Sapphire HD5770 Viper-X: high GPU utilization watching HDTV

Just a question...

I recently upgraded my hardware because of a replacement for my old 21" CRT 1280x960 monitor with a new 24" LCD 1920x1200 monitor, being used as Monitor #2 in a 2-monitor setup. Monitor #1 is also a 24" LCD 1920x1200 monitor and remains unchanged.

To properly support the second digital monitor, I upgraded the video card from a DVI/VGA HD4850 to a DVI/DVI HD5770.

When I was using the HD4850 card in the machine Aida64 reported about 20% GPU utilization when I was watching 1080i HDTV (in a window on monitor #1). Now, with the HD5770, Aida64 reports much much higher GPU utilization... around 65%. GPU utilization is lower when watching 720p, so I'm certain what I'm seeing is a reasonably accurate representation of true hardware utilization related to HDTV.

It's hard to tell about ordinary CPU utilization when watching these HDTV programs, as it was and is relatively low. So if it's now lower (because the HD5770 hardware is taking on the load), it's not that obvious. Even with the HD4850 there was not a high CPU utilization when watching HDTV.

I don't know if this is anything I need to be concerned about, or whether this is good or bad, or whether this is simply an indication that the HD5770 hardware is actually performing much more of the HD-related processing whereas with the HD4850 that same required processing was handled by software in the drivers.

So... can anybody enlighten me as to whether what I'm seeing is perfectly normal and to be expected? Could this simply be a different way that Aida64 sees "GPU Utilization" from the HD5770, and is perhaps just more accurate than it was with the HD4850?

Ivan the SoSo said:

The 5770 is only one step above the 4850 Graphics Card Hierarchy Chart : Best Graphics Cards For The Money: February 2012 I am a bit confused about the card and monitor swap....If the 4850 was driving the 1280x960 vs the 5770 driving the 1920x1200 I don't really see any problems:)

The 4850 drove both 24" monitors at 1920x1200, but one was digital (DVI) and the other was analog (VGA). That card only has one DVI output. That bothered me, as I wanted digital performance and presentation from the new monitor.

The 5770 is dual-DVI, and that solved my objective. Both monitors are still at 1920x1200, but now both in digital mode. The second monitor just looks better in digital mode than in analog mode, especially at that resolution.

Ivan the SoSo said:

The 5770 is only one step above the 4850 Graphics Card Hierarchy Chart : Best Graphics Cards For The Money: February 2012

Nevertheless, the 5770 is certainly "stronger". See output from GPU-Z in below screenshots.

But that's still not relevant to my question as to why the "GPU Utilization" should be so much higher using the 5770 than using the 4850, when the "CPU Utilization" is essentially the same on both machines running WMC for playback.

Here is a demonstration, of the exact same "CSI: Original" played back on two very similar machines (both Intel DualCore 3.0Ghz CPUs, 4GB RAM). One now has the new HD5770 (dual 24" Eizo LCD monitors) and the other has the old HD4850 (single 24" Eizo LCD monitor). WMC running on both systems. Catalyst 12.1a preview drivers running on both systems. 24" Eizo LCD monitor (DVI) used for HDTV window on both systems.

(a) HD4850 machine, showing about 21% GPU utilization:



(b) HD5770 machine, showing about 65% utilization.

Looking at GPU-Z output, I notice that the HD4850 uses DirectCompute 4.1, 256bit bus width and the HD5770 uses DirectCompute 5.0, 128bit bus width.

HD4850 is 61.4GHz bandwidth, HD5770 is 76.8GHz bandwidth.

HD4850 is DDR3 memory, HD5770 is DDR5 memory.

HD4850 is 640MHz GPU clock, 960Mhz memory clock; HD5770 is 860MHz GPU clock, 1200MHz memory clock.

HD5770 is certainly "stronger". So why is it "so busy" with the exact same task?


(interestingly, comparing the two supposedly identical screenshots from "CSI", it looks like the image from the HD5770 is actually noticeably superior. Sharper, better contrast, better color. The HD4850 image is a bit washed out in comparison. I wonder if that is real or not. I wonder if the HD5770 is actually doing a better job of presenting the HD image. Or maybe I'm just on a slightly different frame, which is genuinely slightly different. Very interesting)

Ivan the SoSo said:

If you look at the graphs the 4850 hits higher peaks than the 5770...at one point almost hitting 100%. The 5770 only gets close to 80%:)

The graph I'm assuming you're looking at is from PERFMON, which is a "CPU utilization over time" display from each of the two [very similar] machines. It's unrelated to GPU utilization on the two video cards. I don't think this should be terribly significant, as it also reflects what else I might have been doing over the past few minutes getting screenshots set up, starting GPU-Z, etc. So it isn't an accurate indicator of WMC-caused CPU utilization, unless I let things stabilize and take a screenshot. For sure, it has nothing to do with GPU utilization.

In other words, it is the Aidi64 "hardware monitor" information on the left-lower corner which is most relevant, showing the "instantaneous" values for CPU utilization and GPU utilization while WMC is presenting the video frame shown in the screenshot.

Now one other piece of interesting information. The GPU utilization seems directly related to the bitrate in the underlying HD program transport stream. In other words, here in LA KCBS-DT (which is where that "CSI" program is carried) is the only network not using sub-channels, but instead dedicating essentially of their bandwidth to their one-and-only 2.1 CBS programming content. So it's up at about 15Mb/s bitrate.

In contrast, if I watch one of the basic cable channels where TWC/LA may have either recompressed (to achieve lower bitrate) or the original content from that network is simply at a much lower bitrate than KCBS-DT, the GPU utilization shown on the HD5770 is also much lower... say between 30% and 55%. This does seem reasonable and consistent, if the hardware on the HD5770 is being asked to deal with a lower bitrate datastream rather than a higher bitrate datastream.

So the question then is why would the HD4850 NOT also reflect this same "lesser demand on GPU" when fed lower bitrate programs?


Which brings me back to the original question: why does the HD5770 show so much more GPU utilization than the HD4850, when watching the same datastream with WMC? If the result is seemingly better picture (which to my eyes is undeniable, as revealed by my two accidental screenshots which demonstrate a clear improved result from the HD5770) then I have no complaints and will just keep my mouth shut.

Otherwise, this is really just a question for intellectual curiosity, to come up with an explanation why these two video cards seem to behave differently as far as GPU utilization while watching HDTV.

Thinking this might be a "12.2 preview" driver issue, last week I upgraded to the latest official ATI drivers. Didn't make any significant difference.

Anyway, most interesting is that the problem seems tied entirely to 1080i HDTV, where the GPU usage sits at around 65%. Interestingly, I also seem to have a fairly common lip-sync audio problem when playing 1080i that varies in degree but is almost always present. The lip-sync 1080i problem is more prominent with higher bit-rate 1080i (e.g. OTA 1080i channels, rather than TWC/LA 1080i channels which are re-compressed and thus at lower bit-rates).

But if I watch 720p HDTV, the GPU usage can appear at between 0% (yes, 0%, for a short time anyway) to no more than 25%, with a typical value of maybe 10-15%. Also, zero lip-sync audio problem.

For example, here is 720p showing only 1% GPU usage. Also note the very very low CPU usage for 720p: