Desktop for video editing

[Brian Stirling]

Ah yes, USB 10Gb/s... The MOBO I originally suggest for being a good value actually has a bundle with a USB 3.1 card:
http://www.newegg.com/Product/Product.aspx?Item=N82E16813157600

Or you could buy your own add in card for a different MOBO. They run $25-50.

My issue with the 980ti is that it's not a compute card. None of Nvidias current cards are built for professional applications, and are actually much slower than their older cards that were. The old Titan had 1.5 TFLOPs and Titan Black had 1.7 TFLOPs the newest Titan X and 980ti have about 0.2 TFLOPs.

For comparison the 290x has 0.7 TFLOPs, and the cheaper 280x actually has about 1.0 TFLOPs. I had suggested the 290x because it was on the recommended list, and is newer and usually faster than the 280x (only slightly behind in FP64) and is readily available at retail. There were other considerations I made as well, relative and total costs and the fact that these cards are being replaced soon.

However, all that said I'll admit that I don't know exactly what/how your software runs. If people who do use this software are recommending a 980ti then maybe it's OK. I'll have to try to find benchmarks of this specific program. Even if Nvidias currently weaker FP64 performance is not an issue, I would not suggest the 980ti. It's to expensive to close to end of life, cheaper Nvidia cards have better value.

Yeah, the 980Ti is not a new board so one would think there would be a replacement with improved performance before long. The Titan boards are way over $1K and are way beyond the 980Ti's already high price of $650ish. There are, of course, quite a few of the more serious video editors that spend way north of $5K for a box and might have more than one GPU -- I can't justify going that far at this point and going from $1.5K to $2.5K is about as far as I'm willing to go.


Brian

 

[EliteRetard]

You don't seem to be reading what I'm writing.

Anyway I've come to find that this comment:

General advise
No AMD
For the moment ONLY nVidia GeForce cards are advised. CS6 does not support AMD cards and AMD is still considered the 'New kid on the block' as far as CC goes. AMD is still in its infancy, the drivers and Adobe support is still too new to effectively compete with nVidia's CUDA acceleration, which has been proven in the past years and have stable drivers from nVidia and stable support from Adobe. nVidia CUDA support is a mature technology with a proven track record. Add to that the recent price increases of AMD cards and they have effectively outpriced themselves in the competition with nVidia

Was written for CS6 which was a product of 2012. I've already posted information showing that AMD had almost full compliance with that version back then. I'm still digging but everything I'm finding is suggesting AMD is as good a Nvidia TFLOP for TFLOP. And while Nvidia used to have more FP64 performance, AMD actually now leads FP64.

I still haven't found what I'm looking for, but it looks like you either get an older used Nvidia card or a new AMD card if you want the best performance in this software. The newest Nvidia 9xx series and the newest Titan X have far less FP64 performance than all other cards. Again I'm still looking...

 

[EliteRetard]

Quadro M6000 is based on GM200, same as 980ti ($650 not approved) and Titan X ($1,000 not approved)
Quadro K6000 is based on GK110, same as the old Titan (used $450 approved) and Titan Black (used $450 not approved)

Already you can see the older Nvidia card is better. This is what I've been saying. The 980ti and Titan X are certainly faster for gaming and are generally more efficient with resources. However they have such a massive deficit in FP64 that they cant make up the difference with efficiency and are slower than the old cards in professional applications.

Now the other points:

FirePro W9100 is based on Hawaii like the 290x ($300 approved)
FirePro W9000 is based on Tahiti like the 280x ($200 approved)

Immediately it's clear MPEG-2 performance is a loss for AMD in this case (note this is an older 2014 version). I'm not going to argue on this one, Nvidia wins.

However, when it comes to the tougher Blu-Ray processing the 290x class card performs very closely to the 980ti Titan X class card. The 290x is half he cost. Now look at the benchmark chart I posted earlier, where they have dual GPU tested. You can see there is a serious performance gain by adding a second card. I don't have the proof, but I suspect 2x 290x cards will be faster than 1x 980ti and be cheaper as well. Or if you consider the 280x at $200, you could have 3 cards...and I feel certain that 3x 280x card could beat a single 980ti in this software. Potentially even eliminating the MPEG-2 performance deficit (again i have no proof for this).

I'll keep digging around more tomorrow, it's midnight here now and time for a nap.
I want to close by saying the 280x looks like it might actually be the winner in all of this. It performs well, and is relatively much less expensive than the other options...and I think it could make a great holdover GPU to use until the new cards release.

 

[NAC]

This link is the best I've found on benchmarking video cards for Premiere:
http://www.studio1productions.com/Articles/Premiere-Benchmark.htm
There are a lot of charts - I find the last 4 most useful.

The main conclusion of the benchmarking is that if you use a huge amount of transitions and effects which are supported by the GPU, the GPU will increase speed substantially. If you use a "normal" amount of transitions and effects, you are better of spending the money on a a 8 core CPU. I believe color grading is GPU driven. If you plan on color grading, and it is GPU driven, then it would be a no-brainer to spend on the GPU. You can predict if you'll be editing simple drama/comedy/etc, or doing more graphics intensive things or color grading.

Also - that benchmark also indicates that the 780 is faster than the 980 in Premiere, so if you are comfortable buying used you can save about $300.

 

[NAC]

Also - x99 motherboards are quad channel memory. So you'll want 4*8 instead of 2*16. Plus it is cheaper - it should be about 150.

 

[EliteRetard]

Yep, more confirmation that the old Nvidia GPUs that had much higher FP64 performance are better than the newest "fastest" Nvidia GPUs that have almost no FP64.

But that's the problem, you'd have to get an old and likely used card if you wanted Nvidia (with the best performance). Why do that when you can get a brand new AMD card that competes in FP64.

I also found this:
https://twitter.com/dimitry49/status/614169100617650176

So far in my @Adobe CC 2015 encoding tests, there's no difference btwn GTX 980 Ti, 980, R9 290X, 390X and Fury X

Yet of all those cards, only the 290x is on the official recommended list, and is $300 new (sometimes much less)

Also, when working at 4k, the AMD cards seem to close the gap on the Nvidia cards...this also seems to hold true in gaming.

Again you can see the 280x class card following very close to the 290x class card. But the 280x is 2/3 the price of the 290x, and both are officially recommended. Looking at the benchmarks posted by NAC, you can see that even the weakest GPUs can reduce time significantly. I think one of these cards (I'm especially leaning towards the 280x now) could provide a very decent boost for now, and hold over for the significantly newer cards.

I know I keep harping on about those new cards, but they're going to be a on a new manufacturing node and likely with new memory types. Nvidia needs a new FP64 compute card, they know this is a fact, they simply didn't have the die space to fit FP64 into their current lineup. With the node shrink they'll have plenty of room to add the FP64 back in. A 40% boost in FP32 speed is a totally plausible scenario...but they could easily double or more their FP64 performance on consumer cards. Also potentially worthwhile is more and faster memory on the GPU, which so far has been important in 4k performance.

And these new cards are about 6 months away.

 

[Brian Stirling]

Also - x99 motherboards are quad channel memory. So you'll want 4*8 instead of 2*16. Plus it is cheaper - it should be about 150.

Good point.


Brian

 

[Brian Stirling]

This link is the best I've found on benchmarking video cards for Premiere:
http://www.studio1productions.com/Articles/Premiere-Benchmark.htm
There are a lot of charts - I find the last 4 most useful.

The main conclusion of the benchmarking is that if you use a huge amount of transitions and effects which are supported by the GPU, the GPU will increase speed substantially. If you use a "normal" amount of transitions and effects, you are better of spending the money on a a 8 core CPU. I believe color grading is GPU driven. If you plan on color grading, and it is GPU driven, then it would be a no-brainer to spend on the GPU. You can predict if you'll be editing simple drama/comedy/etc, or doing more graphics intensive things or color grading.

Also - that benchmark also indicates that the 780 is faster than the 980 in Premiere, so if you are comfortable buying used you can save about $300.

Good info, sadly it did not include the GTX 980 Ti just the slower GTX 980.

At this point I'm not planning on a lot of elaborate effects but definitely color grading and a number of other GPU intensive tasks. For sure, if you're just using it to cut a GPU is almost unnecessary whereas the CPU is the thing and having a 8-core CPU would be the way to go -- if you could handle the price of nearly $1K for CPU alone.

So, I've come quite a ways from my first expectations but I'm still not looking to be a benchmark king. If I OC it will be mild OC as I'd rather not push the longevity or risk having a box that's too loud due to the fans screaming. I'll spend some money to water cool the CPU and have given some thought to water cooling the GPU and again, the reason is to keep the temps and noise down.


Brian

 

[Brian Stirling]

Quadro M6000 is based on GM200, same as 980ti ($650 not approved) and Titan X ($1,000 not approved)
Quadro K6000 is based on GK110, same as the old Titan (used $450 approved) and Titan Black (used $450 not approved)

Already you can see the older Nvidia card is better. This is what I've been saying. The 980ti and Titan X are certainly faster for gaming and are generally more efficient with resources. However they have such a massive deficit in FP64 that they cant make up the difference with efficiency and are slower than the old cards in professional applications.

Now the other points:

FirePro W9100 is based on Hawaii like the 290x ($300 approved)
FirePro W9000 is based on Tahiti like the 280x ($200 approved)

Immediately it's clear MPEG-2 performance is a loss for AMD in this case (note this is an older 2014 version). I'm not going to argue on this one, Nvidia wins.

However, when it comes to the tougher Blu-Ray processing the 290x class card performs very closely to the 980ti Titan X class card. The 290x is half he cost. Now look at the benchmark chart I posted earlier, where they have dual GPU tested. You can see there is a serious performance gain by adding a second card. I don't have the proof, but I suspect 2x 290x cards will be faster than 1x 980ti and be cheaper as well. Or if you consider the 280x at $200, you could have 3 cards...and I feel certain that 3x 280x card could beat a single 980ti in this software. Potentially even eliminating the MPEG-2 performance deficit (again i have no proof for this).

I'll keep digging around more tomorrow, it's midnight here now and time for a nap.
I want to close by saying the 280x looks like it might actually be the winner in all of this. It performs well, and is relatively much less expensive than the other options...and I think it could make a great holdover GPU to use until the new cards release.

Come on, none of these boards are even remotely in my price range with the cheapest one being over $2K and one going for $4K -- these boards are irrelevant to this discussion. Yeah, a full time video editor working in Hollywood and not paying for it himself will have no problems plunking down that kind of money, but mere mortals aren't spending $8K-$12K for a PC before adding monitors.

I decided that my early expectations of $1k-$1.5K were unreasonable for more than casual editing of 4K video but the next step up, where I'm at now, is about as good as I can expect and to get noticeably better I'd probably be looking at $5K -- not happening...


Brian