Quad Core vs 8 Core vs ?

[rob fetters]

...advice hunting here:

I'm going to buy a new Mac Pro, and an audiophile hardware / software creator friend of mine

says a Quad Core 3.2GHz will be faster & more efficient than an 8 Core 2.4GHz for

my work with DP, plugs and VIs. Anyone care to weigh in with an opinion? The cost differential

(after I swap in an SSD and buy 16 GB RAM ) is about $500, so that isn't a major factor in my

decision. I want the device to last as long as possible, of course.

Thanks in advance.

[HCMarkus]

I think the 8 core would be a substantially faster than the 4 core, but the 6 core machine is actually the cat's pajamas for DP.

[kenrinc]

For the most part I would agree. Actually all the reviews point to that fact. But understand the reviews are only running a "typical" standard battery of tests, speed tests (mem read, write), photoshop, itunes, etc... Now, if you look at any app that is coded to utilize multi cores it will always score better with a machine that has more cores (duh). If price is not an issue than I'd say the 8 core is probably the way to go.

I'm not convinced that DP takes advantage of those cores (yet) or that it even needs to. It seems many VIs do but again I'm not entirely convinced even if our little performance meters say they are. Maybe it's just me but I find it interesting that the multi core debate isn't raging hotter with many developers.

It's amazing to me that Apple has now gone back to the days of the IIfx where a personal computer can actually cost you nearly $10k. It's sad.

$.02

Ken

[VitcoMusic]

Ken - cutting edge technology is always expensive at first. You should know this.



Sent from my iPhone using Tapatalk

[wavelength]

Gang,

Hey I'm this guy ....

audiophile hardware / software creator friend

.

The deal is that the Quad Core is typically running much faster than the standard 2.4GHz 8 Core unit off the shelf. This was verified by MacWorld.

The question Rob poses is the number of cores going to be more beneficial to his work than the speed.

Most of these speed tests really do not test the shear power of muticore processors. They are merely a set of programs or executions with timed tested output. They throw that into a spread sheet and come out with a number.

So the question is and if there are any programmers for Digital Performer reading this then maybe he (we) can figure this out.

As a programmer, digital hardware designer and someone working on a 32 bit adc/dac with a 8 core processor with little hair left... I can tell you there is very little ways of determining who what when and how each of these cores is performing when they share a pool of memory and try and swap info to and from each other.

Thanks
Gordon

[kenrinc]

Ken - cutting edge technology is always expensive at first. You should know this.

It's not cutting edge....

Most of these speed tests really do not test the shear power of multicore processors. They are merely a set of programs or executions with timed tested output. They throw that into a spread sheet and come out with a number.

As I noted in my post... Although most will at least run Mathmatica which should technically test multicore performance. As much as we'd like to hear it, I truly doubt anybody from MOTU would post anything regarding DP architecture.

Ken-

[HCMarkus]

Now, if you look at any app that is coded to utilize multi cores it will always score better with a machine that has more cores (duh).

Do you really think processor speed doesn't matter? Do the math, and you'll note that, even assuming no penalty for passing data from core to core, the 6 core Mac Pro offers theoretically better performance than the 8 core...

2.40 gHz x 8 = 19.20 ghz Price: $3499

3.33 gHz x 6 = 19.98 gHz Price: $3699

Now, when you consider the user may use apps that don't suport multi threading, as Gordon notes, moving data from core to core introduces at least some inefficincies, the 6 core looks like the clear winner unless you go to the fastest 12 core.

The biggest negative with the 6 core is the smaller number of RAM slots, but by using three 8 gig sticks, one can hit 24 gigs in memory with a three-stick configuration, the config preferred by Intel's Westmere.

As far as pricing, yes, the machines are expensive; but before casting stones, be sure you are comparing Apples to... Dell or HP, not a home-built PC using lower cost processors and parts. And please don't quote prices with maxed-out, (admittedly way overpriced) Apple RAM and HDs. Mac Pros are so ridiculously simple to add drives and RAM to, only a fool will buy either from Apple.

[wavelength]

Gang,

2.40 gHz x 8 = 19.20 ghz Price: $3499

First you cannot just do math to determine this. There is a lot more going on here. Local cache, program architecture etc...

Remember each core still uses a the same chunk of memory as the other cores in that processor. So when one is accessing the ram the others are locked out. The arbitration on something like this is sometimes complex.

In general none of the tests are really testing the cores and in some way if you energize a ton of cores that have extensive memory access like DP would have... it's actually diminishing returns. This is because when a core is paused... it's not doing anything and that degrades the system.

So this is why say a 2.8/3.2 quad is faster than a 2.4-8 core. If the 8 core was at 2.8 or 3.2 then it would probably be pretty equal using current testing methods. The problem there is the diminishing returns on $'s. There is really only a 2.4G 8 core which to me is the dog in the MacPro department. Why only 2.4 is beyond me, so really doing a QUAD 2.8 or 3.2 makes so much more sense.

Ok talked to Rob... he is considering the 6 core. Ok so let's take a look at the models again:

Quad 2.8/3.2
Six 3.33
Eight (dual Quad Core) 2.4
Twelve (dual Six core) 2.66/2.93

Ok with every Core you have a chunk of memory. Dual Quad/Six cores have therefor two sets of memory and the Quad/Six has one chunk of memory.

So one of the disadvantages of going with the six core is going to be memory arbitration. With a quad core you have 4 going into one set of memory and with Six core you have 6 going into one set of memory.

Of course this is where cache really pays off. Each of the processors in the Core has 2MB of cache.

Main memory each Core can have up to 16G per or four slots (sorry Rob I was wrong here). Opps wrong again... OWC says 32G, gang sometimes Apple only specifies a certain amount due to availability, testing current requirements etc.. in most cases and unlike the PC's I have designed they are very conservative. Ok so single core quad/six units can take 32G using 8G in all 4 slots.

Personally I think the best bang for the buck is the 3.2 Quad Core. At $1100 cheaper than the Six core I don't think the Six will be significantly more powerful.

Thoughts?

Gordon

[HCMarkus]

Gordon, I offered my admittedly simplistic math to rebut an even more simplistic statement that "more cores = always faster."

Although the 4-core may prove to be a better bang for the buck, I have no doubt that the 6-core, which has a nearly identical clock speed to the 4-core, will be at least as fast with all apps (including the ones that don't multi-thread), and appreciably faster with many.

Also, the 6-core uses the Westmere architecture, whereas the 4-core is Nehalem.

DP has been using multiple cores very efficiently for some time, so the additional cores will definitely bring some extra useable horsepower to the studio. Worth the money? That's the buyer's decision.

[rob fetters]

The 6 core 3.33Ghz is $800 more than the Quad 3.2GHz. I'll be going elsewhere to purchase

a SSD and RAM. I really appreciate the feedback. Still weighing options....

[wavelength]

Gang,

Here is the real point in the review of the 12 Core unit from MacWorld:

As shown in our "Fastest Macs money can buy" report, the 12-core Mac Pro was not the overall speed king in our tests. It was outperformed by a $3699 built-to-order (BTO) Mac Pro with a 3.33GHz 6-core Xeon Westmere processor, which was faster in 10 of our 17 tests, and matched the 12-core Mac Pro’s scores in two other tests. Testing the 12-core Mac Pro with 12GB of RAM (six 2GB DIMMs provided by Crucial) showed very little improvement over the 12-core Mac Pro with 6GB of RAM—just one Speedmark point.

Ok so using their test we basically come up with the following numbers:

Six Core 3.3GHZ, 3G memory: 263
12 Core 2.66GHZ, 6G memory: 261
Quad Core 3.2GHz, 3G memory: 223-231 interpolated data from 2.66 & 2.8 numbers.
8 Core 2.4GHz, 6G memory: 216
Quad Core 2.8GHz, 3G memory: 207

All of these run the same 1066MHz memory and therefore the same speed. But as you will see below, none of these are sold with optimal memory amounts.
~~~~~~~

Some people have asked since DP is only a 32 bit application, will using more than 8G really matter.

DP as well as logic, garage band heck even iTunes will require allot of what is called Virtual Memory (VM). There is a pool of memory set aside by the kernel for VM use in all applications. This pool is ever expanding and contracting depending on what is required etc... but say an application asks for X-VM and other applications are asking for Y-VM. If the application for Y-VM is running and has access to the pool (real memory), then the application for X-VM is really sitting on a hard drive. When application X goes to access X-VM, then a lot of things happen which really drag down a system. First X-VM app is paused then application (or thread) Y-VM is paused and the kernel writes Y-VM to disk. Then reads out X-VM into the pool and lets application X-VM do it's stuff.

Now if we have a ton of freaken memory then the pool is very large and what happens is the OS does not need to hit the disk. Everything is local... it does not matter than DP is 32, much of it's memory storage is VM and therefore more memory is better.

Intel & Memory... ok.
It is always best with the Intel system to make sure all your sticks are the same size. Using different sized memory is really bad from Intel sake.

You are really best to increment memory as follows:
1G, 2G, 4G, 8G, 16G, 32G... etc.

Using disimilar sizes can lead to slow downs in the access of memory which is always bad considering the memory here is a slow bottle neck.

Activity Monitor is for kids, if you think that this is telling you anything you are totally wrong. Sure it gives you basic stuff... and will show red line... but really if you are in the red you are really way past it as the AM utility is really conservative.

If you want a better perspective open the terminal (applications->utilities->terminal). This will load the UNIX terminal application. Type in "top" this is the system process display utility in UNIX it will show real time processes, threads, application, memory use, VM, process active, CPU usage per process etc... like this:

Processes: 61 total, 4 running, 57 sleeping, 286 threads 08:37:50
Load Avg: 1.79, 1.86, 1.75 CPU usage: 58.13% user, 8.83% sys, 33.2% idle
SharedLibs: 5568K resident, 7664K data, 0B linkedit.
MemRegions: 18083 total, 1766M resident, 23M private, 741M shared.
PhysMem: 976M wired, 3250M active, 1911M inactive, 6136M used, 2057M free.
VM: 132G vsize, 1042M framework vsize, 11497162(8) pageins, 747(0) pageouts.
Networks: packets: 2893530/3376M in, 2515235/916M out.
Disks: 3538988/84G read, 3671678/681G written.

PID COMMAND %CPU TIME #TH #WQ #PORT #MREG RPRVT RSHRD RSIZE
16376 top 6.4 00:04.23 1/1 0 24 33 1204K 264K 1780K
16372 tcsh 0.0 00:00.02 1 0 17 25 604K 244K 1152K
16371 login 0.0 00:02.84 1 0 22 53 480K 244K 1584K
16369 Terminal 2.2 00:00.56 6 2 109+ 114+ 3232K+ 28M 9880K+
16186 mdworker 0.0 00:42.46 3 1 51 88 9452K 10M 63M
15500- firefox-bin 12.5 34:22.58 11 1 225 1760 89M 101M 306M
14858- DesignWorks 0.1 02:22.97 3 1 127 319 9492K 64M 19M
14849 Grab 0.0 00:07.60 2 1 159 173 8336K 38M 16M
13349- thunderbird- 0.0 16:42.64 13 2 216 821 68M 106M 167M
13302 hdiejectd 0.0 00:03.84 2 1 34 43 596K 524K 1624K
13297 diskimages-h 0.0 00:03.03 3 1 75 69 9548K 10M 13M
2443 AppleSpell 0.0 00:01.66 2 1 35 62 3520K 9344K 9416K
1913* LaunchCFMApp 0.6 36:36.26 4 0 135 350 25M 31M 38M
1890* LaunchCFMApp 0.9 46:53.51 8 0 193 443 42M 47M 59M

I have basically running right now Thunderbird, Firefox, Adobe Reader, HP style calc, Drop Zip, Grab and Design Works (electrical schematics and stuff).

You can see that my VM is actually 132G in size.

I have seen activity monitor show like 5% usage with iTunes/Pure Music or Amarra. Then I run top in the terminal and see the CPU usage jump to almost 55% when outputting 24/192 stereo material. Granted this is only a mac mini 2.4G/8G/200G SSD but still you get the idea.

Thanks
Gordon

[kenrinc]

Using dissimilar sizes can lead to slow downs in the access of memory which is always bad considering the memory here is a slow bottle neck.

Yes, matched memory sets are preferred but not mandatory. Supposedly, you take a latency hit, however, any other form of disk storage will be FAR slower than any latency induced by disimilar sized memory configurations. Many people have not been able to show any performance differences when using dissimilar sized memory chips (specifically on the MacPro line).

Ken-

[rob fetters]

Thanks for the feedback all you smart people. I love MotuNation.

Coupla clarifications:
Although Apple doesn't say so, you can go up to 32G of RAM in the 4 slots on the Quads and 6 Core machine via 4 X 8G sticks.
The RAM speed on the 6 core is 1333Mhz.
The links to benchmark tests that appear on some of the excellent threads started by Rusty Shackleford back in August are worth reading if you're trying to make the choice I'm making.

I just ordered a 6 Core, and I'll tweak with an OWC SSD, WD HD, and 2X8G of OWC RAM. I've got some work I need to finish before I jump to the new 'puter, but hopefully by mid-December I'll be experienced enough to offer an opinion besides, "OMG, it's fast!". (Unless I get more work - which would be just swell,thank you - because I learned the hard way to never ever update hardware or software in the middle of a gig.)

[HCMarkus]

Congrats on the purchase... methinks you made a great choice.

No. I'm not jealous.

One note: I understand the best memory config for Westmere is three sticks of RAM so, if within budget, you might want to go 3x8 for 24 gigs total. I look forward to your reports!

[i7user]

Long as it's fast enough... that's ALL that matters because the tools are so good now that it really just comes down to... are the performances and the music, inspirational and artistic?