EEVblog #726 - Dual Xeon Video Editing Machine Build

Dave builds and tests the 24 core dual processor 2.6GHz Intel Xeon E5-2630 v2 video editing machine and benchmarks against his current 8 core i7 3770K machine at 3.5GHz.
This is one hour of PC building and software testing, if you find this stuff boring, DON'T WATCH IT!

CORRECTION:
The i7 3770K is actually running at 3.7GHz
UPDATE:
Even doing a 1 hour render at 240W consumption, the heatsinks do not even get warm to the touch.

Support the EEVblog through Patreon!

EEVblog Amazon Store (Dave gets a cut):
Donations:
Projects:
Electronics Info Wiki:

[MasterControloriginal]

In Vegas/Movie Studio check the "Maximum number of rendering threads" option, you can find it in Preferences>Video tab

By the way the PSU is upside down...

PS: as someone else said maybe the other 8 pin connector is there for a reason so connect it to the PSU.

[Zyphera]

Dave! Do NOT let Sony do the rendering. Sony is awful at rendering.

Just export it uncompressed and let handbrake do the job! And connect the 8 pin.

[Zoey_yea_boom]

Hi Dave

You probably find that, Sony isn't really optimized for more than 8 core, this is why you find turning off Hyper Threading speed things up a little bit.

Windows 7 can't handle the amount of cores you have, the cpu job scheduler for is a bit rubbish. they have fixed it in windows 8 and 8.1.

Also faster memory doesn't really, there are plenty results that show running 1333 vs 1600 vs 1800 give almost the same result.

There maybe BIOS update for that board 

Couple of things to try
   - see if there is a BIOS update.
   - Try and get a copy of windows 10 preview ( windows 10 will be a free upgrade if you have a valid  windows 7 or 8.x license, install Sony on top of this - see if that make an improvement, if so re-enable hyper threading. 

Tony

[SantaClaw]

EEVblog Hi Dave, If I where you, I'd try:
1. Windows 10 tech preview. (it's a LOT better than win 7 on multicore applications)
2. Bios Upgrade. (it may provide better hyper threading support)
3. Plug in the extra 8 pin cpu power cable (as modern cpu's will throttle down if they don't get enough power, I mean, what do you got to loose?)
4. I don't think faster memory will affect your test all that much, based on personal experience.5. Put equal amounts of memory on each cpu/same configuration for optimal performance.6. How about a small overclock on those CPU's ?

[RaStrNL]

Hi Dave, regards from Prague, Czech Republic. I am a bit too late now with my comments, but because I really love to watch you videos about electronics I am a bit shocked how little you know about the “PC computer and server stuff” J ! I have over 15 years of experience with building fast while reasonably expensive servers for my business and so maybe you can find my ideas useful and it will help you to get max. performance from your hardware ! So here are my points:
as already stated multiple times here in comments, you must go for 4 channel memory configuration ! This Intel XEON CPU is highly optimized for 4 channel memory, all it’s caching strategy, and memory access is optimized for quad channel, so using just dual channel you get even less than 50% of max. RAM performance ! CPU supports 1600MHz RAMs, so you should go for 4 channel 1600MHz RAMs there. Of course the lower latency the better too. I am not sure if non-ECC RAM is supported by your motherboard/CPU here, if yes, then even non-ECC RAM could help you to improve performance a bit (while giving-up a bit of reliability, but that’s rather questionable, I haven’t seen memory error for non-malfunctioning models for a decade now, so ECC is not so important as it used to be some years ago), but non ECC server RAM could be a bit hard to get don’t bother about it.
Hyper-threading issue – all Windows versions since Windows XP supports hyper-threading, but in reality it means that Windows XP only knew that hyper-threaded CPU is not a real core and they did not count it so for a purpose of licensing (you know that Windows were/are licensed per socket/physical core and not virtual core). Since Windows Vista, there are other improvements there, process scheduler actually knows that  physical cores are running faster when their corresponding virtual counterpart is unused, so it tries to use physical cores first and then when there is still need for more power it allocates some time on virtual core too. But that’s not 100% working, some multi-core optimized applications are not aware of physical/virtual cores and try to use all of them evenly, which is some cases can actually drop the performance when compared to non-hyper-threading configuration. But my experience is that with those new latest generations of INTEL CPUs hyper-threaded virtual cores have almost the same performance as physical cores, so for most applications hyper-threading is a real benefit, of course the rest of the system must be optimized too, in your case it could be the RAM bottleneck, what caused degrade performance with hyper-threading enabled.
multi-cpu setup: I am not sure if you know that you current setup is using NUMA model, which can degrade performance severely for applications not aware of that, especially when you have even non-uniform memory size (and maybe even speed mismatch). In this case actually using only all cores from a single CPU (event the virtual ones) could lead to much higher performance since shuffling data between those two CPUs when all cores are involved could (and especially in your case with degraded RAM performance) degrade the speed severally too.
power-optimization – while power optimization is generally good, in your case you should choose top performance setting in your BIOS and in your OS power setup, otherwise you might lose significant speed boost available from turbo-boost mode of your CPUs. If you do not let you CPU to boost, you are losing 20% of its possible speed, with a proper cooling and full power enabled it should go to 3.1GHz from a default 2.6Ghz easily since you poorly multithreading application from SONY is not able to load all its cores to 100%.
 
So my conclusion is that you should first try to install optimized RAM configuration, as suggested by someone else here, 8x2GB of 1600MHz unbuffered (if supported otherwise registered ECC), preferably using two sets of 4 DIMM packs, or at least 8 identical modules, with lowest latency available. Of course if you would need more than 16GB of RAM, then you should go for 8x4GB modules, etc., never mix different sizes and speeds ! Then you might experiment with different amount of cores used by your application, you can keep hyper-threading enabled and instead try using Set Affinity option from  Task Manager, to assign different core configurations to your SONY application to find out, what gives the best result. Of course upgrading you software to a version which supports better multithreading, native 64-bits, etc. could be a good way to go too.
 
Good luck to you, I believe that you can squeeze much more performance from your dual XEON setup !

[frnage]

R9 290, great space heater for the upcoming winter!

[dmnt]

Thank god you didn't spread the thermal compound out. The PC Master Race people would of gone crazy.

[ludo]

A small tip for your handbrake pass: upgrade to the latest version and use the x265 / HEVC codec. This should compress your videos even better without impacting the quality.

[electrodacus]

one core with HT is not the same as two real processors it may help 10 to 30% than one without HT depending on application.

[Esthreel]

That Sony software is pretty inefficient encoder. Just export uncompressed and handbrake. 

[aMondia]

Firstly, using dimms of different models is not really optimal.
Secondly, you should at least try to connect the other power connector on the motherboard.
And lastly, I think you could speed up your rendering quite a bit by not doing the bulk of it in movie studios, just render it uncompressed and use some other encoder.

[kidkool]

A couple comments:

Vegas has completely disregarded the thousands of dollars of hardware you've thrown at it. I'd either replace it or render solely using Handbrake.

The server utilizes quad channel memory, meaning memory works best when used in pairs of 4. The more similar each RAM stick is, the better.

It's really a shame to see thousands of dollars of hardware not go to its full potential. Good luck with the system!

[Tangobaldy]

Do you really need huge heat-sinks for processors?

[aerofart]

Dave why are you still dicking around with MovieStudio?  Get yourself the full Vegas product and run it in 64-bit.  Unlike MovieStudio, it's made to handle this sort of stuff and get the horsepower out of your machine with multi-threading, etc.

[TheWolfinator]

EEVblog Have you turned off the Power Efficiency mode in the BIOS?? 

[xboxRob]

Before watching this im guessing the rendering wont load up all 24 cores. I had a 24 core rig once and could only get it to use about 16 cores and only 20% usage on them and was slower than my 8 core i7

[billysgeo]

So... the back CPU cooler is sucking hot air from the front cooler? I guess with a lot of air flow in there there shouldn't be any problems...

[xboxRob]

Your ram was as fast as ddr2 / pc2 ram not good. Get 8 sticks and fill all the blue slots so your cpu is using quad channel. Just get 8x2gb sticks as 16gb is plenty.

[LRTOTAL]

Hi, Dave EEVblog!
It would be really nice to see the thermal images of the pc running.
Like when you used the thin plastic to have the same air flow as when the case is closed.

Thank you for your wonderful contribution to the YouTube community!

[MaZderMind]

Oh dear, I don't think I want to enter the YouTube-Comment-Shitstorm but well… your Video-Rendering-Pipeline basicly consists of three Parts. Part 1 is Source-Material Decoding, Resampling, Scaling, Rendering the Overlays and generally producing a Stream of uncompressed Frames from your Production. This depends mostly on how good your Video-Editor can offload the different Tasks to multiple Cores. Part 2 is encoding these Frames into your intermediate Codec, which obviously mostly depends on how well the intermediate Codec can scale across multiple CPUs. Part 3 is then transcoding to h264 with Handbrake which also depends on how good the codec can scale across the Cores. As you already said, the last Part is incredibly fast while the first two are a bit problematic. When it comed to Multicore-Encoding, different Codecs behave vastly different. For Example WebM can only be encoded Single-Threaded, there is no WebM-Encoder that uses more that one Core. h264 on the other Hand scales well up to around 6 Cores but more Cores won't help much. I don't how good your Intermediate-Codec scales but I do know that h264 is basicly finished; there is no headroom of improvement left in the development of that codec, it is currently the fastest and best scalable codec available, what you basically proved by your Handbrake-Encoding. So I'd suggest trying to a) directly encode to h264/mp4 and skip the intermediate state or b) render to an uncompressed Codec like ProRes onto your SSD (Spinning Metal will not be capable of writing that amount of Data). The second test would give an indication whether your Video Editor is not generating the uncompressed Frames fast enough to saturate the Video-Encoder or whether the Video-Encoder is not able to Scale onto the bigger number of Cores. Good Luck!