1 person found this helpful
Simple answer: ALL
The more CUDA cores, the faster. The article you referenced may be misleading for the uninitiated, because there are so many other factors involved to use all the cores to maximum capacity. That caveat was clearly made in the article you referenced.
The performance differences between the 560 and the 580 are small in real life. See http://ppbm5.com/MPE%20Charts.php
This is what I bought:
HP Pavilion Elite AMD Phenom II Six Core 1045T Computer (HPE-500F):
AMD 6 Core@ 2.7, 8GB of RAM, 1.5TB hard drive(7200rpm) Windows 7 64
I upgraded the PowerSupply to a 700Wat:
Changed the Video card to a GTX 570: (thinking about changing this to a 460 TI
Added 2 X 2TB hard drives on sale(7200rpm, 32 GB Buffer:
Would you mind commenting on it?
Go to http://ppbm5.com/DB-PPBM5-2.php to see where AMD Phenom 6 cores end up. The best heavily overclocked one comes out at rank 195 with a CUDA card, without CUDA they all rank below 326. It can be used, but don't expect stellar performance, due to the limitations of the AMD support for SSE 4.1+ instructions.
On the Model CPU tab, select Phenom, on the Phys. Cores tab select 6 cores and you will see where your system will likely end up.
Here are my old test results using CS5.
I do believe that CS5.5 the results will show more dramatic improvements. For instance in a new Sandy Bridge build that I am working on when I went from a GTX 285 to a GTX 480 my MPEG2-DVD dropped 60%. Now I do not know that everything was identical or why this major change compared to the CS5 results. Next major study will be a similar table with the SB platform and my array of GPU cards. But definitely CS5.5 uses more GPU processing than CS5 did. It appear that your reference used to preceeding Premiere version which makes his data also outdated.
I am the person that wrote the article that you are referencing. A long time ago, I had stated that CS5 wasn't using more than 96 cuda cores, however, that has been corrected a long time ago. What the article states is I recommend at least 96 cuda cores.
CS5 and CS5.5 will use ALL of the cuda cores, however, other factors such as the way your computer is configured you may not see much of a performance difference with more than 96 cuda cores. I show in the article tests we ran on our computer system using different video cards. As you can see, on our system, we did not see that much of a difference between 96 cuda cores and 480 cuda cores. The first chart shows using the PPBM5 benchmark program. As you can see the 96 cuda cores gave us 34.2 seconds vs 480 cuda cores was 31.5 seconds. Not that much of a difference due to the computer system I ran the tests on.
If you have any questions, please feel free to call me at 386-788-6075. I am in between 10am and 5pm Eastern time.
I will be happy to answer any questions you have about the article or the tests I ran.
Studio 1 Productions
For the most part, I agree. But due to the market changes with newer GPUs, you might as well get a card with at least 192 CUDA cores at this point: All of the GT 430s and most of the GT 440s use slow DDR3 RAM. (There are GT 440s that use (G)DDR5 RAM, but those are relatively rare and priced too close to a GTX 550 Ti for comfort.) And of the GeForce 5xx-series line, the retail-boxed cards that use the faster (G)DDR5 RAM have 192 or more CUDA cores in their GPUs. (A GT 545 that has recently begun shipping into the retail channel has 144 CUDA cores but slow DDR3 RAM.)
I do recommend the 500 series. This is exactly what the article says:
"I really can't say what you should buy, but I recommend the GT440 or one of the GT500 series of video cards as long as they have 96 cuda cores or more. The GT500 series are the newest series and they do run cooler and a little faster than the 400 series."
The only reason I mention the GT440 is for the low price that you can pick them up for now, with DDR5 memory. I get about a dozen emails a day from people asking me what video card they can get for $50 to $70, so that is why I mention the GT440 in the article. When someone emails me, I do recommend the GT500 series of video cards. Most people emailing me don't have a lot to spend on a video card right due to the economy.
My tests chart includes, the GT240, GT440, GTX470, GTX550 Ti and the GTX570 video cards all are running DDR5 memory, as these were the video cards I have in my systems here and had easy access to when I did the tests.
With a GTS450 512RAM I got nearly same result that with a GTX460 without O.C. +- 6 sec base on WWW.PPBM2.COM.
Conclution: Focus in you CPU CLOCK SPEED, RAM freq/lat and w/r Hdd speed.
Video cards will NOT make Big difference on PP performance.
Try to sale you AMD and buy a i7, read and learn about OC, them you will get the best performance you can pay.
Cristobal Salas OC/PC.
I think there's better preformance with a gts 450 with 1Gb memory, if you work with PrePro/Ae 5 or 5.5 because with 512 Mb you will not have benefits the Mercury engine.
"Video cards will NOT make Big difference on PP performance."
In the article, all of the benchmarks were done with Premiere CS5.5
Today, I updated the article to make things a little more clear for everyone.
As a side note to Randall, I just got in an EVGA GT545 Card with DDR5 memory so they are available with DDR5 memory now.
Thanks for the update.
So that means that the GT 440 and GT 545 are available in both DDR3 and (G)DDR5 versions, where the DDR5 version is about $20 more expensive than its DDR3 sibling. Under these circumstances, the DDR5 version is well worth the $20 premium if Premiere Pro performance is important, especially since the cheaper DDR3 version is nearly twice as slow as the DDR5 version in the export/render portion of the PPBM5 benchmark test with MPE set to GPU acceleration mode (although one might not notice such a difference on an old Core 2 Duo system, the difference becomes noticeably evident on even an i3-2100, whose performance is roughly on a par with the fastest of the quad-core AMD processors in CS5.x).
Where I wrote PrePro/AE, I should say Premiere Pro/ After Effects and my remarks about 0,5 or 1 GB on the video card is a quote from Dave's, very useful, article:
'''To use the Mercury Playback Engine in Adobe Premiere CS5.5 and CS5 in the CPU Acceleration mode,
you need an NVIDIA graphics cards (aka. video cards) with at least 896 megs of video ram..
Peace and many thanks Dave, for your great research!
I don't recommend DDR3 memory on a video card to anyone, I always recommend DDR5 due to the performance difference.
However, there are some people that already have a video card that has DDR3 memory and they don't want to buy a new one, but they just want to do the "hack" so Premiere sees it. I do explain in the article the benefits of using DDR5 memory over DDR3.
I am going to run some tests with the video card this weekend and add the results to the charts in the article.
Without Clearing the GPU importance, my point is: Porcessor "Brand and clock-speed", Ram speed Freq/Lat. and HDD config "SSD, RAID controllers, SAS, ETC..." THEY WILL make the difference in performance.
In this case we are talking about a AMD "6core". Do not matter if you buy a "GTX600 2GRAM" a simple I7-920 "4cores" with a GTX260 will give you a mucht better performance and if you know how to tunning your pc a SIMPLE I7-920 with the right O.C., the right RAM and the right HDD congif, will do better than a multi xeon PC.
I guess you know what Im talking about .
Another benefit of higher end cards is better quality previews. GT240 DDR5 gives a choppy HDMI output - GTX560 in same box is very close to a proper HDMI I/O device.
Also, due to the fact that the AME CS5.5 queue allows the use of CUDA GPU acceleration in the MPEG-2 DVD portion of the PPBM5 test, it's no wonder why my i7-2600K system with a reference GTX 470 underperformed somewhat in CS5.5: I could muster only a 110-ish second result in the MPEG-2 test while similarly-clocked 2600K systems with a GTX 570 could achieve a result of 70-ish seconds on that same test.
Gosh, I need a new graphics card (or at least one with better cooling)...
Mm, i´m happy with 128 on the MPEG-DVD test with a 2600k OC on 4.4 Ghz and a GTS 450 2 fan OC2 card.
But.. I see on number 4 a 2600k with a GTX470 (28 sec on MPeEG test) and on number 5 a GTX 460 (24 sec.on MPEG test)I
Is this only because they´re OC to a 5.2 Ghz (questionmark)
Those scores are only because they are running 5.0.3 rather than 5.5.0. 5.0.3 loads up the memory cache, so subsequent reads during the MPEG-2 DVD test are coming from the RAM itself. 5.5.0 does not read that cache; however, the GPU performance difference becomes a bit clearer in 5.5.x. The i7-2600K system results in the MPEG-2 DVD test with CUDA MPE-enabled GPUs ranged from 67 seconds (with a GTX 570) to 344 seconds (with a GT 440). Part of the cause for such a difference is the tuning and other parts of the PC - but a large part of the difference may be placed on the GPU (lower-end GPUs like the GT 440 are not as efficient at MPEG encoding as higher-end GPUs). In that case, there is a sizable performance difference between the GT 440 DDR5 and the GTS 450 with respect to MPEG-2 DVD transcodes from HD material.