[Hardware] Comparing two system AE/AME Render times. Within norms?

[BLSmith21122]

My company ordered me an HP 640 Workstation in 2016 that, at the time, appeared to be surprisingly slower than my 2013 custom built workstation (during initial tests). Those tests lead me to create a forum post here regarding that issue. That lead me into purchasing a new system. Once I received it three weeks ago, I re-ran most of my initial tests, and now the results seem more in line with what I would expect. I'd like to know if my new 2017 system should be performing much faster than the 2016 system, or if I should return it, and save my company $5,000. The purpose of this machine is all inclusive: 60% Premiere, 15% After Effects, and within the next 6 months they want me to start learning animation and to dedicate 25% of my time on that (looking into Cinema 4D).

Hardware:

1. 2013 System: Speccy Info.
   1. GPU: Nvidia GTX 660 Ti
   2. CPU: i7 3930K @ 3.2GHz
   3. RAM: 32GB
   4. OS: Windows 7 64-bit
   5. OS SSD: 500GB
2. 2016 System: Speccy Info.
   1. GPU: Nvidia M5000 (x2)
   2. CPU: Xeon E5-1680 V3 @ 3.2GHz
   3. RAM: 64GB ECCC
   4. OS: Windows 7 64-bit
   5. OS PCI-E SSD: 500G
3. 2017 System: Speccy Info.
   1. GPU: Nvidia GTX 1080
   2. CPU: i7 6900K @ 3.2GHz
   3. RAM: 128GB
   4. OS: Windows 10 64-bit
   5. OS SSD: 500GB

Testing: I've tested each system with Cinebench CPU and OPEN GL, PC Mark 7. an After Effects render that consists of solids that, and a Premiere Project (both AE/PP pull all raw files & cache from the OS drive, and also render to OS drive - in this test only.)

1. 2013 System:
   1. PC Mark 5,327.
   2. Cinebench CPU: 894 cb
   3. Cinebench OpenGL: 98.97fps
   4. Personal AE File AME Render Time: 2 Hours, 16 Minutes, 06 seconds.
   5. PPBM:

• Benchmark: Disk I/O Time: 25 Seconds.
• Benchmark: Disk Write Rate: 1483.68 MB/second.
• Benchmark: Time to Encode H.264 Timeline = 64 Seconds.
• Benchmark: MPE Gain: 316/13=24.3
1. CrystalDiskMark:
   • Seq Q32T1: Read: 2170. Write: 1499
   • 4K Q32T1: Read: 317. Write: 267.2
   • Seq: Read: 1583. Write: 1466
   • 4K: Read: 45.51 Write: 144.0
1. 2016 System:
   1. PC Mark 6,153.
   2. Cinebench CPU: 1399 cb
   3. Cinebench OpenGL: 171.54fps
   4. Personal AE File Render Time: 3 Minutes, 37 Seconds.
   5. Personal AE File AME Render Time: 20 minutes, 38 seconds.
   6. Personal PPro File Encoding Time: 40 Minutes, 22 Seconds
   7. PPBM
   • Benchmark: Disk I/O Time: 754 Seconds.
   • Benchmark: Disk Write Rate: 49 MB/second.
   • Benchmark: Time to Encode H.264 Timeline = 150 Seconds.
   • Benchmark: MPE Gain: 515/37=13.9
   • CrystalDiskMark:
     • Seq Q32T1: Read: 499.6 Write: 258.4
     • 4K Q32T1: Read: 126 Write: 119.8
     • Seq: Read: 492.6 Write: 117.2
     • 4K: Read: 22.55 Write: 48.21
2. 2017 System:
   1. PC Mark 6,893​
   2. Cinebench CPU: 1552 cb
   3. Cinebench OpenGL: 157.31fps
   4. Personal AE File Render Time: 19 Minutes, 36 seconds.
   5. Personal AE File AME Render Time: 1 Hour, 14 minutes, 25 seconds.
   6. Personal PPro File Encoding Time: 44 Minutes, 46 Seconds
   7. PPBM
      • Benchmark: Disk I/O Time: 57 Seconds.
      • Benchmark: Disk Write Rate: 650.74 MB/second.
      • Benchmark: Time to Encode H.264 Timeline = 99 Seconds.
      • Benchmark: MPE Gain: 287/37=7.8
   8. CrystalDiskMark:
      • Seq Q32T1: Read: 550.1 Write: 512.5
      • 4K Q32T1: Read: 326.4 Write: 319.1
      • Seq: Read: 515.8 Write: 502.5
      • 4K: Read: 35.51 Write: 134.2

Conclusion: The 2017 system may have a higher PC Mark and Cinebench CPU score, but it looks like the 2016 system pulls ahead in just about every practical real-world test. I'm beginning to think the Xeon / M5000 combo was a much better editing purchase. However, based on other posts I've made in the past. The impression I got before purchasing the 2017 workstation was:

• The additional CPUs provided by this Xeon did not provide any additional performance after a certain number of cores.
• The Workstation GPU / ECCC RAM were "server grade" and were not that helpful for a single individual working on a single project,
• The M5000s inflated price mean't little to nothing for editing performance, and the 2017's GTX 1080 would benefit more due to the number of CUDA cores.

The question remains: I have one week to return the 2017 system for a refund and stick with the 2016 system. Do I return it, or keep both? Thank you for your time.

[Mylenium]

I've written it a million times: coreXi systems will always beat Xeons, since their per core efficiency by ways of higher initial clock speeds, TurboBoost and extra cores not sitting around idly simply due to the fact that for mathematical and technical reasons a lot of video processing is not and will never be multithreaded beyond a certain point, including GPU processing even.

Everything else is not really relevant. You seem to have plenty of time to run all those synthetic tests, but in the end they are all meaningless. There's just a million factors to every individual project and using a specific feature or effect that you never used before can make even the most beefy system break down and working across different programs in different versions will literally shift the balance with every new update. I would strongly suggest you stop your search for the holy grail. Imagine if Adobe actually fixed those quintillion bugs in AE and even a ten year old system would suddenly run smooth as butter...

Sorry, but such things really have become a pointless exercise and you can put Cinebench and all that nonsense wherever you like. In the end it might be more useful to evaluate what work you actually do and trust your gut feeling based on your experience. Otherwise I'm sure many people would just be glad if they actually could afford a 5000+ computer, so for all intents and purposes, what is the real issue here? Some comp that we know nothing about rendering lousy on your Xeon? While that's certainly something that may or may not bug you, I'm sure someone who's working on a measly notebook would use certain language and wonder... If your company can afford it, then go ahead, but I'm afraid you won't get a definitive answer you are looking for and nobody can take the decision of your shoulder.

Mylenium

[BLSmith21122]

I appreciate the reply Mylenium.

My main point with this is: It seems the i7 6900k/128gb ram/gtx 1080 in virtually every practical test performs slower than the workstation grade Xeon E5-1680 V3/64gb ram/2x M5000 in every rendering environment. The only place the i7 system performs better is with a PC Mark score and Cinebench CPU score - which by all accounts doesn't really matter because the practical renderings of both After Effects and Premiere are all slower on the i7 system. So the case to be made that the i7 6900k would beat the Xeon doesn't seem to hold water here. What I'd like to know is: If there is something I've done wrong, and if not, if I should return the i7 system for a full refund.

[Dave_LaRonde]

I think two other questions might be more germane:

Will the type of After Effects work you do change appreciably in the next year or two?

Do you have an IMMEDIATE NEED for the new system that the current one can't fulfill?

If the answers are "no" and "no", send the new one back and get a refund.  Earn a little good will with the bean-counters and higher-ups.  They'll take you more seriously when the time comes that you REALLY need a new system.

Which, I think, would be worth more to you than some barely-noticeable improvement in speed.

[RjL190365]

I just took a look at those results, and you were wrong on one of your impressions. The 2016 system has an abysmally slow disk subsystem, performing much, much slower than even a slow hard drive in the PPBM Disk I/O test! Sequential writes in CrystalDiskMark are also slower than a slow hard drive, as well.

And you will not be able to produce reliable results if you use only a single disk for absolutely everything. You see, the reason why your 2017 system underperformed in your real-world tests is because none of the systems appear to be using a second disk at all, and the OS disk must handle OS housekeeping chores in addition to what you're trying to perform with it. Those housekeeping chores effectively made your disks a whopping four times slower than they should have been.

[RjL190365]

I forgot to mention that your oldest system appears to use a hardware RAID while the other two systems use only a single disk.

[Bill_Gehrke]

As the architect of PPBM with your great testing data there is one test which I see as most significant and it is sort of buried but if you look at the MPE Gain the first number in the results is the CPU power.  Here are those numbers for each system.

2013 System -- 316 seconds

2016 System -- 515 seconds

2017 System -- 287 seconds

That one test is a strictly a CPU exercise using Premiere Pro and since most of Adobe software mentioned is primarily CPU driven,   clearly your 2017 System would be my choice especially with the new demands you will be facing.  Your 2017 system is about twice as fast as the 2016 system.  Also it clearly wins with the 128 GB of RAM

I think you have not tuned the CC2017 system and from the Speccy I believe you have to many unnecessary processes running.  Take a look at our now dearly departed Harm's Tuning Guide.