Some Reasons Why 8core Isn't Going To Make You Workflow Faster!

[Michele Gavazzeni]

8-core processor bandwidth is up to 21.3 GB/s

That says that if the task is pure CPU, the 8-core is
twice as fast as the 4-core.

But if your task has to do a lot of interaction with memory (RAM or HDD Writing/Reading), the advantage drops almost to nothing.
Video editing has a lot of interaction with RAM and HDD.

Memory bandwidth is inadequate for both 8-cores and 4-core.
Memory copy speed is at best 2.9GB/sec on the Mac Pro while maximum processor bandwidth is up to 21.3 GB/s. More than double the memory copy speed.

NOTE: 2.9GB/sec is the max RAM bandwidth independently from processor's number or Ram modules installed.


HOPE THIS CAN HELP
Michele

[Rob Lohman]

I assume you're talking about the new 8-core Mac Pro machines? In the PC world you can put different 8 cpu machines together.

In the end it all boils down to real-world performance. Obviously we will be testing and tweaking on the 8-core Mac Pro's.

[Michele Gavazzeni]

YES 8 core Mac but the same reasons can be extended to PC world

[istvanttt]

Here are some speed comparisions btw the 4 and 8 core Mac

http://www.barefeats.com/octopro1.html

[PaulClements]

So by that account a duo core with 16GB RAM would be as good for editing 4k footage as an 8 core with 16GB RAM would?

[TimothyD]

To me it sounds like just the increased speed when running multiple apps is worth it.

I often will have a render going in FCP, and be wishing I could be working on something or rendering in AE at the same time. Now I will be able to with no problem. I can easily see myself having renders going in FCP and AE at the same time as I am working on something else in photoshop.

Tim

[PaulClements]

But that's only three applications, according to the OP and other sources a quad ought to suffice because you have the capability to run 4 processes.

[TimothyD]

Sure, but if you have a few applications that are efficently using the multiple cores, then you may be able to make use of more than four cores at a time pretty easily.

I just always believe in more horsepower than I predict needing, because I've burned myself too many times by trying to cut corners. For instance, I have never been able to run Motion on my machine at home because I cut corners on the video card. :(

[Thomas Mathai]

To me it sounds like just the increased speed when running multiple apps is worth it.

I often will have a render going in FCP, and be wishing I could be working on something or rendering in AE at the same time. Now I will be able to with no problem. I can easily see myself having renders going in FCP and AE at the same time as I am working on something else in photoshop.

Tim

Running mutiple apps especially while one is rendering isn't always ideal. That's why it's best to have different workstations or a small renderfarm to deal with the heavy rendering while you work.

[istvanttt]

To me it sounds like just the increased speed when running multiple apps is worth it.

I often will have a render going in FCP, and be wishing I could be working on something or rendering in AE at the same time. Now I will be able to with no problem. I can easily see myself having renders going in FCP and AE at the same time as I am working on something else in photoshop.

Tim

Well in theory with OS 10.x if one applications crashes the system still alowes to work on the other application. But when the crash comes on a Sytem call, well then all the application will go down the river. T happened to me 2-3 times in the last years. Personally I will never use again an other application during an important work like rendering or data-transfers like dupplications.

[Rob Lohman]

So by that account a duo core with 16GB RAM would be as good for editing 4k footage as an 8 core with 16GB RAM would?

Please don't conclude such things by simply looking at some specs & theory.

Once the software is available RED and other will do testing and you'll see what the performance difference is.

[David Newman]

While memory bandwidth is an issue, these new CPUs to have a lot of on memory cache. The more CPU work that can be done within the cache the more you see the advantage of n-cores. Of course processing 4K frame happen to be larger than the cache yet there are ways around that. Remember for real-time 4K 4:4:4 your external memory I/O only needs to be about 1GB/s, while decompression and debayering will use up enough CPU clocks, memory bandwith will not be the bottleneck.

[Vincent Rice]

Applications such as Modo, Compressor and After Effects CS3 that can spawn notional separate apps for each processor will see an almost linear increase in rendering speed.

[Poi Boy]

I am getting one for sure.
Aloha
-A

[Jeff Kilgroe]

There's already a discussion on this one around here somewhere... Probably a page or two back. Anyway, one of the big problems with the current Intel quad-core CPUs is that they're really just two dual-core processors stuck onto one CPU die. Hopefully Intel will have unified quad-core CPUs available by the end of the year...

Anyway, several factors come into play with the performance of multiprocessor and multi-core systems. With current quad-core systems, since there are two separate processors on the same CPU chip, you have additional bus snooping operations that wouldn't be necessary with a unified quad-core design. Transferring data from one CPU core to another can become problematic if you're going from core #1 to core #3... Bein non-unified, the data must pass out onto the front-side bus to get to the other CPU rather than just being swapped through L2 cache.

Bandwidth of these systems can be exceeded or saturated, sure... But that's not the real concern as far as performance goes. We need much better CPU and thread management from the operating systems. We need much better multithreading and process scheduling from application designers. A lot of this won't come until the hardware vendors get involved too and right now it's too early in the game for that. Intel is selling all the CPUs they can make just by cramming more cores in there, regardless of who can or can not really use them. Even with the above mentioned changes at the OS and application levels, it's still going to be difficult to understand the topography of CPUs within a system unless the hardware vendors and system integrators provide CPU and threading profiles or software to help dictate how tasks should be divided up the most efficiently. For example, if the OS and applications can be made aware that on one system it's perfectly fine pull data generated from a thread on core #1 and use it in a new thread on core #4 because those cores use the same L2 cache. Whereas on another system, the same operation would waste an extra 8 clocks plus transfer time over the FSB to do the same thing, so the application can account for that.

This is going to be the next big thing that plagues multi-core systems. For years, us multiprocessor users have struggled with finding software that's actually properly written to use more than one CPU. Often we have to get by with running multiple instances to process different portions of our data. 3D animators love multiprocessor systems because it's easy to set individual renders or dedicate portions of a render to a single CPU. The gains are very linear... But this won't be the case as applications grow to span multiple CPU cores. Soon we'll be struggling with getting optimal performance from multiple cores because of generational differences between CPU designs and core implementations. Eventually these issues will be addressed... If not, then there is no way software can efficiently scale to use 16, 24 or even 32 cores or more. I think we will eventually see a day when applications intelligently allocate cores for their use just as they would allocate memory.

[hmurchison]

Excellent points Jeff.

I believe the timeline thus far is this in a nutshell.

Today's 8-core Mac Pro will improve with Leopard as that OS has improved facilities for denoting priorities and dependencies. If you must buy a 8-core now Leopard is your key to fast improvements.

If you can wait Q4 of this year will bring the 45nm Penryn procs. These are still MCM (2x Dual Core CPU on a die) configs but Intel has beefed up the cache size (6MB per DC chip) and the snoop filter now snoops all 4 caches.

True Quad Core isn't coming until late 2008, with Nehalem, and it's going to be a threading beast. You'll have "many core" CPU up to 8-cores per chip and each core will have the ability to run 2 threads.

Intel is working hard to improve threading performance linked to compilers.

http://www.intel.com/cd/software/products/asmo-na/eng/294797.htm

By Nehalem timeframe Intel will likely be on version 2.x and I'm sure Apple will link this to their Xcode IDE for use in compiling.

My recommendations are screw the memory bandwidth. The functions where you "really" need grunt tend to be CPU limited features.

After Effects 8 kicks arse on Octo Core.

http://www.barefeats.com/octopro4.html

Now THAT's an improvement. Buy the most cores you can afford.

[Bruce Allen]

My recommendations are screw the memory bandwidth. The functions where you "really" need grunt tend to be CPU limited features.

After Effects 8 kicks arse on Octo Core.

http://www.barefeats.com/octopro4.html

Now THAT's an improvement. Buy the most cores you can afford.

...and buy 16gb RAM apparently. Bare Feats said that their After Effects tests used 15 out of the 16 available gigs...

Bruce Allen
www.boacinema.com

[hmurchison]

Oh yes can't forget about the RAM. 15GB...that's like running the whole app in a RAM Disk. Well hopefully the DRAM suppliers will actually start supporting these bigger modules. The limitation before was supposedly pincount but since FB-DIMM is serial in nature the pincount of larger modules shouldn't be an issue. I'll be hotter than Hades if 2GB modules are still priced out of whack on a $/Per Gigabyte ratio by summer 2008.

[Michele Gavazzeni]

When you'll be using AfterEffects for some 4K or 2K you'll easly fill the 16GB RAM space.
At this point the aplicatin starts writing on the HDD and ...all this power are there for noting.

[Antoine Baumann]

Personnaly I see 8 cores workstation vers usefull for quick 3D test rendering. While it is an absolut necessity to have access to a render farm to render 3D full res or long shot, it is also very nice to have a very quick view at what you are doing and with 8 cores on your workstation it will be fast. At the moment I am running a dual core workstation, it is nice, but some time it take too long to render even a very small test render, so with 8 cores it should take much less time. Isn't it? And also on my current project (and certainly future too), I am using online render farm, and it takes time to upload project and even more to download the frames...

But I do not think it will be usefull for editing.

Thanks Michele to point out this. I think it is very good to know as much as possible before buying tools, and always think at what will be your needs and workflow.

About REDCINE, what people think about the advantage of 8 cores workstation?

antoine.

[Ace]

This is why I like AMD chips.. Too bad their taking so long.

[hmurchison]

When you'll be using AfterEffects for some 4K or 2K you'll easly fill the 16GB RAM space.
At this point the aplicatin starts writing on the HDD and ...all this power are there for noting.

No that is not true. Your app is either limited by bandwidth or it is CPU limited. 8 cores and proper threading will make a visible impact in CPU dependent application and less so with apps that have to shuttle small bits of data in and out of RAM.

AMD Opterons kicked the bejeebus out of Intel Xeons in Database apps because that style of app is sensitive to memory bandwidth but even more so to memory latency.

If you have the money and need for the fastest computer you can buy there's no issue with buying an 8-core computer.

[Jeff Kilgroe]

This is why I like AMD chips.. Too bad their taking so long.

I've been an AMD fan since they walked all over Intel in FPU performance starting with the Athlon CPU. Unfortunately, with Intel's Core2 design, AMD is playing catch-up and I find it hard to recommend an AMD system these days. Although the Opteron still makes for a good render node CPU in some situations. I'm hoping that AMD will come out with something really good soon because their current roadmap isn't anything to get excited over.

[PaulClements]

Please don't conclude such things by simply looking at some specs & theory.

Once the software is available RED and other will do testing and you'll see what the performance difference is.

I wasn't concluding anything Rob, as my original post suggested. I put a question mark on the end of it, which was basically there to suggest I simply don't know whether what I was saying is correct or not. I'm not an expert on these things and as such want to learn about the impact of multiple cores for applications used with red footage.

Please don't conclude such things by simply looking at my reply.

[Rob Lohman]

I must've missed that question mark then. Internet forums can be funny things :)

Sorry about that!