Nvidia Video Card guide

[GTwannabe]

Originally Posted by lobosrul

Ehh... a 7800GTX smokes a 7600GT.

I remember when those babies came out last summer ('05) they were a giant leap forward in video card speed.

If you dont belive me: http://www23.tomshardware.com/graphi...=529&chart=231

Look for the two lines in blue.

Edit: my bad sort of (the GTX selected above is a 512MB OC'd modeled). See this chart: http://www23.tomshardware.com/graphi...=529&chart=231 Still much higher though. And yes the clock speed of the GTX is lower, it has way more pipelines though im sure.

In 3D games, yes. However, PureVideo uses dedicated silicon on the GPU. That's why its performance scales directly with clockspeed and not the number of pixel pipes or memory bandwidth.

[autoboy]

We are not debating the 3D performance of these cards. 3D performance does not always scale with video performance. The 7300GT is an ok low range card for 3D but it gets smoked by a crappy 7300LE in 2D.

Also, I found that my Nvidia card has 2 different clocks that can be adjusted seperately. In Ntune, my 7300LE has a 2D clock at 450mhz, and a 3D clock also at 450mhz. When I do a manual overclock, I only adjust the 3D clock. When i do a custom automatic overclock, ntune adjusts both clocks seperatly. With a 20min test, ntune got my card to 521mhz in 2D, and 524mhz on the 3D test. Memory went to 850mhz.

This is interesting and could possibly help us experiment with purevideo speed. We can try increasing the 3D clock and see if anything improves with Purevideo.

That's why its performance scales directly with clockspeed and not the number of pixel pipes or memory bandwidth.

We can't say for sure whether or not the shaders are used for some advanced deinterlacing features. We also don't know whether memory performance is a factor. We know that H.264 acceleration is done in the 2D silicon and scales with clockspeed, but we don't know how memory bandwidth factors in.

In fact, we could be focusing on the wrong thing for the 6150. Perhaps the memory bandwidth is the limiting factor preventing smooth 1080p playback and not the limited pixel shaders. If the shaders are being used, shouldn't the power requirements rise? We should check the power useage under video load compared to 3D games.

[stanger89]

Originally Posted by autoboy

In fact, we could be focusing on the wrong thing for the 6150. Perhaps the memory bandwidth is the limiting factor preventing smooth 1080p playback and not the limited pixel shaders.

One user (don't remember if it was here or AVS, here I think) reported that going from single to dual-channel config on their 6150 board made a significant difference in high-resolution VMR9 performance. So memory bandwidth apparently is an issue.

If the shaders are being used, shouldn't the power requirements rise? We should check the power useage under video load compared to 3D games.

Probably depends on how their used, among other things.

[lobosrul]

You guys are telling me the fill rate of a card doesnt affect VMR9 playback? It only relies on the clockspeed?

I wouldve thought 2d clockspeed would only be relevant for Overlay playback.

Anyways, I had smooth playback with a 6600GT (when using purevideo). I only bought a 7800GT for gaming.

[Kirby]

I'm not saying that. AFAIK, VMR9 uses the 3D pipeline in the GPU.

[malore]

Originally Posted by Kirby

So who has tried the 7600gt passive cooling? How's it doing for you?

I'm using the GIGABYTE GV-NX76T256D-RH Geforce 7600GT in my main/game computer and it's been doing fine, no problems with overheating. However, it is very large, taking up three slots. I can't comment on it's HD ability, because I haven't tried it.

[stanger89]

Originally Posted by lobosrul

You guys are telling me the fill rate of a card doesnt affect VMR9 playback? It only relies on the clockspeed?

I think what we can conclude, is that hardware decoding accelleration is not affected by fillrate, only by clockspeed (and 2D clockspeed probably).

Fill rate, does or probably does come into play for VMR9, including advanced deinterlacing/film detection, etc.

[blade]

According to walford over on the AVS forums you need 10 GB/sec of memory bandwith to handle full resolution 1080i. I have no idea how he came up with that number.

I did some testing with my 6600gt last night running at 1080i underscanned to 1688*1004. I used fraps and what I considered smooth was when I got the full fps. Playback would appear smooth at lower settings, but lacked full framerates. I had AF set to 16x and everything at the highest quality settings, also AA was off. I needed the following memory speed to get smooth playback with VMR9.

390mhz (128 bit bus) - 20 mbps, 1080p, film (24 fps)
780-980mhz (128 bit bus) - 18mbps, 1080i, video (60 fps)

For video content it varied a lot from clip to clip so I included a range. Notice that video runs at 2.5x higher fps and I needed 2.5x faster memory speed for most of my clips. I'm trying to download a clip that claims to be 40 mpbs and another at around 12mbps to see if the bitrate affects memory speed requirements.

Maybe someone else can confirm this, because I'm not all that great at testing this junk. Also does anyone have any thoughts to explain exactly what is going on?

[peternm22]

I e-mailed nVidia a few days ago asking why the AGP variants of the 7600GS and 7600GT weren't listed on their chart. I haven't received a reply from them yet, but the chart has been updated http://www.nvidia.com/page/purevideo_support.html

It now has the 7600GS AGP listed (however no 7600GT).

The supported features of the 7600GS AGP are identical to the PCIe version. Which is good news for us AGP folks.

-Peter

[lobosrul]

Originally Posted by blade

According to walford over on the AVS forums you need 10 GB/sec of memory bandwith to handle full resolution 1080i. I have no idea how he came up with that number.

Hmm that sounds high. Even after decoding the uncompressed video isnt nearly that big.

[blade]

Originally Posted by peternm22

The supported features of the 7600GS AGP are identical to the PCIe version. Which is good news for us AGP folks.

Hopefully it is good news, but it may not be. They re-added WMV9 acceleration to the chart for the agp versions of the 6600gt and it still doesn't work. Also the h264 acceleration still doesn't work unless you have a SSE2 capable cpu.

[peternm22]

Originally Posted by blade

Hopefully it is good news, but it may not be. They re-added WMV9 acceleration to the chart for the agp versions of the 6600gt and it still doesn't work. Also the h264 acceleration still doesn't work unless you have a SSE2 capable cpu.

My enthusiasm has now been tempered.

-Peter

[blade]

Well it seems everyone has lost interest in this, but just in case anyone still cares.

A quote from this thread.

To perform advanced motion adaptive deinterlacing with PureVideo at 1080p resolutions requires - last time I checked, which was a while ago - 10GB/s of graphics memory bandwidth minimum, which made the 6600GT entry level for true HD deinterlacing.

If you compare the cards that do Spatial-Temporal De-Interlacing for HD Video with their memory bandwith according to this chart it does appear that the only cards capable of doing it have over 10 GB/sec of memory bandwith. I have no clue if the guy knows what he's talking about or if it is just a coincidence.

[blade]

FiringSquad: How much of PureVideo is done with the dedicated VPU components of the chip versus the 3D pipeline?

Scott Vouri: I’m really glad you asked this because it is one of the coolest things about our programmable video processor. Our PureVideo technology actually does processor load-balancing across all the video cores and the 3D rendering engine. That way we can process multiple tasks at once or process different stages of the video pipeline at the same time.

[Alan's comments: That wasn't as much detail as I was interested in hearing, but in a follow-up question, NVIDIA suggested the example of performing hardware accelerated decode and some post-processing effects on the internal VPUs and then doing additional color-enhancements or post-processing on the 3D engine.]

From this old interview.

[malore]

As is usually the case for me, I find out what I really need after buying something and using it for a while. I purchased a cheap Geforce 6200 and was very happy with it until I tried to watch my first 1080i HDTV show and the stuttering was head ache inducing. It's outputting underscanned 720p. After much trial and error with different decoders and overlay versus VMR, I've found I can get reasonably smooth playback by overclocking the videocard. I highly recommend Fraps and wish I had used it sooner, because it easily verifies what you think you are seeing. I have my card overclocked to 415/700 and Fraps shows a fairly steady 60 with only an occasional blip which might be from noise in the recording. When I returned it to the stock 350/650 the frame rate was in the 50s. I'm using the nvidia purevideo decoders (223) with VMR9 and FSE. I have the 92.91 beta drivers installed.

1080i playback definitely stresses the GPU which runs hotter than during 720p playback. If you choose to overclock, test the playback while nothing critical is recording, because I've had the video lockup after watching a show for a while when I had it overclocked ever higher.

[blade]

Originally Posted by malore

I highly recommend Fraps and wish I had used it sooner, because it easily verifies what you think you are seeing. I have my card overclocked to 415/700 and Fraps shows a fairly steady 60 with only an occasional blip which might be from noise in the recording. When I returned it to the stock 350/650 the frame rate was in the 50s.

I agree about fraps. I've been using it for months and it is very useful especially when over or underclocking things to see how they affect playback.

My 6200 has no problems with 1080p film (24 fps), I only run into problems with video (60fps). You should see the same thing. This may be why some shows playback smooth and others stutter for some people.

[blade]

Out of boredom I keep tinkering. I only tested this on 1080i video content because the requirement for film is so low there isn't any point. Again this was on my 6600gt and I underclocked to the lowest speed that still allowed me the full 60 fps for video content according to fraps for vertical stretch deinterlacing. Then I switched back to Per-Pixel Adaptive to see what sort of frame rates it would get at the same speed.

550 mhz GPU / 500mhz (128 bit) Memory
Vertical Stretch Deinterlacing - 60 fps
Per-Pixel Adaptive - 40-50 fps

335 mhz GPU / 1 ghz (128 bit) Memory
Vertical Stretch Deinterlacing - 60 fps
Per-Pixel Adaptive - 40 fps

To get smooth playback with Per-Pixel Adaptive I needed 500-550 mhz depending upon the video clip. Some were more demanding than others. Since many other cards operate at a slower clock speed and can still do per-pixel adaptive it seems the deinterlacing is mostly dependent on fillrate and memory bandwith. Since h264 and wmv9 is broken for my card I can't test acceleration, but others have already confirmed it is dependent on gpu speed. I've never noticed any difference in cpu usage for hd mpegs whether I'm at 350 or 550mhz so I guess it only really applies to decoding wmv and h264.

I've played around with my 2d clock speed and have yet to see any differences with it. Not sure if it plays any role or not.

[AtariJeff]

Originally Posted by malore

As is usually the case for me, I find out what I really need after buying something and using it for a while. I purchased a cheap Geforce 6200 and was very happy with it until I tried to watch my first 1080i HDTV show and the stuttering was head ache inducing. It's outputting underscanned 720p. After much trial and error with different decoders and overlay versus VMR, I've found I can get reasonably smooth playback by overclocking the videocard. I highly recommend Fraps and wish I had used it sooner, because it easily verifies what you think you are seeing. I have my card overclocked to 415/700 and Fraps shows a fairly steady 60 with only an occasional blip which might be from noise in the recording. When I returned it to the stock 350/650 the frame rate was in the 50s. I'm using the nvidia purevideo decoders (223) with VMR9 and FSE. I have the 92.91 beta drivers installed.

1080i playback definitely stresses the GPU which runs hotter than during 720p playback. If you choose to overclock, test the playback while nothing critical is recording, because I've had the video lockup after watching a show for a while when I had it overclocked ever higher.

Are you happy with the PQ of the 6200 with the newer drivers? I had stair-stepping so bad I had to, yet again, back out to version 84.21. Now its great.

[autoboy]

According to walford over on the AVS forums you need 10 GB/sec of memory bandwith to handle full resolution 1080i. I have no idea how he came up with that number.

This was from the Windows Media Player website. It was a kinda arbitrary number that was supposed to eliminate the slower video cards across all GPUs. This # is why I chose to use a 9600xt because the memory was just about 10Gb/sec. It turned out to be less than ideal. There are so many more factors than just memory fill rate in 1080i video.

This is good testing. Once you guys get a finalized fraps test with all the info i'll update the guide. It has been wrong too many times for me to keep putting my own views up there. I need facts supported by evidence for your bandwidth issues. BTW, i've still never had any problems with my 7300LE in 1080i video except when i ran out of main memory. I'm only using 512MB and sage can use 160MB sometimes.

[Tighr]

So what are the best options for us AGP folks that want good HD playback?

I've currently got an ATI Radeon 9600, and was looking for a good AGP card to upgrade to since my HD playback isn't very smooth. The cheapest 7xxx on newegg is a 7600GS for $125.