NVIDIA 技术支持就关于 NVIDIA HDMI 视频输出存在色阶损失问题回复的汇总

Edison

这个问题从递交到 NVIDIA 客户支持网站到现在大概历时一个半月，由于中间时不时需要我这里拍摄重新取证，而我这里拍摄屏摄需要大量的准备（午夜、三脚架、搬机器、换显卡、重新安装驱动、进行各种可以确定排除各种干扰的设置），加上有时候我都快忘记这件事情，因此拖沓了不少时间，而相比较之下 NVIDIA 的技术支持倒是回复的比较快。

备注：NVIDIA 的官方技术支持网站是 nvidia.custhelp.com 。

这里只张贴 NVIDIA 方面的回复：

Response (Ray)

05/09/2011 11:57 AM

Hi Edison, The case was escalated to me. The next step will be to escalate this to our quality lab and see if they can replicate the failure. But judging from the data this doesn't appear to be something that our lab can easily replicate. I can submit the data you provided and see if they can extract anything from there. I did discuss your case with other agents and they suggest checking to make sure your desktop resolution is set to a PC mode and not a SD, HD mode. According to engineering full color range is only available when desktop is set to a PC resolution. If you go into the NVIDIA Control Panel > Change Resolution, you should see a list of resolutions. The first section will be "SD, HD" but if you scroll down you should see a "PC" list of resolutions as well. Most HDTV have separate mode table for PC. If PC resolutions is available try setting to one of the PC modes and see if that helps. Although I would expect the result to be the same for both 260.99 and 270.61 but according to your data it appears the 270.61 quality is worst then the 260.99? If yes then that would point to a possible driver issue, but may not be the same full range color issue with SD, HD resolution. Best Regards, NVIDIA Customer Care

Response (Ray)

05/10/2011 01:14 PM

We escalated the bug to one of our engineers and he believe the driver is using Limited Range and not Full Range due to the 1920x1080x60Hz/P mode . Per HDMI spec 1.4a chapter 6.6 "Video Quantization Ranges" at page 116: "Black and white levels for video components shall be either "Full Range" or "Limited Range." Limited Range shall be used for all video formats defined in CEA-861-D, with the exception of VGA (640x480) format, which requires Full Range." He believes in your case, the problem was in 1920x1080x60Hz/P which is one of CEA-861-D defined spec. That's why we use "Limited Range" by default. In order use "Full Range" by default you may add a regkey "SetDefaultFullRGBRangeOnHDMI"(dword)=1 as the workaround. The correct RGB limited range is 16-235 in RGB/8bpc mode or 256-3760 in RGB/12bpc (deep color) mode. Can you use the attached "DispRegEdit.exe" tool and add the "SetDefaultFullRGBRangeOnHDMI" registry key and see if that helps. The tool should take you to the area in the registry to add this key. Add the key and then restart Windows and check again. The engineer would like to know if you're using a HDMI monitor that support deep color and how the range limit was measured as 32-219. But he like you to try the registry key to confirm.

Response (Ray)

05/12/2011 11:04 AM

Thanks for the update. The bug has been escalated to engineering and they're investigating. They did request for monitor EDID and how the range limit was measured as 32-219? Can you provide that information? Here are steps below to extract the monitor EDID data so our engineering team can review. To get the EDID, please do the following: 1) Download the program here: http://www.entechtaiwan.com/util/moninfo.shtm 2) Install and run the program 3) Click on File->Safe As 4) Safe it as a *.Bin file and attach it to your case. -Ray

Response (Ray)

06/03/2011 03:49 PM

Hi Edison, I just received any update from engineer who is investigating the issue. With latest 275.33 driver and the same GeForce GTS 450 in HDMI limited range RGB/8bpc mode, our engineer measured the dynamic range as 16-234 which is 1 off from the HDMI spec (16-235). But it's still a lot better than the you're reporting which is (32-219). Can you provide more detail on how you are measuring the pixel values? In our lab, we have a tool to output a gray scale from 0~255 and we used the Astro analyzer to measure the pixel data from the HDMI output. Could be the way we are measuring the pixel value. If we can better understand how you measure then the lab can hopefully duplicate and re-analyze the data. -Ray

Response (Ray)

06/08/2011 01:04 PM

Thanks for the update. Since DSLR was involved in the methodology, there are more factors involved such as if the panel truthfully represent the received pixel data and if the DSLR does not introduce any extra artifacts. But overall our methodology which takes the pixel data directly from the HDMI port by an analyzer is much more reliable. We used your test image (gray_0-255.png) and measured [16-234] from the analyzer in RGB mode, which is expected level. The engineer believes this could be related to the specific HDMI monitor and how well it represents the color from the received pixel data. Other display setup like dithering could also contribute to the final result. But there is no way for us to verify for sure without this particular monitor in house to verify. Now you had indicated that when using the 260.99 you don't see the same quality issue? What was the 260.99 range measured? Was this using the same DSLR methodology to measure the range?  If yes than that would suggest a change in the driver if all variables same but the driver version. Our internal measurement using an analyzer directly to the HDMI port did not yield any quality concerns, in fact based on measurement all is as expected range at [16-234]. So in the end we may have to get this HDMI TV in house and apply some experiment on it to find out what may improve the display quality on this panel. But overlay there is no wrong-doings so far on the graphics driver side from what he can measured. Can you give me the brand and exact model of this monitor so we can check if we have this particular model in any of our labs?

Response (Ray)	06/17/2011 10:15 AM
Hi Edison, Our engineering team did more measurement and confirm that the output range is within specification. We are measuring directly from the HDMI output and all data suggest normal and expected RGB color range. They suspect the final display on the HDMI TV is enhanced by the TV internal video filter which is likely responsible for the loss of the RGB color range. Unfortunately we don't have this TV so we are unable to confirm 100%. As far as they can tell there is no quality issue based on directly HDMI output measurement. -Ray

到此，这个问题已经从 Researching 设置为 Unresolved，我也不可能把电视扔过去给他们研究，而他们使用专业仪器测试出的 HDMI RGB 输出是 16-234，基本符合 HDMI 规范，所以我也就懒得继续跟进，大家如果遇到类似的问题并且你的电视机型号是国际市场上通用的型号版本，可以考虑上去 nvidia.custhelp.com 试试看。

zxl7288436

好长·······

Edison

对事情不是很了解的玩家可以翻阅我之前的帖子，我也在他们的官方玩家讨论区发过一贴：
http://forums.nvidia.com/index.php?showtopic=198883

voidshatter

也就是说我的U2410最好别用HDMI了？

junychen

N卡接电视是没戏了啊。估计这个硬件的Bug吧

whindsonh

这问题我老早就发现了，家里两台电视，灰阶1肯定丢失（RGB都是），xp，win7也都是。。。。
不过可以用adobe gamma校正

whindsonh

7800gt 发表于 2011-6-21 15:52
接电视看片有丢么

有的，看哈6，这片子很bt，很多画面的暗部信息有大量的0和1，有加载adobe gamma和没加载还是有比较明显区别的。

erhang

回复 Edison 的帖子

能说的通俗一点吗？

后来我测过INTEL显卡的HDMI和NV是一样的，也丢色，
只有ATI显卡才能让HDMI效果和DVI保持一致。

用TV其实不好比较，毕竟TV没DVI口。
用带HDMI、DVI的显示器，最容易看出来。

附上我用Monitor Asset Manager收集到的数据。

万恶之灵

唉~~~~~~~~

Viperknight

抓住问题核心就可以了，这个问题NV研究以后未能解决，也就是n卡的HDMI输出是可能丢失色阶，如果你用N卡，就不要轻易用HDMI输出，如果你非得用HDMI，那么最好不要考虑N卡，无论国内国外，似乎未能完全按照HDMI的电视机都很多，显示器的话，基本上都没有按照那个规范了

erhang

又测试了一下ＮＶ的ＨＤＭＩ效果，发现新驱动改善了，不再像以前我拍照片的那样黑白渐变愣子严重了，感觉跟ＤＶＩ没什么区别了，现在ＮＶ视频已经没什么缺陷了。

（以上为错觉，nv新驱动HDMI还是看黑白渐变楞子比dvi明显。）

luckissy

erhang 发表于 2012-6-1 00:46
又测试了一下ＮＶ的ＨＤＭＩ效果，发现新驱动改善了，不再像以前我拍照片的那样黑白渐变愣子严重了，感觉跟 ...

那就是修复咯

Edison

erhang 发表于 2012-6-1 00:46
又测试了一下ＮＶ的ＨＤＭＩ效果，发现新驱动改善了，不再像以前我拍照片的那样黑白渐变愣子严重了，感觉跟 ...

建议用旧驱动对比一下，这样才好确认。

lagoogle

这种问题。

只有在极端使用情况下，才会比较明显吧？

日常使用情况下，很难看出来啊

erhang

Edison 发表于 2012-6-1 01:43
建议用旧驱动对比一下，这样才好确认。

用旧驱动拍照的，当初就是我。

昨天又测了一下，感觉渐变的细腻程度还是不如dvi。
NVIDIA，ati区别其实还是测试片的那种细线条的表现有点区别，还有就是ati的红色总是往右下偏移一个像素。

erhang

火云喷神 发表于 2012-6-1 07:58
本区不是说n卡高清全面超越amd，笑了
接高清电视只有独立a卡，集成都不行。测试总是测试，intel，nv差尿了 ...

NVIDIA/ati，视频有一点区别，但是不截图放大比较，很难察觉。
谈不上谁更好。
要想画质变好，得买个好的显示设备。