10 Replies Latest reply on Mar 1, 2012 9:05 AM by ECBowen

    Quadro 2000 v. gtx 570 v. Quadro 4000

    jamesp2 Level 1

      Can anyone comment on the MPE performance difference between the Quadro 2000, the gtx 570 and the Quadro 4000?  I'll be using hacked GH2 footage with CBR intra-frame coding, which (I'm told!) can and should be put on AVC Intra time-lines (not AVCHD). 

       

       

      I'm concerned most with time-line responsiveness and playback performance.  DVD encoding, exporting footage, etc., won't be happening very often, so that's of less concern. 

       

      I'm aware that the gtx 570 is probably the best buy of the three; but that's not the question.  Many thanks.

        • 1. Re: Quadro 2000 v. gtx 570 v. Quadro 4000
          Jim_Simon Level 8

          While I'm not aware of any Direct comparisons, I'd venture that the 570 will outperform both Quadros, simply because it has significantly more CUDA cores than the other two.  So it's not only cheaper, it's also better.

           

          Having said that, none of the cards will help with simple playback of your media.  That is entirely CPU dependant.  Only effects, scaling, deinterlacing and such are accelerated.

          • 2. Re: Quadro 2000 v. gtx 570 v. Quadro 4000
            RjL190365 Level 4

            As far as performance in Premiere Pro CS5.5, the GTX 570 is the fastest of the three, with the Quadro 4000 being moderately slower than the GTX 570 - and the Quadro 2000 being much slower than the other two. This is because the Quadro 2000 is based on an underclocked GeForce GTS 450, with only 192 CUDA cores and only a 128-bit DDR5 memory bus. The Quadro 4000 is based on the previous Fermi flagship design, the GF100 (also used on the GTX 480, GTX 470 and GTX 465, as well as its higher Quadro siblings 6000 and 5000), but with a significant number of texturing units disabled (the Quadro 4000 has only 256 CUDA cores versus 448 CUDA cores in the GTX 470 and 480 CUDA cores in the GTX 480) and only a 256-bit DDR5 memory bus (versus 320-bit in the GTX 470 and 384-bit in the GTX 480). The GTX 570 is based on a newer (but still a year old) chip, the GF110 with 480 CUDA cores and a 320-bit DDR5 memory bus.

            • 3. Re: Quadro 2000 v. gtx 570 v. Quadro 4000
              jamesp2 Level 1

              The trouble with the 570 is that if you're not building your own system, most off-the-shelf workstations don't appear to have sufficient PSUs to run it....  Meanwhile, the Quadro 2000 is frequently found bundled in discounted HP workstations and has modest power requirements.  Swapping out the PSU isn't beyond my screw-driver handling ability, but it's added cost and trouble.

               

              Based on the only benchmarks I could find here, there's very little difference between any number of cards, except with respect to DVD encoding, which isn't important to me.

               

              Meanwhile, places like videoguys seem to prefer the 4000 to the 570.  But there are no benchmarks to be found anywhere.  One card may have greatly improved specs ov er another, but are we talking 10% faster or 50% faster?

               

              Unfortunately, the answer isn't to be found here, so far.

              • 4. Re: Quadro 2000 v. gtx 570 v. Quadro 4000
                RjL190365 Level 4

                In any case, performance wise the Quadro 2000 is a waste of money: It costs almost $400, yet it performs equally as slowly as a $100 card. And in Premiere Pro CS5.5, the encoding performance becomes significantly slower with lesser GPUs. Look up posts by Bill Gehrke and you may find a list of GPUs along with their performance charts in the PPBM5 benchmarks. Bill tested a wide range of GPUs from a GTX 580 all the way down to an old 9500 GT. Pay particular attention to the MPEG-2 DVD scores. You will find that even on an overclocked i7-2600K system, the system with a GTX 550 Ti took more than twice as long (146 seconds) as the GTX 580 (60 seconds) or even a GTX 560 Ti 448-core (68 seconds) in that test. The Quadro 2000 would have performed even slower than the GTX 550 Ti in that same test (heck, the GTX 550 Ti itself is slightly slower than a first-generation GTX 260 in this test despite having an equal number of CUDA cores due to the 550 Ti's slightly lower total memory bandwidth). The Quadro 4000 would have performed roughly on a par with Bill's tested GTX 285 (117 seconds) in that same test.

                 

                On the other hand, if you're encoding to H.264, then the Quadro 2000 would have been only slightly slower than the GTX 570; you would have had to downgrade further to Quadro 600 (GeForce GT 430) level to see a significant degradation of H.264 encoding performance.

                 

                Secondly, the Quadro 2000 has only 1GB of RAM total. With your footage, it is possible that any effects that you apply will eat up more than the amount of memory on the card. If a scene needs 1.5GB of VRAM to render using MPE GPU mode, then the 1GB card will run out of RAM. And when the rendering job runs out of VRAM, that entire frame or scene will default entirely to the MPE software-only mode, which will result in slower performance and may also degrade image quality.

                 

                And I strongly recommend avoiding the purchase of off-the-shelf PCs or workstations to begin with: Those systems are way too expensive for such bottom-of-the-barrel performance, and upgrading such a system via the manufacturer would have cost you three to four times more than if you bought those same parts elsewhere. If you can't build an editing workstation yourself (or find it too much of a bother), consider contacting a vendor who specializes in custom-configured editing systems such as ADK.

                • 5. Re: Quadro 2000 v. gtx 570 v. Quadro 4000
                  jamesp2 Level 1

                  I think we're looking at the same benchmarks:  the ones which show virtually no difference between the tested cards, except for DVD encoding (which, as noted, I don't care about).   And if Adobe goes to the trouble of certifying these cards, you'd also hope they'd supply information about performance.  If the Quadro 2000 lacks sufficient memory to be of any use in typical work, it would be nice to know Adobe is certifying a useless card....

                   

                  I take your point on the workstation, but they do go at a discount from secondary suppliers, and I wasn't planning to invest in much more than a bare box -- processor, ports, cpu, hard drive, Windows.  I'll install additional RAM and probably the GPU at newegg prices.

                   

                  In any case, the question remains open:  just how much difference does the choice of a GPU card make?  The benchmarks both of us are looking suggest, "not much", unless you're encoding DVDs.

                  • 6. Re: Quadro 2000 v. gtx 570 v. Quadro 4000
                    RjL190365 Level 4

                    It also depends on the CPU that you are using. The lower-end cards would still be an improvement if, say, you are using a cheapo dual-core CPU in that system. But the really cheapo or slow cards can bottleneck a higher-end performance CPU, sometimes preventing it from performing at its best.

                     

                    And "running out of VRAM" is an issue if you're going to be doing multiple layers of HD material or even a single layer of RED 4K. Remember, the higher the resolution of the source material, the more VRAM it eats up for MPE GPU acceleration.

                     

                    And going back to H.264 Blu-ray encoding, only the very slowest of those cards in Bill's list would have slowed down such encodes significantly.

                    • 7. Re: Quadro 2000 v. gtx 570 v. Quadro 4000
                      JEShort01 Level 4

                      James,

                       

                      Check out the following links and focus on "dropped frames" in the "spreadsheet graphic":

                      http://blogs.adobe.com/genesisproject/2011/10/diving-into-nvidia-gpus-and-what-they-mean-f or-premiere-pro.html

                      http://blogs.adobe.com/genesisproject/2011/12/nvidia-maximus-and-premiere-pro-cs5-52.html

                       

                      While I don't have any hard numbers comparing timeline performance of the exact three cards you are asking about, I can say from playing with various setups, video cards (only GTX 460 and 480 though), CPUs, memory, and lots of different hard drive configurations that your GPU definitely does play a role in scrubbing. Generally speaking:

                      - Get at least 16GB of RAM

                      - CPU makes the most difference

                      - you must have at least enough "drives" to feed the CPU (balance is important, can't have a weak link anywhere)

                      - GPUs do make a difference for playback and scrubbing (and I don't know much about your codec!?)

                       

                      I had to start with general comments, but that being said, I would suggest:

                      - select a strong workstation (agree ADK if at all possible)

                      - get a lesser power hungry GTX to start with (i.e. GTX 460)

                      - use you system and play with full, 1/2, and 1/4 playback resolution and only upgrade if you need something more

                       

                      I would not suggest either of the Quadros; too expensive and you don't have a "need" for that line.

                       

                      Regards,

                       

                      Jim

                      • 8. Re: Quadro 2000 v. gtx 570 v. Quadro 4000
                        Scott Chichelli Level 3

                        here is a very simple answer

                         

                        assuming no animation (AE not included as it runs great on GTX) such as solid works and a very few select other porgrams who can actually use the Quadro firmware (the ONLY difference with these cards, they are all GTXs)

                        ANY Quadro card is a complete waste of money, and will offer no benefit (other than 10bit color).

                        so the 570 is the better buy as well as the better performer for all things Adobe.

                         

                        Scott

                        ADK

                        • 9. Re: Quadro 2000 v. gtx 570 v. Quadro 4000
                          jamesp2 Level 1

                          Scott, I don't doubt the GTXs are better values, but there are other issues in this case -- matters of convenience noted above. 

                           

                          The question remains -- cost aside, just how much difference is there between the cards mentioned above, in daily operation -- filters, scrubbing, etc., in long-form filmmaking?   Encoding to other formats is something I wouldn't be doing often, so gains there aren't important.

                          • 10. Re: Quadro 2000 v. gtx 570 v. Quadro 4000
                            ECBowen Most Valuable Participant

                            If you look at the links Jim posted you would see what the MPE acceleration includes such as scaling, interpolation, and many effects. Since the MPE acceleration handles processing for those components of video, you will see significant improvement for realtime playback regardless of the codec being used if you are using an Adobe Preset and also for encoding regardless of the codec because the encoding is being handled by the CPU. The Specs of the video card such as Cuda cores and memory bandwidth decide the overall performance gain of the MPE engine provided your CPU is fast enough to push the video card. In other words if you have a slow cpu or a dual core cpu then your video card will be sitting there waiting on the cpu to get data to it to process. So in that scenario, you would see less improvement from the better video cards. However if you have ideal hardware for the MPE engine then the 570GTX and the 580GTX give you the best performance with the MPE engine because their specs are currently far superior to the Quadro cards accept for the Quadro 6000 which is closer to their specs.

                             

                            Eric
                            ADK