Welcome to our community

Be a part of something great, join today!

Register Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Resolve vs PP Multicore support?

Tom Lowe

Tom Lowe

Well-known member
Joined
Jan 5, 2007
Messages
8,520
Reaction score
1
Points
0
I was planning to build a new overclocked 8-core i7 rig, but lately I have been using Resolve more, and am wondering if I should instead be looking at a dual-CPU rig? How does Resolve compare to Adobe PP in its ability to utilize multiple cores?

For Resolve, would I be better with an overclocked i7 8-core running at 4.5GHz, or a dual CPU system running at lower clock speeds?
 
Premiere, and to an even greater extent, Resolve, both can use a lot of cores/ threads. You will get more out of these applications with the larger number of cores and increased throughput of a dual Xeon workstation, if you build it up right. Premiere does well with scaling across multiple threads up to about 20 threads or so, then the returns of more cores/ threads just offers little in terms of increased value and performance. Also, as we ramp up the number of cores in a CPU, the trade off is in clock speed. Resolve works the same way, but it better utilizes both CPUs and GPUs so you can go a bit further. With this in mind, the real strength of a dual CPU system is that we can increase the number of cores while keeping up the clock speeds and I/O transaction speeds across multiple cores by distributing them across two CPU interfaces rather than one. The real drawback here is the price of building these multiprocessor workstations and how that affects your overall price to performance ratio. I often advocate that it's better to have two or three systems suited to certain workflow tasks than it is to try and do everything with one monster system.

Intel now offers 14 cores in a single CPU, which can be placed onto an X99 motherboard. But the cores are slow when all are working. It makes far more sense for these applications, and our typical workflows in this industry, to go with 16 or 20 cores via dual 8 or 10 core CPUs running at a much higher clock speed.

When it comes to GPUs, Premiere still only uses one GPU effectively. It recognizes more and occasionally will see some boost from a second GPU. However, there are also times when having two of them active in Premiere can actually affect performance in a negative way. Adobe did a great job of coming out with GPU-accelerated editing, but they have since lagged behind Resolve and some other high end applications in terms of multi-GPU support.

Resolve fares much better in this regard, it can use all the GPU power you throw at it, as long as the rest of your system is up to the task of feeding the GPUs. In a E5-v3 dual Xeon setup with ample DDR4 memory, I would recommend two or three Titan Black GPUs or the new GTX980 6GB cards. The rumored 980Ti 6GB card looks like it might arrive next month and should provide the same CUDA performance as a Titan black at 35% lower cost.
 
Jeff Kilgroe said:
Premiere, and to an even greater extent, Resolve, both can use a lot of cores/ threads. You will get more out of these applications with the larger number of cores and increased throughput of a dual Xeon workstation, if you build it up right. Premiere does well with scaling across multiple threads up to about 20 threads or so, then the returns of more cores/ threads just offers little in terms of increased value and performance. Also, as we ramp up the number of cores in a CPU, the trade off is in clock speed. Resolve works the same way, but it better utilizes both CPUs and GPUs so you can go a bit further. With this in mind, the real strength of a dual CPU system is that we can increase the number of cores while keeping up the clock speeds and I/O transaction speeds across multiple cores by distributing them across two CPU interfaces rather than one. The real drawback here is the price of building these multiprocessor workstations and how that affects your overall price to performance ratio. I often advocate that it's better to have two or three systems suited to certain workflow tasks than it is to try and do everything with one monster system.

Intel now offers 14 cores in a single CPU, which can be placed onto an X99 motherboard. But the cores are slow when all are working. It makes far more sense for these applications, and our typical workflows in this industry, to go with 16 or 20 cores via dual 8 or 10 core CPUs running at a much higher clock speed.

When it comes to GPUs, Premiere still only uses one GPU effectively. It recognizes more and occasionally will see some boost from a second GPU. However, there are also times when having two of them active in Premiere can actually affect performance in a negative way. Adobe did a great job of coming out with GPU-accelerated editing, but they have since lagged behind Resolve and some other high end applications in terms of multi-GPU support.

Resolve fares much better in this regard, it can use all the GPU power you throw at it, as long as the rest of your system is up to the task of feeding the GPUs. In a E5-v3 dual Xeon setup with ample DDR4 memory, I would recommend two or three Titan Black GPUs or the new GTX980 6GB cards. The rumored 980Ti 6GB card looks like it might arrive next month and should provide the same CUDA performance as a Titan black at 35% lower cost.
Click to expand...

As usual, up-to-date info from REDusers very own Systems Whisperer. '-)

Very fortunate we are to have access.
 
Jeff Kilgroe said:
Premiere, and to an even greater extent, Resolve, both can use a lot of cores/ threads. You will get more out of these applications with the larger number of cores and increased throughput of a dual Xeon workstation, if you build it up right. Premiere does well with scaling across multiple threads up to about 20 threads or so, then the returns of more cores/ threads just offers little in terms of increased value and performance. Also, as we ramp up the number of cores in a CPU, the trade off is in clock speed. Resolve works the same way, but it better utilizes both CPUs and GPUs so you can go a bit further. With this in mind, the real strength of a dual CPU system is that we can increase the number of cores while keeping up the clock speeds and I/O transaction speeds across multiple cores by distributing them across two CPU interfaces rather than one. The real drawback here is the price of building these multiprocessor workstations and how that affects your overall price to performance ratio. I often advocate that it's better to have two or three systems suited to certain workflow tasks than it is to try and do everything with one monster system.

Intel now offers 14 cores in a single CPU, which can be placed onto an X99 motherboard. But the cores are slow when all are working. It makes far more sense for these applications, and our typical workflows in this industry, to go with 16 or 20 cores via dual 8 or 10 core CPUs running at a much higher clock speed.

When it comes to GPUs, Premiere still only uses one GPU effectively. It recognizes more and occasionally will see some boost from a second GPU. However, there are also times when having two of them active in Premiere can actually affect performance in a negative way. Adobe did a great job of coming out with GPU-accelerated editing, but they have since lagged behind Resolve and some other high end applications in terms of multi-GPU support.

Resolve fares much better in this regard, it can use all the GPU power you throw at it, as long as the rest of your system is up to the task of feeding the GPUs. In a E5-v3 dual Xeon setup with ample DDR4 memory, I would recommend two or three Titan Black GPUs or the new GTX980 6GB cards. The rumored 980Ti 6GB card looks like it might arrive next month and should provide the same CUDA performance as a Titan black at 35% lower cost.
Click to expand...

Thanks, Jeff. Price for dual CPUs is not a major concern for me.. I'm mainly worried about performance. I still am not clear on whether I would be better off with one overclocked i7 CPU or dual Xeons for a Resolve machine? When I have tested dual Xeon CPU vs i7 OC CPU last year, the i7 CPU totally destroyed the dual Xeons in PP and AE. I'm wondering if Resolve is different? What are the best Xeons right now for a dual CPU setup for Resolve? Is Resolve able to utilize more CPU cores than Adobe?
 
Render speed will be faster on the overclocked I7 8 core in most applications except for AE sometimes and C4D most of the time. However the realtime playback perform ceiling will be higher ie complexity of timelines with the Dual Xeon. If Davinci is your primary goal then I would suggest the I7 8 Core clocked at 4.5GHz unless you are planning to add more than 2 video cards. If so then it's time for a Dual Xeon.
 
This is true of render times in most apps as the rendering is not taking advantage of multiple threads or not very many threads. Often times the bottlenecks can be the codecs themselves and even though a system may be capable of spitting out 4K+ frames in real-time, when you have a codec plugin spinning its wheels on two threads out of an available 20, all the CPUs and GPUs in the world are not going to help speed that along. That also comes back to my comment above about often recommending a few good systems rather than a single monster system. I often marvel at some of the frankenstein workstations people cobble together and how much money gets spent on PCIe expansion and other such solutions just so they can have one computer that does everything, although never really excels at any one task.

It's not always about rendering, either. Generally speaking, most all software will perform better on the over clocked i7 system. Very few applications are multithreaded and even fewer of those do it well or utilize more than the few threads that common desktop CPUs offer.

FWIW, I'm building and repurposing computers all the time. I have only built one dual Xeon in the past year -- a dual 10-core -- that serves as my coloring and finishing system. It runs Resolve like a beast and is a total waste of resources for just about everything else, but it's also a beast when it comes to renders I distribute across my network. Everything else is a single CPU system. I'm waiting on RAM and GPUs to build a replacement for my aging 6-core "beater" workstation which has been a jewel of a system that does most everything reasonably well. I have down-sized my system collection over the past few years. Other than R3D workflows and rendering tasks, we just don't need the computing power in terms of size and numbers of computers, compared to a few years ago.
 
Jeff, your statement above about the 10-core dual-Xeon system makes me think Resolve is able to scale across multiple CPUs and cores/threads better than other software? Is this true? If so, what is the science behind it? How does Resolve utilize multiple cores vs how other software utilizes multiple cores?
 
The unfortunate part of jumping to xeons is that last gen chips seem to rarely discount even after new dies release. Id wager to say E5 V2's in 8/10 core variety will always be around $1800-2000 per chip. Which means a machine is now 3-4x that of a desktop class i7. The ROI calculations become further interesting once you factor in specialty GPU's like Rocket handling the load and whether those are solid replacements for Cores. I think the gap is narrower with the introduction of 6+ core desktop cpu's so individuals really have to analyze their work loads and app utilization to find the best answer for their investment.
 
Jeff Kilgroe said:
I would recommend two or three Titan Black GPUs or the new GTX980 6GB cards..
Click to expand...

Jeff, do you know, if a 780ti would be faster, than the new 980?
At least on paper it should, cause the new cards have less cuda cores and less bandwidth.

thanks, Frank
 
Trying to answer everyone's questions here. Resolve does scale much better than Premiere in terms of both CPU cores and threads as well as GPUs. The rendering issue brought up by Eric above is also a concern and when it comes time to render out in terms of finish or delivery of your project, you may find that the over clocked i7 is still faster.

The unfortunate aspect to all of this is that there is no clear-cut answer as to which system is definitively better. Each has their own strengths and weaknesses and so much is dependent on the individual project and related workflow.

In my opinion, given the mostly stagnant improvements in CPU power over the past couple iterations, combined with the impending jump to 15nm Broadwell architecture next year for the desktop and early in 2016 for the Xeon CPUs, I personally don't have any plans to race out and buy any E5-v3 Xeon systems right now or spend tons of money on something that is little better than what we bought last year or the year before that. The biggest advancement with the current Haswell-E i7 and v3 Xeons is the shift over to DDR4 memory. This significantly improves memory throughput and overall system transaction performance. However, it's new and it's somewhat of a two steps forward, one step back situation. The memory is expensive and scarce and the largest modules readily available are 8GB in size. It's essentially impossible to have more than 64GB of RAM on the X99 desktop platform at the moment and the same goes for the v3 Xeons unless we're talking about multiprocessor motherboards with 16 or more RAM sockets.

So if you want my vote, I say go for the X99 build and stack it with 64GB of RAM clocked at 3GHz and an 8-core CPU clocked as high as you can go while keeping it stable. The CPUs seem to vary in temperament, but 4.3 to 4.5 GHz is achievable on most when properly cooled. Sure it's "only" 8 cores, but it's 12.5% faster for CPU clock vs how fast you'll be able to drive the best 8-core v3 Xeon under load and at 3GHz, the RAM is clocked 40% faster. That is pretty significant.

The one area where I hesitate to make this my final recommendation if this is to predominantly be a Resolve system is as far as RAM is concerned. 64GB just isn't enough in some situations. With 8 cores, that is only 4GB per thread, the bare minimum for performance apps in 64bit space, IMO. I have 256GB of 1866MHz RAM in my Resolve system. That is actually overkill, but 128GB on 20 cores was not enough. 64GB in an 8 core Mac TrashCan, and especially a 10/12 core, is often not enough in apps like AE, Nuke, Resolve, etc..

When it comes to GPUs, the 780ti is still a bit faster than the 980, but it also consumes more power and generates a lot more heat. There should be a 980ti model coming real soon. The 980 is the same core design as the 780, but optimized for greater efficiency and lower TDP. They threw in HDMI 2.0 support (sort of, it's still crippled to 10.2Gbps) and it's somewhat of a stop-gap measure until we get the next iteration of GPUs at 20nm sometime next year.

Titan Black or Quadro K6000 are still the king of the hill in their respective markets. In terms of CUDA performance, the 780ti is right at the Titan Black's heels and at significantly less cost. The 980 is just edging out a 780. So I would say the 980ti is the one to go for if it shows up soon. There are rumors of a similarly tweaked generation of the Titan coming, but that may be questionable when we consider the 20nm GPUs are probably about 7 months out at this point.
 
The 980GTX so far in testing has been almost or the same performance as the 780Ti card. I would not bother with the 780Ti cards unless you just get a really good deal and your on a budget. You can install 2x 980GTX cards with only 100W more than a single 780Ti card req.
 
In terms of utilizing available hardware, I'm on new SuperMicro system. Dual 10-core Xeon, dual Titan Black, Rocket-X with a dedicated RAID controller with about 1200Mb/s read and Resolve is rendering Red 5K/ArriRAW at between 90-120fps while Adobe Media Encoder is rendering just about real time even with the GPU acceleration turned on.
 
Paul, are those V3 10 Cores or V2? How does 6K perform?
 
They are 2x E5-2690 v2 and the Titan's are factory overclocked. I don't have much experience with 6K since I own all PL mount glass and most lenses vignette or fully porthole at 6K.

I find a dedicated RAID controller is a big help to getting the frame rates up and the Rocket-X takes a big load off the rest of the machine so everything just hums along.
 
The Rocket X is major factor with Davinci and frame rates. The Decoding load is offloaded completely allowing far more load on the CPU's for other processing. That means the performance should be similar on a single CPU system depending on the complexity of the projects.
 
Is there any way to make Red Rocket X help with other types of footage, like Phantom Flex 4K .cines?

Can RESOLVE make use of computational cards, like Tesla?
 
I would not recommend using the Tesla cards at all for media content applications. Many do not utilize them correctly. Your much better off with the Geforce cards both for performance and price.
 
You must log in or register to reply here.
Back
Top