may be right. Let me check GPU load under "TrueTheater 3D". I only tested SNB + IGP.

I am not sure what's wrong, but even if I checked the "Enable NVIDIA CUDA" check box next to the TrueTheater 3D check box, it is unchecked and grayed out once video playback with 2D-3D conversion begins, hence CPU usage is pretty high (~60% with Pentium SNB G840).

Love the word choice there

I've having trouble with frames skipping when using Haali Media Splitter for 24fps mkv playback. Is there anything else I can use for this? The output is to MPC-HC if it matters.

Try to change the source/splitter and the video decoder (and maybe the audio decoder). LAV Splitter Source is the best source/splitter filter for MKV.

What program(s) created the MKV? I've seen some programs, like DVDFab, that do not set the default audio duration field in the MKV, and that causes video and audio stuttering with the Haali splitter. If you edit the header with mkvmerge, you can manually set the audio default duration to 10666666 for 23.976 fps. That may only be an issue with some HD audio formats (I'm not sure).

It would be good to have a common frame of reference.

The mediaplayer forum has a nice list
http://www.avsforum.com/avs-vb/showthread.php?t=1181902

Including the killer sample of the birds. (1080i@60 x264 33Mbps 12 Ref Frames)

On my GT 430 + MPC-HC + MadVR 0.61, it drops 15 frames. Video engine load is 57% and GPU load is 30% (measured via GPUz)
This clip is a little weird. If you keep replaying it the dropped frames increase
but if MPC-HC is shutdown and restarted, I get a consistent 15 or so dropped frames.

Another one is the Panasonic Ginza Cat 1080p@60 28Mbps
7 dropped frames. Video engine load 90% Gpu load 57%

http://www.avsforum.com/avs-vb/showt...2#post20356642

You can get the Birds clip there too.

What version of the decoder is that? I am trying on a clean system with TMT5 (regsvr32 on ASVid.ax : v2.27.441.128), but it still doesn't show up in DXVAChecker for VC-1 interlaced. Are there any other codecs registered too, or was it only the ArcSoft decoder?

I have numerous codecs registered, even LAV CUVID shows up in the DXVAChecker list, as did ffdshow (which won't work, of course).
Surprisingly benchmarking lav cuvid also worked.

I think its the TMT5 version, might've been TMT3 though, cannot check right now, but both worked in the past.
Needs the usual checkactivate.dll hack, and if you moved it out of TMTs directory, also that one dll with VC1 in its name.

LAV CUVID shows up on my list too, but none of the DXVA modes are active.. I think the only thing left for me to ask is : What is your DXVA Checker version?

Also, want to confirm whether the checkactivate.dll from 2008 (the one with TMT3) will work for TMT5...

I was using the latest DXVAChecker from a few weeks ago, 2.5.0.

Get this checkactivate.dll: http://forum.doom9.org/attachment.ph...7&d=1271603449

I'd like to know what brand of 450 you use, since I'm going to dump my HD 5450 for a 450 or 550 to use LAV CUVID and MadVR. I like having a very quiet computer. Thanks, by the way, for your wonderful work on all of these projects.

It won't be just HTPC, so Cuda may be a very important thing for me (at least I think so). The 460 has almost twice Cuda cores compared to the 450, I think it should have some benefit in real life or maybe not. I found a used 460 for as low price as is a new 450. This is what I found: http://www.newegg.com/Product/Produc...82E16814121420 Sure, I can find a used 450 for even lower price but somehow I like 460 more.

Actually I had Gigabyte 1GB GTX460 few months ago I bought and sold it for my friend. Well, it was a terrific card both silent and cool running. http://www.newegg.com/Product/Produc...82E16814125352

I have a Gigabyte card, with one of their Dual Windforce coolers. Absolutely silent.

This one: http://www.newegg.com/Product/Produc...82E16814125341

Sparkle lists a fanless GTS 450 on their web page, but I've never used one (or even seen one for sale) so I certainly cannot vouch for it:

http://www.sparkle.com.tw/product_de...112&sub_id=408

I use the MSI GTS 450 Cyclone, and it is also absolutely silent:

http://www.newegg.com/Product/Produc...-521-_-Product