VP11 Benchmark on GT 430

Red Prince wrote on 10/19/2011, 2:59 PM

I downloaded the VP11 Benchmark and tested it with my nVidia GT 430, which is the lowest end CUDA 2.1 (Fermi) GPU supported by VP 11.

I set my preview window to 900x506x32, the same Sony used for their own testing. When I previewed the project, the speed varied, as expected since the project contains different levels of complexity. The highest speed I got was 29.902 fps, the lowest 1.99(something) fps. I got the same speeds when previewing using the CPU and when previewing with the GPU.

I then rendered to the same formats Sony used. The MC AVC format took 4:39 to render, the XDCam 5:46. Both values are considerably higher than those Sony got with their high-end cards. I was curious how much, if any, faster it was than rendering using the CPU only. So I also rendered to MC AVC with the same settings, except I told it to use CPU only. This time the render time was 7:00 (minutes), which means my GPU was 1.5 times faster than the CPU. The XDCam format did not offer the choice to use CPU only, so I could not compare it.

Since I normally render using two passes, I also repeated the MC AVC renders using two passes (all other things being equal). Using my GPU it took 9:16, using the CPU only 14:31. So, in this case the GPU was 1.57 times faster.

File sizes in bytes:

XDCam - 193,242,265
MC AVC GPU - 90,339,385
MC AVC CPU - 89,950,896
MC AVC 2 passes GPU - 92,820,442
MC AVC 2 passes CPU - 99,151,053

Comments

dxdy wrote on 10/20/2011, 11:00 AM

I also had excellent results with the GT 430 on my stock i7-950 (12 GB RAM) system. I ran the Rendertest 2010 with the following results:

200 seconds Vegas 10e
233 seconds Vegas 11 cpu only
157 seconds Vegas 11 GPU

All of the rendered files are the exact same size.
I am a little puzzled over why V11 CPU-only is slower than V10.

Sebaz wrote on 10/20/2011, 11:15 AM

I don't know about rendering speeds, my main concern was the playback speed with filters applied, especially keyframing of levels. So I took the time to remove my GT 430 from my HTPC and put it in my main computer. I wasn't pleased with the results, since it performs worse than with just CPU. On a simple AVCHD clip that I added just the levels filter to, with a minor keyframing in one part of it, with CPU only it gives me full fps at preview Best and Full. But when the card is enabled, it goes down to 22 fps, and in the area where the keyframing is applied, it goes down to 13fps.

So I'm sorry to say, but the GT 430, as far as Vegas 11 goes, is useless. Great card for HTPC though, since it handles the 29/59Hz bug in Media Center perfectly.

Red Prince wrote on 10/22/2011, 1:54 PM

Update:

I searched the web for overclocking the 430 and found this web page. I downloaded the MSI AfterBurner from there. By default my 430 uses the core clock at 700 MHz, shader clock at 1400 MHz, and memory clock at 900 MHz.

After some tweaking I found the core clock of 825 MHz, shader clock of 1650 MHz and memory clock of 950 MHz works without producing visual artifacts on the screen.

With those settings the MC AVC format took 4:03 to render. That is 1.15 times faster than my original test with no overclocking (4:39), and 1.73 times faster than the original CPU test (7:00) I mentioned in the OP.

I also did a 2-pass render at 8:01, which is 1.16 times faster than with no overclocking (9:16) and 1.81 times faster than CPU-only (14:31). During the 2-pass render, the temperature rose to 95°C, but quickly dropped down to 64°C afterwards.

Please note that the web page I mentioned above tested it with a 430 whose default memory speed was 1800 MHz, which is twice the speed of the memory in my Zotac card. So, clearly, not all GT 430 are equal, and YMMV. At the time I bought mine, Vegas was in version 9, and since I am not a gamer, all I cared about was the lowest-price nVidia card that supports 3D Blu-rays and other videos.

John_Cline wrote on 10/22/2011, 3:23 PM

"(though there does seem to be quite a few less-than-happy upgraders here at the moment)"

And a LOT more happy upgraders that have no reason to post anything on the forum.

eightyeightkeys wrote on 10/22/2011, 7:00 PM

Not to forget to optimize VP11 in your Nvidia Control Panel using the "application" setting vs. the "global" settting. Go for best performance where applicable etc...

Randy Brown wrote on 10/23/2011, 10:07 AM

Not to forget to optimize VP11 in your Nvidia Control Panel using the "application" setting vs. the "global" settting. Go for best performance where applicable etc...

I'm guessing these settings don't make a difference unless you are using GPU acceleration right?
IOW, I have a less than 2.0 card (GeForce 210) but these settings could not improve Vegas performance without buying a greater than 2.0 graphics card, is that correct?

megabit wrote on 10/23/2011, 10:19 AM

Not to forget to optimize VP11 in your Nvidia Control Panel using the "application" setting vs. the "global" setting. Go for best performance where applicable etc...

Which specific setting do you mean? Most of them are relevant for graphics-intensive apps (like games and 3D visualization); none applies to OCL....

Of course, the number of CUDA GPUs to use should be set to ALL, but other than that?

Piotr

Munster1 wrote on 10/23/2011, 2:30 PM

@Red Prince:

If they stated the default memory speed was 1800MHz then that probably means the memory was clocked at 900MHz (the same as your Zotac).

Red Prince wrote on 10/23/2011, 3:48 PM

Yes, thank you. I have just finished reading all pages of that review I linked to in my OP, and they talk about the clock speed of 900 for an effective (I think that was the word they used) speed of 1800. Quite confusing.

Red Prince wrote on 10/23/2011, 4:19 PM

Yet another update

I did another test.

It has occured to me that in my prior tests, I was only comparing the GPU or no GPU for rendering, but all the time I kept the video effects set to use the GPU. So to get a better comparison of the effect of the GPU, I went to Preferences and turned the GPU acceleration off for the effects.

I then rendered the Sony benchmark to MC AVC with GPU turned off for rendering as well.It took 8:26 to render, which means that using the GPU acceleration for both, video effects and rendering without overclocking (4:39) was 1.81 times faster than CPU alone. And rendering with overclocking as mentioned above (4:03) was 2.08 times faster than CPU alone.

So, yes, the GT 430 does work faster. Not as fast as some of the more expensive cards, but, when tuned up, it still doubles the total speed.