Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
All We Know About Maximizing CPU Related Performance
- Thread starter Tutor
- Start date
- Sort by reaction score
Say,"Good bye Mac Pro Truck" and ... .
Finally, Apple has announced the 2013 or 2014 Mac Pro [ http://www.apple.com/mac-pro/ ], which is 1/8 the size of earlier models. On a positive note is hawked as having 2x the performance of the 2012 model, but much of that is being attributable to the GPUs. CNet concludes that the only internal components that will be internally upgradeable will be the memory [ http://reviews.cnet.com/8301-13727_7-57588607-263/how-upgradable-is-the-new-mac-pro/ ] and that there will probably be third party companies creating replacement flash memory modules. For me, such limited internal upgradability is a no go. The new Mac Pro will come standard with two ATI Fire Pro GPUs. For me, no CUDA is a no go. Thus, if Apple exercises the will power to price the Cylinder right for what it truly is, i.e., an extension of the iToys, then Apple may be able to grow the base of "Mac Pro" owners. Since Apple has tied the top end new Mac Pro to 12-core Ivy Bridge Xeons, the release date of the new Mac Pro can slip and now be blamed on Intel if the Ivy Bridge Xeon release date slips.
Finally, Apple has announced the 2013 or 2014 Mac Pro [ http://www.apple.com/mac-pro/ ], which is 1/8 the size of earlier models. On a positive note is hawked as having 2x the performance of the 2012 model, but much of that is being attributable to the GPUs. CNet concludes that the only internal components that will be internally upgradeable will be the memory [ http://reviews.cnet.com/8301-13727_7-57588607-263/how-upgradable-is-the-new-mac-pro/ ] and that there will probably be third party companies creating replacement flash memory modules. For me, such limited internal upgradability is a no go. The new Mac Pro will come standard with two ATI Fire Pro GPUs. For me, no CUDA is a no go. Thus, if Apple exercises the will power to price the Cylinder right for what it truly is, i.e., an extension of the iToys, then Apple may be able to grow the base of "Mac Pro" owners. Since Apple has tied the top end new Mac Pro to 12-core Ivy Bridge Xeons, the release date of the new Mac Pro can slip and now be blamed on Intel if the Ivy Bridge Xeon release date slips.
Last edited:
I think Apple has finally pushed me past the brink of building a Hack as my next machine. So when the time comes (not soon), I'll be knocking on your door for some help, Dr. Tutor.
They mention that there will probably be third party companies creating replacement flash memory modules, and I agree with that.
Repurposing old ATI video cards - I want apps that also take advantage of OCL 1.0
At one time (in the not too distant past) I favored ATI video cards solely for their OpenGL [ http://en.wikipedia.org/wiki/OpenGL ] display rendering ability. Although then I also owned Nvidia video cards (such as GTX 295s and GTX 480s, they were no match for the OpenGL ability of my overclocked 4890s and 5970s, especially in my multi OS systems. Not until PunkNugget and others helped to get me really interest in CUDA [ http://en.wikipedia.org/wiki/CUDA ] compute capability did I pull my Nvidia cards out of storage and learn about CUDA for my parallel compute needs.
Regardless what anyone thinks about the recently announced Mac Pro or what one might think that I think of it, for selfish monetary reasons, I truly hope that it, as well as all things Apple, is a booming success. Moreover, it's inclusion of ATI Fire Pro GPUs could give OpenCL [ http://en.wikipedia.org/wiki/OpenCL ] compute development a much needed shot in the arm, beneficial to both OSX and Windows systems. Although both Nvidia video cards and ATI video cards support OpenCL, ATI cards do not support CUDA. However, in addition to ATI cards' excellence at OpenGL, they are also excellent at OpenCL.
After thinking about how I could leverage what I already own in light of Apple's rMP announcement, I reviewed some of my "old" video cards [Radeon 4890s and Radeon HD 5970s] using information contained here: http://en.wikipedia.org/wiki/Compar...essing_units#Radeon_R700_.28HD_4xxx.29_Series . This is how I evaluated those cards for potential future repurposing:
Model - Radeon HD 4890 vs Radeon HD 5970
Launch - Apr 2, 2009 vs Nov 18, 2009
Code name - RV790 XT vs Hemlock XT
Fab (nm)- 55 vs 40
Transistors (Million)- 959 vs 2154x2
Die Size (mm2) - 282 vs 334x2
Bus interface - PCIe 2.0 x16 vs PCIe 2.1 x16
Memory (MiB)1024 / 2048 vs 1024x2; 2048x2
Clock rate: Core (MHz)- 850 / Memory(MHz) - 975 vs 725 - 725 / 1000 - 1000
Config core1 - 800:40:16 vs 1600:80:32 ×2
Fillrate: Pixel (GP/s) - 13.6 / Texture (GT/s) - 34 vs 46.4 / 116.0
Memory Bandwidth (GB/s) - 124.8 / Bus type - GDDR5 / Bus width (bit) - 256 vs 128x2 / GDDR5 / 256x2
GFLOPS (Single-precision) - 1360 vs 4640
TDP3 (W) - 190 vs Idle - 51 Max. 294
GFLOPS/W (Single-precision) 7.16 vs 15.78
GFLOPS (Double-precision) - 272 [x3 = 816; +10% OC = 897.6] vs 928 [x3 = 2,784; +10% OC = 3,062.4]
5970 vs. 4890 double precision: 928 / 272 = 3.41
5970 vs. 4890 single precision: 4640 / 1360 = 3.41
So 3 Radeon 5970s have the compute capability of 10.23 Radeon 4890s.
In three systems, I can repurpose six of my ATI 4980 cards (I'll keep one of my other three in each of my three 2007 Mac Pros) and my three ATI 5970 cards. Each of those three systems use Gigabyte UD5 1366 motherboards (and i7 980X CPUs), each with 3 PCIe dual wide slots [with two being x16 and one being x8]. Compute capability of the system with 3 Radeon 5970s is [4,640 x 3 = 13,920; +10% OC = 15,312 or] 15.31 TFLOPS. {I wonder how Mari will run on this system since it over 2x 7 TFLOPS; better yet, how will Mari run on AlphaCanisLupus0 that has over 16 TFLOPS of double precision peak floating point performance and 50.1 TFLOPS of single precision peak floating point performance.} Base compute capability of each of two systems with three Radeon 4980s is [1,360 x 3 = 4,080; +10% OC = 4,488 or] 4.49 TFLOPS. So you Mac application software developers, let the OpenCL floods flow, but just make sure that those apps also support OpenCL 1.0 cards (my 4980s), even tho' those often mentioned Fire Pros include OpenCL 1.2 support. My 5970s should work just fine because they include OpenCL 1.2 support.
At one time (in the not too distant past) I favored ATI video cards solely for their OpenGL [ http://en.wikipedia.org/wiki/OpenGL ] display rendering ability. Although then I also owned Nvidia video cards (such as GTX 295s and GTX 480s, they were no match for the OpenGL ability of my overclocked 4890s and 5970s, especially in my multi OS systems. Not until PunkNugget and others helped to get me really interest in CUDA [ http://en.wikipedia.org/wiki/CUDA ] compute capability did I pull my Nvidia cards out of storage and learn about CUDA for my parallel compute needs.
Regardless what anyone thinks about the recently announced Mac Pro or what one might think that I think of it, for selfish monetary reasons, I truly hope that it, as well as all things Apple, is a booming success. Moreover, it's inclusion of ATI Fire Pro GPUs could give OpenCL [ http://en.wikipedia.org/wiki/OpenCL ] compute development a much needed shot in the arm, beneficial to both OSX and Windows systems. Although both Nvidia video cards and ATI video cards support OpenCL, ATI cards do not support CUDA. However, in addition to ATI cards' excellence at OpenGL, they are also excellent at OpenCL.
After thinking about how I could leverage what I already own in light of Apple's rMP announcement, I reviewed some of my "old" video cards [Radeon 4890s and Radeon HD 5970s] using information contained here: http://en.wikipedia.org/wiki/Compar...essing_units#Radeon_R700_.28HD_4xxx.29_Series . This is how I evaluated those cards for potential future repurposing:
Model - Radeon HD 4890 vs Radeon HD 5970
Launch - Apr 2, 2009 vs Nov 18, 2009
Code name - RV790 XT vs Hemlock XT
Fab (nm)- 55 vs 40
Transistors (Million)- 959 vs 2154x2
Die Size (mm2) - 282 vs 334x2
Bus interface - PCIe 2.0 x16 vs PCIe 2.1 x16
Memory (MiB)1024 / 2048 vs 1024x2; 2048x2
Clock rate: Core (MHz)- 850 / Memory(MHz) - 975 vs 725 - 725 / 1000 - 1000
Config core1 - 800:40:16 vs 1600:80:32 ×2
Fillrate: Pixel (GP/s) - 13.6 / Texture (GT/s) - 34 vs 46.4 / 116.0
Memory Bandwidth (GB/s) - 124.8 / Bus type - GDDR5 / Bus width (bit) - 256 vs 128x2 / GDDR5 / 256x2
GFLOPS (Single-precision) - 1360 vs 4640
TDP3 (W) - 190 vs Idle - 51 Max. 294
GFLOPS/W (Single-precision) 7.16 vs 15.78
GFLOPS (Double-precision) - 272 [x3 = 816; +10% OC = 897.6] vs 928 [x3 = 2,784; +10% OC = 3,062.4]
5970 vs. 4890 double precision: 928 / 272 = 3.41
5970 vs. 4890 single precision: 4640 / 1360 = 3.41
So 3 Radeon 5970s have the compute capability of 10.23 Radeon 4890s.
In three systems, I can repurpose six of my ATI 4980 cards (I'll keep one of my other three in each of my three 2007 Mac Pros) and my three ATI 5970 cards. Each of those three systems use Gigabyte UD5 1366 motherboards (and i7 980X CPUs), each with 3 PCIe dual wide slots [with two being x16 and one being x8]. Compute capability of the system with 3 Radeon 5970s is [4,640 x 3 = 13,920; +10% OC = 15,312 or] 15.31 TFLOPS. {I wonder how Mari will run on this system since it over 2x 7 TFLOPS; better yet, how will Mari run on AlphaCanisLupus0 that has over 16 TFLOPS of double precision peak floating point performance and 50.1 TFLOPS of single precision peak floating point performance.} Base compute capability of each of two systems with three Radeon 4980s is [1,360 x 3 = 4,080; +10% OC = 4,488 or] 4.49 TFLOPS. So you Mac application software developers, let the OpenCL floods flow, but just make sure that those apps also support OpenCL 1.0 cards (my 4980s), even tho' those often mentioned Fire Pros include OpenCL 1.2 support. My 5970s should work just fine because they include OpenCL 1.2 support.
Last edited:
Gee those ATI figures don't sound half bad until you mention what your AlphaCanisLupus0 is cranking out! Then again it's basically free compute power - minus electricity costs
that would otherwise go unused.Now I have a (possibly stupid) question which may go along with this thread and the future of maximizing CPU performance in custom built OSX running systems.
I know that OSX Mavericks is out in Beta, and we have been given a preview of the Mac Pro Tube... so, does this mean there may be functionality built into OSX Mavericks that ensures compatibility with the new Ivy Bridge Xeons and associated new hardware?
I'm talking about the sleep/speedstep etc issues with 2011 socket processors running in custom OSX builds. Or is this compatibility built in somewhere else?
As the appropriate Ivy Bridge Xeons are not released yet, is it possible there may be improvements for the existing Sandy Bridge line?
Building a dual processor 2011 socket machine has been on my mind lately but I've not done too much research into how successful they are because obviously there has been no comparable Apple machine to use a base reference building block.
It's probably true that Mavericks, when fully released, will contain the medicine for those ailments with both Sandy and Ivy Bridges, as well as with the i7s and the like, but I haven't had the opportunity to verify that yet.
I know this is likely answered definitively somewhere, but my Google-fu is weak today.
Can an X5687 (Quad core 3.6 Ghz) be used in a single processor 5,1 system?
I believe the answer is "Yes", but wouldn't mind a first-hand confirmation of such before I begin to search in earnest for one of these. (Id prefer to have better single-threaded performance on only 4 cores vs getting a hexa-core W3680/90)
Thanks in advance.
- Ck
Can an X5687 (Quad core 3.6 Ghz) be used in a single processor 5,1 system?
I believe the answer is "Yes", but wouldn't mind a first-hand confirmation of such before I begin to search in earnest for one of these. (Id prefer to have better single-threaded performance on only 4 cores vs getting a hexa-core W3680/90)
Thanks in advance.
- Ck
Absolutely! But you usually pay a premium for the dual processor capable 56xx series. I'm guessing that's not a problem for as from memory those processors weren't sold retail so you're probably getting a good deal from somewhere.
If you really wanted to I'm pretty sure you can put an i7-990X running at 3.46 GHz in there.
----------
Ahh Tutor, you just added fuel to my impatience
I know you can squeeze more out of a Windows system but c'mon, you've gotta be a little excited to see how far you can blow the next Mac Pro tube out of the water on it's own OS, right?
... .Ahh Tutor, you just added fuel to my impatience
I know you can squeeze more out of a Windows system but c'mon, you've gotta be a little excited to see how far you can blow the next Mac Pro tube out of the water on it's own OS, right?
When a Mac Pro achieves a 64-bit Geekbench score of 37,837 (which was my 2010 OSX geekbench 2 score), I might get a little excited that my 2011 OSX score of 40,100 on my system (as reconfigured that year) might be in jeopardy of being matched, but until then I'll curb my enthusiasm.
BTW- That reconfigured system is the one that houses the four Galaxy GTX 680 GC (OC'ed at 25%) [ http://www.galaxytech.com/__EN_GB__/Product2/ProductDetail?proID=18&isStop=0 ] 4/Gs referenced in my signature. That system has 15.4 TFLOP/s of GPGPU compute performance. So I'm not having any of that new Mac Pro envy-thing going on at the moment because Apple isn't even matching my yesteryear system performance of 2010(CPU)/2012(GPU).
Last edited:
Tutor isn't as locked into an OS as I am, choosing instead to use whatever tool he sees fit for the job at hand. I'm a bit more stubborn in that regard, requiring that my 'work' machine run UNIX. So depending on funding, I'll be seeing what a pair of V2 E5s can do in a Hack, once they're released and we're sure OS X can boot with them. I expect the pricing for those chips to nearly bankrupt me, but... hey, it's only money, right?
Tutor isn't as locked into an OS as I am, choosing instead to use whatever tool he sees fit for the job at hand. I'm a bit more stubborn in that regard, requiring that my 'work' machine run UNIX. So depending on funding, I'll be seeing what a pair of V2 E5s can do in a Hack, once they're released and we're sure OS X can boot with them. I expect the pricing for those chips to nearly bankrupt me, but... hey, it's only money, right?
I like for my systems to run OSX, Linux and all species of Windows, for as jasonvp points out, I prefer to use whatever tool I see fit for the job at hand. No OS do I love much more than another - they've all got their pros and cons and they reveal themselves to me constantly.
Well it's likely you could do some damage to your own score with a rig running dual E5-2687W V2 CPUs, if my estimations are vaguely on track.
I'm thinking around 51,000 - 53,000 may be a ballpark figure. Not that I know anything particularly technical about CPUs, but I put together a spreadsheet from existing geekbench data.
My estimates were based on:
The existing E5 results, taken from Windows score averages. Single and dual CPU scores of the same CPU model were compared and an average ratio was created between the two.
Mac equivalent scores were estimated by comparing averages from known scores across Win and Mac platforms in the X56xx Nehalem line.
The rumored 2013 Mac Pro E5 V2 single CPU score was used for comparison.
From here, estimates were drawn from the GHz and core count of the rest of the E5 V2 line leaked specs.
Yeh yeh in no way perfect by a long stretch but I'm just curious and my brain isn't happy about these things until I've at least fed it some fanciful numbers constructed out of thin air
It's quite likely if I ever get to build a machine like this that I will end up using it as a Windows box for other hardware advantages... but damn it would be handy to run OSX as well. Otherwise it would be the only Windows workstation out of about 20 Macs in the office!
Still I'd like to ask you Tutor, which score are you more proud of... The Windows 58,027 quad CPU, or your 40,100 OSX score running on underclocked magic that you worked so hard for...
I'm thinking around 51,000 - 53,000 may be a ballpark figure. Not that I know anything particularly technical about CPUs, but I put together a spreadsheet from existing geekbench data.
My estimates were based on:
The existing E5 results, taken from Windows score averages. Single and dual CPU scores of the same CPU model were compared and an average ratio was created between the two.
Mac equivalent scores were estimated by comparing averages from known scores across Win and Mac platforms in the X56xx Nehalem line.
The rumored 2013 Mac Pro E5 V2 single CPU score was used for comparison.
From here, estimates were drawn from the GHz and core count of the rest of the E5 V2 line leaked specs.
Yeh yeh in no way perfect by a long stretch but I'm just curious and my brain isn't happy about these things until I've at least fed it some fanciful numbers constructed out of thin air
It's quite likely if I ever get to build a machine like this that I will end up using it as a Windows box for other hardware advantages... but damn it would be handy to run OSX as well. Otherwise it would be the only Windows workstation out of about 20 Macs in the office!
Still I'd like to ask you Tutor, which score are you more proud of... The Windows 58,027 quad CPU, or your 40,100 OSX score running on underclocked magic that you worked so hard for...
It's a matter of perspective.
My 2012 self-built WolfPackPrime0 (that achieved a Geekbench 2 score of 58,027 when I tested it with Linux), my 2013 self-built WolfPackPrime1 (same configuration as WolfPackPrime0), my 2010 self-built WolfPack1 (that achieved a Geekbench 2 score of 40,100 when I tested it with OSX) and my self-built WolfPack2 (same configuration as WolfPack1) each cost me much less than I have invested in my 2009 Mac Pro. I am more satisfied with the performance of WolfPackPrime0 and WolfPackPrime1 than I am with either WolfPack1 or WolfPack2. If I had to sacrifice either WolfPackPrime0 or WolfPack1, then I would continue to own WolfPackPrime0. But I do consider WolfPack1 and WolfPack2 to have more of my personalization imbedded in them since with WolfPack1 I discovered how to underclock, for maximum performance, Westmere and Nehalem Xeon CPUs more like Sandy/Ivy Bridge CPUs would be clocked (low idle, but high TurboBoost potential) before I ever heard of either Sandy or Ivy Bridge CPUs. Underclocking is beneficial in OSX, Windows and Linux. So if you were to call that feeling of achievement "proud," then I have to say that WolfPack1 would get the nod over WolfPackPrime0.
... . Still I'd like to ask you Tutor, which score are you more proud of... The Windows 58,027 quad CPU, or your 40,100 OSX score running on underclocked magic that you worked so hard for...
My 2012 self-built WolfPackPrime0 (that achieved a Geekbench 2 score of 58,027 when I tested it with Linux), my 2013 self-built WolfPackPrime1 (same configuration as WolfPackPrime0), my 2010 self-built WolfPack1 (that achieved a Geekbench 2 score of 40,100 when I tested it with OSX) and my self-built WolfPack2 (same configuration as WolfPack1) each cost me much less than I have invested in my 2009 Mac Pro. I am more satisfied with the performance of WolfPackPrime0 and WolfPackPrime1 than I am with either WolfPack1 or WolfPack2. If I had to sacrifice either WolfPackPrime0 or WolfPack1, then I would continue to own WolfPackPrime0. But I do consider WolfPack1 and WolfPack2 to have more of my personalization imbedded in them since with WolfPack1 I discovered how to underclock, for maximum performance, Westmere and Nehalem Xeon CPUs more like Sandy/Ivy Bridge CPUs would be clocked (low idle, but high TurboBoost potential) before I ever heard of either Sandy or Ivy Bridge CPUs. Underclocking is beneficial in OSX, Windows and Linux. So if you were to call that feeling of achievement "proud," then I have to say that WolfPack1 would get the nod over WolfPackPrime0.
Last edited:
Want space to grow speedily? No current peripheral interconnect option beats PCIe.
There're lots of threads surmising how TB switches fare against PCIe. As a switch, TB on the nMP, using PCIe x4, is far slower than PCIe x16 and TB can't exist on it's own. If you have a need for speed, PCIe x16 solutions are where it's at. Tyan server boards give you that in spades and you can avoid a snake's nest of interconnect and power supply cables and the financial drain of external peripherals and containers for them. I really like the TYAN FT72-B7015 [ http://www.superbiiz.com/detail.php...3c789f87a210&gclid=CIyixZePlLgCFbNj7AodzxUARA ] - a Dual CPU Nehalem/Westmere LGA 1366 (~$3,700 w/o CPUs, ram and storage) server board. It has ten PCIe 2 x16 slots (two of which have x4 signals and the other eight can cradle eight double wide PCIe cards at x16 signal speed), two PCI-Ex1 slots and one PCI 32 bit slot. But for those with a need for PCIe 3 interconnect, there's the soon to be released TYAN B7059F77AV6R [ http://www.tyan.com/product_SKU_spec.aspx?ProductType=BB&pid=512&SKU=600000346 ], with eight PCI-E Gen3 x16 slots, two PCI-E Gen3 x8 slots (one for mezzanine card), three PCI-E Gen2 x1 slots, and one PCI 32-bit slot. It appears that this Sandy/Ivy Bridge version will cost about $5,200 w/o CPUs, ram and storage.
There're lots of threads surmising how TB switches fare against PCIe. As a switch, TB on the nMP, using PCIe x4, is far slower than PCIe x16 and TB can't exist on it's own. If you have a need for speed, PCIe x16 solutions are where it's at. Tyan server boards give you that in spades and you can avoid a snake's nest of interconnect and power supply cables and the financial drain of external peripherals and containers for them. I really like the TYAN FT72-B7015 [ http://www.superbiiz.com/detail.php...3c789f87a210&gclid=CIyixZePlLgCFbNj7AodzxUARA ] - a Dual CPU Nehalem/Westmere LGA 1366 (~$3,700 w/o CPUs, ram and storage) server board. It has ten PCIe 2 x16 slots (two of which have x4 signals and the other eight can cradle eight double wide PCIe cards at x16 signal speed), two PCI-Ex1 slots and one PCI 32 bit slot. But for those with a need for PCIe 3 interconnect, there's the soon to be released TYAN B7059F77AV6R [ http://www.tyan.com/product_SKU_spec.aspx?ProductType=BB&pid=512&SKU=600000346 ], with eight PCI-E Gen3 x16 slots, two PCI-E Gen3 x8 slots (one for mezzanine card), three PCI-E Gen2 x1 slots, and one PCI 32-bit slot. It appears that this Sandy/Ivy Bridge version will cost about $5,200 w/o CPUs, ram and storage.
Last edited:
The B7059F77AV6R looks very interesting. Seems pricey, but I guess the 2+1 redundant power supplies are in there already right?
Would that be ready to run several Titans or would you need to cram another power supply in there?
The array of PCIe options is incredible!
Tyan and Supermicro barebones systems are the new xMac, xLinux, xWindows systems
Pricey in the abstract, unless you don't need that many slots and went with a lower priced system with 3 or 4 single wide PCIe slots. But if one wanted room to grow, this would be the way to go and grow. It appears that for the near future, GPGPU (e.g., Teslas and Titans) and Co-Processor (XeonPhi) cards will dominate the CPU for performance improvement. Whereas each new Xeon CPU, which Intel has now locked down from tweaking, adds about 5% to 15% performance improvement to what it replaces, the GPGPUs (which are still extremely tweakable especially in their gaming versions) and Co-Processors add many times those tiny GPU performance improvements. A Titan, when unleashed, has the compute potential of a Tesla K20x, which has many times the compute potential of a pair of E5-2687Ws [ http://www.nvidia.com/docs/IO/122874/K20-and-K20X-application-performance-technical-brief.pdf ]. If we were to face the cold, hard facts, we'd have to admit that a pair of top end Westmeres offer more than enough compute performance for over 90% of computer users for 100% of what they now or will do on a computer for many years to come, but for those few with compute intensive tasks where parallel processing can play a role, GPGPUs and CoProcessors offer, and will offer for the near term, significantly greater advantages than Tick Tock.
If you go to BHphoto&video's site [ http://www.bhphotovideo.com/c/searc...&N=0&InitialSearch=yes&sts=ma&Top+Nav-Search= ], there you'll find the Cubix and Magma 8 PCIe, dual wide slot, headless chassis from about $7K to $10K+, with a one puny PSU (usually about 850W). From below the middle of that price range, you can fully outfit a B7059F77AV6R, loading the head with Sandy or Ivy Bridge CPUs, ram and storage. Moreover, compared to what a similarly configured SB/IB Mac Pro, if there were dualies, would cost, you'd see the B7059F77AV6R as a bargain, just as I view a fully configured TYAN FT72-B7015 as a bargain compared to the 2010 Mac Pro, especially when you consider that you're only buying what you need. Many Mac Pro owners have spent thousands of dollars more to upgrade their systems because they had already paid for CPUs, ram chips and GPUs they didn't really want but were force to pay for and with the added Apple tax. Additionally and curiously, the 12-core 2010 Mac Pros were a bargain when compared to the HPs and Dells of its day (Apple hadn't fully cornered the greed market).
Pricey in the abstract, unless you don't need that many slots and went with a lower priced system with 3 or 4 single wide PCIe slots. But if one wanted room to grow, this would be the way to go and grow. It appears that for the near future, GPGPU (e.g., Teslas and Titans) and Co-Processor (XeonPhi) cards will dominate the CPU for performance improvement. Whereas each new Xeon CPU, which Intel has now locked down from tweaking, adds about 5% to 15% performance improvement to what it replaces, the GPGPUs (which are still extremely tweakable especially in their gaming versions) and Co-Processors add many times those tiny GPU performance improvements. A Titan, when unleashed, has the compute potential of a Tesla K20x, which has many times the compute potential of a pair of E5-2687Ws [ http://www.nvidia.com/docs/IO/122874/K20-and-K20X-application-performance-technical-brief.pdf ]. If we were to face the cold, hard facts, we'd have to admit that a pair of top end Westmeres offer more than enough compute performance for over 90% of computer users for 100% of what they now or will do on a computer for many years to come, but for those few with compute intensive tasks where parallel processing can play a role, GPGPUs and CoProcessors offer, and will offer for the near term, significantly greater advantages than Tick Tock.
If you go to BHphoto&video's site [ http://www.bhphotovideo.com/c/searc...&N=0&InitialSearch=yes&sts=ma&Top+Nav-Search= ], there you'll find the Cubix and Magma 8 PCIe, dual wide slot, headless chassis from about $7K to $10K+, with a one puny PSU (usually about 850W). From below the middle of that price range, you can fully outfit a B7059F77AV6R, loading the head with Sandy or Ivy Bridge CPUs, ram and storage. Moreover, compared to what a similarly configured SB/IB Mac Pro, if there were dualies, would cost, you'd see the B7059F77AV6R as a bargain, just as I view a fully configured TYAN FT72-B7015 as a bargain compared to the 2010 Mac Pro, especially when you consider that you're only buying what you need. Many Mac Pro owners have spent thousands of dollars more to upgrade their systems because they had already paid for CPUs, ram chips and GPUs they didn't really want but were force to pay for and with the added Apple tax. Additionally and curiously, the 12-core 2010 Mac Pros were a bargain when compared to the HPs and Dells of its day (Apple hadn't fully cornered the greed market).
I run all 8 Titans fully tweaked on the factory PSUs without any additional PSU assist. The Tyan that I purchased has three 1000W PSUs rated as follows: 2400W [(2+1) 2400W @200-240V], Max. 12Vdc@ 199.6A / 3000W [3 x1000W @100-127V], Max. 12Vdc@ 249.6A; But note: Only one AC inlet allowed per circuit breaker.
I'm finding it difficult to think of any add-on/accessory that one would desire to add to a system where a direct, internal PCIe add-on connection would not be the preferable route to go. And here with the TYAN FT72-B7015, you begin with 10 open x16 slots. That's, at the very least, 2 to 3 times what you'd normally get for PCIe expansion. So that's like getting 2 to 3 systems (actually many, many more if you're loading in Teslas, Titans and/or XeonPhis) in one. In this context, I submit that the TYANs are dirt cheap because my 8 tweaked Titans have together have a single precision floating point peak compute potential in excess of 50 Tflops and double precision floating point peak compute potential in excess of 16 Tflops. How many CPU-only based systems would I need to have the equal? Nvidia's answer would be about 80 DP E5-2687W seats. What if that (80) number were truly only 1/2 (still a bargain) or 1/3 (still a bargain) or 1/4 (still a bargain) or even 1/5 (still a bargain) as large. So that's why I began, "Pricey in the abstract."
Last edited:
Thought I would stop in and let everyone know that we have working PM for LGA 2011 under 10.9 DP. I did however have to patch AICPM.kext and we do not have working turbo yet.
All chipset drivers including x79/C600 Series SATA are present. Still not understanding why Apple is added support for SATA for this chipset since they are not releasing a Mac Pro with SATA ports unless there is something they did not tell us a WWDC. Just a thought. The SATA driver was the only one missing in DP1 for this chipset and was added in DP2. Used the generic id in DP1.
All chipset drivers including x79/C600 Series SATA are present. Still not understanding why Apple is added support for SATA for this chipset since they are not releasing a Mac Pro with SATA ports unless there is something they did not tell us a WWDC. Just a thought. The SATA driver was the only one missing in DP1 for this chipset and was added in DP2. Used the generic id in DP1.
As always, you're full of helpful and innovation information. And as always, thank you for sharing.
When time is of the essence, Big Bang for the bucks for After Effects animators.
In a recent thread began by MacVidCards here [ https://forums.macrumors.com/threads/1610520/ ], he references Barefeats test of multiple gpu cards [ http://barefeats.com/gpu680v6.html ], where, in turn, Barefeats refrences Danny Princz's performance chart, compiling the render times for up to three GPUs.
Danny's performance chart looks like an a directive to those making a living doing After Effects (AE) animations to buy an 8 dual wide x16 Pcie slotted system and to load it with 8 $1000 GTX 690s. A render that a base Mac Pro 5,1 (12 x 2.66) takes 415 minutes (6 hrs. 55 minutes) to render, is rendered by one GTX 690 card in a little under 10 (9.56) minutes. To the extent that additional GTX 690 cards add linearly to the gain, a second such card would cut the render time to under 5 minutes (4.78 minutes vs. 415 minutes). Four such cards would cut the render time to about 2 1/2 minutes (2.39 minutes vs. 415 minutes) and eight such cards would cut the render time to about 1.2 minutes (1.195 minutes vs. 415 minutes). So for under $15k, with eight GTX 690s in a server, like my Tyan, one could have, in one chassis, the animation rendering prowess in AE to render in just over one minute what a base Mac Pro 5,1 (415/1.195) would take 6 hrs and 55 minutes to render. Talking about a GPGPU victory and Big Bang for the bucks when the deadline is fast approaching - get muliple, fast CUDA cards!
In a recent thread began by MacVidCards here [ https://forums.macrumors.com/threads/1610520/ ], he references Barefeats test of multiple gpu cards [ http://barefeats.com/gpu680v6.html ], where, in turn, Barefeats refrences Danny Princz's performance chart, compiling the render times for up to three GPUs.
Danny's performance chart looks like an a directive to those making a living doing After Effects (AE) animations to buy an 8 dual wide x16 Pcie slotted system and to load it with 8 $1000 GTX 690s. A render that a base Mac Pro 5,1 (12 x 2.66) takes 415 minutes (6 hrs. 55 minutes) to render, is rendered by one GTX 690 card in a little under 10 (9.56) minutes. To the extent that additional GTX 690 cards add linearly to the gain, a second such card would cut the render time to under 5 minutes (4.78 minutes vs. 415 minutes). Four such cards would cut the render time to about 2 1/2 minutes (2.39 minutes vs. 415 minutes) and eight such cards would cut the render time to about 1.2 minutes (1.195 minutes vs. 415 minutes). So for under $15k, with eight GTX 690s in a server, like my Tyan, one could have, in one chassis, the animation rendering prowess in AE to render in just over one minute what a base Mac Pro 5,1 (415/1.195) would take 6 hrs and 55 minutes to render. Talking about a GPGPU victory and Big Bang for the bucks when the deadline is fast approaching - get muliple, fast CUDA cards!
Last edited:
GO CUDA & GOOD FOR YOU DANNY!!! I'm impressed. It's nice to have the power when you need it; especially at those speeds. On my end I actually turned down the UC and OC on my unit and just using it on it's stock speed. I have enough horsepower with both of my GTX 580's 3GB & Tesla 2070 6GB GPU. For the money that I paid for it, I'm grateful that I'm crankin' out the performance that currently have.
I'm still disappointed that Apple didn't put out a SLAMMIN' machine and have regressed a bit in their new MacPro that's coming out here shortly. Sure it's smaller (way smaller), but I believe people want SPEED and PERFORMANCE. Even if they made it half the size of the current MacPro they could've shoved more things in there to make it more powerful and utilized nVidia's performance with where Adobe is going in utilizing CUDA. To me it's such a backwards move on Apple's part to stay with ATI. But that's my thought on it. Oh well, it doesn't matter because I got what I wanted with what I have and I'm happy as I'm sure Danny is with his BEAST !!!
He should post his build here:
http://www.insanelymac.com/forum/forum/295-macmod-of-the-month/
As we always could use a new entry. I'm sure he'd get some great votes in...
I'm still disappointed that Apple didn't put out a SLAMMIN' machine and have regressed a bit in their new MacPro that's coming out here shortly. Sure it's smaller (way smaller), but I believe people want SPEED and PERFORMANCE. Even if they made it half the size of the current MacPro they could've shoved more things in there to make it more powerful and utilized nVidia's performance with where Adobe is going in utilizing CUDA. To me it's such a backwards move on Apple's part to stay with ATI. But that's my thought on it. Oh well, it doesn't matter because I got what I wanted with what I have and I'm happy as I'm sure Danny is with his BEAST !!!
He should post his build here:
http://www.insanelymac.com/forum/forum/295-macmod-of-the-month/
As we always could use a new entry. I'm sure he'd get some great votes in...
GO CUDA & GOOD FOR YOU DANNY!!! I'm impressed. It's nice to have the power when you need it; especially at those speeds. On my end I actually turned down the UC and OC on my unit and just using it on it's stock speed. I have enough horsepower with both of my GTX 580's 3GB & Tesla 2070 6GB GPU. For the money that I paid for it, I'm grateful that I'm crankin' out the performance that currently have.
I'm still disappointed that Apple didn't put out a SLAMMIN' machine and have regressed a bit in their new MacPro that's coming out here shortly. Sure it's smaller (way smaller), but I believe people want SPEED and PERFORMANCE. Even if they made it half the size of the current MacPro they could've shoved more things in there to make it more powerful and utilized nVidia's performance with where Adobe is going in utilizing CUDA. To me it's such a backwards move on Apple's part to stay with ATI. But that's my thought on it. Oh well, it doesn't matter because I got what I wanted with what I have and I'm happy as I'm sure Danny is with his BEAST !!!
He should post his build here:
http://www.insanelymac.com/forum/forum/295-macmod-of-the-month/
As we always could use a new entry. I'm sure he'd get some great votes in...
Hey hold on a sec, if you're talking about me I haven't actually started anything yet!
I'll be waiting for the release of the E5 2687W V2 processors, and perhaps to see if the 2011 Gigabyte board is Mac friendly. If not, a Tyan system running Windows might be the way to go. I'm sure other machines will surface before I build mine but who knows
Just a follow up - I've secured myself an X5687 on a good deal - should be here in a few days. I'll post my results if I can get 'er working in my 4,1 flashed 5,1.
I know Xavi8tor had tried this route before on an X5687 with no luck - but was unsure if the die was bad. Worse comes to worse, the same Vendor I bought from also has a W3690 for almost the same price - so it wont be a total loss if the chip isn't recognised by the MP. (it's a standard Westmere B1 stepping die - so in theory it works out-of-the-box)
/me crosses fingers.