THE beast 32 core HACKINTOSH is ready.....CS6 vs REDROCKET

[Jeff Kilgroe]

Please Jeff or Joseph, can you test resolve performance under windows?, how many fps can you get from r3d 5k full without redrocket?

Will be doing just that probably this weekend.

Because I don't understand how you can function with NO SLEEP. Amazing :)

Sleep? I'm afraid that if I get too much sleep I might miss something... Or not wake up again. haha

Yeah, I signed up on the EVGA site for the GTX690 notification email. Got the email and instantly clicked on the link... sold out. Apparently all gone within the first couple minutes before all the emails had even gone out. Re-signed up and same thing happened to me again. Put in an order elsewhere and I'm about number 18 on the list and probably won't' be in their first shipment, possibly not in the second either. They're guessing at 2 to 3 weeks. :( But whatever, I can wait a bit.

[Jeremy Ball]

So are you getting realtime or faster than realtime transcodes at full debayer using REDcine X with the system you have laid out here? I would much rather spend a good chunk of change on 1 beast of a computer that I can use for everything than just a RED Rocket I can only use with RED footage.

I am curious as well about the performance of RED footage in Resolve for Windows. Can not wait to hear the results.

[DJ Meyer]

Using a 512GB OCZ Vertex 4 SSD for system drive. 64GB Micron 1600MHz ECC RAM via 8x8GB CAS 10. Quadro 6000 video card....

Cinebench Render score is.... 31.09

Wow. The highest I've seen on the web for the dual E5-2687w even at BCLK 105 is 27.xx Curious to hear how you got it that much faster.

[Jeff Kilgroe]

Wow. The highest I've seen on the web for the dual E5-2687w even at BCLK 105 is 27.xx Curious to hear how you got it that much faster.

Fiddling with the turbo and cache. ;)

Actually, I've come to the conclusion that Cinebench is totally worthless as a benchmark on any of these new systems... It's a very poor representation of how the system will perform compared to older hardware as the benchmark does nothing to actually stress the new hardware and make it "kick into high gear". The OpenGL benchmark in Cinebench has always been worthless. Ramping up the Cinebench score isn't a cheat as I'm not interfering with the software in any way, but you don't get decent numbers unless you can force the hardware to ramp up and run at maximum, firing on all cylinders at full speed.

If I just go into Cinebench and run the test, I score just under a 26... If I repeatedly run the test a few times to get the CPUs worked up a bit it goes into the 27.xx range as you point out, but watching a CPU meter and other monitors shows that a few cores don't even activate on each render cycle and it's rare that the CPU is ramped up to full speed and never for the, very short, duration of the render. In other words, the benchmark is pointless. It's not even a good representation of how the system will perform in Maxon's own rendering applications as more complicated renders will push the hardware harder and for a longer time, in turn forcing the CPUs to ramp up and sustain the load.

Performance on my Z820 and the SuperMicro build is identical on all tests I've run, or at least so close it doesn't matter. I have CAS-10 RAM in the SuperMicro and HP ships CAS-11, but it doesn't seem to make a difference in any memory benchmarks I've run.

[Jeff Kilgroe]

So are you getting realtime or faster than realtime transcodes at full debayer using REDcine X with the system you have laid out here? I would much rather spend a good chunk of change on 1 beast of a computer that I can use for everything than just a RED Rocket I can only use with RED footage.

I am curious as well about the performance of RED footage in Resolve for Windows. Can not wait to hear the results.

RC-X doesn't scale well to this many threads. These systems will give much better transcode performance in Adobe CS6. There are still performance advantages to having the Rocket in the system, but in a proper multithreaded application, the system doesn't choke without the Rocket. Realtime transcodes are still elusive with some codecs and settings as, with or without the Rocket, the encoders themselves are often not multithreaded or are poorly implemented in that regard. So certain codecs (I'm looking at you, ProRes) still lock you to one or two CPUs/threads when encoding and that can become the bottleneck. I'll try to post some real-world numbers here this coming week.

I still haven't put Resolve for Windows through it's paces, but performance seems great.

[Fabian Benninghoven]

Question about my two Nvidia GFX cards... how do i install them right for the best performance?

- the GTX285 is inserted in 1. 16x slot as GUI card
- Quadro FX 3800 as Cuda helper card in the other x16 slot

FX 3800 cant handle 2x DELL 2711 with 2560x1440px, only one, because the card has only 1x DVI out! :(

Is this ok? Do i have to install anything or use other settings?

[Jeremiah P]

Has anyone really tested GTX 680s or 690s in Resolve yet? I am building a similar new workstation and need to finalize the video card decision. I hear the 580 3GBs still reign supreme in Cuda and Resolve processing. I was planning on running 3 GTX580 3GB. But I would love to run 3 GTX 680 4GBs or 2 of the new upcoming ASUS Mars III with 2 680s and 8GBs on a single double wide card. Rumored to hammer the 690. Either way I would love to buy current cards instead of dead cards, but everyone says Kepler is nowhere near as good as Fermi for CUDA and GPU computation.

[Joseph S.]

Has anyone really tested GTX 680s or 690s in Resolve yet? I am building a similar new workstation and need to finalize the video card decision. I hear the 580 3GBs still rain supreme in Cuda and Resolve processing. I was planning on running 3 GTX580 3GB. But I would love to run 3 GTX 680 4GBs or 2 of the new upcoming ASUS Mars III with 2 680s and 8GBs on a single double wide card. Rumored to hammer the 690. Either way I would love to buy current cards instead of dead cards, but everyone says Kepler is nowhere near as good as Fermi for CUDA and GPU computation.

I have a pair of 690's and resolve installed, but I have no way of knowing how fast resolve runs. Not sure if there is a revolve benchmark method.

Joe

[Jeremiah P]

Wow two 690's huh!? I'm trying to track down a test scene that was used on Creative Cow for comparing several cards. I will let you know. Though Luigi has done a lot of Resolve testing and may have a better method.

[Jeremiah P]

So apparently there is a project called "Candle Test" and the video clips can be downloaded form BlackMagics site. Just reconnect the clips and you should be golden. Here's a link to some of the results on Creative Cow http://forums.creativecow.net/thread/277/8355#18191. Thanks for taking the time to run some benchmarks! You would possibly be the first to do it with the GTX 690s!