Register FAQ / Rules Forum Spy Search Today's Posts Mark Forums Read
Go Back   MacRumors Forums > Apple Hardware > Desktops > Mac Pro

Reply
 
Thread Tools Search this Thread Display Modes
Old May 3, 2014, 03:58 PM   #1026
Tutor
Thread Starter
macrumors 65816
 
Tutor's Avatar
 
Join Date: Jun 2009
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute
First Native Power Management

After a few tweaks toward getting native power management working, GB 3 score improved by over 1k points [ http://browser.primatelabs.com/geekbench3/multicore ], i.e., from 48,123 to 49,152, moving the monstrosity's OSX score up five places on the first page of multicore scores.
__________________
21 tweaked, multiOS rendering systems - oTitan RD TE > 58 (86,432 CUDA cores). 13,120 ATI Stream PUs. 206 CPU cores. Benches: CB11.5-48.5; CB15-3,791; GB2-58,027; GB3-71,691; LuxMark/Sala-12,330.
Tutor is offline   0 Reply With Quote
Old May 4, 2014, 04:57 PM   #1027
Tutor
Thread Starter
macrumors 65816
 
Tutor's Avatar
 
Join Date: Jun 2009
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute
How many cores are too many?

As shown yesterday, the OS will boot with 32 real cores/threads activated and yield a Geekbench 3 score of about 49,000 for 32 Sandy Bridge real cores running at 2.7 GHz. I had been having problems with consistent crashes whenever I activated hyper-threading. At first it made me wonder whether the OS had issues with the way my system implemented hyper-threading, but I began to wonder what would happen if I activated hyper-threading, but used a bios setting that limited the number real cores activated from 8 to 4 per CPU. My guess was that 32 was a cap. Well, the next re-boot after making this change let me know that the OS wasn't crashing because it was having an issue with how hyper-threading was implemented on my system because my system booted right up with (4x4=) 16 real cores, plus 16 more hyper-threads and achieved a Geekbench 3 multicore score of 32,391 [ http://browser.primatelabs.com/geekbench3/557286 ], currently on page 21 of the top multicore listing. I'm delving further into this matter other to see if I can find a better solution to this limitation.

BTW - I've got high resolution video working and I am posting this update, as was the case with the earlier followup updates, on 10.9.2 on the SuperMicroMacStrosity, using a really cheap USB to Ethernet Adapter that I had purchased some time ago. That's how I have been able to easily post the Geekbench scores. Also, for purposes of comparison, the current configuration yields a Geekbench 2 overall score of 27,774 [ http://browser.primatelabs.com/geekbench2/2457650 ], a Cinebench 15 xCPU score of 1,720 and a Cinebench 11.5 xCPU score of 19.60. Some of you mathematicians out there might be able to put together formulae for the correlations between scores and predict what my WolfPacks1&2 (which I built in 2010 and 2011, which routinely achieve Cinebench 11.5 xCPU scores of 24.7+ and Geekbench 2 scores of 40,000+) would score under Cinebench 15 and Geekbench 3, respectively, after I finish cleaning their parts and putting them in new, large Lian Li cases. I'm projecting they will each score over 45k in Geekbench 3.
__________________
21 tweaked, multiOS rendering systems - oTitan RD TE > 58 (86,432 CUDA cores). 13,120 ATI Stream PUs. 206 CPU cores. Benches: CB11.5-48.5; CB15-3,791; GB2-58,027; GB3-71,691; LuxMark/Sala-12,330.

Last edited by Tutor; May 4, 2014 at 08:46 PM. Reason: BTW
Tutor is offline   0 Reply With Quote
Old May 5, 2014, 01:55 AM   #1028
DJenkins
macrumors regular
 
Join Date: Apr 2012
Location: Sydney, Australia
Quote:
Originally Posted by Tutor View Post
How many cores are too many?
I'm honestly surprised that many cores are recognised at all, considering there's never been (or will be?) an official mac like that.

** of note, apple are probably looking at ways to block your little experiment with their next update haha

I hope not though!!
__________________
SpeedMetal: Asus RIVF | 4930K | H80i | 2x GTX780 Classified | 32GB G.Skill RipjawsZ
16TB ATTO RAID10 | 256GB 840PRO SSD OSX 10.9/240GB OCZ PCIeSSD Win7
HeavyMetal: EVGA SR-2 | OSX10.8.5 | GB3 | GB2
DJenkins is offline   0 Reply With Quote
Old May 5, 2014, 11:13 AM   #1029
Tutor
Thread Starter
macrumors 65816
 
Tutor's Avatar
 
Join Date: Jun 2009
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute
Sometimes experiments give glimpses into the future by revealing all of the present.

Quote:
Originally Posted by DJenkins View Post
I'm honestly surprised that many cores are recognised at all, considering there's never been (or will be?) an official mac like that.
I too was surprised, but only for a moment because of this: maximum # of supported cores = 32, whether the mix is 32 real cores or 16 real cores and 16 hyper-threads. That underlined portion likely doesn't fall into the category of "never will be." In fact, it may hint at the very next phase - the release of a 16 core high end Mac Pro option when Intel releases the E5-2697's successor. Additionally, my experiment shows that the OS does now support at least 128 gigs of ram.


Quote:
Originally Posted by DJenkins View Post
** of note, apple are probably looking at ways to block your little experiment ...
The word "experiment" best sums up this exercise because I will not be sacrificing the 1.4x performance gain that Windows and Linux OS provide to run my heavily CPU-focused applications this way on some of my best hardware. I'd rather have those applications run full tilt. For the applications that I now rely on OSX to do, my four Mac Pros completely suffice. Although for Mavericks - Apple deprecated three of my Mac Pros despite my having purchased the top of the line. Apple's eraser was too big. But Tiamo provided a fix.
__________________
21 tweaked, multiOS rendering systems - oTitan RD TE > 58 (86,432 CUDA cores). 13,120 ATI Stream PUs. 206 CPU cores. Benches: CB11.5-48.5; CB15-3,791; GB2-58,027; GB3-71,691; LuxMark/Sala-12,330.

Last edited by Tutor; May 5, 2014 at 11:33 AM.
Tutor is offline   0 Reply With Quote
Old May 5, 2014, 06:15 PM   #1030
Tutor
Thread Starter
macrumors 65816
 
Tutor's Avatar
 
Join Date: Jun 2009
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute
Its time for the Grand Experiment to end.

As for that which I did not accomplish in modifying my WolfPackPrimes into SuperMicroMacstrosities -
1) no working HMDI audio;
2) no working internal ethernet, however, I did remember a way to get ethernet service easily and cheaply. But as to internal ethernet and HDMI audio, I didn't even get to that at all before fully accepting that it was time for me to give this exercise a death blow because of
3) no performance to match my expectations. I was looking to OSX performance on those systems like that performance that I've come to expect from Windows and Linux, but I found that the Mac OS has lower limits to the number of cores that it'll readily/easily support than either Windows or Linux have.

But I don't view this exercise in any way as a complete failure for the following reasons -
1) the bios settings that I experimented with to try to makeup for the core # limitations, are serving me better under Linux and Windows than those settings that i had made before to emulate moderate underclocking. So this exercise has made my machines' perform faster;
2) I learned a lot more about my Mac hardware;
3) while not reaching the performance levels that would have made me stick with it to finish off HDMI audio and internal ethernet, I did learn more about maximizing the performance of my systems' OSes - such as, but not limited to, gaining a better understanding of the operating system pieces and the roles that they play; and how to better diagnose sources and cures for OS related problems;
4) it gave me the opportunity to see what Apple may have in store for us in the next Mac Pro (but obviously 2 CPUs with a total of 32 real cores will crush 16 comparable real cores in a single CPU package, even with the additional 16 hyper-threads) and
5) it enabled me to get a better grasp of what the benchmark metrics are really testing, how the current versions of those metrics relate to earlier versions, and how to better use those metrics to increase the performance of my systems.

BTW - Here is my latest Geekbench 3 score (49,337) as this experiment draws to a close - http://browser.primatelabs.com/geekbench3/558552 .
__________________
21 tweaked, multiOS rendering systems - oTitan RD TE > 58 (86,432 CUDA cores). 13,120 ATI Stream PUs. 206 CPU cores. Benches: CB11.5-48.5; CB15-3,791; GB2-58,027; GB3-71,691; LuxMark/Sala-12,330.

Last edited by Tutor; May 6, 2014 at 08:29 AM. Reason: BTW -
Tutor is offline   0 Reply With Quote
Old May 7, 2014, 02:24 PM   #1031
mseth
macrumors 6502
 
Join Date: Nov 2012
Location: Canada
What is the point of flashing the 1,1 to a 2,1?

-> Installed 2x X5355s and got over 47% higher score in GeekBench than with 2x 5160s 3GHz.
-> Everything works just as expected.
-> Edited .strings to change "2x 2.66GHz Unknown" to "2x 2.66GHz Quad Core".
mseth is offline   0 Reply With Quote
Old May 7, 2014, 06:02 PM   #1032
Tutor
Thread Starter
macrumors 65816
 
Tutor's Avatar
 
Join Date: Jun 2009
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute
Quote:
Originally Posted by mseth View Post
What is the point of flashing the 1,1 to a 2,1?

-> Installed 2x X5355s and got over 47% higher score in GeekBench than with 2x 5160s 3GHz.
-> Everything works just as expected.
-> Edited .strings to change "2x 2.66GHz Unknown" to "2x 2.66GHz Quad Core".

Mainly, it offers a change in the model number and some additional processor support. This is the definitive thread on that mod: http://forum.netkas.org/index.php/topic,1094.0.html . Did you edit .strings to insure that the 2x X5355 worked as you desired? There's an old thread dedicated to that system identity change: http://forums.macrumors.com/showthread.php?t=269553 .
__________________
21 tweaked, multiOS rendering systems - oTitan RD TE > 58 (86,432 CUDA cores). 13,120 ATI Stream PUs. 206 CPU cores. Benches: CB11.5-48.5; CB15-3,791; GB2-58,027; GB3-71,691; LuxMark/Sala-12,330.
Tutor is offline   0 Reply With Quote
Old May 7, 2014, 08:29 PM   #1033
mseth
macrumors 6502
 
Join Date: Nov 2012
Location: Canada
Quote:
Originally Posted by Tutor View Post
Mainly, it offers a change in the model number and some additional processor support. This is the definitive thread on that mod: http://forum.netkas.org/index.php/topic,1094.0.html . Did you edit .strings to insure that the 2x X5355 worked as you desired? There's an old thread dedicated to that system identity change: http://forums.macrumors.com/showthread.php?t=269553 .
Thanks. Lots of pages to dig through. Google turns up "it fixes crashes for some", "no perf difference", "reports x5355 correctly (cosmetic)", "microcode has changed", "there is no difference"… and so forth.
-> Can't seem to find a solid reason to change to 2,1.
-> Will stick with it being as original as possible.

That is literally what I did. Change what "About This Mac" says w.r.t. touching up cosmetics (Unknown -> Quad Core label). I don't think it has any tangible effect. Applications (such as GeekBench) detect the correct processor anyways.

Last edited by mseth; May 7, 2014 at 08:39 PM.
mseth is offline   0 Reply With Quote
Old May 8, 2014, 04:13 PM   #1034
Tutor
Thread Starter
macrumors 65816
 
Tutor's Avatar
 
Join Date: Jun 2009
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute
Further Thoughts Concerning SuperMicroMacstrosity - Some Luck Has Its Place;

or Why Might There Currently Be No Self-built Dual E5-2697 System Running OSX To The Hilt1*/;
or Math Too Has Its Place;
or How You've Been Blocked From Doubling In One System What's Been Released As The Best Of The Best, Unless You Figure Out What's Been Done And How To Get Around It;
or They Won't Have To Change CPU Core/Thread Capacity For The Very Next One When the 16 Core Xeons Are Released;
or All Of The Above.

Mathematically, what are snug fitting Sandy or Ivy socks? Luckily, E5-4650 QBEDs fit well, but just not perfectly.

Have you ever worn socks that were too tight? I have and they made my feet feel uncomfortable. Have you ever worn socks that were over stretched or otherwise too big. I have and they kept falling into my shoes. Fit as to socks is important as is fit into max core count if you’re trying to get the maximum in CPU performance from OSX. If I am correct in my belief, as I indicated above in my last few posts, that OSX currently can handle only 32 threads, whether they be 32 real cores or 16 real cores and 16 hyper-threads, regardless of the number of CPUs present, then that has implications for certain other current self-builds. To test my belief further I did this search on Geekbench 3: http://browser.primatelabs.com/geekb...ulticore_score to find out whether I’d see any 2 CPU systems that were identified as being a MacPro6,1 with E5-2697s (12-cores)2*/, and guess what? There were currently 16 pages of them (382 results), but not one of them listed more than 12 real cores, although obviously some of them were Hackintoshes (given their model names and/or core speeds). While it could be that the high price of two E5-2697s has prevented any outfitted systems of that configuration from making their appearance on Geekbench 3, I sort of doubt it and now know that someone other and sooner than me has come up against the 32 thread limit. I.e., I suspect that someone else’s socks were too small because they were trying to fit 24 real cores and 24 additional threads into a 32 core/thread limit (whether they realized that there is a cap and what the cap is and that the cap doesn’t discriminate between real cores vs. hyper-threads, or they just gave up trying without ever figuring out what was the exact dimensions of the problem). As my luck would currently have it in my case, there are not 32 or 16 core Xeon CPUs for 1/2/4 CPU systems and the least amount of cores for any Ivy Bridge Xeon is the 6 cores in the E5-2643 (3.5 GHz base and 3.8 GHz max turbo) that is for up to 2 CPU systems. My CPUs in my 4-CPU WolfPackPrime systems are all 8 cores. Thirty-two is, of course, equally divisible by 32, 16, 8, 4, 2, and 1. Hardly anyone who would experiment, as I did, with a 32 CPU slotted system (32x1=32), a 16 CPU slotted system (16x2=32), or an 8 CPU slotted system (8x4=32). Moreover, there are no 4, 2, or 1 core Ivy Bridge Xeons. But I do know that there are a few owners of 4 CPU slotted systems. So what socks currently fit better than mine for 4 CPU slotted systems? Its the E5-4627 V2s [ http://www.cpu-world.com/CPUs/Xeon/I...4627%20v2.html, but for my four E5-4650s I did spend less a quarter of what four E5-4627 V2s would now cost new and, in fact, for my entire system, fully configured, did I spend less than what four E5-4627 V2s would now cost ]. The E5-4627 V2 is an 8 core CPU has a base frequency of 3.3 GHz (whereas mine have 2.7 GHz) and has a turbo speed of 3.6 GHz (whereas mine have 3.5 GHz [solely because they are the QBED C1 stepping - otherwise it’d be 3.3 GHz]). Since my system gets a Geekbench 3 score of 49,337, I would project that a similarly configured system with 4 E5-4627 V2s would achieve a score of about 60,000+, especially, as to older systems like mine, after having applied the Ivy Bridge Bios Flash and installed faster (1866 MHz) memory : (2.7 GHz x 4 = 10.8 = my systems; 3.3 GHz x 4 = 13.2 = 4xE5-4627 V2s; 13.2 / 10.8 = 1.22; 49,337 x 1.2 = 60,300 ). Moreover, the E5-4627 V2s have no hyper-threads. Thus, there would be no hyper-threads wasted because they wouldn’t fit in the 32 thread cap anyway.




BTW - Looking at systems with a CPU with greater than 8-cores (and admittedly my Geekbench 3 searches, in and of themselves and as worded, are somewhat under-inclusive [see example of under-inclusiveness below in footnote 1*]) -
1) I found one single CPU E5-2696 v2 (12 cores) MacPro6,1 - http://browser.primatelabs.com/geekb...v2+MacPro6%2C1 ;
2) I found not one w/ E5-2695 v2 (12 cores) MacPro6,1 - http://browser.primatelabs.com/geekb...v2+MacPro6%2C1 ;
3) I found two pages of single CPU w/ E5-2690 v2 (10-cores) MacPro6,1 listing a couple of anomalous base speeds - http://browser.primatelabs.com/geekb...v2+MacPro6%2C1 and
4) I found not one w/E5-2680 v2 (10-cores) MacPro6,1 - http://browser.primatelabs.com/geekb...v2+MacPro6%2C1 .


1*/ By using the term "Hilt," I mean that no cores or threads have been turned off, like I and, e.g., he/she obviously had to do with each of those 10 real cores + 10 hyper-threads on the two E5-2690 V2 CPUs - http://browser.primatelabs.com/geekbench3/365335 . Here is a broader search showing a few of those who have Ivy Bridge Xeons and have confronted the 32 thread cap and how they dealt with it: http://browser.primatelabs.com/geekb...ulticore_score .


2*/ Parentheticals containing core counts weren’t part of the search terms.


P.S. - No final released Sandy Bridge Xeons had greater than 8 real cores and no more than one additional hyper-thread per core.
__________________
21 tweaked, multiOS rendering systems - oTitan RD TE > 58 (86,432 CUDA cores). 13,120 ATI Stream PUs. 206 CPU cores. Benches: CB11.5-48.5; CB15-3,791; GB2-58,027; GB3-71,691; LuxMark/Sala-12,330.

Last edited by Tutor; May 8, 2014 at 08:45 PM. Reason: Added P.S. ... . & add URL for info on E5-4627 v2s.
Tutor is offline   0 Reply With Quote
Old May 19, 2014, 06:57 PM   #1035
Rampage Dev
macrumors member
 
Join Date: Dec 2012
Email me and I can make a SSDT for your board to help enable a few things.
Rampage Dev is offline   0 Reply With Quote
Old May 19, 2014, 10:20 PM   #1036
Tutor
Thread Starter
macrumors 65816
 
Tutor's Avatar
 
Join Date: Jun 2009
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute
Quote:
Originally Posted by Rampage Dev View Post
Email me and I can make a SSDT for your board to help enable a few things.
Thanks Rampage Dev. I'll do so.
__________________
21 tweaked, multiOS rendering systems - oTitan RD TE > 58 (86,432 CUDA cores). 13,120 ATI Stream PUs. 206 CPU cores. Benches: CB11.5-48.5; CB15-3,791; GB2-58,027; GB3-71,691; LuxMark/Sala-12,330.
Tutor is offline   0 Reply With Quote
Old Jun 17, 2014, 05:20 AM   #1037
nikezzz
macrumors newbie
 
Join Date: Jun 2014
Have you tried OS X 10.10 Yosemite - I think there's a new kernel (at least it has a new name) maybe it'll support more than 32 cores now
nikezzz is offline   0 Reply With Quote
Old Jun 17, 2014, 02:27 PM   #1038
Tutor
Thread Starter
macrumors 65816
 
Tutor's Avatar
 
Join Date: Jun 2009
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute
Quote:
Originally Posted by nikezzz View Post
Have you tried OS X 10.10 Yosemite - I think there's a new kernel (at least it has a new name) maybe it'll support more than 32 cores now
Not yet.

Quote:
Originally Posted by Rampage Dev View Post
... .
Rampage Dev,

Do you know whether Yosemite supports more than CPU 32 threads, and if so, how many?
__________________
21 tweaked, multiOS rendering systems - oTitan RD TE > 58 (86,432 CUDA cores). 13,120 ATI Stream PUs. 206 CPU cores. Benches: CB11.5-48.5; CB15-3,791; GB2-58,027; GB3-71,691; LuxMark/Sala-12,330.
Tutor is offline   0 Reply With Quote
Old Jun 30, 2014, 06:28 AM   #1039
Tutor
Thread Starter
macrumors 65816
 
Tutor's Avatar
 
Join Date: Jun 2009
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute
Cinema 4d (C4d) V. 15 Team Rendering: Bucketing - A big positive for gigantic frames.

C4d Team Rendering makes multiple fast, low Vram GTX cards shine when rendering large frames.

I have five systems that each have four GTX 590s in each of them. On the positive side, a fast, low Vram GTX Fermi card such as the GTX 590 (which is roughly the equivalent of two GTX 570s) can be had for under $400 on Ebay and in Octane render a GTX 590 renders faster than the oGTX Titan. But on the negative side the advertised 3G of Vram for the GTX 590 is actually 1.5G per card (it has two of them), which seemingly is a real bummer for rendering a large, complex frame, such as a 4k frame. It will not fit into the Vram of a single GTX 590.

C4d Team Rendering and the Octane render C4d plugin have, however, come to the rescue. The Team Render feature in C4d, which has built-in network rendering that supports Octane render, is described, in summary, by Maxon as follows:

"Team Render is an all-new network rendering concept that uses peer-to-peer communication to distribute render tasks. Because there's no bottleneck at a central server, the assets required to render your scene get to each client quicker. This means clients spend time rendering instead of waiting for assets.
Utilize the power of your team to easily render the frames of an animation or to distribute the buckets of a still frame. Machines share the workload to calculate caches for Global Illumination, Ambient Occlusion and Subsurface Scattering so you can utilize advanced rendering features and still meet the deadline."

What this means is that if you set up bucket rendering in C4d while using Octane as the renderer, then the scene gets split up. So the more GTX systems that you have participating in the Octane render, the smaller the scene data sent to your GPUs. Just to be on the safe side when I recently rendered a 4k 3d demo, I used four systems, each with four GTX 590s, participating in the render. The render job was accomplished with every GPU card having lots of memory to spare throughout the render.

P.S. Serendipity and the Samsung 28-Inch Ultra High Definition LED Monitor are wonderful.
__________________
21 tweaked, multiOS rendering systems - oTitan RD TE > 58 (86,432 CUDA cores). 13,120 ATI Stream PUs. 206 CPU cores. Benches: CB11.5-48.5; CB15-3,791; GB2-58,027; GB3-71,691; LuxMark/Sala-12,330.

Last edited by Tutor; Jun 30, 2014 at 07:14 AM. Reason: added PS ... .
Tutor is offline   0 Reply With Quote
Old Jun 30, 2014, 08:54 AM   #1040
sirio76
macrumors regular
 
Join Date: Mar 2013
What is the content of your 4K demo? How many polygons(real, non instance) and texture did you use in your scene?
sirio76 is offline   0 Reply With Quote
Old Jul 3, 2014, 11:58 PM   #1041
Tutor
Thread Starter
macrumors 65816
 
Tutor's Avatar
 
Join Date: Jun 2009
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute
Quote:
Originally Posted by sirio76 View Post
What is the content of your 4K demo? How many polygons(real, non instance) and texture did you use in your scene?
Automotive parts manufacture; textures = 54; and polygons = 2,672.
__________________
21 tweaked, multiOS rendering systems - oTitan RD TE > 58 (86,432 CUDA cores). 13,120 ATI Stream PUs. 206 CPU cores. Benches: CB11.5-48.5; CB15-3,791; GB2-58,027; GB3-71,691; LuxMark/Sala-12,330.
Tutor is offline   0 Reply With Quote

Reply
MacRumors Forums > Apple Hardware > Desktops > Mac Pro

Tags
over clocking, overclocking, swap cpu, under clocking, underclocking

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Similar Threads
thread Thread Starter Forum Replies Last Post
maximizing CPU and other resources mbp150 OS X Mavericks (10.9) 3 Nov 10, 2013 07:19 PM
A7 performance - equivalent desktop cpu? heifetz7 iPhone 21 Oct 4, 2013 12:07 PM
Preview in ML, maximizing without maximizing picture, change sidebar side LeandrodaFL Mac Applications and Mac App Store 0 Sep 3, 2013 06:17 PM
Anandtech's review on RMBP 13 up, no surprises on CPU performance magbarn MacBook Pro 22 Nov 14, 2012 12:58 PM
Slow performance on MBP - Flash related (I think) bsod999 MacBook Pro 1 Sep 3, 2012 05:27 AM

Forum Jump

All times are GMT -5. The time now is 05:47 PM.

Mac Rumors | Mac | iPhone | iPhone Game Reviews | iPhone Apps

Mobile Version | Fixed | Fluid | Fluid HD
Copyright 2002-2013, MacRumors.com, LLC