Core i7 - 6 core versus 4 core performance with Premiere CS5

[Steve Pesenti]

Does anyone have experience of Premiere CS5 performance running on an i7 six core vs. i7 four core processor clocked at the same speed (e.g. i7-970 vs i7-960 each running at 3.2 GHz)?

Assuming the set up is MPE enabled - say a GTX 470 hacked - what sort of performace improvement does the six core deliver over the four core:

1) Would rendering speeds approach 50% faster?

2) Would timeline playback, in portions of the timeline not requiring GPU accelleration, run as smoothly with up to 50% more layers?

Has anyone done a direct comparison test?

[Steve Kalle]

You will find many answers here: PPBM5 Benchmark

Notice the testing with i7-980x overclocked to 3.7 or 4.0 GHz and the i7 920 clocked at the same speeds.

For the money, you can't go wrong with the GTX 460, especially because Adobe does not fully utilize all the cores so there is little performance increase going from a GTX 460 to 470 and even a 480.

[Allan Tabilas]

In addition to the great PPBM4.com and PPBM5.com sites, ADK Video Editing did some benchmarks with CS5 with the 6 core Core i7 980x vs its Xeon cousins, the 6 core X5660 (in dual cpu configurations, 12 cores/24 threads), with also overclocking results using AVCHD and Red source footage. While the 980x is around $1k retail, its Xeon equivalents are around $1700 (the Xeon X5680). The 980x generally outperforms the dual Xeon setup!

http://www.adkvideoediting.com/benchmarks.asp

[Bill Sepaniak]

In terms of the "cost - benefit ratio" I went with an i7 930 that I easily overclocked to 3.85GHz and a GTX 285. The i7 930 was on sale at Microcenter for $199.00. The GTX 285 is now EOL, but an i7 930 and a GTX 460 or 470 would provide good value and performance.

[Steve Pesenti]

Thanks Steve - That's a useful data set. It appears to support the overall time saving of 6 cores over 4.

Interesting to note your comments on the relative cost / benefits of GTX 460 vs. 470 - have not seen that called out before.

[Steve Kalle]

FYI, as you will notice with the PPBM CS5, only H264 is faster on the i7 980x vs dual 6 core. This is mainly due to very poor threading in the H264 codec. Look at the MPEG2 numbers and the extra 6 cores really helps. Not to mention that AE really benefits from the extra cores as well.

With what I've just said and looking at the PPBM CS5 results, can you guess which one is me?

[Harm Millaard]

Steve,

Multi-threading is not the issue here. In fact the multi-threading with H.264 is very good. Look at the attached image. The first block is with software MPE and the second block is the hardware MPE.

The problem with dual processors is that they cause a lot a latency with inter processor communication, as explained in the background information page. I suggest you read the Interpretation page and the Background page carefully. http://ppbm5.com/index.html

[Steve Kalle]

Harm,

If we were talking about the 771 chips, then I would agree that there is latency issues. But with the 1366, the QPI solves this problem. Furthermore, the way I have it setup, each CPU has its own memory and can't share it so there is no 'crosstalk'. One of the best examples that show how much QPI has helped is VM testing.

For 100% accurate CPU testing, the CPU needs to be maxed at 100%, and the H264 test doesn't come close to using 100% on my 12(24) cores.

[Harm Millaard]

Then use Prime95 for your comparison.

There can never be a CS5 test that uses the CPU 100%, because video editing/encoding entails using a lot more components.

That your CPU load does not come close to 100% on the H.264 test is quite logical with your memory setup. You have created overhead for algorithm data in favor of latency, while starving the system per processor of memory. That is a choice. In combination with the low clock speed and the QPI limitations of dual processor design, that Intel acknowledged but has not solved, your results are quite acceptable, although maybe not what you expected.

[Steve Kalle]

Originally Posted by Harm Millaard

There can never be a CS5 test that uses the CPU 100%, because video editing/encoding entails using a lot more components.

This doesn't make sense. If you want to test the CPU, you MAX it out. The easiest way is to use MRQ.

And 'starving' my CPUs with 24GB of ram - that is just funny. Premiere/AME uses more ram during rendering/encoding than any of the 980x systems have in total. Furthermore, if they were starving, then I would see a decrease in performance when activating Hyper-Threading but I don't.

[Harm Millaard]

I don't want to test CPU, there are numerous tests around to do that. You obviously want to, hence my suggestion to use Prime95 or similar.

I want to test real life situations using real life applications. I can't help it that you do not like your results. Using Prime95 you may get better results.

What you don't seem to grasp is the workflow during encoding. I again suggest you read the explanations.

[Steve Kalle]

BTW, take the results of the PPBM5 with a grain of salt because there is no way of proving the reported results. People just email a Txt file which is easily editable. I like the guy whose results show faster MPEG2 than Harm even though his CPU is not nearly as fast as Harm's (2.66 vs 3.7 GHz).

Harm, you just cost me $1000. I just ordered another 24GB of ram after seeing the MPEG2 results.