Dual-Core CPU for HD/HDV NLE

[David Newman]

It was a nice gamers motherboard with a Pentium D Extreme -- I don't know the vendor details.

[Sean Livingstone]

So is Opteron widely expected like xeon or are there some problems with amd?

[David Newman]

Opterons work great, there are no problems.

[Régine Weinberg]

That is a mad discussion
in the Unix world and Linux world we talk clustering
can be made with Mac's too
Open Mosix, Beowolf in Linux or SGI on Irix are doing this since
decades, it is used in Holly/Bolly Wood for renderfarms,.
You don't need to switch your mobo all the time, the clients
can be diskless. Commen sense working with Maya

[Anhar Miah]

I know this is probably a long way away, i've already posted this in the industry news section.

But Sony is planning on using the Cell techonology (PS3) in workstations, they have claimed performance to be something like 16 Teraflops, just imagine with that kind of power

(this is pure speculation and not based on any calculations, if anyone wants to do some calculations, please feel free)

1 hour of SD video endocoded to DVD mpeg in seconds (faster than the harddrive can write)

1 hour of HD/ HDV encodeded to DVD mpeg or even transcoded in mere minut es.


Ok well thats all specualtion, one thing is for sure all your CGI 3d animations rendering should love the extra horsepower (16Tb) it would be like having your own render farm!

[Régine Weinberg]

Hi
one thing is for shure ! For the PS3 there will be a disk with a special Linux.
Pthing the Kernel to have Mosix is easy. The PS3 has Gige Ethernet, so 2 or three are a cluster and way faster as any AMD dual core ..........
We will see...Shrek was rendered with Xboxes this way about a hundret,
way way cheaper as an Intel, Amd, Sun or SGI renderfarm!!

[Kevin Shaw]

David: I talked to someone who said the Pentium D Extreme Edition in particular has some advantages which other current Pentium D processors don't have, so I'd be interested to hear what sort of performance you get if you happen to test "standard" Pentium D processors. What I'd really like to know is how a stock Pentium D setup running at, say, 3.0 GHz, compares to a dual Xeon system running at the same clock rate. If they're even just similar in performance and the Pentium D setup is cheaper that's a big plus in my book. But in terms of maximum performance, I'm not sure whether a Pentium D at, say, 3.2 GHz is going to match the performance of two 3.6 GHz Xeons with 2MB cache each on a good 800 FSB motherboard. Decisions, decisions, decisions...

[David Newman]

I do only have access the extreme version of the Pentium D, yet it is so much faster than the dual Xeon systems I've tested it is hard to imagine the standard Pentium D would be slower than dual Xeon. That is a complete guess. My guess continued : unless Xeon has had a more recent chip set upgrade, the Pentium D will be faster (at the same clock speed) as it is running on the most recent chip set (give it the higher memory throughput.) One data point -- the only major difference between a 3.2GHz Pentium D 840 and 3.2GHz Pentium D EE is the support for hyperthreading -- we find hyperthread increase HD style processing speeds up about 10-15%.

[Sean Livingstone]

The Major advantage the Pentium EE 840 has over the Pentium D is that it supports Hyper-Threading. So really the processor is a quad-way CPU as it has two cores and each core runs Hyper-Threading meaning it is able to run 4 threads at once. Saying that you are only going to see the proformace gains when running applications that support 4 processors and use's it's multithreading well.

I found this http://www.digitalvideoediting.com/a...fterinter=true and they compare the Intel Pentium EE 840 to a dual Xeon 3.6Ghz and the Pentium EE was out proforming the Xeon but only when running windows 64bit.

Well just have to wait and see what the future holds.

[David Newman]

Sean,

That is good info for people like Kevin wishing to make a purchase. My own tests are based on CineForm benchmarking tools that work out the maximum encoding and decoding speeds -- the key element to multi-stream HD editing. This is not a very standard benchmark, but it really helps me recommend system for CineForm products. It was though these experiments, that we mostly recommend AMD dual Opteron (for our work -- so much faster than Xeon -- up to 50% faster.) Yet the Pentium D EE has made huge performance gains, I was completely surprised. For Prospect HD we still say Opteron, but for a killer Aspect HD system I'm thinking Pentium D. I must confess that I haven't tried a dual core Althon yet -- so how knows. :)

[Randy Donato]

David, I am glad you pointed out the benchmarks you are using are based on using your products which I have used on dual 3.6 xeons. Frankly I was surprised that 4 layers stuttered and dropped frames using Aspect in accelerated preview mode and 1.51. Using Edius and Canopus HQ of the same footage(only as you know Edius is playing back full frame full resolution and does not use the reduced res preview to gain playback) I was able to easily handle the steady playback of 4 layers. I really was surprised and so you will know I was using the same clips captured both in Aspect and Canopus HQ and used the same defaults for each. Also the drive is plenty fast enough as it is a 4 drive sata raid. I think I have seen you say before that single cpu's are more efficient with aspect and I may give the same test a go on a P4 3.2 machine. The only other difference is the overhead 1.51 uses versus the Canopus pip which I am sure Ppro's is much higher....but then again we are comparing full res playback to accelerated preview so that should make up the difference.. Since you are big on Opterons is the software written in a manner that favors that architecture over Intel's when it comes to multiple processors?

It doesn't bother me since I rarely uses that many pips but I do wonder how the 4 layer claim can be supported at least on dual xeons...the playback was so jerky it couldn't be used for much. BTW with AE and doing some things with mattes I am using Cineform codec HDV and get to see it on both a Sony crt and a 1920x1080 LCD in full res though the Nx video output module for AE and it looks great.

[David Newman]

Randy, basically we need system with a good memory bandwidth, and Xeon's do not have it. We have tests that show P4s outperforming dual Xeons, it is all to do with memory I/O. Opterons have excellent memory I/O, enabling us to process even higher resolutions with Prospect 2K. So it is nothing to do with single proc vs dual proc. All our products are multi-threaded, but if the CPUs are left waiting for data there are just expensive heat generators. :)

[Randy Donato]

That is strange since the newer xeons are running 800fsb but I understand that bandwidth is shared... I also rang up the pip project last night and looked at the cpu utilization and it was in the 80 to 90+ percent for the 4 processors(2 real and 2 virtual) which means to me Aspect is using the horse power. It wasn't like they were waiting for work but I never really understood what the performance and usage info is measuring, In terms of getting 4 layers with Edius and Nx in full res and not using Aspect in ppro I can only attribute that to a combination of the the overhead of motion effect vrs. the pip tool in Edius and the efficiency of the Canopus codec.

Once 64 bit settles in a little more and Intel has its response to DC firmly in place I will take a look at what is available. Right now I am getting close to the RT I need with HDV....plus it looks like unless someone comes up with a reasonably priced hardware HD encoder(once the format wars are settled) we will need all the machines you can get your hands on for render farms.

[David Newman]

There is more to memory speed than FSB numbers. CPU usage meters will include the time the CPU is waiting for memory to deliver the data to process -- so they aren't all that useful.