What about this "killer" R9 390? :)

http://www.computerbase.de/2015-09/devil-13-dual-core-r9-390-powercolors-dual-grafikkarte-mit-zwei-grenada-gpus/

Has anybody done any tests using a network rendering farm type setup? It seems to me that some (5?) mediocre or better rendering systems would be faster than 1 fairly expensive system. The 5 systems may be a little more money wise but also more flexible when it came to how they are used.

[I]Has anybody done any tests using a network rendering farm type setup?[/I]

Do you mean network renders? That feature has been dropped a while ago.

Re: " What about this "killer" R9 390? :)"

According to the Anandtech bench comparisons, it is dead even with the R9 290X

http://www.anandtech.com/bench/product/1439?vs=1559

( Scroll down to:
Compute: Sony Vegas Pro 13 Video Render)

Even the difference between R9 290 and 290X isn't worth the extra money.

According to this chart, it's 1 second slower than the AMD Radeon R9 Fury X and tied with several other AMD cards :)
http://www.anandtech.com/bench/GPU15/1231

Mike

The fury X with HBM has double the single point math capabilities of a 290 or 390. The benchmark file they are testing at Anandtech is rendering to a file. If you download the vegas test file, you would see that the test is much more complex than rendering a file from a to b, it deals with timeline playback performance for various editing scenarios. There is more to editing an MPEG2 or MP4 rendering performance.

If Anandtech update their test file to say a 4K 32-bit Full level .EXR ACES workflow, you would see much greater difference in the GPUs listed at the top.

You GPU does matter regardless of what Musicvid10 says, not incorporating your GPU into the mix is like doing all your higher math on paper vs using a calculator. The is just faster at certain tasks than the other, and Vegas will the devices that are faster a rendering the results. If you use a gimped AMD GPU or NVidia GPU because you think it does not matter, then you will not see the benefits faster math processor is adding.

Encoding is not that hard, the faster your system works the timeline elements and gets the results to the encoder, the faster your render will be. If you do only CPU, then you better have the fast RAM and plenty of physical cores to turn the timeline calculations, and then also handle all the encoder math. With OpenCL you get the compute power of both the CPU and GPU working in tandem. The GPU does not do everything like the CPU is able, but it does certain math much faster than a CPU ever could. Why would you not want that performance gain? Most peoples poor experience with GPU probably comes from their lack knowledge in system hardware, or purchasing a system by price. Then complaining they are not seeing what they thought.

Everything Oldsmoke is saying about PCI lanes and what not should be heeded as great insight. Cheap system are cheap for a reason, figure out why the system is cheap before purchasing.

The response was to the question about the 390x..

There is very little difference between the 390X and the 290X down the whole list of tests..

You would expect it to scale similarly in Vegas..

The only place where the 390X shines, is in power consumption and noise.. :)

As far as new system configuration goes I think the X99 system with the hex core that Oldsmoke recommends, 32GB RAM, and dual 290/390 GPUs would be good. This way one GPU could be monitor/desktop, and the other used only for Vegas timeline compute with no monitors plugged in. I would personal add an Intel 750SSD or use the M2 slot for boot and project depending on how much space you bought. The x99 will support 2 GPUs at x16 + one 8X PCIe SSD.

If you read the VP11 benchmark testing results.PDF you will see this is how Sony configured their system.

"Tests were developed and performed to determine increases in playback performance and decreases in rendering time on a Dell™ Studio XPS™ system, featuring an Intel® Core™ i7 Quad-Core 3.2 GHz CPU with 9 GB of RAM and Microsoft® Windows® 7 Professional 64. Two video cards from AMD and two video cards from NVIDIA were installed with the latest drivers. A clean uninstall of the previous driver was CRITICAL.".

Skylake would be even more compelling than X99 if it had more memory channels, and more CPU Cores. The addition of DMI3.0 (64GB/s) on SkyLake is about the only thing it has over the x99. We just have not seen an x99 class board from the skylake series yet.

The x99 will support 2 GPUs at x16

Careful with that! The 5820 CPU will not support 2x PCIe x16, it has only 28 lanes. I am waiting for the Skylake E before I change my system.

I also doubt that SCS had two cards installed at the same time. I do run 2x R9 290 on my system but the gains are small compared to a single card but it does help when you work with 4 monitors or more.

RE: VideoFreq - what he says here is true. Your "good" dual Xeons with multiple CPUs would be your best bet. Vegas under the hood has setting to dedicate up to 8 threads alone to AVCHD decoding, and rendering threads up to 16. That is like application 24 threads on those features alone. How fast your system can service all those threads determines how quick your system works. While most are trying to get by on 4-8 threads including HT. Vegas will scale the tread count to detected hardware, as well as windows schedules services of threads beyond the CPU count. This is mainly why core count matters so much.

There is also the benefits of the memory architecture on Xeon mother boards.

The problems I see with Xeon is that the "good" Xeons' have a whopping price tag that makes the average video guy puke, but corporate budgets not even flinch.

Xeon plus a good pair of GPU would be the way to go, because you get the benefits of thread count and GPU calculation speeds for the things even Xeons are not as good at.

The problem with the Xeon is the rather low frequency compared to a mainstream CPU and most don't run all cores at max clock.

@ Old Smoke

Hey Old Smoke,

I Pm'd you about running dual GPU's but got no reply, so I'm thinking maybe your email here is out of date.

I added a second 290x to my 5960X system, but saw no noticeable change in render times with codecs that use GPU ( v12 ).. Is that to be expected, or is there something I'm missing?

@ Oldsmoke - So you are saying the Sony published .PDF is lying about installing 2 GPUs?

The Dell Studio XPS was an X58 system with enough PCI lanes to support dual PCIe2.0x16 GPUs. However, 2 6870s are probably less than one 290X today, in term of memory bandwidth and processing power.

I am not saying you are wrong. But why would the guy writing the paper lie about his test configuration?

@oldsmoke - this is true. The Xeons with high freqs and core counts are really expensive.

@astar

Not lying but also not being clear. The "Video Render Profile Results" chart show results for each of the GPUs but not for both of them together. So yes, two each, 2 Nvidia and 2 AMD cards where tested but, imho opinion never installed at the same time in the system. You wouldn't want to mix the GTX570 drivers with the Quadro 5000 drivers.

"You wouldn't want to mix the GTX570 drivers with the Quadro 5000 drivers."

Sorry, that can be done but is tricky. I run here my machine with both a GTX570 and a Quadro 2000D together. Works fine. With earlier versions of the driver it was important to install first the GTX driver and then the quadro driver (to have a 3D-preview by the Quadro for example). But that can be done. And certain driver combinations did not work at all.

But with the latest driver - also for Win10 - it is enough to install the driver for the Quadro. Seems to work fine also for the GTX - beside the fact that we do not have drivers that support nvidia 3D for the older Fermi cards that work also with Win10 by now.

Beside that it is the question if such an installation should be recommended at all. If there are not clear reasons to do that (in my case the GPU support together with the Quadro 3D support in Vegas) I would not recommend that at all.

" this is true. The Xeons with high freqs and core counts are really expensive."

And this is the reason why I think that the i7 5960X is still an expensive processor - but cheaper then the Xeons with high fregs. And it can be overclocked in a great way what means that it will support Vegas in a great way.

So now we're 15 months down the road from the last post.

What's the latest thinking on best CPUs & GPUs for VP?

I need to build a new system for 4K editing pronto.

The fury X with HBM has double the single point math capabilities of a 290 or 390. The benchmark file they are testing at Anandtech is rendering to a file. If you download the vegas test file, you would see that the test is much more complex than rendering a file from a to b, it deals with timeline playback performance for various editing scenarios. There is more to editing an MPEG2 or MP4 rendering performance.

If Anandtech update their test file to say a 4K 32-bit Full level .EXR ACES workflow, you would see much greater difference in the GPUs listed at the top.

You GPU does matter regardless of what Musicvid10 says, not incorporating your GPU into the mix is like doing all your higher math on paper vs using a calculator. The is just faster at certain tasks than the other, and Vegas will the devices that are faster a rendering the results. If you use a gimped AMD GPU or NVidia GPU because you think it does not matter, then you will not see the benefits faster math processor is adding.

Encoding is not that hard, the faster your system works the timeline elements and gets the results to the encoder, the faster your render will be. If you do only CPU, then you better have the fast RAM and plenty of physical cores to turn the timeline calculations, and then also handle all the encoder math. With OpenCL you get the compute power of both the CPU and GPU working in tandem. The GPU does not do everything like the CPU is able, but it does certain math much faster than a CPU ever could. Why would you not want that performance gain? Most peoples poor experience with GPU probably comes from their lack knowledge in system hardware, or purchasing a system by price. Then complaining they are not seeing what they thought.

Everything Oldsmoke is saying about PCI lanes and what not should be heeded as great insight. Cheap system are cheap for a reason, figure out why the system is cheap before purchasing.

Is there a way to follow you here? it seems that you know very well everthing you say 👍👍

well I already did*

I asked a local computer vendor with a good reputation to recommend a configuration optimized for VP 13. I specifically mentioned Xeon processors, because I saw strong recommendation for it over i7x (ound for pound) on this forum. Here's part of the response I got via email:

"We would plan for a Core i7 7th generation 8-Core CPU.  These days, the highest performing Core I7 CPUs are a match for some of the Xeon Server CPUs.   The difference is that Core i7 branded CPUs are intended for media tasks.   Server processors usually are not."

Sound? Unsound?

Sound? Unsound?

Definitely sound.

I vote for the i7 8 core. I have i7 6 core and am very happy .. My next build will be an 8 core i7...