Resolve vs PP Multicore support?

[Jeff Kilgroe]

+1 - no point in using Tesla cards, or Quadro cards for that matter, which are the same as Tesla cards with the additional video outputs. Resolve and most media apps which rely on CUDA for acceleration don't utilize additional features of the Tesla or Quadro cards, like the multi-point precision. And even in that example, the Titan GPUs offer FP64 ability like the Quadro/ Tesla. The GeForce / Titan cards perform better in most situations because they are clocked faster and have faster RAM while supporting 99% of the same instruction set and all of the functionality used by Resolve...

There are instances where a Quadro or Tesla card make sense. Running Resolve and Premiere are not such an instance. A SuperClocked Titan Black is still the best thing going for Resolve. On the latest systems, you can drive two of them at full capacity, even 3 in some cases or on dual CPU systems.

BMD recommends the GeForce/ Titan cards. They do keep Quadro cards on the list because many people think that's what they need or corporate purchase lists dictate so. They don't even bother recommending Tesla cards.


In Resolve you want to make sure all your GPUs are of the same type and configuration. Same number of cores and amount of RAM. Resolve will utilize each GPU within the specs of your lowest common denominator GPU. So if you have two Titan 6GB cards and you throw in a 3GB GTX 760 with half has many CUDA cores. Your Titans will only be used for half of their CUDA cores and RAM to match.

 

[Evin Grant]

Jeff, what is the current best bang/buck Mac trashcan setup, 8core/64GB/D700?
I'll be be doing mostly lightweight 4K editing in PP/FCPX and finishing in Resolve.

 

[Blair S. Paulsen]

Good question Evin. I find it hard to believe that the $1,500 price jump from 6 to 8 cores is worth it (especially with the clock speed drop from 3.5 to 3.0) unless your critical app is Resolve - in which case you probably want 12. The dual D700s are only $600 more than the D500s and by virtue of their greater VRAM (6+6 vs 3+3) alone, they would be the logical choice for dealing with 4K imagery. I would consider 64GBs of system RAM a minimum for what we do. I'm sure Jeff can breakdown the role(s) that RAM plays in system performance far better than I so I'll just leave it at that ;-).

My vote: 6core/64GB/D700s

Cheers - #19

 

[Eric Bowen]

Dual GPU's with more than 2GB of Vram and any 6 core CPU or higher, then you need 64GB of ram to really make use of the hardware with current GPU acceleration applications. If you limit ram then you limit the usage on the hardware right now. It's really that simple.

 

[Tom Lowe]

Eric, if money was no object at all, and you wanted to build the "Ultimate Resolve Machine" for 4K CC and editing, what would you build right now? Literally, no limit to what you can spend. That question is open to anyone.

 

[DJ Meyer]

Eric, if money was no object at all, and you wanted to build the "Ultimate Resolve Machine" for 4K CC and editing, what would you build right now? Literally, no limit to what you can spend. That question is open to anyone.

Z10PE-D8 WS
2x Xeon E5-2697v3 or perhaps E5-2687W v3
8x Crucial 32GB DDR4 CT2K32G4LFQ4213 (256 total)
Samsung XP941 512 GB m.2 for Windows
Use the 10 SATA ports for 7200RPM Enterprise Drives of choice in RAID
PCI-E Setup top to bottom:

Slot 1: K6000
Slot 2: ---
Slot 3: Rocket-X
Slot 4: Intel DC P3700 2 TB (RAID-0 with slot 6)
Slot 5: Either PCI-E Expansion daughter card (more GPU) or Raid Controller
Slot 6: Intel DC P3700 2 TB (RAID-0 with slot 4)
Slot 7: K6000

Should run you around 45-50k depending on drives and case and such, and whether you opt for the extra GPUs.

 

[Jeff Kilgroe]

Jeff, what is the current best bang/buck Mac trashcan setup, 8core/64GB/D700?
I'll be be doing mostly lightweight 4K editing in PP/FCPX and finishing in Resolve.

Very difficult question to answer. While the clock speed on the 8-core is slower than the 6-core and the 12-core slower than the 8-core... They all even out with their turbo speeds when running fewer cores. So there is an advantage to going with the 8 or 12 core system for the times your software can utilize those resources and, theoretically, when running apps that only use a couple cores, the turbo speeds can ramp up and still give you 3.8~4GHz of performance. Now in the real world, that just doesn't happen. The OS's and software do a piss-poor job of invoking the turbo. Under Windows, we can actually lock performance of the CPUs or crank up the turbo frequency to maximum available based on the number of threads. Unfortunately, it still doesn't stop our brain-dead OS's and software from launching threads across the CPU, therefore scaling down our performance. Stupid, huh? Yeah, it is...

The 6-core is a good place to start and have D700's w/ 64GB RAM. Attach good TB2 based storage like the Areca or Pegasus2 units and if you still have budget left over that you would like to spend on the system, an upgrade to 8 cores is going to help on occasion with Premiere and make a definite impact with Resolve. The 12-core system will show a noticeable improvement in Resolve over the 8-core, and at times you may see an advantage in Premiere as well.

I have two of the new Mac Pro systems. Originally I bought an 8-core and that's what sits on my desk at home connected to the Dell 24" 4K monitor. It's a nice combo. I bought a second nMP that I have on a studio/ office cart. For that one I bought the 4-core and upgraded it myself to a 10-core. Since the 10-core is 3GHz just like the 8-core with the same amount of cache, if memory serves, it performs identical to the 8-core system in everything I throw at it, with the exception of Resolve and occasionally if I load up a few layered R3Ds in Premiere, then the 10-core shows a bit more muscle. Looking back on it, I should have just bought the 12-core as it would have been the same price and less headache without voiding my AppleCare. But then again, I've seen plenty of instances with other peoples' 12-core systems where they don't perform as good as the 8 or 6 core on transcodes or renders that don't scale across all the CPU cores.

Eric, if money was no object at all, and you wanted to build the "Ultimate Resolve Machine" for 4K CC and editing, what would you build right now? Literally, no limit to what you can spend. That question is open to anyone.

DJ's config is pretty good to start with. I wouldn't use the K6000's though, they offer zero benefit within Resolve and the Titan Black cards actually perform noticeably faster in Resolve and Premiere 99% of the time. Even if money were no object, I would go for over clocked Titan Blacks over anything else. Two would be good, I would use 3 if I had the lanes for them. Would also consider some of the newer SuperMicro motherboards for this build over the ASUS. They tend to be more solid and offer a lot more configuration choices. I guess I'd have to look at their newest ones before making a decision.

Onboard RAID hosts on these Xeon motherboards are just fine, no reason to take up a PCIe slot to add an internal RAID. We do have to watch out for SAS and SATA intermixing issues if we add things like an LTO tape.

And with money not being an object, I would most definitely include a Rocket-X card if I were going to be working with R3Ds! Which also brings me to another recommendation for R3D workflow for those trying to decide between single or dual CPU systems. I think an over clocked 8-core i7 with a couple fast GPUs and RAM is probably the sweet spot for most things, then pair it with a Rocket-X as that will do far more to boost R3D operations than a second CPU will.

 

[DJ Meyer]

Would also consider some of the newer SuperMicro motherboards for this build over the ASUS. They tend to be more solid and offer a lot more configuration choices. I guess I'd have to look at their newest ones before making a decision.

Love me some SuperMicro but their 2011-3 board selection is pretty disappointing at this point. They also have a habit of putting the SATA ports at the bottom which means you have to choose between using them and being restricted to a single-width PCI-E card down there or not using them and putting another Titan in.

 

[Jason Osterday]

Regarding the nMP question, I've found a BTO 4 Core, 12GB Ram, D700 system swapped to 10 Core 2690V2 with 64GB Ram is a good ROI. If you look around you'll find the RAM and Xeon upgrade ~$2500-2700 bringing the machine to $6900-7400 depending on your PCIe SSD choice. Personally, I'd just op for the 1TB for $400 extra, especially since most shops offer an "Apple Care package discount" with that setup (which would be voided but at least you get a discount on the machine). There's another couple hundred $ worth of parts as well if you choose to recoup some coin on ebay with the leftover RAM and Xeon.

This is a great choice for those wanting to save $1500-2000 vs a 12 Core straight from Apple with 5-10% similar performance and also staying within the OSX ecosystem.

1 Year later after the nMP launch, however, we're sort of paying a premium for OSX and an ultra compact Thunderbolt machine. The 5690X Windows machine I have spec'd with dual GTX980's, 64GB DDR4, Asus WS board, Dual Samsung XP941's and a nice 8 Drive case is under $5000 and at 4+ Ghz, every aspect of that machine is faster than the Mac. However, I can't fit it in my bookbag with my Macbook Pro like I can with the shiny trashcan.

 

[Eric Bowen]

I would honestly go with Jeff's logic here since the market is in major transition with new hardware releasing for the next 2 quarters that will greatly impact performance. The 5960X is a monster cpu at this point for all applications and workflows up to 6K media even. The ability to clock the DDR4 to 2666 and higher is huge. This shows even when the CPU load is 100%. The playback is still seamless. In audio testing as well we were able to push the CPU to 100% and the audio at .75ms latency still wouldnt start breaking up like it normally does until even further load was added. The bandwidth higher frequency DDR4 is offering is showing some significant performance with latency of processing and GPU acceleration. So right now I would ay a 5960X at 4.4or 4.5Ghz is gold money right now. It covers the best of both worlds with codecs and fx that are more limited threading and applications such as AE and Davinci which will often thread out as many threads as you can provide. The ram preview in AE on this chip is just ridiculous coupled with the ram. 64GB of ram though somewhat limiting with some applications is still very viable in almost any workflow out. I would then take the money saved on the system and get a Rocket X card. The load those take off the CPU is massive which allows far more headroom for processing FX along with managing buffers for GPU acceleration. All of which has a huge impact in Davinci with 4K+ red media and above. The clock speed for CPU and ram is very important to the overall performance of this along with the GPU's. I would recommend the 980GTX at this point and maybe only 1 until the coming higher end versions release probably in the next couple of months. Once again the transition in hardware that is going on means getting what works well now and getting the real good option as soon as that becomes available soon. Once the higher end 900 series cards release then I would look at 2 of them.

This however is where caveats come into play. The Rocket X card along with 2x GPU's are going to eat up most of your slots because of 2 spaces per card. Now you have to decide what all you need for storage and if you need the Blackmagic I/O card which you will most likely want. That means your now looking at an expansion chassis with atleast 3 slots. There are good options available but at some point you are going to push the entire throughput capability of the single socket boards. If you only need the I/O card and plan to use SSD's in raid 0 on the internal controllers then you should be fine with the single socket boards. However if you plan to add a SAS raid storage unit, especially 16 bay SAS unit because of DPX file workflow then that is where the board may not have enough total throughput. At that point you really need a Dual Xeon even though the clock speeds are slower for both CPU and ram. The Dual Xeons boards have far greater total throughput for the entire board which means full performance for all buses and devices on the board. At that point I would suggest the 2690V3 chips over the 2697V3 chips simply due to the extra clock speed. At some point with more cores you need higher clock speed for those cores with GPU acceleration applications. Processing latency is really decided by that and ram clock speed on those applications. The extra cores of the 14 Core wont help as much once you reach that many cores with that hardware setup. Then it's about how long it takes data to get from point a to point b and back again.

One final caveat is the clock speed of the 5960X versus a single Xeon. If you dont plan to clock the 5960X up from stock then get the Xeon 2690V3 chip on the X99 board. The advantage of the i7 Chip is clock speed for both CPU cores and ram. If your not going to clock the cpu cores up then get the Xeon 12 Core and go with the DDR4 2133 ram it's forced to use. That will handle a greater realtime load than the 8 Core but render times will be longer including ram preview if you use AE. The Dual Xeon would be the processing choice if the greatest realtime load capability for very complex projects are the absolute key to the new system with Davinci. If that is the case then the Dual Xeon 2690 V3's are the way to go anyway. Just make sure you get 128GB of ram as well. Keep in mind latency will be a little higher on that system for response.

Summarize Eric choice system
Coolermaster or Antec 1300W PSU (forgot to add this so a idea of power req's were there. Antec has the fully modular PSU which is nice. Coolermaster are bulletproof and very quiet.)
5960X @ 4.5GHz
64GB of 2666 ram (beyond that your playing with fire and better know what your doing like Jeff)
Systemboard - Asus board and Deluxe if you want the most stable with the current ram right now. (Amendment - Asrock is releasing the X99 Extreme 11 which finally adds Reg Dimm support to the workstation boards. That board would allow 128GB of ram with the V3 Xeons. If that is important then Asrock is the way to go for X99 right now. Hopefully Asus will break the mold and do the same with the WS board.)
Noctua U12S CPU cooler and Noctua fans. (This new Noctua CPU cooler designed for all profile ram is really surprisingly good so far in efficiency. Warning though is Noctua coolers can have some sporadic productions where heat pipe efficiency tanks badly. Dont be afraid to return yours if it shows poor cooling. Noctua fans though are top notch and always recommended.)
Samsung 850 Pro SSD drive - OS
3 Samsung 850 Pro 1TB drives in Raid 0 on onboard (8x 1TB Samsung 850 Pro 1Tb drives on SAS storage externally or 16 Bay SAS storage with enterprise drives as long form project volumes - must use expansion chassis if 2x GPU's)
1x or 2x Geforce 980GTx card (Zotac, EVGA, or Asus)
Rocket X card
Netstore [SIZE=-1][SIZE=-1]3 Slot PCI Express 2.0 and GPU Expansion Enclosure via PCIe Host Adapter - For Desktops if expansion chassis is required.(The PCI-E 3.0 NA255A is out now or releasing which would be a better option)
Decklink 4K Extreme 12G
Intel 10Gbe X540-T1 Network adapter.
That would be the core of the system. Let me know if you have any questions on the hardware and why?
[/SIZE][/SIZE]
Do not get the Quadro K6000 cards at this point. They are far over priced for their effect on any GPU acceleration application and basically EOL at this point. Save your money. If you have to get a Quadro even though there is no reason with Davinci then wait for the new Quadro cards based on the 900 series GPU's. Dont waste your money on them now. There are no applications in the currently commonly used workflow that show better performance with the Quadro cards over the Geforce cards. Dont let marketing get you.

 

[Jason Osterday]

What do you guys think of the Asus X99E WS vs The Deluxe? It gives the option of Quad x16 slots which would allow either Single GPU/Rocket/BMD/Raid/10GbE or Dual GPU/Rocket/Expansion ---> BMD/Raid/10GbE

 

[Eric Bowen]

The WS is a nice board and the one to get depending on card requirement with expansion chassis. Just keep in mind ram clocking will be more limited on that board. CPU clocking would be the same unless profiles have changed.

Supermicro boards at times have good options but their warranty is really poor comparatively to Asus and others. Last I dealt with them the warranty clock started when the board was sold to distribution. Be very careful if you are looking at Supermicro. I can asure you their failure rate is no less than Asus Z9 or Z10 series boards. We used the Z9 series boards the entire production and the failure rate was far lower than Supermicro's was and even lower than the Asus desktop boards.

 

[Eric Bowen]

I love the pace of the current hardware market. Asrock is releasing the X99 Extreme 11 with Reg Dimm support. The Reg Dimm support with Xeons is a serious break from systemboard manufactures standard policy. Normally they reserve that for the server boards only forcing people to those more expensive boards with less features just for the Reg dimm support. Reg dimms mean 16GB dimms available on X99 which is 128GB of ram. That is huge. I really hope Asus will do the same for atleast the WS series board but I wont hold my breath. BTW keep in mind SAS controllers are crippled when used onboard. The cache that comes with the controller card is an important aspect to the raid performance over the entire volume. Don't rely on that for high performance parity raids.