Does Premiere CS5 Support 8 Cores?
 

I am wondering whether Premiere CS5 support a dual quad-core setup. This might be a completely newb, ridiculous question, but I am wondering whether, for example, a dual quad-core setup with 8 gigs of ram would result in much better performance in premiere compared to a singe quad-core setup with 4 gigs.

Can anybody shed some light on these questions?

Yes. It performs even better on dual hexa cores, but there is no support yet for a simple 4 CPU Westmore deca core system with 40 physical cores and 80 with HT on, I regret to say.

BTW, using a dual quad core with only 8 GB is tantamount to choking this system to death. At least 24 GB is needed for proper operation.

Actually, I wouldn't go quite that extreme. Many quad-core CPUs have only dual-channel memory controllers. However, I have to admit that running only 4GB of RAM per CPU socket would be barely adequate for a 64-bit operating environment even for basic use.

And even if you have 24GB of RAM, a dual-channel memory controller would most likely run its memory in a mixed (hybrid or imbalanced) dual- and single-channel mode. This is because many dual-CPU-socket motherboards that support dual-channel have eight DIMM slots, which would have required 3GB or 6GB DIMMs (which are technically impossible to manufacture due to the fact that the bit width of the DIMMs and the memory ICs and the size of the ICs are all required to be of base-2 (2 to the nth power) in value) to achieve 24GB in proper dual-channel. And based on the results in the PPBM4 list on the PPBM site, the one system with an i7-920 with a GTX 470 and MPE Off stood out from the rest in that the export times are excruciatingly long compared to the other systems with the same CPU and graphics card because that system was running 16GB of RAM in an imbalanced triple channel + single channel mode (in fact, it was much slower in both export speed and MPEG performance than the LGA1366 i7 systems with the same GPU or slower GPUs and MPE off but with 12GB or even 6GB of RAM). Part of the blame is the motherboard in that system: It was a Gigabyte GA-EX58-UD3R (the predecessor of the GA-X58A-UD3R), which had only four DIMM slots (just like the Intel DX58SO motherboard I used to have in my editing rig, and is now on the shelf as a backup) - and with only four DIMM slots, filling all four of those up with equal-sized DIMMs would have forced this imbalanced memory controller mode in which part of the memory runs in single-channel mode. The "Flex" memory mode has a much higher overall latency than running all of the RAM in a balanced mode. (Thankfully, the EX58-UD3R's replacement model has six DIMM slots.)

So in other words, running 8GB or 16GB of RAM in a system that expects 6GB, 12GB or 24GB would actually result in degraded overall system performance. And in the case of running CS5 on a system with a Bloomfield, Gainestown or Gulftown CPU, 6GB per socket is okay but 12GB or more per socket is better.

Some of what others have said is accurate. However, I can give you accurate, fully-tested information.

To begin, all of the current Nehalem & Westmere based 4 & 6 core Intel CPUs are Triple Channel. Furthermore, almost all PC based motherboards, both single & dual socket, have ram slots in multiples of 3. My HP Z800 has a total of 12 ram slots, 6 for each CPU.

I have 2 PCs - an i7 920 w/12GB ram and an HP Z800 with dual-6 core CPUs and 24GB ram. When I first received the Z800, it had only 2GB ram. Since HP's ram prices are very high, I ordered 24GB from Provantage. And since that 24GB order had a problem due to how I billed it, it took 3 weeks to receive it. So, in the meantime, I purchased 4 2GB sticks of DDR3-1066 (designed for Apple) from Microcenter. I removed the original 2GB and installed 4GB on each CPU. I then ran my own tests using Premiere and After Effects CS5 in addition to PPBM5. With Premiere, I never ran into any problems with not enough ram. And in AE, it automatically adjusts how many cores can be used based on the amount of ram with a minimum of 1GB per core.

However, I am NOT suggesting in any way to use only 8GB for a dual-CPU setup. Because Premiere & AE CS5 are now 64bit, they will use any and ALL the ram you have. Even with 24GB in my Z800, Premiere has used as much as 16GB. It has been even higher but I can't recall the exact number and if I had AE open.

I would definitely use more RAM if I had more cores. I was mainly wondering if 64-bit Premiere would utilize more cores.

Yes, Premiere can utilize more cores. Here is an example where a single i7 920 is not enough - even with MPE hardware acceleration. On my i7 920 PC, using a single XDCAM EX layer, creating a 2nd layer with the cicrle effect to create a vignette and then a non-MPE color correction effect (I can't recall which one) - all 8 'cores' are maxed out and playback stutters from the timeline. And the playback resolution was at Full.

Remember that Premiere still uses only the CPU to decode all video layers during playback from the timeline. Only the 'MPE' accelerated effects use the GPU in addition to scaling and I think one or two other things. If you edit AVCHD and/or Quicktime MP4 from the Canon V-DSLRs, you can't have too much CPU power because they are so CPU intensive.

All I would add is that I recently ran on a series of machines doing some training on CS5, and AVCHD and DSLR stuff was not as much a horse as I would have suspected on a quad core single CPU i7 (hyperthread), and an older dual quad core Xeon Mac without any approved Mercury GPU help at all did pretty well as well.

I'm not saying that they were anywhere near as powerful as the kinds of beasts that these guys have been discussing mind you, I could bury the i7 with three layers of AVCHD, or two layers of AVCHD and a couple effects without the Quadro "turned on", but I do think that the threshold at the low end where you fall below some go/no go point is probably pretty modest.

But being "able to run" and "able to crank through the work as fast as you can think" are two distant modes of operation.

Tim, that dual-CPU Mac Pro you used has Nehalem CPUs, which are the Xeon version of i7 (unless I misread your prior posts).

Hmmm... could be.

I thought the Nehalems were a bit newer than that machine was... They told me it was a couple years old.

Nehalem based CPUs have been out for about 2yrs. I thought you stated that they were 2.26GHz CPUs, which are only Nehalem in the Mac Pro.

I wouldn't have put that in a message unless I looked at the system specs on the computer itself as I don't have a lot of Mac background anymore. (I was completely Mac until about 2000 or 2001...)

So...they must be. That would explain why it seemed to go right along I guess.

Tim, yes you did here: "Today I was editing on a year or two old Mac Dual Quad Core Xeon 2.26 GHz with no GPU help" http://www.dvinfo.net/forum/converge...cs4-cs5-2.html

Don't ask me how I remember this $hit.

Wow...that's impressive.

I wasn't disagreeing with you BTW... I trusted that you were correct.

I couldn't remember what I was doing or where I posted it, but that spec sounded like it was about the right numbers...

I can't keep all the processor and buss codenames and brandnames straight frankly... I get these questions from guys sometimes... "Oh...is that the Masada 5150 with the Westhedge buss and the X5000 Chipset in the BMX form factor with a Poseidon controller?..."

"Well...it's primarily green with lots of little metallic pinstripes all over it...and some black rectangular things on it, which one is my Westhedge and which is my Masada...?"


I just break 'em.... I don't build 'em.

:-)

Tim, sorry if I came across as argumentative as I wasn't trying to be.

PS If you need any assistance with your fxphd class, let me know.

Well, it's a good thing I don't do computer configuration... that's for sure.

It's one of the reasons I really need to have an actual integrator with some serious support build my edit machines. I don't know enough to be really dangerous, which means I do know enough to be a catastrophe...

Actually, make that all of the current Nehalem- and Westmere-based 4- and 6-core Intel CPUs for Socket LGA1366 have triple-channel memory controllers. There are a few Nehalem-based 4-core Intel CPUs that have only dual-channel memory controllers (those are the Socket LGA1156 Lynnfield i5-750 and i7-8xx series CPUs plus the related Xeon 3400-series CPUs).

Randall,

I'm pretty sure Steve ruled out the 1156 platform because video editors quite frankly don't use them because you can get more RAM on the 1366 AND have the option of dropping in a 6 core. Arguing semantics of technologies we shy away from takes away from the direction we try to go with conversations, and that is real world performance for video editors.

As far as Harm's first statement that you quoted, I'm on the same page as him. If you can take a RAM-bottlenecked system though and outperform his machine in PPBM I'll maybe start thinking otherwise.

Actually, Randall called the LGA-1156 'Lynnfield', which is correct; therefore it is not Nehalem.

Craig & Harm, I completely disagree that 8GB seriously chokes a dual-quad PC. I have a dual-6 core HP Z800, and I initially installed only 8GB while waiting for 24GB to be delivered. I don't know how reliable the PPBM5 was a few weeks ago when I had 8GB installed but in some tests, I easily beat Harm's numbers.

Crap, that reminds me, I meant to run the new PPBM5 but the power was out for the last few hours and finally came back on 20 mins ago. If anyone is very interested, I can test my pc with 8GB and 24GB to see if there is a difference.

However, in no way am I suggesting to use only 8GB of ram with a dual-quad or dual-hexacore PC. With 24GB of ram, Premiere CS5 has used 16GB so it certainly benefits from having more.

Lynnfield may have a Nehalem architecture, but it's not a full Nehalem chip like the Bloomfield/Gainestown/Gulftown chips. Therefore, I should have clarified my previous post: All of the full Nehalem chips have triple-channel memory controllers. In addition, there has never been a true Nehalem CPU with fewer than four cores.

Oh... well then! I was talking about us mere mortals that don't have access to such extravagant technology =p That's like whipping out the bazooka in a friendly target practice contest. Can't wait to see the PPBM5 benchmarks on that. I'm actually curious to see how my system came in compared to others.

Craig,

Unfortunately, someone else has a similar HP with faster Xeons who ran the new PPBM CS5. I have dual X5660 @2.66GHz and his are X5670 @2.93GHz, and he has a FX4800 and I have a FX3800.

From the PPBM5 numbers, it appears that Premiere benefits more from CPU speed than # of cores. I think the main reason for this is that neither the H264 nor the MPEG2 tests come anywhere close to utilizing 100% of my 24 'cores'.

If the dual Xeon EVGA motherboard was available 2 months ago, I probably would have gone that route because it allows overclocking of the Xeon CPUs. Thus, I could take the X5660 hexacore CPUs and easily push them to 3.6GHz and possibly up to 4.0GHz.

Btw, my dual-6 core HP Z800 was slightly more than the base dual-quad Mac Pro. Z800=$3600 + $800 (FX3800) + $1100 (24GB ECC Registered Ram).

Ok, so having a 2009 MacPro w/dual 2.26 CPU's (with 4 RAM slots per CPU) would perform *optimally* better with 12gb or 24gb rather than 8gb or 16gb?

Yes. Since these are triple-channel Gainestown CPUs, only three of the four RAM slots should be filled for optimal performance. The inclusion of only four RAM slots per CPU is a sign of cost-cutting in these nominally triple-channel platforms. If you fill up all four RAM slots per CPU, the performance might plummet to below that of an older Core 2 Quad.

I wouldn't go anywhere near that far. However, if you are referring to just memory performance, then I do agree, but just not with CPU performance. Tomshardware.com tested various ram setups, and there wasn't much of a performance hit, maybe 2-5% slower. Intel finally moved the Memory Controller to the cpu, which greatly helps all ram setups from single to triple channel.

Randall, I should apologize for getting too picky about semantics and Nehalem/Lynnfield, and because I was partly wrong. Nehalem is actually a 'Microarchitecture', not a code-name for a series of CPUs, which Lynnfield is.

I'm planning a new computer to work with Premiere, and have a situation to consult about.

One option is to buy an Intel i7 930, and get an associated motherboard, although every one I find on newegg that supports this processor only supports 2 6.0gb/sec sata drives.

The other option is to get an AMD Phenom II X6 1090T or the AMD Phenom II X6 1055T, whose motherboards in a similar price range support 5 6.0gb/sec sata drives.

1. What is the deal with these Intel-supporting motherboards only having ports to 2 6.0/gb/sec sata drives? Have people come across this problem?

2. Is 16 Gigs of Ram enough to run either of the two above-mentioned AMD processors? All but one AMD motherboard support a max of 16 gigs of ram.

3. How much ram adequately supports the Intel i7 930?

Natan,

1) Unless you plan on using $500-1000 Solid State Drives (SSD), then there is no need for 6Gb Sata because only a few very expensive SSDs can actually read or write faster than Sata 3Gb. Not even the fastest 15,000 rpm drives can come close to Sata 3Gb. In terms of onboard Sata & Raid, Intel is still the best.

2) I highly recommend Intel so no comment.

3) 12GB of 1333MHz or 1600MHz at a minimum. Preferably, you want 1600MHz so you have plenty of room to overclock the i7 CPU, whether it be the i7 930 now or hexacore i7 later on (i7 970 hexacore is being released within the next few weeks for $800ish). For ram, I prefer Corsair and G.Skill. I do NOT like OCZ as they tend to have more problems which sorta relates to why they tend to be cheaper. Let me tell you that a single stick of bad ram can appear as a small problem at first and then lead to major problems including file corruption. I know as this happened to me last year with OCZ ram in my custom i7 PC.

Another consideration is whether you use After Effects, Encore and/or Dynamic Link a lot because they will gladly eat your ram up. If you do, look into getting 24GB of ram. Or you can get 12GB by using 3x4GB sticks now so you can add more ram later.

When you get your PC up and running, the first thing you need to do is run MemTest86+ - download here: Memtest86+ - Advanced Memory Diagnostic Tool and run this test overnight.

FYI, there is another memtest site that should be avoided because the owner rips people off by SELLING a FREE program.

That is not an option, unless you are willingly looking at a system about 2 or 3 times slower than an Intel system.

Harm, c'mon. The 6 core AMD is not 2 times slower than the i7 930. With both at stock speed, the AMD is finally able to keep up and actually overtake the 930 in some tests. However, the i7 has more headroom for overclocking; thus, it will be faster. Below is a test from tomshardware.com using the Mainconcept H264 encoder (I can't recall if they use Premiere Pro or the actual Mainconcept program for this test).

Many people here respect you (including me) and come to you for advice. This sounds like a 'fanboy' comment, and it does not help anyone.

If you are comparing the two CPUs with an equal number of cores, then yes, some tasks will take nearly 2 times slower on an AMD Phenom II processor than on an Intel i7 processor. And if a Phenom II x4 965 is about as fast as an average Intel Core 2 Quad in CS5, then don't expect the x6 1090T BE to be any faster than a stock-speed i7-930 in CS5.

And yes, the H.264 encoding tests use the actual MainConcept Reference AVC encoder. The Adobe front end favors Intel CPUs over AMD CPUs, however.

Steve, I don't understand your comment about 6Gb Sata drives. Are 6Gb Sata drives not faster than 3Gb ones? And why?

Steve is referring to the physical transfer speed of the hard drives themselves. Currently, the fastest hard drives can barely top 150 MB/s in sequential transfer speed (or put it this way, the fastest SATA hard drives can barely max out even a 1.5 Gbps SATA I interface).

"6Gb" refers only to the interface speed, not the actual speed of the hard drive. For example, the Samsung F3 1TB drive uses the Sata 3Gb interface but is faster than the WD Caviar Black 1TB 6Gb drive. At the moment, 6Gb is only a marketing gimmick designed to sell 'newer' 6gb drives and sata controllers. There are many reviews all over the web testing 6Gb drives on both Sata 6Gb and Sata 3Gb connections and every drive is the same speed on both.

Which goes back to the actual physical sequential speed of the hard drive being the limiting factor (as I implied in my post above).

Steve, just look at the PPBM4 results and you will see that the Phenom II X4 955 comes out around 95 seconds. Assume the X6 is 50% faster in the best case, and the results will still be over 65 seconds.

In the other test look at the HP 9400 results. See where I get these rough estimates.

Remember also that AnandTech and TomsHardware are not very reliable test sites to consult for video editing.

The AMD's all suffer from their limited SSE support, which is what is used extensively during encoding to MPEG and H.264 and that is the reason AMD will always be way behind Intel on these typical video oriented tasks.

SSE4 subsets

Intel SSE4 consists of 54 instructions. A subset consisting of 47 instructions, referred to as SSE4.1 in some Intel documentation, is available in Penryn. Additionally, SSE4.2, a second subset consisting of the 7 remaining instructions, is first available in Nehalem-based Core i7. Intel credits feedback from developers as playing an important role in the development of the instruction set.

AMD currently only supports 4 instructions from the SSE4 instruction set, but have also added two new SSE instructions that is named SSE4a. These instructions are not found in Intel's processors supporting SSE4.1 and alternatively AMD processors aren't supporting Intel's SSE4.1. Support was added for SSE4a for unaligned SSE load-operation instructions (which formerly required 16-byte alignment).[3]

What is the best video card, in the range of 90-140 dollars, that is recommended with the i7 930 and 12 gigs of 1600 ram?

Natan, the 9800GT 1GB ($90-100) w/ 112 coreswhich I tested with the 'patch' is a good ecnonomical choice. I ran a test on my i7 920 at stock 2.66GHz using 3 AVCHD layers (60i from Sony xr500v) with the same 8 MPE-accelerated effects on each layer. I exported a 60s clip, and the 9800GT took about 14 mins while a GTX 275 (240 cores/$220) took under 9 mins. If you want a little more info, find my thread from a month ago where I include this info.

Harm, I have seen your disdain for tomshardware in prior posts, but I have not seen any good evidence proving that their testing is flawed and unreliable. I have come across many sites with flawed testing, but tom's is not one of them. Maybe they weren't very reliable in the past, but over the course of the last 2-3 years, I have found most of their testing to be very reliable, which encompasses CPU and HDD testing. I don't game so I don't pay too much attention to the GPU testing except when they test workstation cards.

Furthermore, I have been searching for a good site that tests various hardware aimed at video editing, but have not found one that does more than a few tests here and there. At the moment, the only one I do like is digitalcontentproducer.com (aka millimeter). They have done a fair amount of testing with the Mac Pro and the last 3 generations of HP workstations, including the Z800 with dual-hexacore. For the HP workstations, they have always used Premiere Pro, and one of their best tests showed how much 8-16GB+ of ram helps with Premiere CS4.

Harm, maybe you can talk to Chris about writing up articles that review computer hardware aimed at video editing. If you needed any assistance, I would be glad to help, such as showing what my computer can do compared to others. Thus, people can make an informed decision on what hardware they need. I have many ideas on testing if you are interested.

I'd put these two together and discovered that neither Anand nor Tom's used the Adobe suites at all. In fact, they might have used older versions of the video editing and encoding software that does not take anywhere near full advantage of even SSE3, let alone SSE4.