Video Editing / Transcoding $$$

[dac7nco]

What would be faster @ H.264: Two Xeon X5650 or two Opteron 6168? I'm into building a transcoding rig and I've never has an AMD... would 24 1.9GHz cores beat 12(+HT) 2.66GHz ones? I don't want the EVGA SR-2 board as I want a ton of ECC memory. I'll probably use a Quadro card for filters and such (CUDA). Are there any 24-thread benchmarks aside from DB/Server stuff? - This will be my main workstation.

Thanks

 

[Markfw]

Well, I don't have any real numbers to prove it, but I think the AMD would win by a landslide.

 

[dac7nco]

What I'm looking for is handbrake/TMPGenc benchmarks on the Magny-Cours vs. X5600-series... Anandtech did a good one, but it was DB stuff. I'm looking at the 24-core Opteron as a KILLER workstation.

 

[zagood]

This is so far beyond my pay grade. But my google-fu is strong tonight.

x5650 vs. Optie 6174
http://www.bit-tech.net/hardware/cpus/2010/03/31/amd-opteron-6174-vs-intel-xeon-x5650-review/1

 

[dac7nco]

Thanks zagood,

I think I have a reason for 1366 - The cinebench had me rooting for AMD, but the intel just killed... Supermicro has some great customer support, and I HATE HATE HATE Tyan.
I Thing I'll get the MBD-X8DAH+-F-O, and a Lian-Li PC-80 Case. 72GB ECC Winte4k can't hurt.

 

[LR6]

You might want to wait until Sandy Bride arrives. It is said that it will have a section of the processor dedicated to video transcoding. Keep in mind that if you do wait you wait you will likely have to wait again to get software support.

 

[craftech]

I am not sure what you are looking to do? I am a video editor who uses Sony Vegas for video editing and Procoder 3 for transcoding and Sorenson Squeeze for QT web video encoding. I all three cases the use of multiple cores is software dependent, not hardware dependent. The most any of them can utilize is four cores. In terms of memory usage, a 64-bit OS can utilize more memory than the 2GB per app. limit of 32-bit systems, but again that too is software dependent. One can use an editor to change the 2 GB limit using free software. That will improve render times. But what you are contemplating will have little return for your investment.

John

 

[Idontcare]

dac7nco, I use TMPGEnc on my Q6600 and with CUDA through my GTX460.

Sandy Bridge has been shown (leaked) to have a pretty healthy improvement in TMPGEnc specifically.

Do you use TMPGEnc already? It is multithreaded but the thread scaling is not perfect. Even on my quad at stock clocks my CPU utilization is not running 100% (it varies between 80% and 95%) and my GPU utilization doesn't get much over 10%.

This is on intensive 4hr transcodes where the disk subsystem isn't really limiting the computations. (I'm in Win7 x64, 8GB ram, and disk storage is behind an Areca 1280ML w/2GB cache)

Based on the experiences I have had with TMPGEnc, if I had to choose between fewer faster cores versus many slower cores I would go the fewer faster core route and I'd wait for Sandy Bridge since it has already been shown to provide another ~20% IPC boost over Intels current chips.

 

[soccerballtux]

dac7nco, I use TMPGEnc on my Q6600 and with CUDA through my GTX460.

Sandy Bridge has been shown (leaked) to have a pretty healthy improvement in TMPGEnc specifically.

Do you use TMPGEnc already? It is multithreaded but the thread scaling is not perfect. Even on my quad at stock clocks my CPU utilization is not running 100% (it varies between 80% and 95%) and my GPU utilization doesn't get much over 10%.

This is on intensive 4hr transcodes where the disk subsystem isn't really limiting the computations. (I'm in Win7 x64, 8GB ram, and disk storage is behind an Areca 1280ML w/2GB cache)

Based on the experiences I have had with TMPGEnc, if I had to choose between fewer faster cores versus many slower cores I would go the fewer faster core route and I'd wait for Sandy Bridge since it has already been shown to provide another ~20% IPC boost over Intels current chips.

20%? Bulldozer's not going to help amd

OP I would wait personally.

 

[T2k]

Get the Xeons with enough memory, they have a lot higher clockspeed, that pretty much settles it, regardless of more cores on Opterons - they are a lot slower at 1.9GHz, aimed at virtualization hosts etc where infinite amount of cores can be utilized simultaneously.

Now if you could afford the 2.3Ghz 12-cores Opterons that would make it more interesting...

Did you look into GPU-based transcoding BTW?

 

[dac7nco]

I/O