Welcome to our community

Be a part of something great, join today!

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 for Windows - GTX vs Quadro

mcKayj

mcKayj

Active member
Joined
Oct 29, 2007
Messages
30
Reaction score
0
Points
0
Location
New Orleans,LA
Anybody care to weigh in on the pros and cons of using a high-powered consumer GTX instead of a Quadro for Resolve? More specifically, how would a GTX590 stack up against a Quadro 6000?
 
I was just looking and doing some research how to build a system for it. I know in the config docs, they propose supermicro chassis and mb. (I'm a happy supermicro user, for my current WS),

just thinking whether going with a simplified asus 2011 mb, then you have 6x pci-e x16, so maby
quadro 6000
gtx580
gtx 580
gtx 580

redrocket
decklink card,



eventually with cubix or pci-e cabel extensions,


Would like to know , whether it's better to spent the money on rockets , cpu-power (eventually with a 4way supermicro which can hold 4x xeon e7 decacores), or even 8way which can hold 8x xeon e7 decas????

the cubix gpu-xpander is not too pricey, so I assume it's better taking one as well...
 
Being able to build such a system, I ask myself , whether how to compare Assimilate Scratch and Nucoda filmmaster to it power- and speedwise????

also would like to know whether it's possible to do proper and fast s3d grading with resolve and windows machine?
 
Sven Seynaeve said:
Being able to build such a system, I ask myself , whether how to compare Assimilate Scratch and Nucoda filmmaster to it power- and speedwise????
Click to expand...
Only Resolve supports multiple GPUs, so comparing it to Scratch and FilmMaster is impractical. But because of a genius caching scheme on Nucoda products, multiple GPU support is not that important nor is it really necessary.
 
mcKayj said:
Anybody care to weigh in on the pros and cons of using a high-powered consumer GTX instead of a Quadro for Resolve? More specifically, how would a GTX590 stack up against a Quadro 6000?
Click to expand...

If you do a little search on this site, you'd see, that GTX 480 blows Quadro 6000 away. Figure GTX 590 at least twice as fast as GTX 480. Your mileage may vary.
 
thx for the hint with nucoda how it works,

would like to have more practical information how al these systems work compared to eachother, but it seems hard to find such info, rather then contacting the companies themselve and have a commercial talk that always explains their product is better
 
Look for Mike Most's blog, it's called Postworld. He did a couple of good posts covering the difference between all the main players.
 
QUadro and GTX cards are very close cousins under their fan shrouds. The only measurable differences are in Double Precision Float performance and overall RAM amounts. The Quadro cards use same GPU, it is just allowed to have higher DPF abilities somehow. You can see this number in the CUDA-Z performance window. I have never found an OSX app where this truly makes a big difference. I would be thrilled if someone from BMD came and said it did.

In the informal testing that I did with Jake Blackstone, we were unable to find Quadro cards any faster in Resolve using NR. The Quadro 5000 & 6000 use GF100 cores like the GTX470/480 cards and they are both specced below the GTX cards. The Quadro 6000 has 6 Gigs of VRAM, but CUDA-Z only sees 4 GB and OSX & GLView only see 2GB. The OS just isn't ready for 6GB. OpenCl could see all 6GB but that was only place.

Sascha Haber did some testing with Resolve which seemed to equate Resolve performance with the Single Precision Float number associated with these cards. A Quadro 4000 is around 700-800 in SPF, while a 5000 or 6000 do right around 1050-1100. A GTX470 does just slightly better. A GTX480 always scores above 1300 here. I will chuck a GTX285 in and see what it does here if anyone wants that number. I seem to recall that is is around the Q4000's score here, so perhaps this isn't the whole picture.
In any case, the GTX480 left even the $2,000+ Quadro 6000 for DEAD. (Jake posted test results at The Cow) Unless Resolve for Windows uses DPF, I can't imagine that the situation there will be any different.

If anyone else in Hollywood wants to do some more extensive testing, I can offer up a Quadro 6000, Tesla C2050, GTX470, GTX480 and a GTX285 (in 1 and 2 GB flavors). These cards work in OSX and Windows. I will also soon have an OSX ROM on a GTX460 2 Win, which has 2 @ GTX460s on one board and could in theory be both a GPU and a GUI card. Sadly, it requires far more power than a Mac Pro could provide, but could be used with external power or in an enclosure.

The simple answer is that GTX cards offer far more "Bang for the Buck", at least in my limited experience.
 
Dave P said:
I will chuck a GTX285 in and see what it does here if anyone wants that number.
Click to expand...

Yes, please, I could still get one or two for our machine.
 
Looking at the official configuration guide, this is the order of performance, according to Blackmagic:

Quadro 6000 > Quadro 5000 > GeForce GTX 580 > Quadro 4000

However, it does not mention if that is a GeForce GTX 580 1.5 GB or 3 GB. Technically, the 1.5GB version is the standard reference design, so it will be interesting to see where the 3GB variant ends up on that list. I suspect Resolve could tap into double precision and/or other functions accelerated by Quadro drivers.
 
I'm probably missing something here, but I was living under the impression that the Windows version works only with Quadro cards. In other words, it won't work with GeForce "gaming" cards at all. I hope I'm mistaken.
 
As I understand it, Quadro has advantages in vector graphics applications like CAD and 3d modelling due to drivers that are optimized for that purpose. For raster based applications, the consumer cards do fine.
 
Petri Teittinen said:
I'm probably missing something here, but I was living under the impression that the Windows version works only with Quadro cards. In other words, it won't work with GeForce "gaming" cards at all. I hope I'm mistaken.
Click to expand...

I'm running Windows Lite-version with single GTX 285, works fine..
 
Just for kicks, I tried running Resolve Lite with an ancient GPU out of the history vaults - 8800 GTX - for both GUI and CUDA. I bought this purely for gaming purposes - GPGPU acceleration was still a nascent idea. I was pleasantly surprised by how well it works - 1080p is still real-time for primary correction, and only starts dropping a couple of frames after a couple of nodes with key/tracking. After a while it ran out of memory - 768 MB seemed overkill in 2006! But the frame rate never dropped below 15 fps. Performance with AMD Radeon HD 6970 is poor in comparison, however. A shame as I have shifted to AMD GPUs on my primary system. HD 7970 releases next month with an astonishing 4 Tflops on tap. Quadro 6000 can only muster 1 Tflop in comparison. AMD also move to a compute oriented GCN architecture. I know there is some OpenCL processing happening on AMD GPUs, but I hope that is extended to match CUDA. There's also OpenACC - not sure where that is headed though. DirectCompute wouldn't be appealing for a cross-platform software.
 
I have a single GTX580 card with 3Gb vram, never seen anything drop in speed except when my harddrive bandwidth and speed isn't enough. Other then that, uncompressed 1080p works like a charm.
 
Christoffer Glans said:
I have a single GTX580 card with 3Gb vram, never seen anything drop in speed except when my harddrive bandwidth and speed isn't enough. Other then that, uncompressed 1080p works like a charm.
Click to expand...

Do you mean, that both GUI and GPU functions on your rig are running off single 580 card at real time speed?
 
You must log in or register to reply here.
Back
Top