View Full Version : madVR - high quality video renderer (GPU assisted)
Olivier C.
10th February 2014, 19:49
I've recently seen 3GB 7950's selling for 120€ but their fans looked pretty low grade and these were Sapphire...I guess the days when you had to buy a TOTL GPU fan and install it yourself for silence aren't quite history. I still got a Zalman VF1000 that's serious business with a 120mm slapped onto it, I wonder if it'd fit.
I don't think Sapphire is the worst :
http://www.hardware.fr/articles/853-16/temperatures-bruit.html
cyberbeing
10th February 2014, 19:57
Actually, when I look at them again, 5 seems to have a much more defined edge than 7.
7 seems to blur the edges together a bit.
As the goal of dithering is not to deband the image, that seems worse.
Well, when you hear me speak of "crosstalk issues" that is what I'm referring to. Don't put too much weight in those images though, as they are literally a small crop of a much larger picture. Overall, ED7 actually fairs a bit better than ED5 in that regard on the full image, and the 'noisy2' build fairs slightly better than both. All of these madVR error diffusion algorithms seem to have crosstalk issues to a certain degree on the some of the exact same edges, even the non-random ones.
@leeperry
I am interested to hear if have the same 'fuzzy' issue as ED7 with any of those new builds. The new random method gives them all a completely different noise profile then previous builds, to an extent that I am unable to even guess which algorithm is supposed to be which.
Olivier C.
10th February 2014, 20:07
Hi,
Maybe i am wrong, but i can not see any difference between an ICM profile created by Argyll (untick disable gamma ramp in madVR) :
http://img11.hostingpics.net/pics/772771icm.jpg
... and a 3DLUT used in madVR created from the same profile
I didn't watched carefully at the black but maybe by ICM (CCM Windows), the bug "black raised" is not present ?
And if ICM works well with madVR, it will not consume a lot of GPU.
I have to make a choice between ICM, 3DLUT and yCMS.
What do you recommand ?
PS : sorry for my english, i am French
drew_afx
10th February 2014, 20:16
Kinda off topic but with all these new algorithms...
I wonder if madvr can calculate temporal dithering for the display?
24fps source to 120hz display with frc can simulate 10bit from 8bit output and work with 8bit native displays...
even 10bit displays will benefit from less artifacts since it'll be 16bits -> 10bits instead of 16 -> 8 spatial dithering.
one problem I see is the rendering time required for 120fps...about 8ms
Ver Greeneyes
10th February 2014, 20:24
And if ICM works well with madVR, it will not consume a lot of GPU.
madVR doesn't support ICC profiles and I don't believe support is planned. If you want to test whether your profile is doing anything, try installing inverted.icm from this folder (https://www.mediafire.com/folder/czveb1ynbox6s/ICC_Profiles) - it should invert all colors in color managed applications.
Olivier C.
10th February 2014, 20:45
madVR doesn't support ICC profiles and I don't believe support is planned. If you want to test whether your profile is doing anything, try installing inverted.icm from this folder (https://www.mediafire.com/folder/czveb1ynbox6s/ICC_Profiles) - it should invert all colors in color managed applications.
I will try.
Could you try this profile if you have some time ? :
link (http://www.mediafire.com/download/ci79qjquj0pia86/madVR_2014-02-08_D6504_2.2_VF-M_3xCurve+MTX.icm)
Install profile and dont change the name to be sure
Ver Greeneyes
10th February 2014, 21:22
Could you try this profile if you have some time ? :
link (http://www.mediafire.com/download/ci79qjquj0pia86/madVR_2014-02-08_D6504_2.2_VF-M_3xCurve+MTX.icm)Why? I use my own profile for Firefox and a 3DLUT in madVR. By the way, the EVR renderer in MPC-HC does support ICC profiles, but of course it's not as nice as madVR in other ways.
wolfman2791
10th February 2014, 21:28
Quick q for you guys... is it worth paying the money for JRiver or should one stick with MPC-HC?
huhn
10th February 2014, 21:36
it's there something jriver provided what you need?
if you don't know you don't need it.
Ver Greeneyes
10th February 2014, 21:38
One of the things that caught my eye when this discussion first started is that 6233638 reported being able to better differentiate between patches of similar brightness, and I was wondering if this might improve my dispcal calibration near black.
Sorry, this was an error on my part.OK, thanks for clarifying :) I'm still curious about my other questions though (especially the 3rd).
wolfman2791
10th February 2014, 21:46
Quick q for you guys... is it worth paying the money for JRiver or should one stick with MPC-HC? Forget it. I don't like JRiver already :D
tschi
10th February 2014, 21:59
There could be something wrong with your system.
Thx, you 're right. I tried the card in other pc, I was able to do an 4x upscale for nnedi luma with 32 neurons and all "trade quality for performance" unchecked on sd material. I think I would need to make clean install on the htpc :(
leeperry
10th February 2014, 22:18
I am interested to hear if have the same 'fuzzy' issue as ED7 with any of those new builds. The new random method gives them all a completely different noise profile then previous builds, to an extent that I am unable to even guess which algorithm is supposed to be which.
The new build 3 looks so much better than the others to my eyes and Bam! madshi told me that 3 = 7 duh, so 7 it is http://forum-images.hardware.fr/images/perso/casper87.gif
compared 7 with a sapphire 7870xt to the gtx-770 and preferred the GTX-770
Did you carefully recalibrate for each graphic card?
The consensus on the HCFR forum has always been that AMD won PQ-wise, I think I'll be trying an Asus or Gigabyte 660 anytime soon.
At least with nvidia you get a great looking double frame rate deinterlacer, AMD's would be way too slow for realtime apparently.
It would be great if we could find a procedure to benchmark GPU's but at this point it's still unclear how NNEDI would perform via DC.
is it worth paying the money for JRiver or should one stick with MPC-HC?
Only if its GUI doesn't make you wanna jump off a cliff as it does for me :D
But it's got a built-in Reclock'ish audio renderer that plays nice with mVR, pretty cool.
ZoomPlayer & PotPlayer come with a truckload of exclusive features(such as seamless playback mixing any kind of container), the latter even has a fully skinnable D3D GUI that doesn't break mVR's FSE...the former might also have, IIRC it used to have one that played nice with VMR9 ages ago so I would guess that it's been made compatible with mVR as well.
KeyserSoze17
10th February 2014, 22:47
Forget it. I don't like JRiver already :D
If you are looking for a media center with madVR as a native player you could try MeediOS it's 100% free.
http://forum.doom9.org/showthread.php?t=167183
http://www.meedios.com/EN/
Olivier C.
10th February 2014, 23:00
Why? I use my own profile for Firefox and a 3DLUT in madVR. By the way, the EVR renderer in MPC-HC does support ICC profiles, but of course it's not as nice as madVR in other ways.
Because it works in madVR for me.
Sadly your profile does not work at all (firefox, windows, etc).
Strange ...
I am using this tool (it's not an installer) in order to test quickly an ICM file :
http://www.xrite.com/downloader.aspx?FileID=539&Type=P&returnurl=%2fproduct_overview.aspx%3fID%3d757%26Action%3dsupport%26SoftwareID%3d539
You can check with this tool that the ICM selected is taken in account during film in MPC+HC + madVR.
I can post some screenshots if needed
Ver Greeneyes
10th February 2014, 23:08
Because it works in madVR for me.
Like I said, madVR doesn't support ICC profiles. If your profile is affecting the output, you're not using madVR.
How are you installing inverted.icm? You can either use dispwin.exe included in ArgyllCMS withdispwin.exe -I inverted.icm(assuming the program and profile are in the same directory), or go into Windows' Color Management settings page, use the add button, browse for it, then set it as the default for your monitor.
Edit: Ah, what you're probably seeing is the effects of the calibration included in the profile (this is the vcgt tag the documentation of that X-Rite tool talks about). This is a set of three 256-value curves that get loaded into your video card, and will affect every application. But the calibration is only half the story - a full ICC profile contains a lot more information than just that.
Edit2: Oh, and to enable full color management in Firefox, you have to go to about:config and set gfx.color_management.mode to 1. At its default setting of 2 it only does anything for images that contain a profile themselves.
Olivier C.
10th February 2014, 23:19
Edit: Ah, what you're probably seeing is the effects of the calibration included in the profile. This is a set of three 256-value curves that get loaded into your video card, and will effect every application. But the calibration is only half the story - a full ICC profile contains a lot more information than just that.
Yes, you are right.
I will try to check if the full ICC is taken in account.
I tried quickly 3DLUT versus ICC and can't see any difference but may be i need to check carefully.
Thanks :)
Ver Greeneyes
10th February 2014, 23:25
I tried quickly 3DLUT versus ICC and can't see any difference but may be i need to check carefully.
For me the difference is pretty subtle, but definitely noticeable if I switch between "this display is already calibrated" and "calibrate this display by using external 3DLUT files" (and apply).
Also see the second edit in my previous post for a note about Firefox - I wish that was the default, but some people use Windows Update supplied profiles which can be pretty broken.
XMonarchY
10th February 2014, 23:36
I use a 3DLUT but I do not have my TV's absolute 0 black raised because I figured out how to remedy that flaw. However, I do get dithered blacks (just not the absolute black) in any build other than DC3, ED4, and ED2. I can see a definite difference even in ED9 and ED10 in any movie with blacks (deep or shallow). After experiencing error diffusion without dithering I can't get myself to watch an entire movie with noisy/dithered blacks... ruins the experience. Isn't the point of error diffusion to NOT dither anything?
I also use BT.1886, which elevates blacks and makes black dithering even more obvious. IMO blacks and their perception is of utmost importance to get that feeling of contrast, depth, and high definition. Plasma owners may not care much for dithered/non-dithered because plasma dithers by default and has blacks deep enough where dithering does not have any negative effects on perception of those blacks. However, at least one friend of mine with ST60 plasma prefers DC3 sharpness, even after seeing the banding and the worm patterns screenshot...
As an LCD owner, I find that black dithering is very visible, distracting, and even makes recent HD content look like a re-mastered from SD to HD movie from 1980's. I personally can't take it - not after seeing the effect without dithering. I only see banding in pattern tests - never, not once, in a movie in motion or not in motion. Non-dithered methods look clean, sharp, precise, and make blacks look rich, with more of that inky effect... I guess I'm like that family from AVS Forums who hated the picture when their daddy bought a high-end ZT60 plasma and refused to believe that the picture was HD due to all the noise... It really does take away the clean HD effect.
I can only hope that the final version will include simply 2 versions to pick from - noise-less (DC3, ED2, ED4), and noisy (ED whatever).
If not, then please please please make the noise-less madVR.ax files continue to be compatible with future releases and update those files to include whichever new fixes or features will be updated in the official noisy files...
I think we need way more people to test and rate these methods to find out what most people would prefer. Many opinions are based on what is supposed to look better to a knowledgeable expert, which isn't always what the case for a non-expert, who also has eyes and can judge. Its like Movie ratings or game reviews - critic reviews vs. user reviews - huge difference between is often present.
Finally, my 3DLUT is actually more accurate using noise-less DC3 madVR.ax file! Very slightly more accurate, but still...
Shiandow
11th February 2014, 00:32
I use a 3DLUT but I do not have my TV's absolute 0 black raised because I figured out how to remedy that flaw. However, I do get dithered blacks (just not the absolute black) in any build other than DC3, ED4, and ED2. I can see a definite difference even in ED9 and ED10 in any movie with blacks (deep or shallow). After experiencing error diffusion without dithering I can't get myself to watch an entire movie with noisy/dithered blacks... ruins the experience. Isn't the point of error diffusion to NOT dither anything?
According to my implementation of the algorithm of ED10 it should be possible to solve this by changing the weights to the following:
modified Floyd-Steinberg (multiplied by 0.9623):
0.5172
0.0601
0.1804
0.3007
-0.0962
madshi (unchanged):
0.7770
0.0000
0.0090
0.4861
-0.3098
This should lower the response on very black colours without changing the result much on lighter colours. Although it does lose the ability to express the difference between total black and only slightly black, but currently that seems to be more of a problem than a feature.
kstuart
11th February 2014, 02:16
Hi,
Just upgraded from IGP to HD6450 in order to try out madVR (and see if I want or need a further GPU upgrade).
FYI for later thread readers with HD6450, I find that Chroma scaler Bicubic75 and Upscaler Lanczos 3 AR work fine with that card, as long as nothing else is using the GPU. This means decoding needs to use the CPU. And I've found that using Open-CL Error Diffusion instead of Random Dithering, is a huge performance hit on this card, and can't be used.
I find that with Exclusive Mode and Aero enabled, I don't get juddering, which is good, because Smooth Motion seems to often use enough GPU to cause... juddering... so it is self-defeating on my system.
Lastly, I've found that checking Automatic Deinterlacing in madVR also significantly increases the GPU load (when the source is interlaced). If I uncheck it, then the OSD says it is off, but playback seems normal.
Does this mean AMD Catalyst is doing the deinterlacing ? Or is it passing the task all the way to my HD TV ?
I also noticed that madshi recently moved "... use half frame rate deinterlacing... " to a special "not recommended" section of "trade quality for performance". Are there any youtube videos that demonstrate the difference ? (And is anything other than madVR going to be using half frame rate deinterlacing by default?)
Thanks !
huhn
11th February 2014, 03:23
Hi,
Just upgraded from IGP to HD6450 in order to try out madVR (and see if I want or need a further GPU upgrade).
FYI for later thread readers with HD6450, I find that Chroma scaler Bicubic75 and Upscaler Lanczos 3 AR work fine with that card, as long as nothing else is using the GPU. This means decoding needs to use the CPU. And I've found that using Open-CL Error Diffusion instead of Random Dithering, is a huge performance hit on this card, and can't be used.
is dxva forcing your gpu in mid powerstate?
I find that with Exclusive Mode and Aero enabled, I don't get juddering, which is good, because Smooth Motion seems to often use enough GPU to cause... juddering... so it is self-defeating on my system.
even a intel hd4000 can do smoothmotion is error diffusion used at the same time?
Lastly, I've found that checking Automatic Deinterlacing in madVR also significantly increases the GPU load (when the source is interlaced). If I uncheck it, then the OSD says it is off, but playback seems normal.
of cause deinterlacing normally creates 60/50 fps all need to be rendered a not interlaced stream or ignored stream is treated as 30/25 fps this is way faster. bd are normally 23p and some dvd too this is even faster.
Does this mean AMD Catalyst is doing the deinterlacing ? Or is it passing the task all the way to my HD TV ?
madvr uses ivtc(film) or deint(video). the deint is used from your gpu and normally creates a lot of gpu load. out puting interlaced is pretty unusually with a htpc. you can add interlaced output modes but else progressive is outputted.
I also noticed that madshi recently moved "... use half frame rate deinterlacing... " to a special "not recommended" section of "trade quality for performance". Are there any youtube videos that demonstrate the difference ? (And is anything other than madVR going to be using half frame rate deinterlacing by default?)
tv recording can use 60 real fps in 60i like sports. so you can lose half picture information and frame rate
StinDaWg
11th February 2014, 04:01
Hello all. Long time user of madVR 1st time poster here. I am looking to upgrade my graphics card, because my current card (AMD 5450) can't handle 30+ fps content in madVR (even though it works fine in MPC-HC EVR). What is the cheapest AMD and Nvidia card that can play 1080p60 or 1080i30 deinterlaced to 1080p60 without any issues? I would also like to be able to take advantage of the new debanding filter if possible, but it's not a deal breaker depending on price.
It also needs to be low profile and preferably have passive cooling.
cyberbeing
11th February 2014, 04:18
The new build 3 looks so much better than the others to my eyes and Bam! madshi told me that 3 = 7 duh, so 7 it is.
Weird. I double checked, and "noisy3" definitely has more chunky low frequency noise and crosstalk than the old "ED7" build. If you've come around to liking the old "ED7" build as well, I'd say just have madshi go back to the old random method and be done with it.
What I personally liked most about the new "noisy2" build, is that it has the least amount of chunky low frequency noise, crosstalk, and distortion out of all the builds. This seemed to give it the most perceptually solid colors, and the smoothest and most accurate dithering transitions because of its almost sole use of low intensity high frequency noise.
For comparison purposes only, here are those two bar saturation enhanced crops from the new builds:
Noisy1 | Noisy2 | Noisy3 | Noisy4
http://i.imgbox.com/kMVbu5jK.png http://i.imgbox.com/53eXdLeF.png http://i.imgbox.com/i8D9sSvu.png http://i.imgbox.com/w1E2RohT.png
I'd be interested to see bacondither repeat his Big Buck Bunny test use with these new builds (tip when taking non-fullscreen screenshots: MPC-HC preset minimal, Navagate -> GoTo frame, ALT+PrtScn), to get a better idea how they behave in more extreme dithering scenarios. "Noisy3" feels like it has the most heavy noise of these builds, and it would be interesting to see if that was actually beneficial or harmful.
huhn
11th February 2014, 05:09
Hello all. Long time user of madVR 1st time poster here. I am looking to upgrade my graphics card, because my current card (AMD 5450) can't handle 30+ fps content in madVR (even though it works fine in MPC-HC EVR). What is the cheapest AMD and Nvidia card that can play 1080p60 or 1080i30 deinterlaced to 1080p60 without any issues? I would also like to be able to take advantage of the new debanding filter if possible, but it's not a deal breaker depending on price.
It also needs to be low profile and preferably have passive cooling.
2 slot low profile 630 gt passive version cost about 50€ in germany 730 points (http://www.videocardbenchmark.net/gpu.php?gpu=GeForce+GT+630) or a passive low profile 6570 752 points (http://www.videocardbenchmark.net/gpu.php?gpu=GeForce+GT+630) about 55 €
with software deinterlacing from lavfilter (yadif) this should work.
but you should go for a aktive r7 250/7750 or gt 640 WITH GDDR 5! the gt 640 cost 65 €+ the r7 250 71€+ and the 7750 80€+
but they can do it for sure.
kstuart
11th February 2014, 05:34
with software deinterlacing from lavfilter (yadif) this should work.
Is that "full frame rate deinterlacing" (and not half frame rate deinterlacing) ? If so, how do I set things so that the deinterlacing is done by LAV in software, and not by madVR or Catalyst ?
Hopefully a user knows this, since nevcairiel is very busy these days with development.
huhn
11th February 2014, 06:36
http://abload.de/img/deintlavfilterd4riz.png
that's all.
i don't know how much cpu power this takes so not sure if this is usable with 1080i60 on your cpu. this should work with dxva copy back. didn't tried it i'm sorry.
try it out with bilinear filtering in madvr or EVR CP
StinDaWg
11th February 2014, 08:27
2 slot low profile 630 gt passive version cost about 50€ in germany 730 points (http://www.videocardbenchmark.net/gpu.php?gpu=GeForce+GT+630) or a passive low profile 6570 752 points (http://www.videocardbenchmark.net/gpu.php?gpu=GeForce+GT+630) about 55 €
with software deinterlacing from lavfilter (yadif) this should work.
but you should go for a aktive r7 250/7750 or gt 640 WITH GDDR 5! the gt 640 cost 65 €+ the r7 250 71€+ and the 7750 80€+
but they can do it for sure.
Wouldn't the Nvidia be able to use CUVID for deinterlacing rather than yadif, while still being able to switch over to "film mode" deinterlacing for telecined stuff? I'm not really familiar with Nvidia cards, I've always owned AMD.
leeperry
11th February 2014, 09:21
Weird. I double checked, and "noisy3" definitely has more chunky low frequency noise and crosstalk than the old "ED7" build.
To be honest I haven't retried 7, it's just that 3 looks outstanding on my rig and madshi said that 7 = 3....many ppl also voted for 7, so there ya go :)
I watched OBLIVION yesterday and was blown away by the pop effect, sharpness and soundtrack.
"Noisy3" feels like it has the most heavy noise of these builds, and it would be interesting to see if that was actually beneficial or harmful.
More noise can increase the pop effect, not sure why and how but it works as long as it's not completely random.
turbojet
11th February 2014, 09:24
StinDaWg: You could deinterlace with cuvid but it adds gpu load. A passive and/or low profile gpu is not going to very powerful for madvr and you want to limit gpu usage as much as possible.
You may want to see if the nvidia 800 series will have a more powerful low profile gpu, it may only be about a month if the rumors are true.
Thunderbolt8
11th February 2014, 10:03
I wonder whats the point to buy a new GPU now anyway. In order to use everything like NEDI doubling, OpenCL/DirectCompute etc. youd need the most powerful GPUs you can get and then you can almost be sure they will create a lot of noise which is not really what youd want when watching movies. so wouldnt it make more sense to wait at least one more year so that those GPUs will become more mainstream than highend and therefore they also will become more quiet over time? (or better said then mainstream cards which are not that noisy could then be as fast as nowadays high-end models)
ryrynz
11th February 2014, 10:19
What I personally liked most about the new "noisy2" build, is that it has the least amount of chunky low frequency noise, crosstalk, and distortion out of all the builds. This seemed to give it the most perceptually solid colors, and the smoothest and most accurate dithering transitions because of its almost sole use of low intensity high frequency noise.
Agreed, I think ideally this is what we're looking for.
I'd be interested to see bacondither repeat his Big Buck Bunny test use with these new builds
Yeah...get on it bacondither, please capture the same frame for each one :)
Ver Greeneyes
11th February 2014, 10:26
I wonder whats the point to buy a new GPU now anyway. In order to use everything like NEDI doubling, OpenCL/DirectCompute etc. youd need the most powerful GPUs you can get and then you can almost be sure they will create a lot of noise which is not really what youd want when watching movies. so wouldnt it make more sense to wait at least one more year so that those GPUs will become more mainstream than highend and therefore they also will become more quiet over time? (or better said then mainstream cards which are not that noisy could then be as fast as nowadays high-end models)The question of upgrading is the same as it always has been, really. If you have a lot of money to throw around or it's an actual investment, get the newest offering whenever you can. Otherwise, just upgrade if you really need a new card (which is something only you can decide). If you're at the end of a graphics card cycle, the existing cards will be cheaper than the ones just around the corner - if you're at the start of a cycle, then there won't be anything new for a year anyway.
Having said that, I am at least looking forward to seeing if the release of the PS4 and Xbox One cause the development of GPUs to accelerate at all. The latest offerings are already more powerful than what the new consoles bring, of course, but at least the difference is no longer as ridiculously big as it was with the PS3 and Xbox 360. I don't think Nvidia's Maxwell will be that generation, since it's too soon after the new consoles' release, but maybe the generation after that. The same probably goes for AMD, though they seem to be focusing more on integrated graphics these days (with the consoles being a big example).
DragonQ
11th February 2014, 10:49
2 slot low profile 630 gt passive version cost about 50€ in germany 730 points (http://www.videocardbenchmark.net/gpu.php?gpu=GeForce+GT+630) or a passive low profile 6570 752 points (http://www.videocardbenchmark.net/gpu.php?gpu=GeForce+GT+630) about 55 €
with software deinterlacing from lavfilter (yadif) this should work.
but you should go for a aktive r7 250/7750 or gt 640 WITH GDDR 5! the gt 640 cost 65 €+ the r7 250 71€+ and the 7750 80€+
but they can do it for sure.
I'm pretty sure the GT 630 is a rebranded GT 440, which might be cheaper to find. Don't expect to be able to use any high quality settings in MadVR with this kind of card. Bicubic75 AR chroma, Lanczos3 AR upscaling, and CR AR downscaling should be possible. Deinterlacing is good and you can get passive ones that are silent, which is nice for HTPCs.
I'd love to use all the great new MadVR features in my HTPC but (aside from having to change media centre software) it'd require a beefy GPU and I doubt any of them are quiet enough for an HTPC environment. :(
ryrynz
11th February 2014, 10:52
Having said that, I am at least looking forward to seeing if the release of the PS4 and Xbox One cause the development of GPUs to accelerate at all.
I think all the rage is going to be directcompute and OpenCL going forward, so expect huge performance gains in this area which will be great for madVR.
Intel is also going to be focusing a lot more on graphics which will also boost competition. Just watch as the GPU crushes the x86 CPU into near insignificance.
Now if only we could get silent HD7870+ performance.. always seems the next generation can never come soon enough.. Thanks Madshi. :)
I'd love to use all the great new MadVR features in my HTPC but (aside from having to change media centre software) it'd require a beefy GPU and I doubt any of them are quiet enough for an HTPC environment. :(
I think a HD7770 is fine for Jinc3AR + Bicubic 75AR + ED isn't it? That's fairly powerful and doesn't draw that much power.. It's when you get into NNEDI that the GPU requirement starts to hit the roof.
Skwelcha
11th February 2014, 13:10
Again I have a problem which could be an issue with MadVR or MPC-HC... With some files I can't go directly into exclusive mode. When I enable fullscreen with ALT + Return, in some files, the picture enlarges but the header from MPC-HC is still seen and exclusive mode isn't running. Example: http://www7.pic-upload.de/thumb/11.02.14/68tec14cizs4.png (http://www.pic-upload.de/view-22234975/1.png.html)
After that I have to press ALT so that the menu bar shows up like here -> http://www7.pic-upload.de/thumb/11.02.14/bt54oxdw4kpp.png (http://www.pic-upload.de/view-22234982/2.png.html)
If I press now outside the window on my first display, it finally goes in to the real exclusive mode... I have this problem since a few MPC-HC versions, so I guess it's more likely an MPC-HC error?
leeperry
11th February 2014, 13:36
Now if only we could get silent HD7870+ performance
The more I think of it, the hotter I am for an Asus 660 as these cards are really dead silent: http://www.techpowerup.com/reviews/ASUS/GeForce_GTX_660_Direct_Cu_II/26.html
the GTX 660 Direct CU II offers phenomenally low noise levels. ASUS apparently knows how to make quiet cards
They often show up on the used market as hardcore gamers want a 700 board now and these come with a 3y warranty.
What remains to be seen is their NNEDI performance as I would really hate being forced to turn down the 64x NNEDI 720p@1080p with CR LL AR + J3AR chroma & ED my o/c 7850 allows me to enable :o
nevcairiel
11th February 2014, 13:42
I got one of those ASUS GTX 660 DirectCU II, never heard any noises from it.
Didn't bother to test NNEDI though, because I was too lazy to install ages old drivers.
leeperry
11th February 2014, 13:45
yep, I briefly owned this board (https://www.asus.com/Graphics_Cards/GTX660DC2OCPH2GD5) and it was so amazingly quiet :)
but can it do 64x NNEDI 720p@1080p with CR LL AR + J3AR chroma & ED?
OTOH my 7850 GPU can go all the way up to 90°C (http://www.techpowerup.com/reviews/Powercolor/HD_7850_SCS3_passive/30.html) so I can just throttle down its fan et voilà ))
Xaurus
11th February 2014, 13:58
leeperry,
I have a 660 Asus and to be honest it struggles with NNEDI. 64x 720p@1080p is a no go.
I am currently using 32x 720p@1080p with no error diffusion and Jinc3 for luma/chroma.
leeperry
11th February 2014, 14:10
Oh bummer, hopefully DC will save the day :)
kasper93
11th February 2014, 14:55
@leeperry: It won't. Madshi already said that the DC version is 10x slower. Hopefully he will improve that, but I don't have much hope. For me DC is crucial because OCL interop kills performance on HD5870 ;/ I wasn't able to use ED with OCL...
leeperry
11th February 2014, 15:49
well maybe the nvidia drivers engineers will do their homework for a change.....but AMD destroys Nvidia at Bitcoin mining, can the gap ever be bridged? (http://www.extremetech.com/computing/153467-amd-destroys-nvidia-bitcoin-mining)
GCN executes INT32 code at the same rate as double-precision floating point.
The Radeon 7790, a $149 GPU, offers 80% of the GTX Titan’s performance for 15% of its price
OpenCL-based Luxmark 2.0 benchmark is now running under the Titan [..] The Titan is a huge improvement to the GTX 680, but it’s still half the performance of the HD 7970.
OpenCL tests demonstrate that while the Titan crushes the HD 7970′s performance in gaming, it can lag by up to 50% in other tests, despite a far larger transistor budget, more cores, and twice the price.
Further CUDA optimizations might improve the overall performance scenario slightly, but there’s no miracle kernel with 100% increased performance waiting in the wings. Even if there were, it would scarcely matter — GK104′s performance would need to quintuple for the GTX 680 to even be competitive.
Should you mine if you have an Nvidia card? You can, but be aware that power costs make this a losing proposition
They lost the OpenCL war on pretty much every front apparently.
Ver Greeneyes
11th February 2014, 16:08
To be fair, bitcoin is a pretty recent development - even if the current architectures don't perform well, that doesn't really say anything about future ones. What's also interesting is that as I recall, Nvidia was all over GPGPU a few years ago - but Kepler was actually a significant step back in terms of double precision performance compared to Fermi (with the exception of the Quadro-based Titan which came out later). Of course double precision isn't everything, but if we're talking about GPGPU rather than gaming, it's telling that they neutered it so much. Meanwhile, AMD raced to catch up and seems to have thoroughly overtaken them now. I think Nvidia made a mistake in underestimating the continuing advance of GPGPU, especially with what looks like a lot of rich BOINC and Folding@Home users buying cards specifically for crunching. I still prefer their drivers, though, aside from the current OpenCL interop problems ;)
leeperry
11th February 2014, 16:28
Interop or not, they apparently completely lag behind for GPGPU...so much for all the CUDA commercial fluff, plain pathetic.
Ver Greeneyes
11th February 2014, 16:34
Well like I said, CUDA was the only thing out there for GPGPU for a while. AMD's CAL never really caught on from what I can tell, but they got OpenCL working while Nvidia sat on its laurels, and now their hardware is superior. It's just kind of ironic given that Nvidia had a head start.
leeperry
11th February 2014, 16:35
Long story short, nvidia will always be a terrible choice for advanced mVR processing apparently...tbh I'm impressed by how much this 90€ o/c 7850 gives in mVR :)
Ver Greeneyes
11th February 2014, 16:45
Long story short, nvidia will always be a terrible choice for advanced mVR processing apparently...
I think that's a silly conclusion, to be honest. Let's see how things look in a few years. Of course for right now, AMD does look the better choice for these purposes, though it's a pity they run so hot.
GREG1292
11th February 2014, 16:47
yep, I briefly owned this board (https://www.asus.com/Graphics_Cards/GTX660DC2OCPH2GD5) and it was so amazingly quiet :)
but can it do 64x NNEDI 720p@1080p with CR LL AR + J3AR chroma & ED?
OTOH my 7850 GPU can go all the way up to 90°C (http://www.techpowerup.com/reviews/Powercolor/HD_7850_SCS3_passive/30.html) so I can just throttle down its fan et voilà ))
I will say that the GTX-770 needs EVGA K-Boost to do do
Needi3 64,CR LL AR + J3AR chroma & ED and doubling.
720P Matrix I watched last night looked stunning on
build 3/7:) To be quite honest am just going to enjoy watching movies for awhile.
Until Nvidia Maxwell comes out and I can do
needi3 256 I am one happy madVR user:D
leeperry
11th February 2014, 17:07
I think that's a silly conclusion, to be honest. Let's see how things look in a few years. Of course for right now, AMD does look the better choice for these purposes, though it's a pity they run so hot.
What I meant is that there is no "sweet spot" to be expected in their current or previous lineups, but they more than likely want to keep the high GPGPU performance for their megabucks pro boards.
This one seems pretty silent: MSI R9 270X Gaming 2 GB (http://www.techpowerup.com/reviews/MSI/R9_270X_Gaming/23.html)
The card should be almost inaudible in both idle and full-on gaming, even for extended periods of time.
As that BTC link said power consumption will also be much lower than with nvidia, making their boards crunch GPGPU is a major waste of money as it would appear.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.