All We Know About Maximizing CPU Related Performance

LBattis said:

... .RAM has me in a bit of a conundrum. Speed/Timing/Latency . . . I am not fully adept at manipulating these variables. Among others I am looking at the following:

$229.99 CORSAIR Dominator Platinum 16GB (2 x 8GB) 240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Desktop Memory Model CMD16GX3M2A1600C7
Cas Latency: 7
Voltage: 1.5V
Multi-channel Kit: Dual Channel Kit
Timing: 7-8-8-24
Model #: CMD16GX3M2A1600C7
_________

$182.99 CORSAIR Vengeance Pro 16GB (2 x 8GB) 240-Pin DDR3 SDRAM DDR3 2400 (PC3 19200) Desktop Memory Model CMY16GX3M2A2400C11R
Cas Latency: 11
Voltage: 1.65V
Multi-channel Kit: Dual Channel Kit
Timing: 11-13-13-31
Model #: CMY16GX3M2A2400C11R
_____

My sense of it is that the Dominator DDR3 1600, might well outperform the Corsair DDR3 2400 and be worth its price (which I will shop harder). Is there any good rule of thumb or hard-equation for solving this?

Click to expand...

I recommend the Vengeance Pro over that particular Dominator ram. Your CPU will support that Dominator ram, but it also supports 1866 MHz ram. That Dominator won't get you to that 1866 MHz Ivy Bridge standard because that particular Dominator ram is 1600 MHz ram.

Ram timing and latency tend to increase as MHz increases to help prevent errors. Using 2400 MHz ram (like that Vengence ram) means that you can clock the ram (as I have down with my Dominator Platinum) at 2133 MHz in bios and still have some overhead to tweak its first few {3-5} parameters in bios (lowering their values to two steps above where stability begins to suffer) to get the benefit of faster timing and lower latency. Moreover, if you have the funds, you could get (what I use) the Dominator Platinum [16 GB - http://www.newegg.com/Product/Product.aspx?Item=N82E16820233347 or 32 GB - http://www.newegg.com/Product/Product.aspx?Item=N82E16820233394 ], but it is more expensive. Although I hope that you know that I'll help you to tune whatever ram that you get, the Dominator Platinum would tend to make it a bit easier since that's the ram I have in a number of my systems. Still, I have no significant reason to caution you against getting that Vengeance ram (perhaps to save some cash) and, as always, my offer of assistance to help you to tune it still stands.

LBattis said:

To answer your question about the 21st/18th Century comparison. Two quick examples that come to mind are the increasing civil unrest and questions about the viability of the food supply. My mind is prone to wandering off in any variety of directions. This is just something I was musing the other night.

-III

Click to expand...

Nothing's wrong with that mind of yours. Those are two excellent examples. Plus, musing (and dreams) so long as you remember/record them -> creativity.

DJenkins said:

The NHD14 is awesome but ridiculously huge.
I too agree with avoiding custom water loops, just for the difficulty it would cause if you need to swap components out for replacement or testing etc.

I have however moved over to an all in one Corsair H80i which performs just under that of the Noctua but takes up way less space.
It is slightly noisier as well because of the pump - and the Noctua is just really well designed.

So are you saying the Noctua will block one of your PCI slots? You would have to consider what it might cost in limiting future upgrades. A lot of video editing platforms are starting to leverage dual GPUs - what software are you using?

As far as RAM goes I think frequency might have more of an effect over timings… 1600 is getting a bit outdated and leaves no headroom. Something higher like 1866 can be enabled in BIOS using XMP profiles without the possible headache of overclocking.

I think tightening the timings of your ram is a bit easier than over clocking past it’s rated frequency.

Anyway, that’s just my opinion and somewhat limited experiences. Tutor has been an amazing guide for me and will correct anything I’ve been mistaken on and without doubt add a whole bunch of other useful info!!

It would help to know exactly what kind of video work you are doing and with what software. I spent a lot of time working out the best way to handle editing & monitoring, After Effects, DaVinci grading, Cinema 4D, Octane rendering and ProTools audio production all in one box running OSX. So I definitely agree with your title of ‘Balancing Act’ and hopefully I can help in some way as well

Click to expand...

Hello DJenkins,
I agree with you on all points. Also, I too use the Corsair H80i for cooling almost all of my systems and it does not block any PCI-e slots. I'm using the same motherboard that he's planning to purchase. The version of its bios that I'm using (which is not the latest one - simply haven't taken the time to upgrade all 8 of my GUP4 systems yet and my habit is to do all such upgrades at once) allows you to select speeds much higher than 1866 MHz if memory's top speed allows it.

Future Proofing My A**

When we use the phrase,” future proofing” in connection with a computer system, what do we truly mean? I don't believe that any computer system can withstand time's march taking giant steps. The 4 core/ 8 thread i7-975 CPU was discontinued by Intel in 2011 [ http://www.cpu-world.com/CPUs/Core_...tion I7-975 AT80601002274AA (BX80601975).html ] (about 3 years ago). Can anyone guess the approximate multi-core Geekbench 3 score of my i7-975 system? I built that system in the fall of 2009 (almost 5 years ago) and tested it with Geekbench 2 on December 21, 2011 [ http://browser.primatelabs.com/geekbench2/532301 ]. It then got a score of 15,101. I have yet to test it with Geekbench 3. That system is one of my 21 tweaked/multiOS rendering systems.

This is how the 4 core/8 thread E5-1620 systems now stack up in Geekbench 3 [ http://browser.primatelabs.com/geekbench3/search?dir=desc&q=e5-1620&sort=multicore_score ]. The highest muticore score is listed first. It's 16,743 as of 4:13 p.m. (CST) on August 22, 2014.

The Illusion Of Future Proofing

Future proofing is an impossibility. In 2009, my i7-975 system (aka “WolfCub1”) scored much higher in Geekbench 2 than other 4-core systems, then achieving a Geekbench 2 CPU score of 15,101. That placement is no longer the case under Geekbench 3.

I’ve attached a pic of how my WolfCub1 now performs in Cinebench 15. I also wanted to test the GPU that I had originally placed in it, i.e., a 2009 PowerColor 4890, so I removed my CUDA GPUs and installed that 4890 video card. Among other things, my 4890 video card attained an OpenGL score in Cinebench 15 of 66.49 fps and the i7-975 attained a CPU score is 685. On August 23, 2014 in Geekbench 3 it achieved a single-core score of 3,361 and a multi-core score of 13,157 [ http://browser.primatelabs.com/geekbench3/723996 ]. Using the ranking search (today consisting of 116 pages of scores) that I set forth in my last post (# 1054, above), my WolfCub1 would be listed today on page 11 on those 116 pages of scores if the scores are ranked by the highest single-core score being listed first (and, of course, if the i7-975 was named "E5-1620"). My WolfCub1 would be listed today on page 16 of those 116 pages if the scores are ranked by the highest multi-core score being listed first. If I narrow the search to list only E5-1620 v2 systems, my WolfCub1 would be listed today on page 11 on those 110 pages of scores if the scores are ranked by the highest single-core score being listed first. My WolfCub1 would be listed today on page 15 of those 110 pages of scores if the scores are ranked by the highest multi-core score being listed first. Thus, a four-core system that once achieved top Geekbench scores no longer has its no.1 ranking, providing a real life example showing that future proofing is truly an impossibility, unless by using the phrase “future proofing,” one really means - “short-term future proofing.”

What One Does With Computing Technology Affects The Illusion of Future Proofing

Hello Aiden,

Your posts (because of their insights, relevance and sound bases) are always welcomed greatly.

AidenShaw said:

I run from $500K to $1M per year for hardware spend at work. We'll soon be dropping E5-x6xx v2 systems (the only ones in the MP6,1) to the "near obsolete" category - because by the end of the calendar year we'll have dozens of E5-x6xx v3 systems with AVX2. AVX2 will be huge for our projects, and nobody will want to use the current generation of Xeons once they arrive.

[Much less important is raw CPU speed on non-SIMD code - since that's been rather slowly improving.

New instructions (like AVX) are important - because depending on the app that can make newer systems much faster than generic benchmarks predict.]

The current Xeons are doomed - totally not "future-proof" at work.

Click to expand...

I wish that I had the budget for the hardware that you run! For now, I can only dream about the systems that your posts describe.

I agree with you completely that the sauce that Intel has been basting on it’s CPUs (new feature / instruction sets like AVX and their further/future advances) means that CPU advances have and will have something tasty for making future purchases inevitable. Because I apply my systems to tasks such as 3d and video rendering, my new/later purchases are, however, supplements, not replacements. That's dictated by what I do with my systems.

AidenShaw said:

Personally, I bought a Dell T3610 with the E5-1650 v2 CPU (same hex that Apple uses in the MP6,1) in April. For every system that I've personally purchased - the only reason for replacing it was that it could not support the amount of RAM that I needed.

The T3610 (which I bought with 80 GiB - supports 128 GiB and should support 256 GiB when the 32 GiB DIMMs are here) replaced a Core i7-940 system that was maxed at 24 GiB. The Core i7-940 replaced a Q6600 that maxed at 8 GiB. The Q6600 replaced a Pentium III system that maxed at.....which replaced a Toshiba Satellite Pro that maxed at 40 MiB (8 MiB soldered and a 32 MiB SO-DIMM).

Get the picture?

Click to expand...

You've painted that picture clearly and beautifully. Concerning CPUs (whether one purchases or builds a system), future significant CPU growth potential is significantly dictated by whether the initial purchase is for a Tock or Tick model (and, of course, the motherboard and its bios) because of looming socket changes. Unfortunately concerning upgrade potential, the nMP was released just before a looming socket change.

AidenShaw said:

The most important part (IMO) of future-proofing is that you can add memory. The MP6,1 is a big fail here, with only four slots.

Click to expand...

The nMP requires me to make too many compromises in lots of important areas. For what I do, advances in GPU compute technology have the greatest ability to give me somewhat of an illusion of future proofing. The ability to add more and faster GPUs offers advantages that help to minimize the limitations inherent in Intel’s certain changes in socket design. GPU computing means that I can continue to make past systems faster compute units because newer GPUs, although built to accommodate faster PCIe technology {take PCIe V3 for example}, can still be harnessed to provide faster render units in systems with even PCIe V1 or V2 slots. So I vote with my dollars, in order of importance to me, for (a) more powerful GPU processors, including faster and larger memory amounts on those GPUs, (b) more PCIe slots for those GPUs, (c) faster, more powerful, and more CPUs offering the most advanced feature sets, (d) faster, larger capacity and more numerous memory units and, of course, more memory slots for the motherboard and (e) faster and larger capacity storage units. Costs, as always in my case, can be deal breakers.

AidenShaw said:

Notice how things like wristwatches are called "water resistant" instead of "waterproof"?

All we can do with computers is "future resistant". All computers will become obsolete eventually - but a "future resistant" computer have a longer useful life before it "hits the wall". (I realize that an old computer is just as fast on old software as when it was purchased - but most of us upgrade applications and add applications, and those often assume a more modern, more powerful system.)

WolfCub1 has lasted 5 years - that's pretty good. If you had purchased the Core i7-940 like I did, you probably would have replaced it earlier. (I replaced the Core i7-940 not because it was slow, but because I hit the wall with only 24 GiB of RAM.)

Click to expand...

Aiden,
Love it! A very close analogy and a lot more accurate phrasing - "future resistant." That nails it. I'm burning that phrasing into my grey matter right now. Thanks again for sharing your wisdom.

P.S. I replace things too when they no longer meet my needs.

Future Resistant Nehalems/Westmeres

More Geekbench 2 Results Shown By Ranking Scores By System CPU


1) Achieving the highest i7-975 listed system score is my WolfCub2*/ (my only system of this kind), then running OSX.
http://browser.primatelabs.com/geekbench2/search?dir=desc&q=i7-975&sort=score
http://browser.primatelabs.com/geekbench2/532301

2) Achieving the highest W5580 listed system score is my WolfCub1 (my only system of this kind), then running OSX.
http://browser.primatelabs.com/geekbench2/search?dir=desc&q=W5580&sort=score
http://browser.primatelabs.com/geekbench2/418647

3) Achieving the highest i7-980X listed system score is my WolfPack11, then running OSX. My other two similarly configured and tweaked systems (WolfPacks12 & 13) have not been tested.
http://browser.primatelabs.com/geekbench2/search?dir=desc&q=i7-980x&sort=score
http://browser.primatelabs.com/geekbench2/482108

4) Achieving the highest 2xX5680 listed system score is my WolfPack1, then running OSX,
a) http://browser.primatelabs.com/geekbench2/search?dir=desc&q=x5680&sort=score
a) http://browser.primatelabs.com/geekbench2/500630
a) http://browser.primatelabs.com/geekbench2/334835

and next is my WolfPack2**/, then running OSX,
b) http://browser.primatelabs.com/geekbench2/search?dir=desc&q=x5680&sort=score
b) http://browser.primatelabs.com/geekbench2/500492
b) http://browser.primatelabs.com/geekbench2/416919

and my 2009 —> 2010 MacPro scores higher than all other 2xX5680 listed real MacPro systems (note that a real MacPro with 2xX5680s runs at 3,333 Mhz). My hacked, real 2xX5680 MacPro also scores higher than lots of 2x5690 listed real MacPro systems (note that a real MacPro with 2xX5690s runs at 3,460 Mhz).
c) http://browser.primatelabs.com/geekbench2/search?dir=desc&q=x5680&sort=score
c) http://browser.primatelabs.com/geekbench2/422811
c) http://browser.primatelabs.com/geekbench2/search?dir=desc&page=5&q=x5690&sort=score

WolfPacks 3-10 are seven 8-core and one 6-core SandyBridge systems. WolfPacks14-16 will soon be three similarly configured and tweaked W3670 systems.

*/ Here’s how I now classify my systems: A WolfCub has fewer than 6 real cores; a WolfPack has 6 to 24 real cores; a WolfPackAlphaCanisLupus has at least 12 cores and 8 internal double wide GPUs; and a WolfPackPrime has at least 26 real cores (My WolfPackPrime0 and WolfPackPrime1 each have 32 real cores).

**/ The Geekbench 2 scores for WolfPack1 and WolfPack2 are also higher than the highest listed X5690 score of any system running any OS [ http://browser.primatelabs.com/geekbench3/search?dir=desc&q=x5690&sort=multicore_score ].


BTW - All of the GPUs in my self-builds have also been tweaked for maximum performance: OpenCL (with the help of LuxRender to do so in the case of my AMD/ATI GPUs), CUDA (with the help of Cinebench to do so in the case of my GTX GPUs) and OpenGL (with the help of Cinebench to do so for all of my GPUs). I’ve found that I get the best performance from the CPUs and GPUs in all of my systems by using cooler air conditioning settings and additionally (1) in the case of my cMacPros (a) using SMC fan control and (b) flipping the system by 45 degrees (face down - see pic, below) and (2) in the case of my self-builds using Corsair 80s/100s H20 coolers and SilverStone Raven cases, to keep each system’s internals as cool as possible. The SilverStone Raven case orientation of fans and GPUs help to promote better cooling. I'm trying to mimic it in the pic below. I used support helpers - two of my two firewire drive cases in the pic, below. The Raven's supports, handles and front panel access points are better placed/oriented to pull this off elegantly/gracefully (see 2nd pic). Cooler air comes in at the bottom of the system and warmer air completely exits the system faster because warmer air wants to rise. Doesn't this thermal design also mimic the nMP and nature (cooler air is lower than warmer air) ? Also, that black cable coming out the back of my MacPro, powers the FSP BoosterX5 (in the top optical drive bay - see, also, post #972 here - https://forums.macrumors.com/threads/1333421/ ) that helps to power the GTX 480 and GTX 590, now in each of my three 2007 MacPros.

Flipping a MacPro 2006-2012

Similar to the increased cooling benefit of the SilverStone Raven [ See further discussion - here: https://forums.macrumors.com/showthread.php?p=19517364&redshiftrender#post19517364 ] is the benefit achieved by setting the cMacPro with the face oriented to the floor (using supports for either end, of course). Also note that this is very similar to the cooling system works in the nMP and in a SilverStone Raven case. Having done this to my 2007 MacPro resulted in the following during Cinebench15 testing:
1) Idle temperature drooping from an average of 36 degrees C to a max temp of 31 degrees C during idle;
2) a consistent increase of the OpenGL score by 1.5 frames per second on average (temps did not rise during OpenGL testing);
3) a consistent increase in the CPU (multi-core) score by 8 to 12 CBs (System temp now rises by an average of only 3 degrees during testing - max temp was 34 degrees C during CPU (multi-core) tests );
4) a consistent increase in the CPU (single-core) score by 1 to 2 CBs (System temp now rises by an average of only 1 degree during testing - max temp was 32 degrees C during CPU (single-core) tests); and
5) a consistent MP ratio increase from an average of 7.1x to to an average of 7.36.

I would expect that the effect of flipping the case on CPUs in a 2009-2012 cMP would be greater because the CPUs in those systems have turbo boost potential and one of the factors affecting if, how much and how frequently turbo boosting occurs, is the temperature of the CPU's cores. I haven't tested this yet on my MacPro 4,1 -> 5.1 because I've removed that system's motherboard to do some work on it. But when I test it, I'll post an update of my findings.

LBattis said:

As it goes, the morning after I ordered the video etc, I find a price on the i7-5930K. I did the swap w/ Amazon this morning. I was nice to be inside the 30-day envelope on that development.

-III

Click to expand...

The i7-5930k [ http://ark.intel.com/products/82931 ] requires a 2011-v3 motherboard (unlike the Sandy and Ivy Bridge Es which use the plain 2011 motherboard). I've been considering the following two motherboards:

(1) http://www.evga.com/Products/Product.aspx?pn=151-HE-E999-KR (if you intend to over clock it significantly) and
(2) http://www.newegg.com/Product/Product.aspx?Item=N82E16813128751 (if you intend to over clock it minimally);

and, at least, 32 GB of DDR4 2800 (PC4-22400) or faster memory.

Old dogs can not only be taught new tricks, they can master them and devise new ones.

LBattis said:

... . My age ... make that a non-starter. It was me 2 decades ago; not now. ... .

-III

Click to expand...

All the reasons that you give, except this one, I fully understand. I sprang forth in the early 1950's. For me, advanced age is but an asset in these endeavors - acquiring knowledge and wisdom take time. Moreover, if patience is a virtue, then age makes us virtuous for we tend more to take things in stride.

Some Recent i7-5690X Reviews

1) http://www.legitreviews.com/intel-core-i7-5960x-8-core-haswell-e-processor-review_149560
2) http://www.tomshardware.com/reviews/intel-core-i7-5960x-haswell-e-cpu,3918.html
3) http://www.anandtech.com/show/8426/...review-core-i7-5960x-i7-5930k-i7-5820k-tested
4) http://thepcenthusiast.com/intel-haswell-e-features-specs-price/

A Sample i7-5690X System

An Under $6,500 Sample i7-5690X Highly Overclockable 8-core System with 64 GB of DDR4 Ram, 512 GB PCIe SSD and 4xCUDA GTX 780 Ti GPU Rendering Capability (w/o Keyboard)

1) SilverStone RAVEN Series RV03B-W Matte Black Steel / Plastic ATX Full Tower Computer Case [ http://www.newegg.com/Product/Product.aspx?Item=N82E16811163180 ]
1 IN STOCK
$159.99
$139.99
Save: 12.50%
Subtotal: $139.99
Shipping Options
$9.99
Super Eggsaver (4-7 Business Days)
$12.99
Newegg 3 Business Days


2) EVGA 03G-P4-2888-KR G-SYNC Support GeForce GTX 780 Ti Dual Classified/ EVGA ACX Cooler 3GB 384-bit GDDR5 PCI Express 3.0 [ http://www.newegg.com/Product/Product.aspx?Item=N82E16814487007 ]
4 IN STOCK
$759.99 ea.
$699.99 ea. (4x = $2799.96)
Save: 7.89%
$30.00 Mail-in Rebate Card


3) EVGA 120-G2-1600-X1 1600W ATX12V/ EPS12V SLI 80 PLUS GOLD Certified Full Modular Power Supply Intel 4th Gen CPU Compatible PSU [ http://www.newegg.com/Product/Product.aspx?Item=N82E16817438033 ]
1 IN STOCK
$349.99
$344.99
Save: 1.43%
$15.00 Mail-in Rebate Card


4) G.SKILL Value Series 32GB (4 x 8GB) 288-Pin DDR4 SDRAM DDR4 2133 (PC4-17000) Desktop Memory Model F4-2133C15Q-32GNT [ http://www.newegg.com/Product/Product.aspx?Item=N82E16820231785 ]
1 IN STOCK
$469.99 ea. (2x = $939.98)


5) EVGA X99 Classified (151-HE-E999-KR) LGA 2011-3 Intel X99 SATA 6Gb/s USB 3.0 Extended ATX Intel Motherboard [ http://www.newegg.com/Product/Product.aspx?Item=N82E16813188163 ]
1 IN STOCK
$414.99

6) GIFT #2 EVGA TORQ X10 901-X1-1103-KR Wired Laser Gaming Mouse [ http://www.newegg.com/Product/Product.aspx?Item=N82E16826188002 ]
1 IN STOCK
$89.99
$0.00
Save: 100.00%

7) Intel Core i7-5960X Haswell 8-Core 3.0GHz LGA 2011-3 140W Desktop Processor BX80648I75960X [ http://www.newegg.com/Product/Product.aspx?Item=N82E16819117404 ]
1 IN STOCK
$1,049.99

8) Microsoft Windows 7 Ultimate SP1 64-bit - OEM [ http://www.newegg.com/Product/Product.aspx?Item=N82E16832416808 ]
1 IN STOCK
$189.99


9) CORSAIR Hydro Series H80i High Performance Water/Liquid CPU Cooler 120mm [ http://www.newegg.com/Product/Product.aspx?Item=N82E16835181031 ]
1 IN STOCK
$114.99
$94.99
Save: 17.39%


10) Plextor M6e PX-G512M6e M.2 2280 512GB PCI-Express 2.0 x2 Internal Solid State Drive (SSD) [ http://www.newegg.com/Product/Product.aspx?Item=N82E16820249047 ]
1 IN STOCK
$459.99
$419.99
Save: 8.70%

Subtotal: $6,254.88
Shipping Options

Shipping Options

SAVE $9.04 TODAY AND GET FREE 3-DAY-OR-LESS SHIPPING ON YOUR ORDER (ELIGIBLE ITEMS ONLY) TODAY WHEN YOU SIGN UP WITH NEWEGG PREMIER. Learn More
Choose a delivery method What's this?
$9.04 Newegg 3 Business Days
$90.68 Newegg 2 Business Days
$91.92 Newegg Next Business Day
free 2-day shipping & free returns learn more | sign in
Shipping: $9.04
Newegg Digital Download Delivery Service Important Shipping Information
NVIDIA Gift - Borderlands: The Pre-Sequel (Up to $60 Retail Value)
Gift Item GIFT #1
NVIDIA Gift - Borderlands: The Pre-Sequel (Up to $60 Retail Value)
Digital Deliver Gift Card/Credit Return Policy
4 IN STOCK $239.96 /$0.00/ Save: 100.00%
($0.00 ea.)
Choose a delivery method
FREE Online Services Shipping: $0.00

No Payments + No Interest if paid in full in up to 12 Months. Minimum purchase required.
Subject to credit approval. See Terms
Grand Total: $6,413.90

Reduce the number of GTX 780 Ti to one and system ram to 32 GB takes total price to under $3,900.

P.S. - One could also save an additional ~ $154 by using this motherboard - [ http://www.newegg.com/Product/Product.aspx?Item=N82E16813128751 ]. Gigabyte motherboard's are renowned for the multi-OS capabilities, but see post # 1064, above. Also, large format video editors may want to add a couple or three or more of these Plextor M6e PX-AG512M6e PCI-E 512GB PCI-Express 2.0 x2 Internal Solid State Drives [ http://www.newegg.com/Product/Product.aspx?Item=N82E16820249044 ] for $460 each; or in a mix of 3d rendering and video environment (using the EVGA motherboard), add three of the above Plextor cards to the motherboard based Plextor M6e PX-G512M6e M.2 2280 512GB PCI-Express 2.0 x2 Internal Solid State M.2 Drive and two GTX 780 Ti's in the two outer PCIe slots, yielding > 2,000 MBps writes and > 2,500 MBps reads in a 2T raid 0 array + the Octane rendering capability of 2.64 original reference design Titans, plus whatever other SATA 6 storage needed (EVGA motherboard supports 10). However, the Gigabyte board has a 1 x Thunderbolt add-in card connector.

LBattis said:

And the 'future resistance' is covered nicely.



So my question here centers on the present. Are these platforms that will support OSX? That's where I can't afford to spend time hammering out collimation. If so I have time and enthusiasm to make some transitions. If not then I need to step back into the hardware set that runs the environment at present.

Thank-you for that splendid review.

-III

Click to expand...

Given what you desire, I suggest that if you want a system that's the easiest on which to deploy and upgrade OSX, that you get a new MacPro. Building a system and deploying various OSes on it is a process that requires study, time and patience. If you feel that you can expend the time, want to learn a lot about how MacPro hardware and OS interface to deploy that learning to another system, and have and maintain patience, then you can build a custom system. But, I'd be remiss if I didn't let you know that the learning process has to continue for the life of your system unless you never further upgrade its hardware and OS. To have AND TO KEEP a Hackintosh requires that you have and maintain some C. Columbus and R. Amundsen in you, i.e., having a Hackintosh is a whole lot easier than maintaining it over time.

A Haswell system would offer the best performance. However, since the high performance systems have only recently been introduced, it appears to be better for you to work on building an Ivy or Sandy Bridge system if you prefer not to buy a new MacPro because on the earlier Windows systems the deployment wrinkles have been ironed out. That is not to say that for one who takes this path that the issues commonly involving OS upgrades will not arise and cause frustration. But, that is not very different from what happened to many of us who own early (pre-2008) MacPros, when Apple decided to deprecate the pre-2008 MacPros for Mavericks (and subsequent OS) installation. On pre-2008 MacPros, many users who want to run Yosemite on them are, possibly without even realizing it, using the some of the same tools and methods to run Yosemite on those early MacPros, that are used to install the OS on Hackintoshes. Moreover, a significant number of MacPro owners have, over time, upgraded their MacPro's ram, storage, CPUs, GPUs and PSUs and, in those processes, have had to do most of the tasks that are involved in the physical process of building a computer system. So even over the long haul, you could be forced to make a real MacPro future OS resistant by using the same techniques and tools for hardware and OS upgrades that Hackintosh builders use.

With that in mind, the X79 Gigabyte motherboards (and Sandy and Ivy Bridge CPUs that those motherboards accommodates) offer the easiest route to building a system that is customized and runs the OSes which you mention. I suggest that you read as much as you can on RamPageDev's website [ http://rampagedev.wordpress.com ] before you spend anymore money to get a better idea about what's involved. Although Rampaged is partial to Asus X79 motherboards, his site has resources for building most of the most popular X79 systems, including systems based on the Gigabyte X79 motherboard that I referenced earlier.

Regarding the X99 motherboards (and Haswell CPUs that they accommodate), now is too early to expect that a relatively painless/easy build can be accomplished fully successfully. You'd have to be hardcore to go that route. If you're hard core, just let me know.

The benefit of building your own system is that each subsequent build becomes easier. Your ability to trouble shoot/repair/upgrade your system will grow greatly. For me, it's been truly liberating.

The 18-core Xeon E5-2699 v3 Is A Multicore Rendering Monster

See [ http://www.chiploco.com/xeon-e5-2699-v3-dual-cpu-benchmarks-35897/ ] and to top it off it's clock tweakable [ http://www.chiploco.com/haswell-ep-xeon-e5-2699-v3-benchmarks-35799/ ]. Also see my post #67 here: https://forums.macrumors.com/showthread.php?p=19561564#post19561564 . They may need to be underclocked properly to yield their highest level of performance. Thus, we may have a new, but greatly more expensive, option for my Sample i7-5690X System layed out in post no. 1068, above. One of them costs a mere $4,058.

Update - A pair of them achieves Geekbench 3 scores of 3,354 (Single-Core) and 80,023 (Multi-Core) [ http://browser.primatelabs.com/geekbench3/662846 ].

FurryBall

FurryBall [ http://furryball.aaa-studio.eu/aboutFurryBall/index.html ] is a 3d GPU renderer that is said to be up to 15x to 100x times faster than a CPU renderer. It supports Maya, 3ds Max and Cinema 4d on Windows (and is said to be Mac OSX compatible with dual boot [i.e., bootcamp]). I've downloaded the demo and it seems very promising. Additionally, like OctaneRender, FurryBall has a benchmark test.

My FurryBall 4.8 Benchmark results can be found at - http://www.aaa-studio.cz/furrybench/benchResults4.php?

Here are my test IDs, corresponding cards tested, high scores and tweaks:
1) ID 4801 - 7 GXT Titans (6.8/11.06/76.33 secs)(memory overclock = 100 MHz);
2) ID 4800 - 7 GXT Titans 6.91/11.28/90.55 secs)(No OC);
3) ID 4797 - 4 GTX 780 Ti (5.73/13/97.45 secs)(memory overclock = 75 MHz);
4) ID 4796 - 4 GTX 780 Ti (7.74//)(memory overclock = 100 MHz);
5) ID 4795 - 4 GTX 780 Ti (5.91/15.78/119.18 secs)(No OC);
6) ID 4794 - 4 GTX 780 Ti (5.76//)(memory overclock = 25 MHz); and
7) ID 4792 - 4 GTX 780 Ti (5.83/10.12/75.92 secs)(memory overclock = 50 MHz);

OverAll Best Time Ranking
1) GTX 780 Ti - 1st Place = ID 4792 - 4 GTX 780 Ti (5.83/10.12/75.92 secs)
2) GTX TITAN - 2nd Place = ID 4801 - 7 GXT Titans (6.8/11.06/76.33 secs)

Detailed Benchmark Results -
Rasterization
avg best
1) 4797 - 2nd Place 4797 - 2nd Place
4xGTX780Ti
2) 4796 - 3rd Place 4796 - 3rd Place
4xGTX780Ti
3) 4794 - 5th Place 4794 - 4th Place
4xGTX780Ti
4) 4792 - 6th Place 4792 - 6th Place
4xGTX780Ti
5) 4795 - 7th Place 4795 - 7th Place
4xGTX780Ti
6) 4801 - 24th Place 4801 - 24th Place
7xGTXTitans
7) 4800 - 25th Place 4800 - 26th Place
7xGTXTitans

Detailed Benchmark Results -
Raytracing
avg best
1) 4792 - 1st Place 4792 - 1st Place
2) 4797 - 3rd Place 4801 - 2nd Place
3) 4800 - 4th Place 4800 - 3rd Place
4) 4801 - 5th Place 4797 - 5th Place
5) 4795 - 7th Place 4795 - 6th Place
6) 4796 - Not tested ————————————
7) 4794 - Not tested ————————————

Detailed Benchmark Results -
High Raytracing
avg best
1) 4792 - 1st Place 4792 - 1st Place
2) 4801 - 2nd Place 4801 - 2nd Place
3) 4800 - 4th Place 4800 - 4th Place
4) 4797 - 5th Place 4797 - 5th Place
5) 4795 - 7th Place 4795 - 7th Place
6) 4796 - Not tested ————————————
7) 4794 - Not tested ————————————



Analysis to follow.

You'd better watch out. Hopefully, Santa will be coming to town - mine.

Sometimes, it's the little things that matter. This might be one of those special times -

Santa Claus,
Please come to me, dropping off a SuperMicro GPU SuperWorkstation 7048AX-TR (mind you - I know that it doesn't yet exist, but, of course, you're Santa with those special powers) and the additional components that I've listed below.

This is what does now exist:
1. Supermicro makes servers optimized for extreme low-latency, designed for day traders:

A. SuperWorkstation 7047AX-TRF
http://www.supermicro.com/products/system/4U/7047/SYS-7047AX-TRF.cfm and

B. SuperWorkstation 7047AX-72RF
http://www.supermicro.com/products/system/4U/7047/SYS-7047AX-72RF.cfm $1,757.99 + $11.99 Shipping @ http://www.superbiiz.com/detail.php?name=SY-74772RF .

See, also:
a) http://www.supermicro.com/products/nfo/files/Hyper-Speed/Low-Latency-White-Paper.pdf .
b) http://www.supermicro.com/products/nfo/files/Hyper-Speed/f_Hyper-Speed.pdf .
c) http://www.supermicro.com/products/nfo/Hyper-Speed.cfm .


Both of these systems could be overclocked by up to 1.3X if the CPUs would allow it. The rub is that the E5-2600 V2 Xeons that they support, can be overclocked at most by 1.0755x. Also, note that those two V2 systems have an "AX" in their modeling. That's crucial because the DAX motherboards (that's where the "AX" comes from in the bare bone system modeling) allow clock tweaking of dual Xeon systems much better.

Supermicro does make a barebones system for E5 2600 V3s (although it'll require that I use some of that one hundred thousand check that you'll also be dropping off (to me) to pay for some electrical rewiring, given that 2000W+ requirement for the completed system). That particular barebones system that I have in mind, will make a perfect home for four of those new Titan Blacks, one terabyte (16x64) of DDR4 2133+MHz ram, twelve Samsung 845DC EVO Series 960GB 2.5 inch SATA3 Solid State Drives, and two E5-2699 V3 Xeons that you'll also be bringing to me. The high-end V3 Xeons can be overclocked a whole lot more than the V2s. The high-end V3s have much higher bins and they're unlocked.

This is Supermicro's summary of a recent barebones system (but watch the model no. carefully, more on that later) most like the one you'll be hauling to me:

GPU SuperWorkstation 7048GR-TR -
1. Dual socket R3 (LGA 2011) supports IntelÆ XeonÆ processor E5-2600 v3 family; QPI up to 9.6GT/s,
2. Up to 1TB ECC DDR4 2133MHz; 16x DIMM sockets,
3. 4x PCI-E 3.0 x16 slots, 2x PCI-E 3.0 x8 (1 in x16) slots, 1x PCI-E 2.0 x4 (in x8) slot; Up to 5x GPU,
4. I/O ports: 1 VGA, 2 COM, 2 GbE, 10 USB 2.0, & 1 IPMI Dedicated LAN,
5. 8x 3.5" Hot-swap, 3x Fixed 5.25," and 1x Fixed 3.5" Drive Bays,
6. 4x Heavy Duty Fans, 2x Exhaust Fans, and 2x Active Heatsink with Optimal Fan Speed Control,
7. [2x] 2000W Redundant Power Supplies Platinum Level (94%),
8. Tower or Rackmount [Santa, I want the tower] and
9. GPU Kit for passive GPU/Xeon Phi support (MCP-320-74701-0N-KIT).
http://www.supermicro.com/products/system/4U/7048/SYS-7048GR-TR.cfm .

Note that this one doesn't have an "AX" in its modeling (it has "GR"). So just be sure that my barebone system model is a SYS-7048AX, because those two letters, i.e., the "A" and the "X" mean that its a DAX motherboard based system and that I can significantly clock tweak the performance of those E5-2699 V3s by 1.3x rather than by merely 1.0755x maximum. I thinking about a 36-core system that has a base speed of 2.9+GHz, a 1st stage turbo of 3.1+GHz for all 36 cores and a max turbo of 4.6+GHz (of course with many fewer cores - most likely 2, or at best 4, cores spread over both CPUs), yielding a Geekbench 3 score of about 102K+ and a Cinebench 15 CPU score of 5.8K+, and that handles 4k video as an appetizer to an 8k course meal. I like my systems future resistant.

Thanks Santa,

P.S. - I've been nice this year and fully on my Ps & Qs. And I've improved my patience - I wouldn't even shed a tear if you told me that you had to defer your delivery to me of all that I've requested, until next summer.

Asides: I've also learned to give credit to others: Thank you day traders and to SuperMicro for coming to the rescue of you day traders, as if you day traders needed more rescuing. And, Santa, in case you forget the CPU feature that I want in a Supermicro V3 barebones (4 double wide PCIe) based system, just remember that I want to cut my CPU render times with the metaphorical equivalent of an axe, i..e., Dee ax. Just poking some fun, Santa, so don't hold these asides against me.

Analysis of FurryBall Benchmarks

In the context of FurryBall 4.8 Benchmark -

1) An original (o) GTX Titan’s GPU and memory can be profitably overclocked (with EVGA PrecisionX) by more MHz than a GTX 780 Ti’s GPU or memory. In the end however, the 780 TI’s will still run faster.

2) A moderate memory overclock improves the performance of a oTitan and a 780 Ti. The memory of either card can be tweaked profitably by a larger absolute numerical amount than can the GPU.

3) A higher memory overclock of either a oTitan or a 780 Ti can improve rasterization performance, but may degrade High Raytracing performance.

4) One slightly overclocked 780 Ti, rather than two or more of them, yields the best performance for FurryBall Raster/Biased renders.

5) FurryBall does not scale like OctaneRender does for various CUDA GPUs. In OctaneRender, seven oTitans would always surpass the rendering performance of four 780 Ti’s and, unlike Furryball raytracing, Octane render times improve even with the addition of more CUDA GPUs, although their speed and core count may differ. However, four GTX 780 Ti’s render the FurryBall benchmark scene faster than seven oTitans.

Note Well - Whatever I observed regarding the GTX 780 Ti’s would also apply (likely more so) for the GTX Titan Black and GTX Titan Z.

In addition to the tests that I spoke of in post #1072, above, I ran additional tests under ID nos. 4925, 4949, 4951, 4957, 4958 on which to base these observations. See, e.g., [ http://www.aaa-studio.cz/furrybench...orderByTime=best_time&orderByID=1&version=4.8 and http://www.aaa-studio.cz/furrybench/benchResults4.php ].

Good News for 2013+ MP Users: High Quality 3d OpenCL Rendering Is Under Development

Heading towards squeezing every bit of compute performance one can now have in a computer is Thea Render's OpenCL Port for Thea Presto (A Work In Progress) [ http://www.thearender.com/forum/viewtopic.php?f=17&t=15516 ]. It now appears that Thea Render may be the first CPU, CUDA, OpenCL and Xeon PHI renderer. The OpenCL port (under active development) may support AMD and Nvidia OpenCL, as well as Xeon PHI. The results so far appear promising. Thea Render already supports CPU and CUDA rendering and has plugins for 3ds Max, Blender (no cost), Cinema 4d, Rhino and SketchUp.

Also, you might also like to take a look at a new Thea Render applied application - TileLook:
1) http://www.thearender.com/cms/index.php/news/featured-artists/343-tilelook-featured-project.html
2) https://www.youtube.com/channel/UCtUw3ZdEyVwxV705DDAZ7eQ .