Are you Mac or Windows?
First read http://forums.adobe.com/thread/773101 which has a link to the current list of supported cards
For "some" others, with at least 1Gig of video ram, use the nVidia Hack http://forums.adobe.com/thread/629557 - which is a simple entry in a "supported cards" file
Unless you need 10bit video output, the Quadro is overpriced compared to MANY other nVidia cards
Go to the CS5 Benchmark http://ppbm5.com/ to see what people are using
I am not personally in the market for a new card, but if I was (Windows) I would mostly likely go for the nvidia gtx 670 as today's "best bang for the buck"
My personal experience: I obtain CUDA support using a Geforce GTX 550Ti with PPro 5.5 on Windows. This card isn't on the list of officially-supported cards, and I therefore had to add it to the list as described in the nVidia hack (see above). Why isn't it on the list ? Probably because Adobe are not prepared to devote the resources to test every card with CUDA that nNvidia bring out. I've had no problem up to now which could be attributed to the use of CUDA. You can tell you've got CUDA support when there's a yellow line along the top of the sequence you're working on.
If you use any, several or many of these Adobe features or effects you might want to use something besides the GTX 550 Ti since your budget seems you may be a serious user.
• color space conversions • Alpha Adjust • Basic 3D • Black & White • Brightness & Contrast • Color Balance (RGB) • Color Pass • Color Replace • Crop • Drop Shadow • Extract • Fast Color Corrector • Feather Edges • Gamma Correction • Garbage Matte (4, 8, 16) • Gaussian Blur • Horizontal Flip • Levels • Luma Corrector • Luma Curve • Noise • Proc Amp • RGB Curves • RGB Color Corrector • Sharpen • Three-way Color Corrector • Timecode • Tint • Track Matte • Ultra Keyer • Video Limiter • Vertical Flip • Cross Dissolve • Dip to Black • Dip to White • Film Dissolve • Additive Dissolve • Invert • Directional Blur • Fast Blur • frame rate differences • field order differences • pixel aspect ratio differences • frame size differences • media with different alpha channel representations
Here is a chart that I have just created with CS6 when I compared the time to encode and 1920 x 1080 60i AVCHD timeline with quite heavy use of the above features and effects with and without the GPU MPE hardware acceleration. (I did encode to DVD at 24 frames per second.)
Unfortunately I do not have the new GTX 660 Ti but I think it might now be the Best Buy for the Buck for a regular user. If you are only a weekend occasional user then the $100 GTX 550 Ti probably is fine. Besides the newer model is a next generation PCIe 3.0 card. From my testing I would guess it would test slightly better than the GTX 580. If I did not already have the GTX 680 and were buying another graphics card today. it would be the GTX 660 Ti.
Message was edited by: Bill Gehrke
Sorry Jim that GTX 570 is old power hungry technology
Thanks for your post, Bill.
Some questions concerning your benchmark: in the Adobe staff post concerning the MPE, Tod Kopriva states:
"It's worth mentioning one set of things that Premiere Pro CS5 doesn't process using CUDA: encoding and decoding."
But in your post you state that your tests consist in encoding an HD timeline using PPro CS6. Does this mean:
- that PPro CS6 now uses CUDA for encoding, contrary to CS5 ?
- that the acceleration is obtained just in the rendering phase ?
In any case, graphics card performance is not the only parameter determining the choice of a configuration for running PPro; other parameters include CPU processing power, power consumption, and the presence or absence of official Adobe support for the graphics card. In this string so far we have recommendations for a GTX 570, a GTX 660, and a GTX 670, showing the diversity of appreciations concerning all these factors.
Those hundreds of parallel processors on the graphics card need to be fed by a sufficiently powerful CPU, and the optimum choice when you're spending around 1000$ concerns the CPU/Graphics card pair. If your CPU is an i3 a GTX 680 is likely to be a waste of money, it'll be waiting around for the CPU.
What would be interesting would be Bill's benchmark executed on a range of different CPU/GPU pairs to identify balanced configurations as a function of budget. You could then make your choice after factoring in power consumption, Adobe support, and, naturally, whether or not you're serious.
Good point Nicholas. My processor (12 core AMD), my power and Raid 0 could be enhanced with a card, which takes advantage of mercury's enhanced potential. I appreciate everyone's input.
You are 100% correct in that the actual encoding done in the Mainconcept encoder is all CPU. But before it goes to the encoder all those effects and features you select are done before the actual encoding. I guess I should have made my post above emphasize the "......with quite heavy use of the above features and effects....."
For instance just going from AVCHD to MPEG2-DVD just automatically requires scaling. Both in Pixel Aspect Ratio and Frame Size.
I fully agree about matching the CPU and the GPU but the OP was think of investing $1000 just in the graphics card and he does have a 12 core processor so my suggestion was to buy the latest GTX 600 series and I picked out what I though was a reasonable price version to save him $700.. Therefore if he ever goes to a new computer it will be more suitable and less power than the GTX 570 and be PCIe 3.0 compliant. We do have proof that the GTX 660 Ti works well in the PPBM results..
I do have results on that same system running with hyperthreaded six real cores as above and:
- different CPU clocks
- non hypertreaded
- fewer real cores
- less RAM
- different project disk read/write speeds
I just do not know how to effectively organize it and I need to make additional runs to verify results.
I did a bit of search on forum in order not to start another topic. Right now I am on low budget, and i think I will buy 2700k cpu and cheapest P67 / Z77 motherboard.
I used Premiere GPU hack page: http://www.studio1productions.com/Articles/PremiereCS5.htm
as a source of estimate of how well GPU works in Premiere.
2700K is not overly fast, and I chose it because of cost, so it wouldn't be wise to overly invest in GPU now. Later when I upgrade my CPU and GPU, i will keep slower GPU as a replacement if my newer GPU will break down.
From that page:
"Your video card should have 96 CUDA cores or more."
and should have more than 800MB memory.
I shortlisted few 300W GPU's that I hope would not slow down my work:
Card CUDA cores Power (W) DDR type Interface Graphic MHZ Processor MHZ GT 240 96 300 watt DDR3 / DDR5 128 Bit 550 Mhz 1340 Mhz GT 430 96 300 watt DDR3 128 Bit 700 Mhz 1400 Mhz GT240 DDR5 better GT 440 96 300 watt DDR3 / DDR5 128 Bit 810 Mhz 1620 Mhz GT 530 96 300 watt DDR3 128 Bit 700 Mhz 1400 Mhz OEM Card GT 640 384 300 watt DDR3 128 bit 797 Mhz 797 Mhz „Version A” GT 640 144 300 watt DDR3 192 bit 720 Mhz 1440 Mhz „Version B” GT 640 384 300 watt DDR5 128 bit 950 Mhz 950 Mhz „Version C”
Could we make some silly formula to give rough estimate of how such cards work in MPE?
DDR5 is 1.4 times faster than DDR3
192bit is 1.4 times faster than 128bit
= CudaCores * (1.0 DDR3 / 1.4 DDR5) * (1.0 128bit / 1.4 192bit) * (?Mhz Graphic) * (?Mhz Processor)
that would give us a strange number to use and roughly compare such older / low-end GPUS.
I know it is rather difficult technical question, because You guys do not use this cards... these are oldies or low-end.
I am only interested in MPE smooth playback/review when editing Premiere project - but render times seem to me a good estimate how well GPU works in Premiere. I know guys with gtx 470 / gtx 570 and their GPU usage is about 25% most of the time.
(For example right now maybe i could buy Gigabyte GF GT240 1024MB DDR5/128b for less than 75$) Or how this A/B/C version GT640 compare to each other?
Apparently you are not familiar with our Premiere Pro Benchmark (PPBM5) as of yet. Here is some data that I took using PPBM5.5 for CS5.5 with a very similar processor but I overclocked it considerably and it does not have any GTX 6xx series cards. After running the data I sorted on the MPEG encoding time as that best shows the effects of the GPU. You can see "cores are king" but I have not tested any of the real low end GTX 6xx series. I do have a GTX 660 and it does very well in current testing. No experience yet with the 650 or 640 boards so I have no verification as to how well they will do in similar testing.
You can forget about both the GT 240 and the GT 640 at this point: In my testing with Premiere Pro CS5.5 release 5.5.2 and my main overclocked i7-2600K system (with 16GB of RAM at the time I ran those tests), my old GT 240 (even with 1GB of GDDR5 RAM) needed all of 249 seconds to complete PPBM5's MPEG-2 DVD test (versus 107 seconds with a GTX 470 I used to have and 69 seconds with my current GTX 560 Ti 448). Of the three GT 640 configurations that you listed, none of those are available at retail (all three of those are OEM-only cards, available only to large system-building companies). There is a retail GT 640 available - but it is basically "Version A" with a higher GPU clock speed (900 MHz versus 797 MHz). The memory throughput of the retail GT 640 is an abysmally low 28.5 GB/s. If you absolutely must limit your spending to no more than $110-ish on a GeForce GPU, the GTX 650 (non-Ti) is a better choice for an extra $10 over the GT 640 even though it has only 1GB of RAM versus 2GB on the GT 640: At least the GTX 650's performance isn't totally crappy like everything in the GT 640 and lower range is (since at least the GTX 650 offers 80 GB/s memory throughput due to its use of GDDR5 RAM instead of the DDR3 RAM that cheaper GeForce GPUs use).
By the way, I looked at the PPBM5 MPEG-2 DVD results list for systems equipped with both the GT 640 and the GTX 650. While the GTX 650 results are inconclusive due to the fact that the only two GTX 650 results on that list are running only first-generation dual-core Intel i3 (Clarkdale) CPUs, the GT 640 results are more clear cut: All three systems using that GPU are running quad-core Sandy Bridge or later i7 CPUs, and the GT 640 needs at least twice as much time as the mid-to-high-end GPUs just to complete the MPEG-2 DVD portion of the PPBM5 benchmark. That alone tells me that the GT 640 is itself the limiting factor in systems with such higher-end CPUs.
As such, if your total system budget is that limited, you'd be better off downgrading the CPU to an i5-3570K and then putting the money that you saved with the CPU downgrade towards the purchase of a GTX 660 (a much faster GPU than anything 650-level and below). That would give you a better balanced PC (performance-wise), unlike your original plan which gives you too much CPU for such a woefully inadequate GPU.
THANKS for your informations.
"As such, if your total system budget is that limited, you'd be better off downgrading the CPU to an i5-3570K and then putting the money that you saved with the CPU downgrade towards the purchase of a GTX 660 (a much faster GPU than anything 650-level and below). That would give you a better balanced PC (performance-wise), unlike your original plan which gives you too much CPU for such a woefully inadequate GPU"
We work right now only with H.264, and few people told me about only "25%" usage when in edit mode of their GPU (when looking at system monitoring programs). Looking at your table the differences are quite big, but numbers in table H.264 show only minor differences. (The important factor to me is performance when editing and previewing ). Thanks guys for help and take care and you earny lot lot good karma for lifetime for helping here
And my question is when editing, could I just some older GPU until I'm editing H.264, and only later buy GTX460/470/560/570. I does not really matter to me when buying used GPU which one I buy. But - you are righ