Thanks for the introduction to Rampage Dev and any part you may have played in his coming to this thread and to this forum.
Last edited:
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
Dude. Bible quotes?
Tutor has brought a ridiculous amount to this thread. Don't derail it with your off topic indulgences.
Totally agree. That was really out of place Punk. (AKA inappropriate!) In fact your entire challenging line of comments are out of place. Please take remarks like these over to christian-ethic-debate-forums dot com and let us get back to All We Know About Maximizing CPU Related Performance.
Last edited:
If you're into rendering 3d animations, here's something to consider.
Below is a pic of how various video cards have performed on the Lux Mark OpenCL rendering test [ http://www.luxrender.net/luxmark/ ]. The Lux Mark test is OpenCL (OCL) only. OpenCL [ http://en.wikipedia.org/wiki/OpenCL ] and CUDA [ http://en.wikipedia.org/wiki/CUDA ] are two avenues for harnessing the power of the video card (really the graphics processing unit {GPU} in it) to do things that once only central processing units (CPUs) like the Intel or AMD chips in your computer can do. GPUs are, to date, more powerful than CPUs. CUDA empowers only the GPU. OpenCL empowers GPUs, CPUs, and some digital signal processing units (DSPs). ATI does only OCL. Nvidia's video cards do both OCL and CUDA. There are many more 3d applications that benefit from CUDA's acceleration than from OCL's. Your mileage in CUDA would likely differ, but the relative standing I would expect to stay the same.
Below is a pic of how various video cards have performed on the Lux Mark OpenCL rendering test [ http://www.luxrender.net/luxmark/ ]. The Lux Mark test is OpenCL (OCL) only. OpenCL [ http://en.wikipedia.org/wiki/OpenCL ] and CUDA [ http://en.wikipedia.org/wiki/CUDA ] are two avenues for harnessing the power of the video card (really the graphics processing unit {GPU} in it) to do things that once only central processing units (CPUs) like the Intel or AMD chips in your computer can do. GPUs are, to date, more powerful than CPUs. CUDA empowers only the GPU. OpenCL empowers GPUs, CPUs, and some digital signal processing units (DSPs). ATI does only OCL. Nvidia's video cards do both OCL and CUDA. There are many more 3d applications that benefit from CUDA's acceleration than from OCL's. Your mileage in CUDA would likely differ, but the relative standing I would expect to stay the same.
Attachments
Last edited:
What? WTF?
A 580 beats two 590's?
Two 590's beats two 690's?
A 480 beats a Quadro 7000?
That list is mux!
A 580 beats two 590's?
Two 590's beats two 690's?
A 480 beats a Quadro 7000?
That list is mux!
I included the URL for the curious. I picked the top score for each kind of card. These aren't necessarily running at factory speed(s). Owners can overclock them. Also, at factory speeds some cards, like the 580's, are faster than others with the same category designation (not all GTX 580s are marketed to be equal) and usually sell for a bit more. Moreover, some rendering pipelines do not scale linearly for each GPU when rendering.
That's one 590 that has 2 GPUs. But the end result doesn't surprise me at all. I believe that the difference is due to overclocking and having a fast 580.
That's one 590 that has 2 GPUs and one 690 that has 2 GPUs. But the end result doesn't surprise me at all. I believe that the difference is due to Fermi vs. Kepler, with Fermi being much better at rendering. One 580 cleans a 680's clock when rendering, but for gaming or video display the reverse would be true.
Does the Quadro 7000 use Fermi or Kepler cores ( I can't find info on that right now)? If Kepler, then no surprise here either. If Fermi, then I too would be surprised.
Last edited:
Hmmm, Umm-kay.
I haven't been able to find anything that specifically tells me whether the Quadro 7000 is a Fermi card, but given that it's (1-4) GPUs are 512-cores each, that makes me think that it is a Fermi card (the GTX 500 series cards like the 580 have 512 cores); thus, that a GTX 480 (480-cores) could render a scene faster than a Quadro 7000 w/2 or more GPUs would be a big surprise. But if it was a Quadro 7000 with only 1 of it's 4 GPUs, I be just slightly surprised. Maybe that's why it doesn't show as a 2x (or 3x or 4x) card like the other multi GPU video cards show in Lux Mark's database..
Last edited:
no sleep hackintosh help
hey guys,
reaching out here since i haven't been able to get an answer in another forum regarding my hackintosh build. not sure if i will get it here, either, but thought i'd try anyway. well, the problem and only problem i have with my hackintosh is that i can never let it sleep or i will have to hard reset the machine. i basically can't wake the computer once it goes to sleep. everything else works. well, my usb3 ports doesn't work but is livable since i don't have any usb3 devices, anyway. what is sort of frustrating, though, is that i can't leave my computer alone since it can't sleep, so, i have to either turn off the computer or go to windows land where i go to 90% of the time anyway to play pc games. os x is relegated for editing videos, photos and illustrations. i am an aspiring editor/filmmaker/artist and built a hackintosh for the extra horsepower that an i7-3770k will give me, compared to my mid-2010 MBP, which is just a dual-core i7 (1st gen) with hyper threading. my mobo is an asus sabertooth z77 and my gpu is a gtx 680, which is recognized. so, the only thing it lacks that i wish it could do is sleep. can anyone point me to the right direction regarding a fix? or is this just something i have to live with? thanks in advance everyone.
hey guys,
reaching out here since i haven't been able to get an answer in another forum regarding my hackintosh build. not sure if i will get it here, either, but thought i'd try anyway. well, the problem and only problem i have with my hackintosh is that i can never let it sleep or i will have to hard reset the machine. i basically can't wake the computer once it goes to sleep. everything else works. well, my usb3 ports doesn't work but is livable since i don't have any usb3 devices, anyway. what is sort of frustrating, though, is that i can't leave my computer alone since it can't sleep, so, i have to either turn off the computer or go to windows land where i go to 90% of the time anyway to play pc games. os x is relegated for editing videos, photos and illustrations. i am an aspiring editor/filmmaker/artist and built a hackintosh for the extra horsepower that an i7-3770k will give me, compared to my mid-2010 MBP, which is just a dual-core i7 (1st gen) with hyper threading. my mobo is an asus sabertooth z77 and my gpu is a gtx 680, which is recognized. so, the only thing it lacks that i wish it could do is sleep. can anyone point me to the right direction regarding a fix? or is this just something i have to live with? thanks in advance everyone.
Try looking at URLs below and if the solution's not there, google this: i7-3770k hackintosh asus sabertooth z77 won't wake from sleep
http://www.projectosx.com/forum/index.php?showtopic=2565
http://www.osx86.net/guides-tutorials/19291-guide-clover-efi-bootloader-36.html
http://www.osx86.net/guides-tutorials/19291-guide-clover-efi-bootloader-140.html
http://www.nvidia.com/object/quadro-fermi-overview.html
And this really long ULR.
And this one too.
Yeah, pretty sure it's Fermi.
Here's the 7000 spec sheet+ http://www.nvidia.com/object/product-quadroplex-7000-us.html I didn't think there were any variants... Are there?
Last edited:
hey tutor,
thanks for the swift reply. i will check out those links but quickly glancing at them already gives me a daunting feeling. and yeah, i have googled the heck out of asus sabertooth z77 hackintosh sleep issues....
thanks for the swift reply. i will check out those links but quickly glancing at them already gives me a daunting feeling. and yeah, i have googled the heck out of asus sabertooth z77 hackintosh sleep issues....
Its two - two - two really good cards in one.
A big thanks goes out to ATI and it's Tahiti and to Intel and its Xeon Phi. Both are putting pressure on Nvidia to be more competitive. Some think that NVidia's latest GTX card - the Titan - has no real place in the GTX lineup. Those critics are wrong. The $999+ GTX Titan video card has been configured to satisfy two audiences: (1) high end gamers and (2) CUDA programmers and developers needing better Double Precision floating point performance from a General Purpose Graphics Processing Unit (GPGPU). Titan brings some high end Tesla features and performance back into the GTX line that were lost with the GTX 600 series break from the compute performance model that the GTX 500 Fermi series had given many a taste for. If you're a programmer or developer needing more compute performance, but lacking the fat wallet to buy a Tesla card, Titan might serve your needs. If you're a high end gamer, GTX Titan may not play games as rapidly as the GTX 690, but the Titan does play them smoother (one heavyweight GPU carrying the load vs. two where that baton pass doesn't always go perfectly smooth).
The Double Precision floating point performance of a GPGPU compute device (i.e., video card) is generally more accurate, but also slower, than the device's Single Precision floating point performance. Double Precision floating point performance is vital where the numbers operated on have many digit places. The Double Precision floating point performance of a video card is important to those whose systems will be used for tasks such as 3d rendering, climate and weather modeling, CAE, CFD, computational physics, biochemistry simulations and computational finance. The card's Single Precision floating point performance is generally more important to those who use their systems for tasks such as audio/video signal processing and video analytics. Games mainly depend on Single Precision floating point performance. The only other Nvidia card that has a higher Single Precision floating point performance rating than Titan (4.5 Teraflops)'s is the Tesla K10 (4.58 Teraflops).
When a GTX Titan video cards is first removed from its original packaging, it's come set to provide the Double Precision floating point performance of 233.592 Gflops/s. That 233.592 Gflops/s figure is not at all shabby. Titan's performance against this measure is higher than any other single GPU in the GTX 600 series lineup. In fact, straight out of the box Titan's performance matches the 230+ Gflops/s Double Precision floating point performance of a GTX 580. (The GTX 580 is, thus, more powerful in that category than any GTX 600 series card, except maybe for the GTX 690 card that is really two GPUs, with each having about 130 Gflops/s of Double Precision floating point performance.) But depending on how the application that the GPU is processing, was coded, or a few other factors, a GTX 580 may outperform a GTX 590.
GTX Titan can be made to increase its Double Precision floating point performance in two ways. As indicated in post# 537, above, one is to go to the NVidia Control Panel software that you should have already installed and with the GTX Titan installed - access the "Manage 3D settings" selection under "3D Settings," then locate the feature called, "CUDA - Double precision" and select the Geforce GTX Titan option. Doing that will boost the Double Precision floating point value from 233.592 Gflop/s to 1,578.51 Gflop/s. So by making that simple selection you've increased the Double Precision floating point performance by 6.76 times (rounded - 1578.51 / 233.592 = 6.75755162848043) above factory setting. That's your first paycheck. But you have to pay taxes on those earning, so making that selection will down clock your GPU's speed. But you can immediately get that tax money back with your ability to add overclocking to that mix. The GTX Titan settings depicted in post # 546, above, are an example. With those settings, I was still able to have my Titan stay well under it's thermal and voltage ceilings, without any crashes, to achieve another increase in the Double Precision floating point performance, but by how much I cannot even tell using CPU-z (those figures in the pic indicate that the true value is out of range). However, the new level of performance caused my Lux Mark rendering score to go from about 1,500 to 1,760 (for an increase of a little over 17%). Thus, there's amazing potential in the GTX Titan for increasing CUDA performance to even much higher levels than the Double Precision floating point performance of the Tesla K20s ($3.2+K for the base card) and the Tesla K20X ($3.5+K for the base card).
But the Titan differs from the Tesla cards because the Tesla cards have ECC memory. That EEC memory operates about 12% slower than when the ECC memory function in Tesla is disabled. ECC memory in Tesla not only protects the DRAM, but also protects the register files, L1 and L2 caches and shared memory. Titan does not have ECC memory.
In Titan, certain other Tesla card features have been disabled (i.e., the Titan drivers don't activate them).
RDMA for GPU Direct is a Tesla feature that enables a direct path of communication between the Tesla GPU and another or peer device on the PCI-E buss of your computer, without CPU intervention. Device drivers can enable this functionality with a wide range of hardware devices. For instance, the Tesla card can be allowed to communicate directly with your Mercury Accelsior card without getting your Xeon or i7 involved. Titan does not support the RDMA for GPU Direct feature.
Hyper-Q for MPI is a Tesla card feature that enables multiple CPU cores to simultaneously utilize the CUDA cores on a single Kepler GPU, significantly increasing GPU utilization. MPI is a message-passing system designed to function on a wide variety of parallel computers. So Hyper-Q for MPI helps increase performance for thousands of legacy MPI applications without requiring significant code rewrites, by enabling multiple CPU threads or processes to launch work on a single GPU at the same time. Titan supports Hyper-Q for CUDA streams, but not for MPI connections. So the absence of Hyper-Q for MPI connections on Titan should not have a significant impact on programmers and developers who are more likely to need and will have Hyper-Q for CUDA streams.
Dynamic Parallelism enables a CUDA kernel to create work hierarchies, synchronizing new nested work. Basically, one CUDA Kernel calls another one within that parent CUDA kernel, to do some work on (computation) and then the parent receives that work on its completion, then uses it, all without CPU involvement. Titan has Dynamic Parallelism.
So all-in-all there's not much that Tesla has that Titan doesn't have that a programmer or developer needing more compute performance would need. Again, a big thanks goes out to ATI and it's Tahiti and to Intel and its Xeon Phi for helping Nvidia better serve its customer base. Ain't competition great?
A big thanks goes out to ATI and it's Tahiti and to Intel and its Xeon Phi. Both are putting pressure on Nvidia to be more competitive. Some think that NVidia's latest GTX card - the Titan - has no real place in the GTX lineup. Those critics are wrong. The $999+ GTX Titan video card has been configured to satisfy two audiences: (1) high end gamers and (2) CUDA programmers and developers needing better Double Precision floating point performance from a General Purpose Graphics Processing Unit (GPGPU). Titan brings some high end Tesla features and performance back into the GTX line that were lost with the GTX 600 series break from the compute performance model that the GTX 500 Fermi series had given many a taste for. If you're a programmer or developer needing more compute performance, but lacking the fat wallet to buy a Tesla card, Titan might serve your needs. If you're a high end gamer, GTX Titan may not play games as rapidly as the GTX 690, but the Titan does play them smoother (one heavyweight GPU carrying the load vs. two where that baton pass doesn't always go perfectly smooth).
The Double Precision floating point performance of a GPGPU compute device (i.e., video card) is generally more accurate, but also slower, than the device's Single Precision floating point performance. Double Precision floating point performance is vital where the numbers operated on have many digit places. The Double Precision floating point performance of a video card is important to those whose systems will be used for tasks such as 3d rendering, climate and weather modeling, CAE, CFD, computational physics, biochemistry simulations and computational finance. The card's Single Precision floating point performance is generally more important to those who use their systems for tasks such as audio/video signal processing and video analytics. Games mainly depend on Single Precision floating point performance. The only other Nvidia card that has a higher Single Precision floating point performance rating than Titan (4.5 Teraflops)'s is the Tesla K10 (4.58 Teraflops).
When a GTX Titan video cards is first removed from its original packaging, it's come set to provide the Double Precision floating point performance of 233.592 Gflops/s. That 233.592 Gflops/s figure is not at all shabby. Titan's performance against this measure is higher than any other single GPU in the GTX 600 series lineup. In fact, straight out of the box Titan's performance matches the 230+ Gflops/s Double Precision floating point performance of a GTX 580. (The GTX 580 is, thus, more powerful in that category than any GTX 600 series card, except maybe for the GTX 690 card that is really two GPUs, with each having about 130 Gflops/s of Double Precision floating point performance.) But depending on how the application that the GPU is processing, was coded, or a few other factors, a GTX 580 may outperform a GTX 590.
GTX Titan can be made to increase its Double Precision floating point performance in two ways. As indicated in post# 537, above, one is to go to the NVidia Control Panel software that you should have already installed and with the GTX Titan installed - access the "Manage 3D settings" selection under "3D Settings," then locate the feature called, "CUDA - Double precision" and select the Geforce GTX Titan option. Doing that will boost the Double Precision floating point value from 233.592 Gflop/s to 1,578.51 Gflop/s. So by making that simple selection you've increased the Double Precision floating point performance by 6.76 times (rounded - 1578.51 / 233.592 = 6.75755162848043) above factory setting. That's your first paycheck. But you have to pay taxes on those earning, so making that selection will down clock your GPU's speed. But you can immediately get that tax money back with your ability to add overclocking to that mix. The GTX Titan settings depicted in post # 546, above, are an example. With those settings, I was still able to have my Titan stay well under it's thermal and voltage ceilings, without any crashes, to achieve another increase in the Double Precision floating point performance, but by how much I cannot even tell using CPU-z (those figures in the pic indicate that the true value is out of range). However, the new level of performance caused my Lux Mark rendering score to go from about 1,500 to 1,760 (for an increase of a little over 17%). Thus, there's amazing potential in the GTX Titan for increasing CUDA performance to even much higher levels than the Double Precision floating point performance of the Tesla K20s ($3.2+K for the base card) and the Tesla K20X ($3.5+K for the base card).
But the Titan differs from the Tesla cards because the Tesla cards have ECC memory. That EEC memory operates about 12% slower than when the ECC memory function in Tesla is disabled. ECC memory in Tesla not only protects the DRAM, but also protects the register files, L1 and L2 caches and shared memory. Titan does not have ECC memory.
In Titan, certain other Tesla card features have been disabled (i.e., the Titan drivers don't activate them).
RDMA for GPU Direct is a Tesla feature that enables a direct path of communication between the Tesla GPU and another or peer device on the PCI-E buss of your computer, without CPU intervention. Device drivers can enable this functionality with a wide range of hardware devices. For instance, the Tesla card can be allowed to communicate directly with your Mercury Accelsior card without getting your Xeon or i7 involved. Titan does not support the RDMA for GPU Direct feature.
Hyper-Q for MPI is a Tesla card feature that enables multiple CPU cores to simultaneously utilize the CUDA cores on a single Kepler GPU, significantly increasing GPU utilization. MPI is a message-passing system designed to function on a wide variety of parallel computers. So Hyper-Q for MPI helps increase performance for thousands of legacy MPI applications without requiring significant code rewrites, by enabling multiple CPU threads or processes to launch work on a single GPU at the same time. Titan supports Hyper-Q for CUDA streams, but not for MPI connections. So the absence of Hyper-Q for MPI connections on Titan should not have a significant impact on programmers and developers who are more likely to need and will have Hyper-Q for CUDA streams.
Dynamic Parallelism enables a CUDA kernel to create work hierarchies, synchronizing new nested work. Basically, one CUDA Kernel calls another one within that parent CUDA kernel, to do some work on (computation) and then the parent receives that work on its completion, then uses it, all without CPU involvement. Titan has Dynamic Parallelism.
So all-in-all there's not much that Tesla has that Titan doesn't have that a programmer or developer needing more compute performance would need. Again, a big thanks goes out to ATI and it's Tahiti and to Intel and its Xeon Phi for helping Nvidia better serve its customer base. Ain't competition great?
Last edited:
http://www.nvidia.com/object/quadro-fermi-overview.html
And this really long ULR.
And this one too.
Yeah, pretty sure it's Fermi.
Here's the 7000 spec sheet+ http://www.nvidia.com/object/product-quadroplex-7000-us.html I didn't think there were any variants... Are there?
I take your word as being 100% correct. I can't find the exact article that I looked at, having looked at so many of them, that gave me the impression that one could purchase up to 4 GPUs in a Quadro 7000 system. However, since you originally posed the concern, two other possible explanations come to mind: (1) in many software packages that I've seen and Lux Render is one of them, you have to select each CPU and/or GPU for rendering assist and if your card has more than one GPU (like my GTX 690), you can select one or both of them to assist and maybe this tester selected only one if not all get selected by default or deselected one or (2) I suspect that when you are marketing a device in this price range that you don't put it on the market without spending a lot of time tuning the drivers for the specific software that the expected/projected customer base would be using. Lux Render is free and not a likely candidate that I would expect Nvidia to be at all concerned with when it was tuning its drivers for such an expensive video card system. Surely, their customer base who spent many thousands of dollars on a video card system would not most likely be intending to run free software such as Lux Render or Blender on it. So it would not surprise me at all now if the hardware that does best on certain applications is more in line with the price of the application that the hardware manufacturer thought that it's customers would be using. Even if someone at Nvidia had raised the concern about tuning the drivers for Lux Render or Blender, someone with final authority over that cost expenditure probably would have responded just as you did earlier, by saying, "WTF."
With that said, I believe the render rankings are what they are and that is, for example, that a particular GTX 480 has the facility to render a specific scene in Lux Mark faster than a particular Quadro 7000. Would all of the relative rankings be the same in Softimage, Houdini, Maya, Lightwave, Modo, Cinema 4d, Z-Brush, Bryce, Terragen, Groboto, 3-d Coat, Poser, Cheetah 3d, Vue, Electric Image, etc. (and lets not forget that a specific scene probably cannot be rendered at the same speed by the same GPU in all versions of a particular software package)? So, I'm sure that the relative rankings wouldn't be exactly the same across all software packages no matter the version, but I would expect that for the vast majority of them that the overall hardware rankings wouldn't be vastly different in the near term, but will almost certainly change over time. At some point in my history, I've used all of the ones that I've listed by name, but at various points in time and on vastly different hardware, with Lightwave and Cinema4d being the ones that I learned to use first on hardware manufactured by companies who are no longer players in this market and containing CPUs few would now recognize. Also, I recognize that extrapolation from one instance can be faulty, but you have to have one before you get to two and so on. Unfortunately by the time you go completely through the list evaluating each one as carefully as possible, the one that you started with has been changed and the hardware has evolved. If this is your livelihood, you'll never be able to make a perfect decision because you'll never have perfect information. The reality is that you're always shooting with a different weapon at a moving and variable target. But that's not just the nature of this one endeavor, but it manifests itself now in almost everything that we do. But throw up my hands in defeat I am yet unwilling to do. So as I get information about how CPUs have been changed or how other hardware improves on what we have traditionally relied exclusively on the CPU to do, I feel compelled to share it even though it may benefit no one, because it might benefit someone.
Last edited:
I googled it too and got that very same feeling, but the one bright spot is that your computer and I have something in common. I also suggest that you send a private message to Rampage Dev (see post 515) who may be better able to help you out.
A few things in list format cuz I suck at tie-in sentences.
- Don't trust me I don't pretend to understand anything about modern Quadro cards. I thought I did once upon a time for the first few model releases but they have changed and revised the attributes which make a quadro a quadro so many times and I haven't really payed attention so.
- There's more to a card than the chip (number, model, and feature masking) tho. Just the layout, material, solder mask, number of layers, amount and speed or RAM, component quality, and so on can together (or alone) have a pretty huge impact on performance (user experience). I know every time I get the opportunity to use one it feels smoother than it's G counterpart. Maybe not always faster but smoother: as in the timing under my mouse so to speak, is more consistent. This all leads to lots of "WTF" moments for me anyway - I mean when comparing the two classes.
- You're probably right about Lux Render / Blender concern at NVidia but that's a dynamic and follows the potential customer profile (probably). If we listen to NVidia themselves and follow the bulk of their Quadro adverting, it seems the target base are primarily professional CAD/CAM users and secondarily the CG/3D crowd - including Blender operators BTW.
- Hehe.. Yeah... But also I think the pro user base (the targets of the Quadro class cards) are sold a more mission critical component array. So the Quadro isn't the renderer. It's a Quadro card + 2 to 4 Titans if GPU render assisted output is the goal. So just profiling a single Quadro card is probably only useful to a freelancer on a budget. And I guess that's not really what the Quadro is specifically about - although of course it certainly can be. In this case I think the Quadro cards are a waste of money and the freelancer in question would be better off configuring their workstation or desktop with G components - which I think is more their target area. Of course configuration options and on-line configurators from order taking web-sites don't really make this very clear and seem more tuned to make the seller a little extra profit rather than producing a properly tuned and balanced system for the intended purpose of the user. Salesmen... Heh!
Last edited:
Update for Sandy Bridge dual+ E5 self-builds
EVGA SR-2 motherboards for Nehalem and Westmere CPUs are still the best for dual CPU builds. EVGA SR-X was for Sandy Bridge E5 CPU builds, but the SR-X motherboard has been discontinued. This is apparently due to the lack of sales resulting from the inability to tweak the Sandy Bridge E5s to any significant degree no matter what the motherboard. Asus has Sandy Bridge E5 dual CPU challengers, but I do not recommend them - the originals and replacements I've had were all poorly manufactured and all were unstable. So the last credible suppliers of motherboards for Sandy Bridge E5 dual CPU builds appear to be Gigabyte (See post #1, above) and Supermicro. I'd suggest looking to Supermicro first if you're building dual CPU Sandy (and in the future Ivy) Bridge system(s). No Supermicro motherboard has ever failed to impress me as being 100% solid/well built, reliable and possessing the most important up-to-date features, despite their consistent lack of flash and pizazz.
BTW-WolfPackPrime0 and WolfPackPrime1 have as their foundation the Supermicro System-8047R-7RFT+ 4U LGA2011 E5-4600 PCIE DDR3 SAS Sat 1400W 80+ RTL [ http://www.provantage.com/supermicro-sys-8047r-7rft~7SUPB08K.htm ].
EVGA SR-2 motherboards for Nehalem and Westmere CPUs are still the best for dual CPU builds. EVGA SR-X was for Sandy Bridge E5 CPU builds, but the SR-X motherboard has been discontinued. This is apparently due to the lack of sales resulting from the inability to tweak the Sandy Bridge E5s to any significant degree no matter what the motherboard. Asus has Sandy Bridge E5 dual CPU challengers, but I do not recommend them - the originals and replacements I've had were all poorly manufactured and all were unstable. So the last credible suppliers of motherboards for Sandy Bridge E5 dual CPU builds appear to be Gigabyte (See post #1, above) and Supermicro. I'd suggest looking to Supermicro first if you're building dual CPU Sandy (and in the future Ivy) Bridge system(s). No Supermicro motherboard has ever failed to impress me as being 100% solid/well built, reliable and possessing the most important up-to-date features, despite their consistent lack of flash and pizazz.
BTW-WolfPackPrime0 and WolfPackPrime1 have as their foundation the Supermicro System-8047R-7RFT+ 4U LGA2011 E5-4600 PCIE DDR3 SAS Sat 1400W 80+ RTL [ http://www.provantage.com/supermicro-sys-8047r-7rft~7SUPB08K.htm ].
Last edited:
Need more GPU rendering power but lack open PC/Mac PCI-e slots and enuf power?
Consider studying your choices here: http://www.cubixgpu.com/Files/Products/4UupJSfqq.pdf and buying it here: http://www.bhphotovideo.com/c/buy/Shop-by-Brand-Cubix/ci/4/phd/4242329457/N/4294255798.
For a cost/benefit analysis, see also post #24 here: https://forums.macrumors.com/threads/1565650/ .
Consider studying your choices here: http://www.cubixgpu.com/Files/Products/4UupJSfqq.pdf and buying it here: http://www.bhphotovideo.com/c/buy/Shop-by-Brand-Cubix/ci/4/phd/4242329457/N/4294255798.
For a cost/benefit analysis, see also post #24 here: https://forums.macrumors.com/threads/1565650/ .
Last edited:
How can I power a Titan or what if I want to install more than one of them?
If you want information on how to power a Titan or if you want to install more than one of them, see posts beginning at no. 21 in this thread: [ https://forums.macrumors.com/threads/1565650/ ].
If you want information on how to power a Titan or if you want to install more than one of them, see posts beginning at no. 21 in this thread: [ https://forums.macrumors.com/threads/1565650/ ].
Last edited:
The FSP one looks good. 450W, 85%, 2x6pin and 2x6+2pin for $90.
Would you want to use something like that in a MP1,1 (w/6 HDDs and 2 X5355s) for running a GTX680?
Also do you think there would much of a performance cap introduced by limiting the 680GTX to the MP1,1 PCIe version 1.x speeds?
--
Sorry to be contributing to the OT discussion.
Purely based on my personal perception, there isn't any difference in running my Galaxy 680s 4Gb in my Mac Pros 2007 vs. my Mac Pro 2009 that I've upgrade to 2012 status. But my three Mac Pro 2007s are overclocked to run at between 3.4 to 3.6 GHz vs. 3.0 GHz factory (your mileage varies) , using this: [ http://www.zdnet.de/39192217/exclusive-zdnet-overclocking-tool-enhances-performance-of-mac-pro/ ] which does speed up the PCI-e bus and unfortunately the time clock.The FSP one looks good. 450W, 85%, 2x6pin and 2x6+2pin for $90.
Would you want to use something like that in a MP1,1 (w/6 HDDs and 2 X5355s) for running a GTX680?
Also do you think there would much of a performance cap introduced by limiting the 680GTX to the MP1,1 PCIe version 1.x speeds?
--
Sorry to be contributing to the OT discussion.
Yeah, that's kinda what I thought too. From the benchmarks I've created and looking over the GFX I/O stack it seems like a similar situation to the SATA II vs. SATA III debate. In a few rare cases the slower buss speeds caps performance but in over 90% of cases it doesn't.
How about on the power issue? What say you? Would you go for the additional PSU or just use native power?