I just returned the ati 6870. It seemed to work fine but would not gpu accel saphire edge which is why I upgraded my card to begin with using vegas 11. So... I am off to nvidia 550ti land or 560 ti.

Like David, I tried the HD 6870. It was actually slower than either my CPU or my older nVidia (GTS 240) card. I returned it and bought the nVidia GTX 560 Ti. Happy with the speed on GPU-capable functions. I just wish MB Looks was GPU-capable!

I have tried both the Radeon HD6970 and the nVidia GeForce GTX570 (similar price), and I've done some benchmarks. I've decided that the GTX570 is going back. At least part of that is the fast that the video output of the GTX570 I have was horrible... I looked like analog video with bad/mismatched termination. Don't even know how you'd get that with a digital device... but that didn't affect the benchmarks.

My system is otherwise an AMD1090T (x6, 3.2GHz) with 16GB RAM. I'm pretty certain that the CPU performance and even the system performance (actual PCIe speeds) will have an effect on your results, too, which is why I listed this here. I do plan to look at any possible BIOS tweaks to PCIe performance at some point.

I started with a simple project: AVCHD video (HMC40 camcorder, "PG" mode, 720p60) , one crossfade, with the Sony Brightness/Contrast plug-in applied globally, and a straight render. For preview, I found Vegas 10 did 51.6fps, Vegas 11 non-GPU did 57.4fps, and with GPU I only got 40fps. Not a good start.

Rendering to MainConcept MPEG-2 (25Mb/s, 720p60, VBR 2-pass), I found Vegas 11 no-GPU rendered 9.1% faster than Vegas 10, while with the GPU, 20% slower! I then rendered to Sony AVC (16Mb/s, 720p60, CBR) and found, versus Vegas 10, VP11 did 13% faster without GPU, 25.6% faster with GPU.

Ok... so next, I layered four 50Mb/s 720p60 MXF clips in two parts. In the first, I did a animated break out into four quadrants, in the second, it was just four layers with transparency. No plug-ins. Vegas 10 and 11 did 2.4fps average playback, so did Vegas 11 with GPU on the GTX570. However, the HD6970 did 5fps. Not amazing, but still a 108% improvement.

In rendering to MainConcept again, I saw a 2% slowdown, VP11 vs. VP10, without the GPU, but a 43% speedup with the HD6970. The GTX570 crashed with some kind of OpenCL or CUDA error. I rendered to Sony AVC, same as before, and saw VP11 render 6.6% faster without GPU, 27% faster on the HD6970, and 15.9% faster with the GTX570.

That project was set to best quality and 32-bit pixels. I ran it again with 8-bit pixels and "good" quality. Very different. I found I got 18fps from Vegas 10, only 13fps from Vegas 11, no GPU, and only 10fps from Vegas 11 with the HD6970. I also found that VP11 rendered to MPEG-2 19% slower without GPU, 86% slower with GPU. To Sony AVC, I sw a 7.9% speedup with Vegas 11, but a 17% slower render once I enabled the GPU. Go figure.

Next, I ran a section of an actual animation project I was working on; this lots of PNGs, some raw AVI with alpha transparency, not many plug-ins, but lots of motion keyframing. I only ran this in Vegas 11, and only on the HD6970. I saw 5.1fps without GPU, 14fps with the GPU. In rendering, I saw a 25% speedup to Sony AVC, a 37% speedup to Main Concept AVC.

Finally, I ran Sony's VP11 benchmark. Like my animation, this has a mix of video, keyframe animation, but lots more effects. Without the GPU, I saw 8fps on my system. With the HD6970, I saw 28.5fps ...near full speed. With the GTX570, I saw 27.5fps, still very respectable.

I rendering, I got a whopping 118% speed up to Sony AVC on the HD6970, a real close 107% improvement with the GTX570. Rendering to XDCAM EX 1080i60, I saw a 210% improvement with the HD6970, a 175% improvement with the GTX570.

One curious thing is the CPU consumption. In the Sony benchmark, for example, I see 96% average CPU when rendering CPU-only... that's a pretty well tuned system. On the HD6970, I see CPU at 58% and GPU averaging 30%... and yet, this is the fastest render. The GTX570 gives me 75% CPU and 50% average GPU. I'm using GPU Shark to measure GPU, it has never crashed on my system.

I'll also point out I'm using the very latest AMD drivers. And that it's a royal pain installing these. I saw errors reported when I first installed... this was actually bogus -- they report an error if their attempt to install C++ framework stuff fails due to there being a newer version of that already in place. Major consumer hostility fail. So I tried again, which apparently boggered up their control panel software -- it gets wonky if installed more than once. So I had to run a couple of driver cleaners on my system to scrape away every bit of old AMD, ATi, and (while I was at it) nVidia stuff from my system. Seems good now.

If you're playing with effects, I wager that most run in realtime or close to with a CPU if you're applying them to standard PAL DV. And also keep in mind, only GPU-enabled effects are going to be accelerated, even when they can be. You can find the GPU accelerated FX listed, so this is easy to assure.

I'm a little underwhelmed, I'll admit, at the lack of preview performance on simple things. As shown, my system can preview 720p60 pretty much on the CPU alone. If I run video in a GPU-accelerated player (which uses DXVA 2.0, not OpenCL, so sure, it's more efficient), I can easily play four 1080p60 videos on my system, now, in full 59.97fps display. Even the fairly wonky VLC can run at least three 1080p60 videos at absolutely full speed. So I do wonder why I'm actually losing performance on very simple things, even though I actually do see a speedup on some projects. And for me, since I do videos with a ton of layers, lots of animation, etc. I think I will, at least for some of my work, see a big win from the HD6970.

AMD Driver Info:
Driver Packaging Version 8.892-110914m-125030C-ATI
Catalyst Version 11.9
2D Driver Version 8.01.01.1186
Direct3D Version 7.14.10.0860
OpenGL Version 6.14.10.11079
AMD VISION Engine Control Center Version 2011.0908.1355.23115

nVidia Driver Info:
Version: 285.62 WHQL
Release Date: 2011.10.24
Operating System: Windows Vista 64-bit, Windows 7 64-bit
Language: English (U.S.)
File Size: 147 MB

Hazy,

Very well-done benchmarks. A lot of useful information there.

In the GT line of nVidia cards, it doesn't seem like the really high-end cards are performing any better than the cards lower down in the line. There was another post (in this seemingly infinite series of GPU posts) from a developer indicating why this might be so. It seems like there may definitely be a "sweet spot" of price/performance one or two levels down from the top end of the GT or GTX line.

It sure will be very nice when we get some more information like yours so that some of us can make some decisions on whether to buy V11 and then which card to add. I have an 800 W power supply so I can put in just about anything, but I'd prefer not to have to stress the system if it's not going to buy me anything.

Again, thanks for the great set of measurements.

thanks too hazy.....

i agree with jm, a couple of steps down from top end (in my case 550ti, with soon to be tested 3 monitor out).

there's been some very informative posts among the endless card threads here, especially the one linking to cards visa vie ppro5.

i'm just dl'ing the sony bench mark to see how my new card rates.

Hazy, did you have both cards installed while testing? Or is that even possible?

I have the nVidia GTX560ti, does the "ti" reference make a difference? I'm getting excellent reproduction on my monitors with no artifacting. I did have this of built by specialists who make them for video editors.

Thanks for sharing your hard work. Very unselfish of you.

Cheers

Grazie

Ah, I see you both cards installed simultaneously! I'm just wondering, with all the variations of GPU and CPU, on/off, whether or not having both cards installed would affect the "number" outcomes? Does the amount of onboard memory on each card affect outcomes?

Helluva tangle this one.

Grazie

talk about a gordian knot......

from what i gater vegas only uses one card (?) - i did have a 250 and 210 installed prior to my new card and though they both had cuda only the 250 was used.

yeah, what's the 'ti' stand for - totally irrelevant?

google is my friend.....

TechPowerUp has spotted the "Ti" marker used in a recently-retained set of leaked 266.44 GeForce drivers. The "Ti" marker is surprising given that Nvidia previously used it nearly a decade ago to differentiate GeForce 3 and GeForce 4 GPUs with programmable shaders (Ti) from the mainstream (MX) GPUs without the shaders.

No, I didn't have both cards installed simultaneously. That might in theory lead to complications between the two. And I don't think my power supply would be happy with that.

Hazy, did you have both cards installed while testing? Or is that even possible?

In theory, I could have tried to put both in, since I have two 16-link PCIe slots. But I didn't know for a fact they'd both be treated equally, or even if the drivers would coexist. And my power supply would most likely voice a strong objection.

The 560Ti has 384 CUDA cores, clocked by default at 822MHz, the 560 has 336, with a range of clock speeds. Both 560s are on a 256-bit memory bus.

The 570 I tested, the EVGA "Superclocked" edition, has 797MHz core clock and 480 CUDA cores on a 320-bit-wide memory bus.

You would expect performance to go up with the number of CUDA cores, the clock speed, and the memory bus width.

At least one person has reported better results with an earlier driver version.

Quite possible. I found the drivers included in the box -- didn't record the version -- to be slower than the latest, so I grabbed the latest. I did find I got CUDA bugs/crashes with this release, both on MainConcept renders (AVC or MPEG-2) and on some of the OpenCL benchmarks.

Not sure I have the time & energy to back and check with older drivers. I pretty much had to reject the nVidia card just based on bad display quality. I'm sure that's not a general problem, or no one would be using these.

Purely FWIW...

For me in win7 with a 6870 results vary according to driver version, video format on the timeline, render target format, & whether I'm using 32/64 bit Vegas Pro 11. In the last couple weeks I've used the official 11.9 Cat drivers, the 11.10 preview 3 drivers, & the official 11.10 drivers that came out the 31st, though not much time testing on those yet. Because of the difference in 32 vs 64 bit I think the rest of the installed video support files in win7 quite possibly have an effect -- in the next day or so (i.e. as they become available) I plan to add modded 11.10 drivers to win8 prev., [the modded official drivers] & test there since there's nothing much installed yet, & that might give some sort of confirmation.

RE: GPU assist -- some of Sony's code could be better, comparing Vegas Pro 11's ATI assist with other apps that have it I'll say more right. [Reading the release notes, the mention of Nvidia problems if Cuda drivers are installed with a non-Cuda card was/is a bit surprising, & IMHO hints that the coders may be a bit new to this GPU assist stuff.] At any rate, & overall, GPU assist is not always better, regardless the app, & as the goal is most often increased speed, the code is often written with a speed rather than quality bias. That makes comparisons harder than they might be otherwise.

For me so far, & I'm still working on testing usability rather than new features [I found v. 10 unusable], mpg2 render is a wash with smart encoding still broken (just not as bad) being balanced by slightly better encoding speed. Mpg2 encoding only works for me with GPU assist off, though there's still a fair amount of DXVA used with the 6870. The Sony AVC encoder OTOH is for the 1st time what I'd consider usable, with reasonable encoding time in Vegas Pro 11 64 using GPU assist, & quality looks good so far -- that said it's still ~2.5 X slower than the GPU assisted encoder I use, & almost 2X slower than another AVC encoder I use that doesn't have ATI hardware assist. Quality is comparable for all 3. Despite it's nasty lack of bit rate settings [e.g. 22, 23 etc. won't work], using the Sony AVC encoder saves me from going to an intermediate so may work out sometimes. My biggest concern is is that with 2 of the driver versions I've tried I get the 1st frames green screening, both in the preview & rendered -- I was waiting until yesterday's driver release to see if I can nail it down & hopefully fix it.

If it helps, AMD/ATI driver installation can be a bit of a PITA... The OpenCL stuff has a tendency to not install, & the ATI files in the Common Files folder(s) [e.g. Avivo] have a tendency to not get installed/updated either, even when the ATI installer reports success. [I use Universal Extractor on the expanded msi files in the c:\ATI or AMD folder to double check.] Those files work in concert with the drivers, & to some extent I've had luck mixing/matching versions in the past.

What a great benchmark document.

I'm thinking of upgrading my GPU. The cards I'm looking at are all in the $350+ range. Does that sound about right?

I have been rendering to MPG2 for DVD (MC template, highest quality, 9M max, 4M min, reduce interlace flicker) from AVCHD media with Boris BCC 7 Chroma key applied. Computer = i7-950, 12 GB RAM, PNY GTX 560TI, three hard disks.

A six minute clip takes 180 minutes CPU only, and even though the GPU rarely goes over 11% utilization, it takes 141 minutes with GPU, a 20% improvement. With GPU, the CPU didn't go over 60% utilization.

My current project will require 24 similar renders, so a saving of 24 x 39 minutes is substantial.

I Run my own videography company and on the three systems I have, it takes longer to render using gpu accelaration.

this is a shockingly bad product and its hard to contain my loathing of the problems i now have,

my first system is a i7 980x 4ghz hex core with nvidia 580gtx 12gb 1600 ddr3 ram
using 4 cameras multitrack it cant even play smoothly, time line surges and the render times to mpeg2 for dvd are over an hour slower on a 1 1/2hr project

second system is a i7960 with nvidia 580gtx card same issues 24gb 1600 mhz dd3 ram

third is an alienware m17 r3 laptop with a i7 2860 cpu 16gb ram and 580 gtxm card same issues.

reverting back to cpu and turning off gpu accelartion made the software useable but no better than vegas 10.

since the new build the cpu only now has the same playback and rendering issues.

1 track mxf 2 tracks achd and one mpeg in the project.

in my Proffesional OpinIon this software is unusable and the advertising is false and misleading!

totally shocking!