I just don't buy that. Can't really say that until we see the benches on the upper models.
You don't buy what? That a spec bumped classic Mac Pro would have been a dual CPU, three slot tower with 8 RAM slots with the latest PCIe speeds and RAM speeds?
That's what a spec bump is. Keep the same configuration and just use the latest motherboard, chips, etc.
Which means one expansion card per port and only at PCIe2 x4 speeds.
The Mac Pro only had an x16 and two x4. This has six x4 and each port can have 6 devices. The main use given for the x16 was another GPU and this already has two.
I imagine the base spec will be (costs listed, not retail prices):
AMD is struggling financially though so I reckon Apple will have some pretty good deals on the GPUs and this isn't really a higher price than before because you'd be able to get the 12-core with the entry FirePros, which would come out around $5k. They'll probably just let you spec it out however you want on a single page.
If you can BTO nVidia Quadros that would be nice and solve some problems.
The main problem is that nobody should be using CUDA because it doesn't run on the CPU. OpenCL runs on NVidia, AMD and Intel. The problem there lies with the developers so they should be getting the complaints.
A spec bumped classic Mac Pro has up to double the CPU cores (24 vs 12), double the RAM slots (8 vs 4), higher GPU Compute capability (2 high end GPUs + Tesla/Xeon Phi/whatever), and more storage (even the ePCI SSD kind if you're willing to drop the number of TB2 ports to get some lanes back in addition to SATA/SAS storage).
It wouldn't have had 24-cores because Apple doesn't use the most expensive chips in the dual models. It would have topped out at 16 and the performance improvement vs the single 12-core wouldn't be nearly as much as the GPUs. When it comes to the GPUs, the FirePro has a 274W TDP so you'd get one in the old Mac Pro and that's it. Everything else would be in a Cubix box, which costs over $2k:
Well, this isn't even a complete sentence. I don't buy that your theory would be faster than what they've done.
Jesus. That's all you can come up with? Which part of "spec bump classic Mac Pro" do you not understand? A 2013 Mac Pro designed like the old one would have been dual CPU, have 8 RAM slots and more GPU and GP/GPU compute options. How is double the CPU and 50% more GP/GPU capability not faster?
The Mac Pro only had an x16 and two x4. This has six x4 and each port can have 6 devices. The main use given for the x16 was another GPU and this already has two.
The old Mac Pro was limited by the chipset. A spec bumped 2013 Mac Pro has the same number of PCIe lanes. Lets count in the new Mac Pro: Two x16 lanes + 6 x4 lanes.
Split that up into the spec bump Mac Pro and you can have thee x16 slots with 2 TB2 ports left over. Or 2 x16 and one x8 and 4 TB2 ports.
Quote:
I think a single SSD would be ok internally, they'll at least allow up to 768GB. I think they'll have up to 1.5TB for $1500 retail.
Given that a 1TB SSD is only $600 retail that's not so great.
The issue however is if you only have a 256GB in the base model then upgrading the 128GB RAM that 10.9 can handle means you lost half the available disk space to the sleep image. 128GB remaining is kinda tight with the size of apps.
512GB really should be the minimum so you're not screwed because you added RAM later.
Well, it was my original point so you should have been "outraged" at it back then.
Which part of "spec bump classic Mac Pro" do you not understand?
The part where it wouldn't be faster than the one they did release.
A 2013 Mac Pro designed like the old one would have been dual CPU, have 8 RAM slots and more GPU and GP/GPU compute options. How is double the CPU and 50% more GP/GPU capability not faster?
We'll have to see when the benchmarks hit, huh. :no:
The old Mac Pro was limited by the chipset. A spec bumped 2013 Mac Pro has the same number of PCIe lanes. Lets count in the new Mac Pro: Two x16 lanes + 6 x4 lanes.
Split that up into the spec bump Mac Pro and you can have thee x16 slots with 2 TB2 ports left over. Or 2 x16 and one x8 and 4 TB2 ports.
That seems to be mixing PCIe 2 and 3. Thunderbolt ports are PCIe 2 x4 equivalent but are driven by PCIe 3 lanes in the new MP. The GPUs may be on PCIe 3 x8 each and have an x4 per TB controller leaving 12 lanes free, some for PCIe storage. I don't really see the benefit of reducing the number of expansion ports in order to increase their speeds. I still haven't seen a single example of a PCI card that saturates an x4 slot so your setup is like having 3 slots free rather than 36. You'd never be in a situation where every one is maxing out in the same direction and worst case you get some slowdown in a transfer somewhere but it's hardly a problem being stuck at just 2.5GB/s.
The issue however is if you only have a 256GB in the base model then upgrading the 128GB RAM that 10.9 can handle means you lost half the available disk space to the sleep image.
What sleep image? This isn't a laptop. Hibernation modes are disabled on the desktops. If for some reason they are enabled, you can disable them manually. Also, this will only go to 64GB RAM as it has 4 RAM slots. When DDR4 comes along next year, the 2014 Haswell model will probably manage 128GB.
512GB really should be the minimum so you're not screwed because you added RAM later.
I agree that 512GB is a good working size to go with for this type of machine but if someone is buying the entry model, they might be using it for photography so massive amounts of storage might not be needed. 128GB would let you store over 42,000 3MB images. They could take a $200 hit on the margins to get the 512GB in the $2499 model. This would make their gross margins about 33% instead of 40%. If they wanted to do that, it would be great but then some might wonder why not have an even cheaper entry point with 256GB. 256GB is the entry on the MBP so I reckon it's ok for that to be the starting point.
*Thunderbolt is great for expansion with the accessory ecosystem i.e. PCI expansion, Hard Drive arrays, networking, XGrid and Displays with thunderbolt will be Awesome! "Apple Thunderbolt to Optical Audio dingy? "
*U.S. assembly plant finally.
*I also can't wait to see the different holders and mounting brackets for it. There is going to be a Wine Rack for sure.
*Virginia Techs super computer would look really neat with these Mac Pro units.
"Parallel computing question" Could someone have two of these MPs and connect all of the thunderbolt together and make them run as one computer? 24cores 4 graphic cards.
I posted this in reply to you before, in the 'throws out the rulebook' thread…
I must have missed that. If that is in fact a supportable place for an additional SSD card then Apple needs to ship the machine with a socket there. This would immediately make for a far more versatile machine.
<span style="color:rgb(24,24,24);font-family:arial, helvetica, sans-serif;line-height:18px;">This image shows the two GPU boards; on the right hand one we see the SDD blade; on the left hand one we can see the pads where a second SSD blade could be added in…</span>
Excellent close up.
<span style="color:rgb(24,24,24);font-family:arial, helvetica, sans-serif;line-height:18px;">And again from the same thread, but only to get your thoughts & also to rile up Tallest Skil…</span>
<span style="color:rgb(24,24,24);font-family:arial, helvetica, sans-serif;line-height:18px;">Over on CGTalk, someone mentioned that the CPU & GPU cards seemed to be just that, cards… As in, connected to a backplane and from there the main logic board. Wondering about BTO configurations if this is true…</span>
Of course they are cards, you need some way to assemble the machine.
<p id="user_yui_3_10_0_1_1370982093914_1016" style="color:rgb(24,24,24);font-family:arial, helvetica, sans-serif;line-height:18px;">No matter the configuration, the boot flash is still in its socket on the GPU card, and if you look at some of the pics out there, you can see the pads where a second SSD Flash could be installed on the second GPU card…</p>
<p id="user_yui_3_10_0_1_1370982093914_1045" style="color:rgb(24,24,24);font-family:arial, helvetica, sans-serif;line-height:18px;">And if Apple will do such a BTO, then they could also have consumer grade CPU cards, and consumer grade GPU cards…</p>
This would be very nice indeed and would be close to the capabilities I want to see in an XMac.
<p id="user_yui_3_10_0_1_1370982093914_1049" style="color:rgb(24,24,24);font-family:arial, helvetica, sans-serif;line-height:18px;">Maybe a quad-core Haswell i7 CPU card (with 16GB RAM), nVidia Geforce GFX 700 series GPU card & dual SSD card with only a single SSD populated (and NO SSD Flash boot drive on the GPU card); giving us the mythical xMac…!!! Shipping by Christmas time, US$1,500.00…!!!</p>
That would be golden.
<p style="color:rgb(24,24,24);font-family:arial, helvetica, sans-serif;line-height:18px;">I cannot wait for these bad boys, I can see one of these, all pimped out with max CPU, max RAM, max SSD, max GPUs; hooked to two 31" 4K ThunderBolt 2 Cinema Displays & a Cintiq 24HD touch, with a huge honking HDTV off of the HDMI, a TB2 A/V I/O box & a TB2 RAID; the ultimate DCC workstation…!</p>
I doubt that I will be able to afford the dual work station GPU models unless Apple gets a very good deal from AMD. So a trimmed model would be very nice indeed. I really want a quad core machine for my next desktop.
I had a dream, where Apple would manufacture an extension chassis for the new MacPro, resembling the base of the Cray one. The cylindrical main unit would just be centered on top of this base and provide for the cooling. Room enough for loads of hard drives, PCI cards, etc.
I was literally thinking the same thing. The base mainly being a disk array and maybe that damn optical everybody whines about.
That is probably just my wild imagination but I do expect Apple to offer a complete solution when these ship. That includes a disk array of some sort and new high res displays. I just don't see them leaving launch success to third party suppliers.
I was literally thinking the same thing. The base mainly being a disk array and maybe that damn optical everybody whines about.
That is probably just my wild imagination but I do expect Apple to offer a complete solution when these ship. That includes a disk array of some sort and new high res displays. I just don't see them leaving launch success to third party suppliers.
Good point about the RAID array, although, this would probably be "side" offered by Promise.
That seems to be mixing PCIe 2 and 3. Thunderbolt ports are PCIe 2 x4 equivalent but are driven by PCIe 3 lanes in the new MP. The GPUs may be on PCIe 3 x8 each and have an x4 per TB controller leaving 12 lanes free, some for PCIe storage. I don't really see the benefit of reducing the number of expansion ports in order to increase their speeds. I still haven't seen a single example of a PCI card that saturates an x4 slot so your setup is like having 3 slots free rather than 36. You'd never be in a situation where every one is maxing out in the same direction and worst case you get some slowdown in a transfer somewhere but it's hardly a problem being stuck at just 2.5GB/s.
TB2 ports are about 2.5 PCIe 3.0 lanes at 20 Gb/s. Each 3.0 lane is 1GB/s. Thinking about it the 6 TB ports would be about 15 lanes worth leaving only 25 out of 40. That doesn't seem quite right unless they don't expect all the TB2 ports to run full speed.
For GAMING performance PCIe x8 is fine today but not OpenCL performance. A Radeon HD 7970 a year ago showed an 17.94% difference in OpenCL benchmark score between a x16 and x8 slot.
That's only one benchmark but it depends on your actual GP/GPU workload behaviors. If you're computationally heavy with little data transfer then yes, you can run in a x4 slot and not notice any difference.
Here's Anand:
"For GPU compute. Improving bandwidth and latency between the CPU and the GPU are both key to building a high performance heterogenous computing solution. While good GPU compute benchmarks on the desktop are still hard to come by, we did find one that showed a real improvement from PCIe 3.0 support on the 7970: AMD's AES Encrypt/Decrypt sample application.
Simply enabling PCIe 3.0 on our EVGA X79 SLI motherboard (EVGA provided us with a BIOS that allowed us to toggle PCIe 3.0 mode on/off) resulted in a 9% increase in performance on the Radeon HD 7970. This tells us two things: 1) You can indeed get PCIe 3.0 working on SNB-E/X79, at least with a Radeon HD 7970, and 2) PCIe 3.0 will likely be useful for GPU compute applications, although not so much for gaming anytime soon."
I'd be surprised (and disappointed) that the new Mac Pro was running those FirePro cards in x8 mode. But assume they are then 16 lanes for the two GPUs, 15 lanes for the TB ports and you have 9 left. Given the performance of the SSD that's likely a 2 lane device. Viable but I STILL don't believe that they are running both those GPU at x8. That's just shortsighted for a dual GP/GPU platform.
Hmmm...12x? That's in the data spec and they control both GPU and MB completely.
Quote:
Also, this will only go to 64GB RAM as it has 4 RAM slots. When DDR4 comes along next year, the 2014 Haswell model will probably manage 128GB.
64GB max would be very small for a workstation class machine. Guidance has been to have three times the system RAM as the GP/GPU boards. Meaning that for two 6GB boards the recommended RAM was 36GB for the Quadro 6000 + Tesla C2075.
128GB is my expected max since there are 32GB ECC DIMMS.
64GB max would be very small for a workstation class machine. Guidance has been to have three times the system RAM as the GP/GPU boards. Meaning that for two 6GB boards the recommended RAM was 36GB for the Quadro 6000 + Tesla C2075.
128GB is my expected max since there are 32GB ECC DIMMS.
I always went with 4GB per core on desktop, or 2GB per core on a laptop. Where did you get this guidance from, just interesting, not doubting or anything.
I always went with 4GB per core on desktop, or 2GB per core on a laptop. Where did you get this guidance from, just interesting, not doubting or anything.
Huh? I put my MP5.1 with 10.8 to sleep instead of shutting down. That's hibernation, right? I'm missing something here...
Indeed, you are missing something. In laptops, when the system boots for the first time, it will write a file on the hard drive having the size of the RAM. So, if you have 4 GB RAM in your system, a file of 4 GB will be written (like /var/vm/sleepimage, if I remember well). In this file the operating system stores its state so that it can be quickly recovered and restored after hibernation.
During normal sleep mode, available for both laptops and desktops, the operating system stores its state into the RAM. If the battery is drained while the machine is still in sleep mode and cannot keep the RAM in standby anymore, then the system will update the contents of the sleepimage with the current state and will shutdown the computer. In the next reboot it will read the content of sleepimage to restore the previous system state. As far as I know this is available only to laptops (which have a battery to begin with).
This means that in systems with much memory it may be beneficial to disable this feature in order to save storage room.
Huh? I put my MP5.1 with 10.8 to sleep instead of shutting down. That's hibernation, right? I'm missing something here...
Indeed, you are missing something. In laptops, when the system boots for the first time, it will write a file on the hard drive having the size of the RAM. So, if you have 4 GB RAM in your system, a file of 4 GB will be written (like /var/vm/sleepimage, if I remember well). In this file the operating system stores its state so that it can be quickly recovered and restored after hibernation.
During normal sleep mode, available for both laptops and desktops, the operating system stores its state into the RAM. If the battery is drained while the machine is still in sleep mode and cannot keep the RAM in standby anymore, then the system will update the contents of the sleepimage with the current state and will shutdown the computer. In the next reboot it will read the content of sleepimage to restore the previous system state. As far as I know this is available only to laptops.
This means that in systems with much memory it may be beneficial to disable this feature in order to save storage room.
Ah, ok, thanks. I indeed don't have 'sleepimage' only 2 swapfiles (67MB each). And I presume since my MP is always powered, it doesn't need this image. Thanks.
Comments
Quote:
Originally Posted by Tallest Skil
I just don't buy that. Can't really say that until we see the benches on the upper models.
You don't buy what? That a spec bumped classic Mac Pro would have been a dual CPU, three slot tower with 8 RAM slots with the latest PCIe speeds and RAM speeds?
That's what a spec bump is. Keep the same configuration and just use the latest motherboard, chips, etc.
Well, this isn't even a complete sentence. I don't buy that your theory would be faster than what they've done.
The Mac Pro only had an x16 and two x4. This has six x4 and each port can have 6 devices. The main use given for the x16 was another GPU and this already has two.
I imagine the base spec will be (costs listed, not retail prices):
Quad Xeon $294
8GB RAM $100
dual FirePro W5000 $600 ( http://www.newegg.com/Product/Product.aspx?Item=N82E16814195119 )
256GB SSD $200
Motherboard, PSU, peripherals = $300
Total = $1494 + 40% margin = $2490 so same starting price.
Top model:
12-core Xeon $2000
64GB RAM $500
dual FirePro W9000 $5000 ( http://www.newegg.com/Product/Product.aspx?Item=N82E16814195116&Tpk=firepro w9000&IsVirtualParent=1 )
1.5TB SSD $1000
same extras = $300
Total = $8,800 + 40% margin = $14,667
AMD is struggling financially though so I reckon Apple will have some pretty good deals on the GPUs and this isn't really a higher price than before because you'd be able to get the 12-core with the entry FirePros, which would come out around $5k. They'll probably just let you spec it out however you want on a single page.
The main problem is that nobody should be using CUDA because it doesn't run on the CPU. OpenCL runs on NVidia, AMD and Intel. The problem there lies with the developers so they should be getting the complaints.
I think a single SSD would be ok internally, they'll at least allow up to 768GB. I think they'll have up to 1.5TB for $1500 retail.
It wouldn't have had 24-cores because Apple doesn't use the most expensive chips in the dual models. It would have topped out at 16 and the performance improvement vs the single 12-core wouldn't be nearly as much as the GPUs. When it comes to the GPUs, the FirePro has a 274W TDP so you'd get one in the old Mac Pro and that's it. Everything else would be in a Cubix box, which costs over $2k:
http://www.sabrepc.com/p-1093-cubix-gpu-xpander-desktop-4-with-16-pcie-channels-and-4-double-wide-slots-retail.aspx
http://www.amazon.com/Cubix-GPU-Xpander-Desktop-Channels-Double-wide/dp/B004W50R70
You get a choice of GPUs but not two high-end ones, certainly not plus a Tesla or Xeon Phi card.
Quote:
Originally Posted by Tallest Skil
Well, this isn't even a complete sentence. I don't buy that your theory would be faster than what they've done.
Jesus. That's all you can come up with? Which part of "spec bump classic Mac Pro" do you not understand? A 2013 Mac Pro designed like the old one would have been dual CPU, have 8 RAM slots and more GPU and GP/GPU compute options. How is double the CPU and 50% more GP/GPU capability not faster?
Quote:
Originally Posted by Marvin
The Mac Pro only had an x16 and two x4. This has six x4 and each port can have 6 devices. The main use given for the x16 was another GPU and this already has two.
The old Mac Pro was limited by the chipset. A spec bumped 2013 Mac Pro has the same number of PCIe lanes. Lets count in the new Mac Pro: Two x16 lanes + 6 x4 lanes.
Split that up into the spec bump Mac Pro and you can have thee x16 slots with 2 TB2 ports left over. Or 2 x16 and one x8 and 4 TB2 ports.
Quote:
I think a single SSD would be ok internally, they'll at least allow up to 768GB. I think they'll have up to 1.5TB for $1500 retail.
Given that a 1TB SSD is only $600 retail that's not so great.
http://www.newegg.com/Product/Product.aspx?Item=N82E16820148696
The issue however is if you only have a 256GB in the base model then upgrading the 128GB RAM that 10.9 can handle means you lost half the available disk space to the sleep image. 128GB remaining is kinda tight with the size of apps.
512GB really should be the minimum so you're not screwed because you added RAM later.
Well, it was my original point so you should have been "outraged" at it back then.
The part where it wouldn't be faster than the one they did release.
We'll have to see when the benchmarks hit, huh. :no:
That seems to be mixing PCIe 2 and 3. Thunderbolt ports are PCIe 2 x4 equivalent but are driven by PCIe 3 lanes in the new MP. The GPUs may be on PCIe 3 x8 each and have an x4 per TB controller leaving 12 lanes free, some for PCIe storage. I don't really see the benefit of reducing the number of expansion ports in order to increase their speeds. I still haven't seen a single example of a PCI card that saturates an x4 slot so your setup is like having 3 slots free rather than 36. You'd never be in a situation where every one is maxing out in the same direction and worst case you get some slowdown in a transfer somewhere but it's hardly a problem being stuck at just 2.5GB/s.
This is PCIe SSD remember:
http://www.amazon.com/OCZ-Technology-RevoDrive-PCI-Express--RVD3X2-FHPX4-960G/dp/B0058RECRM
What sleep image? This isn't a laptop. Hibernation modes are disabled on the desktops. If for some reason they are enabled, you can disable them manually. Also, this will only go to 64GB RAM as it has 4 RAM slots. When DDR4 comes along next year, the 2014 Haswell model will probably manage 128GB.
I agree that 512GB is a good working size to go with for this type of machine but if someone is buying the entry model, they might be using it for photography so massive amounts of storage might not be needed. 128GB would let you store over 42,000 3MB images. They could take a $200 hit on the margins to get the 512GB in the $2499 model. This would make their gross margins about 33% instead of 40%. If they wanted to do that, it would be great but then some might wonder why not have an even cheaper entry point with 256GB. 256GB is the entry on the MBP so I reckon it's ok for that to be the starting point.
*U.S. assembly plant finally.
*I also can't wait to see the different holders and mounting brackets for it. There is going to be a Wine Rack for sure.
*Virginia Techs super computer would look really neat with these Mac Pro units.
"Parallel computing question" Could someone have two of these MPs and connect all of the thunderbolt together and make them run as one computer? 24cores 4 graphic cards.
I was literally thinking the same thing. The base mainly being a disk array and maybe that damn optical everybody whines about.
That is probably just my wild imagination but I do expect Apple to offer a complete solution when these ship. That includes a disk array of some sort and new high res displays. I just don't see them leaving launch success to third party suppliers.
Quote:
Originally Posted by wizard69
I was literally thinking the same thing. The base mainly being a disk array and maybe that damn optical everybody whines about.
That is probably just my wild imagination but I do expect Apple to offer a complete solution when these ship. That includes a disk array of some sort and new high res displays. I just don't see them leaving launch success to third party suppliers.
Good point about the RAID array, although, this would probably be "side" offered by Promise.
Quote:
Originally Posted by Marvin
That seems to be mixing PCIe 2 and 3. Thunderbolt ports are PCIe 2 x4 equivalent but are driven by PCIe 3 lanes in the new MP. The GPUs may be on PCIe 3 x8 each and have an x4 per TB controller leaving 12 lanes free, some for PCIe storage. I don't really see the benefit of reducing the number of expansion ports in order to increase their speeds. I still haven't seen a single example of a PCI card that saturates an x4 slot so your setup is like having 3 slots free rather than 36. You'd never be in a situation where every one is maxing out in the same direction and worst case you get some slowdown in a transfer somewhere but it's hardly a problem being stuck at just 2.5GB/s.
TB2 ports are about 2.5 PCIe 3.0 lanes at 20 Gb/s. Each 3.0 lane is 1GB/s. Thinking about it the 6 TB ports would be about 15 lanes worth leaving only 25 out of 40. That doesn't seem quite right unless they don't expect all the TB2 ports to run full speed.
For GAMING performance PCIe x8 is fine today but not OpenCL performance. A Radeon HD 7970 a year ago showed an 17.94% difference in OpenCL benchmark score between a x16 and x8 slot.
http://www.expreview.com/img/review/HD7970_PCIE3/p3_16vs8.png
That's only one benchmark but it depends on your actual GP/GPU workload behaviors. If you're computationally heavy with little data transfer then yes, you can run in a x4 slot and not notice any difference.
Here's Anand:
"For GPU compute. Improving bandwidth and latency between the CPU and the GPU are both key to building a high performance heterogenous computing solution. While good GPU compute benchmarks on the desktop are still hard to come by, we did find one that showed a real improvement from PCIe 3.0 support on the 7970: AMD's AES Encrypt/Decrypt sample application.
Simply enabling PCIe 3.0 on our EVGA X79 SLI motherboard (EVGA provided us with a BIOS that allowed us to toggle PCIe 3.0 mode on/off) resulted in a 9% increase in performance on the Radeon HD 7970. This tells us two things: 1) You can indeed get PCIe 3.0 working on SNB-E/X79, at least with a Radeon HD 7970, and 2) PCIe 3.0 will likely be useful for GPU compute applications, although not so much for gaming anytime soon."
http://www.anandtech.com/show/5264/sandy-bridge-e-x79-pcie-30-it-works
I'd be surprised (and disappointed) that the new Mac Pro was running those FirePro cards in x8 mode. But assume they are then 16 lanes for the two GPUs, 15 lanes for the TB ports and you have 9 left. Given the performance of the SSD that's likely a 2 lane device. Viable but I STILL don't believe that they are running both those GPU at x8. That's just shortsighted for a dual GP/GPU platform.
Hmmm...12x? That's in the data spec and they control both GPU and MB completely.
Quote:
Also, this will only go to 64GB RAM as it has 4 RAM slots. When DDR4 comes along next year, the 2014 Haswell model will probably manage 128GB.
64GB max would be very small for a workstation class machine. Guidance has been to have three times the system RAM as the GP/GPU boards. Meaning that for two 6GB boards the recommended RAM was 36GB for the Quadro 6000 + Tesla C2075.
128GB is my expected max since there are 32GB ECC DIMMS.
Quote:
Originally Posted by nht
64GB max would be very small for a workstation class machine. Guidance has been to have three times the system RAM as the GP/GPU boards. Meaning that for two 6GB boards the recommended RAM was 36GB for the Quadro 6000 + Tesla C2075.
128GB is my expected max since there are 32GB ECC DIMMS.
I always went with 4GB per core on desktop, or 2GB per core on a laptop. Where did you get this guidance from, just interesting, not doubting or anything.
Huh? I put my MP5.1 with 10.8 to sleep instead of shutting down. That's hibernation, right? I'm missing something here...
You are missing what to do correctly.
The Maximus install guide
Unfortunately, this is for USB 3.0, but I expect some mfr. will make one for Thunderbolt.
Newegg.com - IVIEW Iview HF2-SU3 4 3.5" Drive Bays USB 3.0 / eSATA 4-Bay Storage Box
IVIEW Iview HF2-SU3 4 3.5" Drive Bays USB 3.0 / eSATA 4-Bay Storage Box
D-Link Cloud Router 2000 (DIR-826L), Wireless N600, Dual-Band, Gigabit Ports, USB SharePort, mydlink enabled
$67.99
Quote:
Originally Posted by PhilBoogie
Huh? I put my MP5.1 with 10.8 to sleep instead of shutting down. That's hibernation, right? I'm missing something here...
Indeed, you are missing something. In laptops, when the system boots for the first time, it will write a file on the hard drive having the size of the RAM. So, if you have 4 GB RAM in your system, a file of 4 GB will be written (like /var/vm/sleepimage, if I remember well). In this file the operating system stores its state so that it can be quickly recovered and restored after hibernation.
During normal sleep mode, available for both laptops and desktops, the operating system stores its state into the RAM. If the battery is drained while the machine is still in sleep mode and cannot keep the RAM in standby anymore, then the system will update the contents of the sleepimage with the current state and will shutdown the computer. In the next reboot it will read the content of sleepimage to restore the previous system state. As far as I know this is available only to laptops (which have a battery to begin with).
This means that in systems with much memory it may be beneficial to disable this feature in order to save storage room.
Ah, ok, thanks. I indeed don't have 'sleepimage' only 2 swapfiles (67MB each). And I presume since my MP is always powered, it doesn't need this image. Thanks.