I want to still utilize a Z800 system I have if I can, but before I go through some of the GPU upgrades that would need to happen, I was curious if anyone was using a Z800 Dual Quad Core 3.2GHz machine (or near to). I’d love to know how it’s performing and what kind of transcoding speeds you’re seeing - just to help make a decision.
I think many people continue to use Z800s in production, and we do plenty of testing on Z800s. Here are some benchmarks results from v1.4 on different hardware:
Cortex06 and Dev04 are both Z800s. The other machines are based on i7 CPUs. Depending on the format, CPU, GPU and Storage can all be factors in performance.
Interesting. So it appears the big factors are GPU type and base CPU speed. Number of cores doesn’t seem to affect performance as much (except maybe for Red Epic - but I can’t tell if it’s the 12 cores or the titan card giving high numbers).
Our own informal tests reflect the same: 4-6 cores (with Hyperthreading) seem to be sufficient. Too many cores (above 8) doesn’t help performance much. If buying a new PC, you might as well go with less cores, then take the $$ and get a better GPU. But that’s my opinion. Your mileage my vary.
@runningman - what CPU model do you have? Probably a Xeon W5580? So your speeds would be slightly lower than the red bar above, I guess? Throw down a decent GPU and you should be able to use the Z800 for a while longer. We have GTX670 cards on several machines and get pretty good speeds (some of our computers as slow as 2x 2.4Ghz). As you can see on the graph, you could get by with a GTX580. For a good compromise, Newegg has an EVGA GTX660 with 3GB RAM for just over $200. But as @JDV recommends in another post, the EVGA GTX780ti with 3GB RAM will give you the best bang for the buck (Newegg has it for $720, Amazon for $770).
But again - your mileage may vary.
For Red, we’ve generally seen that even with the GPU SDK, the CPU is currently the bottleneck, so I think in this case its the CPU.
We do have the Xeon W5580 in the Z800. I’ve been delayed responding while waiting for the GTX card to arrive and here’s what we’ve found. We currently have been using a Dell T7600 Dual 8-Core ES-2687W 3.10GHz with the K5000 card and when processing 2k Alexa material we’re able to achieve an average of 80fps. With the newly configured Z800 (Dual 4-Core 3.2GHz) with the GTX 780Ti card we hit an average of 33fps with the same clips. I don’t know if there’s any settings that need to be tweaked on the GTX card, but I was hoping for a little more given that I believe Alexa processing utilizes the GPU (we have a job coming in shooting Alexa that averages 5-6hrs/day). While I won’t be building another Dell for this, it would seem from the graphs that even a i7-4 core would likely be a decent option. Appreciate the help.
Well, to point out the obvious - the Z800 has a 5 year old chip while your Dell has a 2 year old chip. Also, the Dell has twice as many cores as the Z800. Plus, the newer chip has a higher turbo speed than the older.
But I would also ask about RAM in the Z800 and disk speed of the Alexa footage. (RAID? single drive? SAN?) And what kind of material are you creating? (DNxHD, MPEG, h.264). And what kind of disk is the transcoding destination? (NAS, SAN, RAID, same disk as Alexa footage?)
But having said that, 33fps seems a bit slow for 2K material, especially with a 780Ti GPU.
Yeah, curious about the storage in these tests as well - sounds like the likely culprit.
And just to be sure, we’re talking about Alexa ProRes here?
We are talking Alexa ProRes444 processed to DNx36. In both cases I set the storage to a second internal hard drive on both systems to try to best level the test as the Z800 isn’t in a position to be attached to the SAN.
I am still strongly suspecting the storage. Are you reading from / writing to the same internal hard drive?
No - I’m using the same USB3 drive to read from, for easy portability, and on both systems using an internal SATA drive to write to. Both internal drives are 2TB, but not likely the same.
Did you install a USB3 card in the Z800? It almost sounds like its reading at USB2 speeds.
That’s what I thought too - it’s a 4 port USB3 card that claims USB3 per port, and not a single shared across. I copied the material to the second 2TB SATA drive and tried processing from the one internal to the second internal. A slight improvement of 35fps, but not substantial.
Try a few different tests with those two internal drives, switching up which one is reading and writing. (1 -> 2, 2-> 1, 1->1, 2->2). If you keep getting the same results, maybe we can take a closer look with you sometime tomorrow.
If you copy the footage directly from the USB3 drive to the internal 2TB (no transcoding), what kind of speeds do you get? In the copy dialog, you should be able to expand the dialog and see a transfer speed.
Also, do you have the latest USB3 drivers for the 4-port card?
Latest drivers are installed - though there were some initial issues getting it to work. When copying the test material from the USB3 drive to the internal it was in the 90-95MB/s area - using Windows copy>paste. The USB3 drive is a single drive in the inclosure running 7200rpm.
Did some more tests with @runningman are some results:
So this still feels storage related to me, I could be wrong and the issue may be more subtle.
Approximate encoding speeds on your his current Z800 config from ProRes444 to:
- prores proxy: 95 fps
- prores 422hq: 72 fps
- prores proxy 720p: 120 fps
- h264: 58 fps
- dnxhd36: 46 fps (better than the 33 fps reported earlier!)
On a Z800 here at MTI with similar specs (3.33 Ghz vs. 3.2 GHz), I’m getting 72 fps from ProRes4444 to DNxHD36.
I still suspect the difference has something to do with the way the dnxhd codec handles I/O since that is the thing that is the “most different” between our two systems, but I’m not sure. Maybe it is the slightly slower CPU that is really at fault.
Probably can’t tell for sure without hooking it up to some other storage… which @runningman will be trying shortly.
Could it be up to the 150fps?
Transcoding the 2.8K arriraw to the ProRes 422HQ–150fps
You would probably need more than one workstation to get aggregate speeds of 150 fps those formats.