4-Core Ivy Bridge VS my 12-core MacPro

[alyarb]

It is not likely that he will introduce undetectable, deal-breaking errors into his production due to overclocking, as his video is played back and examined many times before being exported to some media or published for consumption. This is not a complex numerical simulation, it's video editing, and then encoding.

Overclocking is simple to test and simple to reverse. It would only take about 30 minutes to settle into a conservative 4.0 GHz overclock and do some rigorous testing and comparisons to see if it affects any of his output. The performance gains from such high frequencies are too easy to obtain and too great to simply ignore. I've encoded on a 4 ghz Yorkfield for many years and it has not varied my output nor introduced instability. If those are your only reservations, they can be addressed by testing.

Not overclocking your machine because someone told you there is a rule against it is almost as foolish as overclocking without testing. Lots of people already use x264 as a burn test. Why not Premiere?

 

[kernelc]

It is not likely that he will introduce undetectable, deal-breaking errors into his production due to overclocking, as his video is played back and examined many times before being exported to some media or published for consumption. This is not a complex numerical simulation, it's video editing, and then encoding.

Overclocking is simple to test and simple to reverse. It would only take about 30 minutes to settle into a conservative 4.0 GHz overclock and do some rigorous testing and comparisons to see if it affects any of his output. The performance gains from such high frequencies are too easy to obtain and too great to simply ignore. I've encoded on a 4 ghz Yorkfield for many years and it has not varied my output nor introduced instability. If those are your only reservations, they can be addressed by testing.

Not overclocking your machine because someone told you there is a rule against it is almost as foolish as overclocking without testing. Lots of people already use x264 as a burn test. Why not Premiere?

I respectfully but strongly disagree.

A workstation is a production machine, with the goal to earn money. You should overclock only if your are really short of performance and don't have enought money to upgrade the system.

Obviuosly situations varies, so a particular system can be overclocked without so much fear, while other should really left at default setting.

I see your point, really: back in the days, I always ask my system admin "why not overclock this wonderful Tualatin-based PIII-S server?". The reply more or less was: "if the server crash and I have the slightest culript in overclocking it, I will lose my work". A pretty good reason to stick with default setting, don't you?

However, basing the entire workstation purchase/not purchase question on overclocking is wrong, in my opinion.

Regards.

 

[alyarb]

In a corporate environment (where actions have consequences) I would never overclock even in the most ripe scenario, but this guy inherited an obsolete workstation with dwindling resale value and he now works from home on whatever hardware he pleases.

I would definitely not advocate ambitious overclocking, but a little can't hurt.

 

[philipma1957]

THIS is the only thunderbolt board for the i7 3770k I can find.



http://www.newegg.com/Product/Produc...82E16813131850

I thought I had seen a msi board.

I have a lot of t-bolt gear and I want a t-bolt board myself.

I also want more choice.


EDIT FOUND THE MSI INFO!


http://hothardware.com/News/MSI-Unleashes-Z77AGD80-Motherboard-With-Thunderbolt/

 

[alyarb]

I tend to favor ASUS over MSI anyway. I also like the black i/o shield. Maybe anodize it next time for the price they're asking.

 

[dsc106]

Alyarb, do you agree with greenhawk's number breakdown? Or do you maintain what you said in your post before that, how the Ivy Bridge should practically be faster than "50%~ish" speed?

I will say this:

* A conservative over-clock doesn't bother me for a home video editing workstation, there's auto-save and file save vault. It's not like I'm running a company wide SAN server where any error is going to be detrimental. I'd be more worried about Adobe software crashing under normal circumstances than a moderate OC causing a problem. If I had a problem it wouldn't be in losing/corrupting a project, it *might* be in rendering a final video out, but that's easy to re-do, and from what I've seen/heard/read I doubt it'll be much of an issue. Besides, if I get a decent performance boost and I'm at slightly more risk for error/crash, that seems like an OK trade-off as the odds of having that many problems are low, and if I did, I don't see it as being one that would cause me to miss a deadline since projects auto-save every couple minutes and I rarely have a rendered output that takes longer than 30 min.

* Hackintosh is a little more risky if I decide I need/want to do this; BUT I did it for 3 years 2007-2010 before I got this MacPro, and it's only gotten better since then. In the past, I never really had a problem. People complain about not counting on Apple these days for pro-level hardware/software updates... the hackintosh hackers who "code when they feel like it" tend to be just as fast if not faster than Apple at getting drivers out. The bigger concern here is if I want a new FCP X software update that requires an OS update that breaks the hackintosh, I have to research this first and wait, which is just a pain.

 

[dsc106]

Also, here are some comparisons in FCPX (which I believe is better programmed to take full advantage of multicore), showing westmere 6-core VS ivy bridge:

http://www.barefeats.com/macs11_01.html
http://www.barefeats.com/fcpx01.html

Hard to say how that would scale to 12-core, but mine is 2.66ghz and the test there is 3.33ghz westmere... and this is comparison to a 3.4ghz Sandy Bridge... so when factoring in non-linear scaling, my slower 2.66ghz cores, more speed from Ivy Bridge, and higher clock speeds from the Ivy Bridge I'd get (even before any OC) than listed here........ seems like the Ivy Bridge is a no brainer?

Plus, since i can't afford a GTX680 or to upgrade to 32gb of RAM in my current MacPro... but get both of those upgrades AND a bunch of cash by selling/trading... I feel like that would widen the gap??

 

[alyarb]

I really wasn't paying attention to those theoretical numbers because all of them assume that whatever compute resources are available will be fully utilized. This assumption by itself negates whatever method either of you were using. Whether it's cores, cycles or whatever you are factoring in with those percentages, they aren't pertinent to real-world operations in Premiere or FCPX and they do not reflect the behavior that you demonstrated in your photostream. Sure we can form speculations from top-level ratios, but they're way too idealized to be applicable to real life and should be avoided (unless the discussion is limited to synthetics such as linpack or prime).

The barefeats data you linked show what I was assuming earlier. Sandy Bridge is faster in everything but encoding, and in encoding, it really is not much slower. You've insisted so far that FCPX is highly parallelized, but this data illustrates the many caveats, even in encoding, that can upset an assumption like that. This is where your and greenhawk's percentage hypotheses break down. With the top-down percentage method, you or greenhawk would've been liable to assume that a 4 core Sandy Bridge would be about 70% as fast as a 6-core Westmere, but the FCPX encode test shows that Sandy actually comes within 9% (single procs, I know, just illustrating that this cannot be modeled by your rudimentary numerics, nor can we speak for exactly what/how they were encoding during that test).

Also note that the Mac Pro in those tests is running 666 Mhz faster than your 2.6 GHz Mac Pro, so without that clock frequency advantage, Ivy Bridge is looking even more economical. Yes, you have more cores than the barefeats Mac Pro, but you can see from the graphs that the non-encoding tests favor Sandy Bridge over Westmere, and a 4 GHz Ivy will do even better.


I have tried to avoid discussing hackintosh builds because you are entirely on your own there. As long as you have some kind of CoreImage and QuartzExtreme working, you should get a comparable experience to what you have now. You don't have a quadro card so don't expect any GPU-accelerated compute to take place on the Mac, but you will definitely get CUDA acceleration under Windows with Premiere CS6.

 

[ALIVE]

Also, here are some comparisons in FCPX (which I believe is better programmed to take full advantage of multicore), showing westmere 6-core VS ivy bridge:

http://www.barefeats.com/macs11_01.html
http://www.barefeats.com/fcpx01.html

Hard to say how that would scale to 12-core, but mine is 2.66ghz and the test there is 3.33ghz westmere... and this is comparison to a 3.4ghz Sandy Bridge... so when factoring in non-linear scaling, my slower 2.66ghz cores, more speed from Ivy Bridge, and higher clock speeds from the Ivy Bridge I'd get (even before any OC) than listed here........ seems like the Ivy Bridge is a no brainer?

Plus, since i can't afford a GTX680 or to upgrade to 32gb of RAM in my current MacPro... but get both of those upgrades AND a bunch of cash by selling/trading... I feel like that would widen the gap??

that is a sandy bridge cpu not ivy bridge
ivy will be 5-10% faster dependign the software so keep the 5% as conservative a 2,66 will have the 80% of the perfomance of a 3.33 if it linear scaled by hz.
100hz more on the cpu it is 3% more from 3.4 to 3.5 +5% from sandy to ivy you are at 8% better than the test
fcpx plur
6core westmere 34,75 /12 core 17,38
4 core ivy 22,87 76% of the 12 core system
fcpx plur sharpen
6core westmere 41.38/ 12 core 20.69
4 core ivy 25.37 81% of the 12 core system
fcpx export h.264
6 core westmere 28.38 /12 core 14.39
4 core ivy 21.39 67% of the 12 core (h.264 is highly threaded)
after effects cs5.5
6 core westmere 88.75 /12 core 44.38
4 core ivy 93.52 47% of the 12 core system

so it can be as low as 50% of the 12core or as hight as 80%
worse case scenario the computer will be half speed without any oc
a 6 core ivy would definetly close that gap