12 Replies Latest reply on May 8, 2012 9:31 AM by Another Photographer

    Sandy Bridge Vs Cuda

    EstebanMartin Level 1

      Hi, I edit in a macbook pro 13' and it is a bit slow.


      I just thinking about what should i buy here the options.


      1- sandy bridge 2600k 16gb = 700€

      2- sandy bridge 2500k 16gb and gt460 = 780€


      My use will not be profesional, just recreational but who nows...  which is the most interesting option in terms of performance and economy, the cuda performance increase is sonotable as cpu (i7) so powerful?



      I ask this, because I've never tried.

        • 1. Re: Sandy Bridge Vs Cuda
          Harm Millaard Level 7

          The Sandy Bridge 2600K is around the same speed as a 920 OC'ed to 3.8. You would definitely need a 460, since the Intel approach does not work with CS5, at least not with hardware MPE, so the performance penalty is around 10 x longer rendering speeds in the PPBM5 Benchmark but that may differ for your usual timelines.


          Personally, I would stay away from Sandy Bridge and wait for Ivy Bridge. The octo-core looks nice. Sandy Bridge has no noticeable benefits over an OC'ed 920 and seriously suffers from the memory architecture for NLE work, combined with an intergrated graphics 'chip' that is unusable.

          • 2. Re: Sandy Bridge Vs Cuda
            ECBowen Most Valuable Participant

            Hehe if the Audio techs would finish their testing on the 2600 I could get the PPBM5 benchmarks done for it. As soon as they do I will get those done and email them to Harm. So far though the 2600 performance is far better than expected. The 2300 was not really any better than the I5 750 or 760 although the turbo is more efficient. The turbo on the 2600, atleast the sample we have is acting completely different though. That is something I am still talking to Intel to find out why. I will have further details soon




            • 3. Re: Sandy Bridge Vs Cuda
              EstebanMartin Level 1

              My first conclusion is that apparently, the use of gpu is required.
              I thought that with a good processor that would not be necessary.


              I've seen ppbm5 but difficult to draw a conclusion, as I can get more performance witha 920 OC, with and without cuda about RT?

              • 4. Re: Sandy Bridge Vs Cuda
                Chuck A. McIntyre Level 3

                Tom's has an article about Sandy Bridge.  I'm sure Adobe has been working with Intel related to its video encoding enhancements.  Makes me wonder how this will effect future releases of Premiere and the Mercury Playback Engine.


                As I read further, Tom's has a benchmark, Premiere CS5 MPE software only.  I'm looking forward to PPMB5 benchmarks comparing the Sandy Bridge to the current top performers.  The Tom's article builds up the new Sandy Bridge early in the article as something that blows away Nvidia. Then when you see the CS5 comparison on their site, the results just don't live up to the hype.  Perhaps earlier in the article they are referencing "video" in a gaming context.

                • 5. Re: Sandy Bridge Vs Cuda
                  Harm Millaard Level 7

                  The main problem with the Sandy Bridge chipsets, P67, Q67 and H67, is identical to the P55 chipset: The lack of PCI-e lanes. This prevents one to use a nVidia MPE capable video card AND a raid controller. For high performance both are needed, but it is impossible on these platforms, due to the lack of PCI-e lanes. In addition only PCI-e 2.0 lanes are supported, not 3.0.


                  The new platforms have the same shortcomings as the P55, making it unsuitable for serious editing. The CPU may be fast enough, the rest is no more than mediocre at best.


                  The H.264 test results with software MPE on Tom's Hardware show 20:55 for the 2600K and 26:02 for the i7-950. 24% difference.


                  The best i7-950 in the PPBM5 benchmark has 93 s in the H.264 test and 127 s for the render test. An overclocked i7-920 has 65 and 73 s respectively. 43% and 74% respectiviely. These tests are incomparable, but the conclusion is clear:


                  Sandy Bridge is not faster than an OC'ed i7-920/930/950, still needs a nVidia card and still needs a raid controller, which is impossible.


                  Thus Sandy Bridge is unsuitable for serious editing, also because of the memory architecture.


                  PS. Keep in mind that the much touted Turbo mode only works effectively with single threaded applications, or applications that use less than the available cores. Luckily PR is very good at multi-threading, so the impact of the Turbo mode will be rather small.


                  PPS. This is worth the wait: Sandy Bridge E and X68: http://www.nordichardware.com/news/76-motherboards/41290-intel-x68-motherboards-being-test ed-by-intel-and-partners.html


                  Quad channel memory (meaning 32 GB capacity with 8 slots), octo cores and 40 PCI-e 3.0 lanes. Now that starts to look very interesting.

                  • 6. Re: Sandy Bridge Vs Cuda
                    Jim_Simon Level 9

                    "...which in turn houses a quad-channel DDR3 memory controller but no graphics.  The processors supports 40 PCIe 3.0 lanes and is geared toward  enthusiasts and workstations with demands for maximal performance."


                    Now THAT is more like it.

                    • 7. Re: Sandy Bridge Vs Cuda
                      Bill Gehrke Most Valuable Participant

                      The plain and simple answer is go to the "Benchmarks Results" and slide over to the option and choose the last one "MPE Results"  Then on that page look at the column under "Render Timeline" and the sub column "MPE ON" and look for the first "none" (that means they did not have a CUDA card with enough memory for MPE processing).  That total score is 385 (where 140 is the best score) and the "Performance Index" is 685 or 6.85 times slower that the best scores.


                      Here is another way to look at the the difference a CUDA enabled score differs from a non-CUDA/MPE card.  These are some samples off the Benchmarks Results page after the contributors upgraded thier video cards.

                      • 9. RAID controller + nVidia
                        Another Photographer Level 1

                        I was not aware that some motherboards do not allow both a RAID controller and a video card.  Can you explain?  I had assumed that if you had the slots, you can install the cards.

                        • 10. Re: RAID controller + nVidia
                          ECBowen Most Valuable Participant

                          The boards allow them but populating the slot required for the raid card slows the video card's slot down.




                          • 11. Re: RAID controller + nVidia
                            JCschild Level 3

                            a better explaination err anology..


                            imagine you want to water your lawn. you have only one hose bib.

                            but on that hose bib is 4 way splitter for 4 hoses.


                            you connect 1 hose to a sprinkler on the south part of your lawn and turn it on.

                            its working great and shoots 35 ft in all directions.


                            now you connect a 2nd hose to the north side and turn it on its working good and now both hoses are

                            shooting 25 ft in any direction. however the north part of your lawn requires you to cover 30 ft in any direction.

                            so now you will have to move it to cover the part not getting water...

                            this will require more time




                            • 12. Re: RAID controller + nVidia
                              Another Photographer Level 1

                              How many lanes are needed?  Would Z77 chipset with a single PCI Express 3.0 x16 cut it, or do you need 2 x16 lanes?