GPU-Z measures GPU decoding load

[Przemek_Sperling]

I just wanna let you know that the newest (v.0.38) GPU-Z http://www.techpowerup.com/downloads/1709/TechPowerUp_GPU-Z_v0.3.8.html measures GPU load during hardware H.264 decoding. It works like a charm with Mirillis Splash Lite and ATI chipsets.

I wonder GPU load on different GPU chipsets - is Radeon 5850 much better than, for example, Radeon 4350?

[CiNcH]

It only shows shader usage which does not have anything to do with H.264 decoding. Some post processing is done on shaders in case of DXVA or if you use MPC-HC with its shaders.

[mark0077]

At the end of the day, what are people trying to achieve with low % load.. Is it to get less power usage? Less noise?

I suggest getting somethign to measure total power usage, or total noise output from your machine if these are of concern. With so many different CPU's and GPU's I don't think measuring their % usage really means anything, compared to the effect they have on power, heat, noise etc. Like 1% usage of my GTX295 (hot and noisy) might be equivalent to 20% usage of a small ATI card in terms of heat and noise.... The 1% and 20% really don't mean anything here if your not concerned about using the remaining % for something else (or your content is pushing usage up near the 100% where you really do care thats its in a comfortable zone).

[Przemek_Sperling]

At the end of the day, what are people trying to achieve with low % load.. Is it to get less power usage? Less noise?

Actually both.:devil:


I know that I compare apples and oranges but you gave a good example. From the point of view of energy your GTX295 makes little sense because the card needs more energy in idle than my Radeon 4550 in stress. So I bet you boughtyour card in order to play games, not to offload anything.


Comparison IMO makes sense. There are different classes of cars and their mileage varies. No one denies, however, that some cars have better mileage than others. I do not compare my ol' good Mercedes 560 SEL with my Toyota Yaris 1.4 D-4D. I just want to see how other.similar cars compare to mine.

PS My machine is almost all passively cooled. There is onlu one fan in my computer - this one which works with my Noctua cooler (@850 rpm). The rest is passive - including my Yesico PSU.

[honai]

That comparison is flawed becaused it suggests that, with gas prices as they are now, a difference in car engine efficiency between a "Mercedes 560 SEL" and a "Toyota Yaris 1.4 D-4D" translates into the same difference with GPUs, i.e. driving one over the other saves you a couple of USD/EUR for every 15 minutes you drive.

That is absolutely not the case with PC systems. Assuming that your machine runs 8hrs/5days/52weeks per year (I'm being generous), a difference of 100W (I'm being generous) in continuous total power consumption, at electricity costs of .20 EUR/kWh ((I'm being generous), translates into only ~ 42 EUR per year! At 4 hours per day (I guess that's much more realistic for anyone playing games or watching movies) we're at 21 EUR per year. And that's assuming a constant power load, whereas in reality the CPUs and GPUs of the last 3 years do indeed throttle significantly when under lower load, thus making the margins even smaller.

However, even a small difference in car engine efficiency of 7 litres vs. 8 litres per 100 km, assuming yearly milage of 10.000 km (still generous), assuming an average gas price of 1 EUR/litre (still generous), translates into 100 EUR per year. A difference of 7 litres vs. 10 litres (a much more realistic figure for new vs old car) already translates into 300 EUR per year.

I point this out because it has become very common to attribute very high costs to typical PC power consumption for the average user, while in reality there are other areas in life that expose much higher power saving potential (in terms of money spent), and swapping a 100W GPU for a 50W GPU will yield no measurable cost savings for the non-server, non-24/7 system setup.

So today a lower load during video playback has nothing to do with "saving money" or "going greener", but only with reducing noise levels by way of eliminating fan noise.

[Przemek_Sperling]

Sure, to some extend I agree. I write "some" because I dare say that HW decoding is also more comfortable than SW decoding. I use Splash Lite and I can run quite a lot CPU demanding apps in background when I watch TV or camera files. I can run them with SW decoding but I experience stuttering when I watch DVB-T or watch Video TS files from my camera.

PS I do not know how much energy is exactly eaten by NV 295 in idle (I think that Anandtech estimated it to 65W, but I am not absolutely sure). Anyway in stress it needs 289W http://www.nvidia.com/object/product_geforce_gtx_295_us.html Radeon 4550 by contrast needs 25W in stress http://ati.amd.com/products/Radeonhd4500/index.html Actually the real consumption of the system (not the cards themselves) is a bit higher because PSUs are never 100% efficient (~80%).

[Cyber-Mav]

ok here is some hard evidence, tested on my gtx260.
http://img229.imageshack.us/img229/9299/videoengineloadtest.jpg

as shown in the image, i can play back 2x 1080p videos + a 720p video simultaneously. 3 x 1080p results in dropped frames on the 3rd video. this is using coreavc cuda as you can see 3 instances of it in the system tray.
performance is not bad at all for a purevideo generation 2 decoder on my card. the 2 1080p videos are my own blueray rips of transformers and iron man, both are 12gb big and the 720p video is a samsung hd demo that came with my tv which i encoded using x264. all encodes done with x264 rev1183 level 5.0 dxva, cabac enabled etc

[Mangix]

this doesn't work for me on windows xp 32bit geforce 8500 gt 195.62 drivers.

[namaiki]

this doesn't work for me on windows xp 32bit geforce 8500 gt 195.62 drivers.

What in particular? Screenshot?

[patch1]

I wonder GPU load on different GPU chipsets

I agree having some idea of the reserve hardware processing capability is useful as it is an indicator of how future proof the hardware is likely to be. Especially with openCL starting to gain acceptance.

Running multiple movies is an interesting way of testing.

Unfortunately software limits can over ride hardware capabilities http://forum.doom9.org/showthread.php?t=150325

[Cyber-Mav]

ok here is few more screenshots to show cpu usage with and without cuda acceleration. cpu is a intel q9650@3.6ghz:


firstly 2 x 1080p films, ironman and transformers: coreavc running software mode (blue orbs in system tray)
http://img81.imageshack.us/img81/936/dual1080pplayback.jpg

as you can see cpu usage is around the 47% mark

now here is a screenshot with cuda enabled:
http://img263.imageshack.us/img263/4646/dual1080pplaybackwithcu.jpg

as you can see cpu usage has dropped a considerable amount, down to 10% for playing back 2 x 1080p videos simultaneously.

[Mangix]

What in particular? Screenshot?

just the load parts.

http://img80.imageshack.us/img80/1966/clipboard01pd.jpg