All We Know About Maximizing CPU Related Performance

PunkNugget · Dec 24, 2012

Tutor said:
Rampage Dev and PunkNugget,
Thanks for the present to all of us. Happy holidays.

PunkNugget,
I'm disinclined to spend much more time tweaking WolfPack5 - the MSI - 3930k 6-core system. I do not expect that I can improve this system's performance by much more. WolfPacks 1 to 4 and 6 to 13, my Mac Pros (and yes, I'm hackintoshing [per netkas's forum] all of my old Mac Pros to give them new lives, but their hardware I'll leave in tack) and WolfPackPrime0, as well as my GPUs, will get the remainder of my free time these holidays. My Mac portables are next.

Using the same unit that you posted in Windows w/a GB of 28,000+, what are the GB results using Mac OS X? And if you have it then, what OS are you using?

Tutor · Dec 25, 2012

PunkNugget said:
Using the same unit that you posted in Windows w/a GB of 28,000+, what are the GB results using Mac OS X? And if you have it then, what OS are you using?

Presently, only Linux and Windows OSes are attached to WolfPack5. Mac OS is attached to my Mac Pros. I made the mistake of upgrading one Mac Pro to Lion. The rest are using what I consider to be the best OS ever made by Apple - 10.6.7. Unless apple plucks the eye [candy] from the lions, I'll stick with whatever other OS is focused on performance, just as was 10.6.7. Apple is pointed towards only closed "pro" systems and that is where I'm not going with them for my trucks.

PunkNugget · Dec 25, 2012

Tutor said:
Presently, only Linux and Windows OSes are attached to WolfPack5. Mac OS is attached to my Mac Pros. I made the mistake of upgrading one Mac Pro to Lion. The rest are using what I consider to be the best OS ever made by Apple - 10.6.7. Unless apple plucks the eye [candy] from the lions, I'll stick with whatever other OS is focused on performance, just as was 10.6.7. Apple is pointed towards only closed "pro" systems and that is where I'm not going with them for my trucks.

I hear that and loved the extra boost without all that heat in the office as a result of UC'ing, but really need to use my SR-2 for vid production and I need to use my 580 GTX for that; especially for Fermi, CUDA and Mercury playback support (10.6.7 won't support that card), and although the CPUs and RAM play some role in that rendering process (using Adobe's CS6), 2/3's of my rendering support will be dedicated to that GPU. That's why I need to run 10.8.2.

Tutor · Dec 26, 2012

WolfPack2 GPU stats

Here are some stats and GPU benchmarks on WolfPack2 [dual W5680's on EVGA-SR2, QPI set to 7118 GT/s in bios].

Tutor · Dec 31, 2012

MacPro 2007_1 GPU stats

The GTX 580's performance in my MacPro 2007_1, under GPU test GiMark, is comparable to that of the same card when overclocked in my underclocked EVGA SR-2 and surpasses, under GPU test FurMark, that of the same card when overclocked in my underclocked EVGA SR-2. Go figure.

Modonaut · Jan 6, 2013

Tutor,

What do you know, or expect from the upcoming Xeon Phi cards? Will they be plug and play or will applications need to be taliored to take advantage of them? Would they work in a current version of MacPro (I have a 2,1) or only in some new macPro design to come?

Also, are they 4 socket configurations that can run OSX? WHat would you recommend? I really need a single machine to render on rather than a small farm as my rendering software is only about 50% efficient when spread across a network. So, I want to build the fastest workstation I can put together to run OSX - even if more than dual CPU is possible. I suppose I could get used to working with Windows 8, but would like to stick with mac if possible.

Thanks.

Tutor · Jan 6, 2013

If your need is pressing go with CUDA - it's here and now vs. Xeon Phi

Modonaut said:
... . I really need a single machine to render on rather than a small farm as my rendering software is only about 50% efficient when spread across a network. So, I want to build the fastest workstation I can put together to run OSX - even if more than dual CPU is possible. I suppose I could get used to working with Windows 8, but would like to stick with mac if possible.

I'm not sure if you're talking about building a computer from scratch or modifying your Mac Pro 2.1; but I'll assume for now that you're talking about modifying your Mac Pro.

Modonaut said:
What do you know, or expect from the upcoming Xeon Phi cards?

I expect developers to recompile CUDA software or write, from scratch, software for the 60 and 61 core Xeon Phi to enable it to achieve over one teraflop of processing power.

Modonaut said:
Will they be plug and play or will applications need to be tailored to take advantage of them?

They will not be plug and play because applications will need to be tailored to take advantage of them.

Modonaut said:
Would they work in a current version of MacPro (I have a 2,1) or only in some new macPro design to come?

If an OSX application is modified to take advantage of the 50+ core Xeon Phi, it is likely that those cards could be forced to run on a Mac Pro2,1, but it won't be optimal. Xeon Phi cards are PCI-e 2 and 2+ cards. [ http://www.intel.com/content/dam/www/public/us/en/documents/product-briefs/xeon-phi-datasheet.pdf ] PCI-e 2 made it's appearance in the 2008 Mac Pro; so a system of that vintage or later would be optimal.

Modonaut said:
Also, are they 4 socket configurations that can run OSX?

From everything that I've read and seen, they look like PCI-e x16 video cards that are inserted into a single x16 slot , but will cover the next adjacent one, just like a GTX 580 for example.

Modonaut said:
What would you recommend?

I need to know what app you're using to render and what your typical render job consists of before I can make a meaningful recommendation, but my off the cuff response would be for you to get a GTX 560-580 or two, plus a FSP Group Booster X5 450W Independent/Supplementary SLI Certified CrossFire Ready Active PFC Dedicated Multi.GPU Power (Newegg -~ $90.00 [ http://www.newegg.com/Product/Product.aspx?Item=N82E16817104054) + Tronixin 15 Pin SATA Male to 4 Pin Female IDE HDD Power Cable [ http://us.dinodirect.com/Tronixin-1...6&source=sem&gclid=CJ-0hers1LQCFQq0nQodGhMARQ (>$3-4) ] if the render application that you are using takes advantage of CUDA.

If you were starting from scratch and had an immediate need for a render box, but it had to be a Mac, then I'd recommend:

(1) Refurbished Mac Pro 2.4GHz 8-Core Intel Xeon [ http://store.apple.com/us/product/FC561LL/A ] - $2,649.00 + 2 - W5680 for $900/cpu [ https://www.eoptionsonline.com/p-2049-594880-001.aspx ] Total ~ $4500
(2) 2 GTX 580 + FSP Group Booster X5 450W Independent/Supplementary SLI Certified CrossFire Ready Active PFC Dedicated Multi.GPU Power (EVGA - ~ $1000 for 2 of those video cards; FSP Group Booster X5 450W -~ $90.00 [ http://www.newegg.com/Product/Product.aspx?Item=N82E16817104054 + Tronixin™**15 Pin SATA Male to 4 Pin Female IDE HDD Power Cable [ http://us.dinodirect.com/Tronixin-1...6&source=sem&gclid=CJ-0hers1LQCFQq0nQodGhMARQ ].

If you were starting from scratch and had an immediate need for a render box, but you were able to build your own PC, then I'd recommend:

(a) an EVGA Classified SR-2 (Super Record 2) motherboard (7 PCI -e slots) [ http://www.evga.com/Products/Product.aspx?pn=270-WS-W555-RX ] - $400 +
(b) 2 - W5680s for $900/cpu [ https://www.eoptionsonline.com/p-2049-594880-001.aspx ] - $1800 +
(c) 48 gigs of ram [ CORSAIR DOMINATOR GT 12GB (3 x 4GB) 240-Pin DDR3 SDRAM DDR3 2000 (PC3 16000) Desktop Memory with Air Fan Model CMT12GX3M3A2000C9 ] - $560 +
(d) 1200 + Watt PSU [ CORSAIR AX1200i 1200W Digital ATX12V v2.31 and EPS 2.92 SLI Ready CrossFire Ready 80 PLUS PLATINUM Certified Full Modular Active PFC Power Supply ] - $330 +
(e) an HPTX case [ LIAN LI PC-D8000 Black Aluminum ATX Full Tower Computer Case ] - $370 +
(f) 2 - CORSAIR Hydro Series H80 (CWCH80) High Performance Liquid CPU Coolers [ http://www.newegg.com/Product/Product.aspx?Item=N82E16835181016 ] - ~ $200 +
(g) 2 EVGA GeForce GTX 580 Classified Ultra 3072MBs (check with evga [ http://www.evga.com/Products/Product.aspx?pn=03G-P3-1595-AR ] for refurbs that carry a 1 year warranty) - < $1000 +
(h) FSP Group Booster X5 450W Independent/Supplementary SLI Certified CrossFire Ready Active PFC Dedicated Multi.GPU Power [ http://www.newegg.com/Product/Product.aspx?Item=N82E16817104054 ] -~ $90.00 +
(i) OCZ Vertex 4 VTX4-25SAT3-256G 2.5" 256GB SATA III MLC Internal Solid State Drive (SSD) [ http://www.newegg.com/Product/Product.aspx?Item=N82E16820227792 ] -~$220 +
(j) 3 Western Digital WD Black WD2002FAEX 2TB 7200 RPM 64MB Cache SATA 6.0Gb/s 3.5" Internal Hard Drive -Bare Drive [ http://www.newegg.com/Product/Product.aspx?Item=N82E16822136792 ] - ~ $540 (for raid 0) +
(k) Microsoft Windows 7 Ultimate SP1 64-bit - OEM [ http://www.newegg.com/Product/Product.aspx?Item=N82E16832116997 ] - ~ $180. So for about a total $5690, you can get a machine that allows you to adjust it's speed pretty much to your liking [ I was able to tweak mine to get a Geekbench 2 score above 40,000 and a Cinebench 11.5 score of 24.7 - that's about 1.4 times the benchmark scores achieved by Mac Pros after swapping dual W5680s into them], with lots of ram and fast storage, enormous CUDA capability, and it has enough room left inside for a Xeon Phi or two. That's what I'd view as " ... a single machine to render - ... build[ing] the fastest workstation ... possible [today for a reasonable price]".

DJenkins · Jan 6, 2013

Quote:
Originally Posted by Modonaut
Also, are they 4 socket configurations that can run OSX?

Tutor said:
From everything that I've read and seen, they look like PCI-e x16 video cards that are inserted into a single x16 slot , but will cover the next adjacent one, just like a GTX 580 for example.

Tutor I think by 4 socket he means a 4 CPU machine.
I know you have built them but from what I remember I think the cost greatly increases and it's not an ideal machine to run OSX on.

Also Modonaut, the benchmark scores Tutor has achieved are after pretty much a lifetime of computer tweaking. If you are willing to learn then it's possible you might get there with a similar setup, but most of us who have followed are getting between 32,000 and 36,000+.

Sorry it sounds a bit like I'm raining on the parade but it just shows that Tutor really is a master and I wouldn't want anyone to get unrealistic expectations if they haven't done this before!

Best of luck, I hope you get a killer box running!

Modonaut · Jan 6, 2013

Thank you for the detailed response. I can add some clarification. I am interested in a new build. I can also update my 2,1 at some point, but initially, I think I'd like to throw my money into some very fast CPUs.

I use Luxology's Modo for rendering. The company has publicly elected to avoid developing for GPU-rendering (at least for the foreseeable future) and all rendering tasks rely on CPU power. Of course, I would expect that I'd budget $5-600 for a fast Graphics card and CUDA for other apps would be good to have.

A second issue I deal with is that Modo's network rendering is buggy and inefficient. A slave machine of equal power only contributes about a 50% of its potential power during rendering. So, I will benefit by having as many CPU cores and threads under one hood.

Having said that, how does your recommendation stand? Would the be even a 4 socket system that could run osx? From what I have read, that means e4500 series? Pretty pricey. What about a dual e2687w system? is that faster than what you recommended?

If I did end up one day with a Phi, would I need to plan now to provide sufficient power in the power supply?

It should be obvious by now that I am rather under-educated on the technical aspects of building this myself. I might be able to plug the hardware together, but there's a lot to understand about the bios and the tuning involved, not to mention coercing OSX to run smoothly.

Thanks for your thread here and your feedback and recommendations!

Dan

DJenkins · Jan 6, 2013

Joy550 said:
Just so I'm clear... this is all Hackintosh-only right now, correct? There's no underclocking solution yet for proper Mac Pro desktops?

Yep, not likely to ever be any over/underclocking scope on a real mac. On hackintosh it's all handled in the BIOS.

Modonaut,
All I can say is if it's your first time building a computer, please take careful handling and anti-static precautions. I'd hate to see you run into troubles with such expensive parts! Especially if you don't have spares lying around to test.

Maybe even build a basic machine and sell it to a friend for practice first.

Then follow a build someone else has done already. Exact parts, exact installation method. I had various issues due to initially ignoring this simple rule... cost hundreds of hours of troubleshooting!

Modonaut · Jan 6, 2013

-

djenkins said:
quote:
Originally posted by modonaut
also, are they 4 socket configurations that can run osx?

Tutor i think by 4 socket he means a 4 cpu machine.
I know you have built them but from what i remember i think the cost greatly increases and it's not an ideal machine to run osx on.

yes, you are right. I was thinking if i could go a step further than dual socket and get 4 cpus under one os rather than in multiple boxes, it would be a more efficient render farm in one machine. But, i also figured that raises the technical hurdles. Nothing is ever easy.

also modonaut, the benchmark scores tutor has achieved are after pretty much a lifetime of computer tweaking. If you are willing to learn then it's possible you might get there with a similar setup, but most of us who have followed are getting between 32,000 and 36,000+.

Sorry it sounds a bit like i'm raining on the parade but it just shows that tutor really is a master and i wouldn't want anyone to get unrealistic expectations if they haven't done this before!

Oh yeah, i can believe it! I have the money and the work to justify this purchase, but not the time to devote to the build. I am more of an artist that uses the machine for production and I think I would be out of my depth doing this build myself. I was thinking of hiring someone who does know what they are doing and try to learn as much as I can to at least have some understanding of what these issues are. This seems like a great place to start!

best of luck, i hope you get a killer box running!

Tutor · Jan 7, 2013

Modonaut said:
... . I am interested in a new build. I can also update my 2,1 at some point, but initially, I think I'd like to throw my money into some very fast CPUs.

If you're interested in a new build, then my recommendations above, still stand.

Modonaut said:
I use Luxology's Modo for rendering. The company has publicly elected to avoid developing for GPU-rendering (at least for the foreseeable future) and all rendering tasks rely on CPU power. Of course, I would expect that I'd budget $5-600 for a fast Graphics card and CUDA for other apps would be good to have.

What Luxology hasn't done yet relative to GPU-rendering, has already been done by others such as Otoy by their Octane software and their free community developed exporter plugins for: Autodesk 3D Studio Max, Autodesk Maya, Autodesk Softimage XSI, Blender, Maxon Cinema 4D, Sketchup and, most importantly for you, Modo for Luxology users [ See http://render.otoy.com/forum/viewtopic.php?f=34&t=22098 and http://render.otoy.com/features.php click each term on the left black panel at the top of the page for more info]. That's why I still would recommend going the CUDA route first.

Modonaut said:
A second issue I deal with is that Modo's network rendering is buggy and inefficient. A slave machine of equal power only contributes about a 50% of its potential power during rendering. So, I will benefit by having as many CPU cores and threads under one hood. Having said that, how does your recommendation stand?

True generally that having many CPU cores and CPU threads under one hood are a good thing, but not so much so if you go the Octane route,"Octane Render uses the untapped muscle of the modern GPU compared to traditional, CPU based engines. With current GPU technology, Octane Render can produce final images 10 to 50 times faster than CPU unbiased render engines, or even more with multiple GPUs (depending on the GPU(s) used)." With Octane, it's more important to have many GPU cores and GPU threads preferably, but not necessarily, under one hood. So my recommendations above, still stand.

Modonaut said:
Would there be even a 4 socket system that could run osx? From what I have read, that means e4500 series? Pretty pricey.

Generally, an E5-4600 series system will run Linux (the fastest OS for the four socket systems {and importantly Xeon Phi runs under a version of Linux installed on the card}) and Windows server (the priciest version of Windows), but not OSX yet. So all in all - the 4 socket systems are usually much more expensive than going the CUDA route first.

Modonaut said:
What about a dual e2687w system? is that faster than what you recommended?

I also have an E5-2687W system. My E5-2687W system is slower than the EVGA SR-2 under OSX, but is faster than most other EVGA-SR2s under Windows. It's just much more expensive for the 5 - 20% increase in performance over tweaked X5680s that you'd see. However, please note that my comparisons of the relative speed of an E5-2687W system compares non-tweaked E5-2687Ws against tweaked X5680s. So the good news is that the E5-2687Ws are fast untweaked, but the bad news is that for their much higher price they can be tweaked just a very small amount (1-4%). In fact for that $1.8k price differential (between dual X5680s and dual 2687Ws), you could purchase 2 GTX 580s and an external chassis from BHPhoto Video to house another PCI-E dual slot encumbering card [ http://www.bhphotovideo.com/c/produ...A211A_NB_3_SLOT_PCIe_EXPANSION_ENC_EXC34.html ] and still pocket some change. Also, remember that CUDA is currently being used by Adobe and others in the creative app market to accelerate compute intensive functions and with apps like Octane, CUDA cards scale linearly.

Modonaut said:
If I did end up one day with a Phi, would I need to plan now to provide sufficient power in the power supply?

Unlike for CUDA cards the Xeon Phi literature says,"There is no auxilliary 2x4 or 2x3 power connector on the card." So the answer to your question in included in my recommendation, above, namely, get a 1200 watt or greater power supply to power the motherboard and get a FSP Group Booster X5 450W Independent/Supplementary SLI Certified CrossFire Ready Active PFC Dedicated Multi GPU Power supply to power {or at least help power} your video cards, thereby, relieving the main PSU of that chore so that it can better support the power requirements of the CPUs, memory, drives and PCI-e cards, like the Xeon Phi, that don't require more power through separate PCI-e card plugs.

My main point might be better made with some questions: How much would it cost you to buy 10 to 50 additional cpus and put them to work in a rendering environment? Could you do so for less than the price of an EVGA GTX 690 ($1K) or a GTX 580 (<$700 for the most expensive ones that I've seen in the last 4 months)? Then, what would it cost you to buy an additional 10 to 50 cpus and put them to work in a rendering environment? Don't forget the space, thermal and electrical requirements and the cost of electricity for all of those systems. Cost is why I recommend going the CUDA/Octane route with an EVGA SR2 build (or an EVGA SRX build if you want to use Sandy [or a little later Ivy] bridge chips*/) until Xeon Phi can stand steadily on its own legs and not lean excessively against the side of your jacket holding your wallet. I'll freely admit it, "I'm cheap."

*/ "Developers ... [have] one of several routes [to program the Xeon Phi]:

[1] Using pragmas to augment existing codes so they offload work from the host processor to the Intel Xeon Phi coprocessors(s)[;]
[2] Recompiling source code to run directly on coprocessor as a separate many-core Linux SMP compute node[;]
[3] Accessing the coprocessor as an accelerator through optimized libraries such as the Intel MKL (Math Kernel Library)[; or]
[4] Using each coprocessor as a node in an MPI cluster or, alternatively, as a device containing a cluster of MPI nodes." [ http://www.drdobbs.com/parallel/programming-intels-xeon-phi-a-jumpstart/240144160?pgno=1 ]

Since the Xeon Phi has it's own Linux OS, it can be programmed to behave as a separate system, in which case whether your CPUs are Sandy/Ivy bridge or Westmeres/Nehalems shouldn't matter. However, if your app developer programs the Xeon Phi to act as a coprocessor, then you might be better off with an EVGA SRX (which uses the E5 chips) for better compatibility.

Modonaut · Jan 7, 2013

In exploring your suggestion of picking up a refurbished MacPro directly from Apple and as I am will to pay a bit of a premium for Apple, I did stumble across these options I could order now. I have a 20% discount with Apple as well, so it might make the most sense (least headache) to just pick up one of these machines and maybe upgrade the CPU at some point if there is any real benefit :cost to doing that.

In looking at these configurations, does one stand out as the best value with some performance enhancement potential I could do on my own? Keep in mind, that my benchmark is what I know from working on a MacPo Dual 3.0 GHz (X5365) Quad-core Intel Xeon for the last 4-5 years.

Best value seems to be 1 by estimating (very unscientifically) a rough cost per point on Geekbench. My guess is choice 3 is best for my performance needs. But maybe the choice 1 + some new cpus (if I could get them for around $1100 each) could get me faster? Could I get two 8 core CPUs into choice 1?

Refurbished Mac Pro 2.4GHz 8-Core Intel Xeon
Eligible for OS X Mountain Lion Up-to-Date Program

Originally released August 2010
Two 2.4GHz Quad-Core Intel Xeon "Westmere" processor
6GB (6 x 1GB) of 1066MHz DDR3 ECC memory
1TB Serial ATA 7200 rpm
18x SuperDrive (DVD±R DL/DVD±RW/CD-RW)
ATI Radeon HD 5770 with 1GB GDDR5 memory
$2,649.00

Refurbished Mac Pro 2.4GHz 12-Core Intel Xeon
Eligible for OS X Mountain Lion Up-to-Date Program

Originally released June 2012
Two 2.4GHz 6-Core Intel Xeon processors
12GB (6 x 2GB) of 1333MHz DDR3 ECC memory
1TB Serial ATA 7200 rpm
18x SuperDrive (DVD±R DL/DVD±RW/CD-RW)
  ATI Radeon HD 5770 with 1GB GDDR5 memory
$3,239.00 GeekBench ~17,000 (~19 cents per Geek point)

Refurbished Mac Pro 2.93GHz 12-Core Intel Xeon
Eligible for OS X Mountain Lion Up-to-Date Program

Originally released August 2010
Two 2.93GHz 6-Core Intel Xeon "Westmere" processor
6GB (6 x 1GB) of 1333MHz DDR3 ECC memory
1TB Serial ATA 7200 rpm
18x SuperDrive (DVD±R DL/DVD±RW/CD-RW)
ATI Radeon HD 5770 with 1GB GDDR5 memory
$4,799.00 GeekBench ~21,000 (~23 cents per Geek point)

Finally, this option at a 20% discount:

Current TOTL Mac Pro 3.06 GHz 12-Core Intel Xeon
Two 3.06GHz 6-Core Intel Xeon (12 cores)
12GB (6X2GB)
1TB 7200-rpm Serial ATA 3Gb/s hard drive
ATI Radeon HD 5770 1GB

$6,199.00 GeekBench ~22,000 (~28 cents per Geek point)

Tutor · Jan 7, 2013

Modonaut said:
... .
In looking at these configurations, does one stand out as the best value with some performance enhancement potential I could do on my own? ... .
Best value seems to be 1 by estimating (very unscientifically) a rough cost per point on Geekbench. My guess is choice 3 is best for my performance needs. But maybe the choice 1 + some new cpus (if I could get them for around $1100 each) could get me faster? Could I get two 8 core CPUs into choice 1?

#1 is what I'd still recommend in a Mac Pro - please revisit post no #457 for recommended addons/prices/sources. Dual 6-core 56xx are the max.

Modonaut · Jan 7, 2013

Tutor said:
My main point might be better made with some questions: How much would it cost you to buy 10 to 50 additional cpus and put them to work in a rendering environment? Could you do so for less than the price of an EVGA GTX 690 ($1K) or a GTX 580 (<$700 for the most expensive ones that I've seen in the last 4 months)? Then, what would it cost you to buy an additional 10 to 50 cpus and put them to work in a rendering environment? Don't forget the space, thermal and electrical requirements and the cost of electricity for all of those systems. Cost is why I recommend going the CUDA/Octane route with an EVGA SR2 build (or an EVGA SRX build if you want to use Sandy [or a little later Ivy] bridge chips*/) until Xeon Phi can stand steadily on its own legs and not lean excessively against the side of your jacket holding your wallet. I'll freely admit it, "I'm cheap."

Yes, I understand the logic. However, CUDA does me absolutely no good as 99% of my income comes from applications that do not yet use GPU for the heavy lifting I am trying to do. It is 100% CPU based. All your info has been very helpful.

Tutor · Jan 7, 2013

Modonaut said:
... . However, CUDA does me absolutely no good as 99% of my income comes from applications that do not yet use GPU for the heavy lifting I am trying to do. It is 100% CPU based. All your info has been very helpful.

I am glad that my info has been helpful to you. However, I regret that CUDA does you absolutely no good. I thought that Modo rendering was important to you. That is why I suggested, in post no. 462, that you use the free Octane plugin to export your Modo graphic and animation projects to the Octane renderer which relies on the presence of, at least, one CUDA card to increase your render speed by, at a minimum, 1,000% to 5,000% and by an additional equal amount for each additional CUDA card present. Sorry for the misunderstanding.

Ianduarte · Jan 16, 2013

Underclocking

Tutor,

First of all I have to apologize for interrupting your topic, but I didn't know how to reach you in here.
I have a hackintosh that I build myself.
My mobo is a gigabyte Ga58xusb3, my processor is a Xeon w3690, a compatible radeon video card with an 27" apple display connected to it, and I have 2 ssd drives with 24 gigs of ram.
My computer is overclocked and I am reaching 17130 points on geekbench.
I don't know if I understood it wrong, but I saw someone saying that you can underclock the computer for better results.
Is it true and if it is, is there a tutorial or something so that I can learn how to do it?

Thankx in advance man.

Tutor · Jan 16, 2013

Ianduarte said:
Tutor,

First of all I have to apologize for interrupting your topic, but I didn't know how to reach you in here.
I have a hackintosh that I build myself.
My mobo is a gigabyte Ga58xusb3, my processor is a Xeon w3690, a compatible radeon video card with an 27" apple display connected to it, and I have 2 ssd drives with 24 gigs of ram.
My computer is overclocked and I am reaching 17130 points on geekbench.
I don't know if I understood it wrong, but I saw someone saying that you can underclock the computer for better results.
Is it true and if it is, is there a tutorial or something so that I can learn how to do it?

Thankx in advance man.

See URL in my signature line {note especially that the lower idle frequency systems tend to outperform the higher frequency ones}, but to sum it up, e.g., I get a Geekbench 2 score of over 21,000 on a 6-core i7-980X, a score of over 15,000 on a 4-core i7-975 and a score of over 40,000 on dual 5680s. So I think that underclocking has, at least, one benefit.

Wrap your head around this -
The following is in essence why you're doing what you're doing, what you are doing and what's going on:

I. Why you're doing it?
What you're doing is you're raising the turbo speed by increasing the number of turbo steps/bins from idle to max turbo [see CPU World for your chip's turbo capability]{similar to that which occurs under Sandy Bridge but with Westmeres, but I do it to an exaggerated amount} and you're increasing the size of each step/bin with a higher BCLK. Analogy time: Think of a centipede with 100 one millimeter feet with a 3 mm gait. Now, think of a millipede with 1000 ten millimeter feet with a 30 mm gait. Which is potentially faster. See, you're increasing your system's speed potential.

II. What are you doing and how it affects what's going on?

Overclock the CPU (with turbo disabled) in Windows to the speed that you want it to run at max turbo. This way you'll get, thru your adjustments, the maximum of the various voltages and frequencies for the CPU, memory and QPI correct for when turbo kicks in. Now write them down. The next step is doing the math (and using that math to guide your system bios settings and your knowledge of the turbo features of your cpu) so that when max turbo kicks in, a reduced frequency multi (this keeps idle low and temps low - I love a setting of 13 on X5680s) gives way to the turbo max value/ranges (for X5680s its 26 and 27, i.e., a range ). When turbo kicks in the turbo values get multiplied by a higher BCLK which you'll set [this stays the same while turbo does its thing]. This should be set to yield, at max turbo, about the same speed attained in overclocking stage, but with fewer cores attaining that speed at any one point in time (because that's the nature of turbo). Then reboot, turning on turbo and applying the math and using it (and your knowledge of the turbo features of your cpu) to guide your system bios settings. So when max turbo kicks in, a reduced CPU frequency multi (this keeps idle low - I love 13 on dual x5680s seated in an EVGA SR2, but slight underclocks get good results on other motherboards with other chips) gives way to the turbo max value/ranges (for X5680s its 26 and 27, i.e., a range) times a higher BCLK [this stays the same once you set it], yielding about the same speed attained under your overclock, but now with cores getting continuously refreshed and being swapped in [the cooler ones] and out of play [the hotter ones]. Then you start reducing voltages to just slightly above the point where stability is solid. The result is that when turbo isn't triggered your system runs slower and cooler (because of the lowered CPU frequency multi), but when turbo kicks in all hell breaks loose. On the one hand because of the nature of turbo, your system will still be taxed when under turbo, but also because turbo isn't continuous, your CPUs/machine will not get as hot, nor stay hot as long, nor require as much electrical juice. Turbo will more likely keep getting triggered while any appreciable load is present.

Now if you switch OSes, read my post # 96 to get native power management working. Don't forget the internal references. You've got to get Apple's native power management working correctly and I can vouch for it working the best only under OSX 10.6.7.

Ianduarte · Jan 16, 2013

Thank you Tutor

Thank you for your answer.

I will try to do what you said and as soon as I get things running or have another question I will talk to you.

Thankx a lot man !!!

Tutor · Jan 16, 2013

Ianduarte said:
Thank you for your answer.

I will try to do what you said and as soon as I get things running or have another question I will talk to you.

Thankx a lot man !!!

No problem. Happy to help. That's what keeps me vibrant.

speters · Jan 22, 2013

Upgrading processors in 2006 & 2010 mac pro?

So I have a 2.66ghz 12 Core Mac pro that I would like to upgrade the processors on. I also have a 2006 3.0ghz 8 Core. That computer is currently not running, but I think it is a 1.1. I took it into the Apple store to find out what was wrong, and after a lot of testing they narrowed it down to the processors. I didn't have them fix it because I new that I wanted to upgrade the processors on that myself. I have been reading over replacing the processors on these on and off for the last couple of months and just stumbled across this thread today. I have a few questions about these upgrades, but first a little bit more info about these machines. Really, the 12 core is the most important, I think that I would use the 8 core mostly as a network rendering machine as well as a home theater computer. Most of my work currently is Photoshop and Cinema 4D, but I also use Realflow from time to time and I plan on starting to get into After Effects pretty heavily. I currently have 48gb Ram, and the ATI Radeon 5770 in the 12 core. I am considering getting a new video card, such as a GTX580 for the 12 Core, but right now I first want to get the processors replaced. I did see tutor's post about the Octane render which I plan to look into more at a later time. From what I gathered, looking into very quickly, is that when it comes to rendering, if I went the Octane route, the gpu would be more important than the cpu? But I think that with After Effects, there is still alot that would be cpu intensive, so the processor upgrade would still improve overall performance? With the 8 core, I need to put in some new processors to get it working again.

Now the questions:
-So for both of the processor upgrades, if I go the used route, do I have to get matching pairs?

-The cpu-world link for the 5.1's no longer exist. But looking at the 1.1 cpu-world links I am a little confused because there are a lot of different processors listed, ie. 5110, 5113, 5120,.......Then for some there is HH, BX, etc. I have understand that the numbers would be different speeds, but could I choose anyone that is listed in that link? Can some one tell me exactly which processor, with all pertinent letters and numbers, that I should look for?

-For the 12 core how much of a difference is there between that 3.33 and the 3.46 when it comes to overall performance for my needs?

Thanks,
Steve

Tutor · Jan 22, 2013

speters said:
So I have a 2.66ghz 12 Core Mac pro that I would like to upgrade the processors on. I also have a 2006 3.0ghz 8 Core. ...Most of my work currently is Photoshop and Cinema 4D, but I also use Realflow from time to time and I plan on starting to get into After Effects pretty heavily. I currently have 48gb Ram, and the ATI Radeon 5770 in the 12 core. I am considering getting a new video card, such as a GTX580 for the 12 Core, but right now I first want to get the processors replaced. I did see tutor's post about the Octane render which I plan to look into more at a later time. From what I gathered, looking into very quickly, is that when it comes to rendering, if I went the Octane route, the gpu would be more important than the cpu? But I think that with After Effects, there is still alot that would be cpu intensive, so the processor upgrade would still improve overall performance? ... .Now the questions:
-So for both of the processor upgrades, if I go the used route, do I have to get matching pairs?

Yes, use matching pairs. Depending on how you use the Adobe apps, you may get more bang for the buck by just getting the GTX 580 first. All of the apps you mentioned can take advantage of the GTX 580's CUDA prowess to varying degrees, so be sure to read the Adobe PDF for their apps: http://www.nvidia.com/docs/IO/121761/adobe-hardware-performance-white-paper-2012-06-27.pdf . You may want to consider downloading the Octane demo as that is the app I'd recommend for Cinema 4d. Apps that can take advantage of CUDA can easily gain 500%+ increases in speed (where CUDA operates) rather than the 25% increase you see from getting the 12-core X5680s @ <$1800 a pair (Intel Xeon X5680 - AT80614005124AA (BX80614X5680) - http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon X5680 - AT80614005124AA (BX80614X5680).html from somewhere like here: https://www.eoptionsonline.com/p-2049-594880-001.aspx .

speters said:
The cpu-world link for the 5.1's no longer exist. But looking at the 1.1 cpu-world links I am a little confused because there are a lot of different processors listed, ie. 5110, 5113, 5120,.......Then for some there is HH, BX, etc. I have understand that the numbers would be different speeds, but could I choose anyone that is listed in that link? Can some one tell me exactly which processor, with all pertinent letters and numbers, that I should look for?

For the 2006/2007 Mac Pro, I'd recommend what I have in all three of my 2007 Mac Pro2,1s - Intel Xeon X5365 - HH80563KJ0808MP (BX80563X5365A / BX80563X5365P) after you do the EFI utility upgrade for the 2006 MP (See post #1 and [ http://forum.netkas.org/index.php?action=dlattach;topic=1094.0;attach=888 ]). But since you mentioned that you have a 2006 3.0ghz 8 Core, you may already have had those CPUs { and if so my guess is that they may not have been completely, properly installed or overtaxed by something like the ZDnet overclocking utility - but that's just mere speculation on my part}.

speters said:
For the 12 core how much of a difference is there between that 3.33 and the 3.46 when it comes to overall performance for my needs?

3.46 (X5690s) / 3.33 (X5680s) = 1.04x or a speed increase of 4%. In light of the above, the GTX 580 (+ later [ unless money is no object] the X5680s) will yield more bang for you buck since each X5680 can usually be purchased for over $500 less than a X5690. That $500+ savings per CPU could go a long way to your getting two GTX 580s plus this: FSP Group Booster X5 450W Independent/Supplementary SLI Certified CrossFire Ready Active PFC Dedicated Multi.GPU Power [ http://www.newegg.com/Product/Product.aspx?Item=N82E16817104054 + Tronixin 15 Pin SATA Male to 4 Pin Female IDE HDD Power Cable [ http://us.dinodirect.com/Tronixin-1...6&source=sem&gclid=CJ-0hers1LQCFQq0nQodGhMARQ ]] to power them [so long as you get the 6x6 or 8x6 pin powered GTX 580s - but I'd recommend sticking with the 6x6 pin GPUs in a dual GPU setup in a Mac Pro unless you live in a cold clim and also need a warmer space heater]. CUDA generally scales linearly with Octane, at least. So a second GPU would be twice as fast and since Octane can, with a GTX 580, yield the equivalent speed increase in rendering of 10 to 50 general purpose CPUs you get that doubled. That's big bang.

DJenkins · Jan 23, 2013

Hi speters, definitely go with what Tutor is saying in regards to the 12 core machine. You will see a greater benefit in directing your cash away from the X5690 CPUs and going for the X5680 CPUs and a good GPU and maybe fixing up your old machine.

For example - I also have the same 2006 8 core upgraded to 2,1 machine with the 3.0GHz processors. It gets above 9000 on geekbench which will add around 30% improvement to your Cinema 4D net render times alongside your 12 core. So this could be worth pursuing if there's not too much cost involved. Better than a 4% increase with the X5690's alone as far as distributed CPU rendering goes.

Now there is a lot of talk around CUDA in the Adobe apps. Maybe I haven't given it a long enough chance but if you are looking to use CUDA to produce 3D graphics inside after effects I found it a bit limited. The initial excitement for me was being able to integrate basic 3D directly inside AE without having to render a C4D scene. However, it's not all that integrated. You can't use blending modes, track mattes, masks and other various handy things... so you will end up having a nest of pre-comps and pre-renders which sort of defeats the purpose. In the end I sort of gave up and thought that I could make it look better in C4D anyway!

To quote an Adobe rep - "This feature was not intended to replace dedicated 3D applications. It was meant to give simple text and vector graphics extrusions to people who don't have or know how to use a 3D application like Cinema 4D or Maya. If you have those programs, use them for what they're great at.

The things that we were trying to replace were the hacky, ridiculous workarounds that we saw people using in After Effects to get 3D-ish text by stacking hundreds of layers or using a quirk in the Shatter effect."

Your situation could be totally different and there would of course be benefits if you were creating graphics within these parameters. And don't get me wrong, you will still see huge improvements in most apps after sticking one or two GTX580 cards in your machine!

Good luck!

speters · Jan 26, 2013

Used 5680's

Thanks guys, Can anyone recommend a good place to get used 5680's and 5635 chips?

Tutor · Jan 27, 2013

speters said:
Thanks guys, Can anyone recommend a good place to get used 5680's and 5635 chips?

I'd search for them on E-bay, especially from Dinsdale, but I'd also recommend that you search for low priced new ones there. Used ones usually have very short warrantees and therefore it's difficult to tell whether you've made a wise choice before you've gotten stuck with a bad one and if you search this forum you can probably find the thread where some members are said to have found some for significantly under $1k.

All We Know About Maximizing CPU Related Performance

macrumors regular

macrumors 65816

macrumors regular

macrumors 65816

Attachments

macrumors 65816

Attachments

macrumors newbie

macrumors 65816

macrumors 6502

macrumors newbie

macrumors 6502

macrumors newbie

macrumors 65816

macrumors newbie

macrumors 65816

macrumors newbie

macrumors 65816

macrumors newbie

macrumors 65816

macrumors newbie

macrumors 65816

macrumors member

macrumors 65816

macrumors 6502

macrumors member

macrumors 65816

Our Staff