cMP 5,1 Budget Upgrade Plan

[usna92]

My office has (finally) started to refresh some older Mac Pro equipment. I managed to snag a 5,1 cMP with 12 Core with running at 2.66 GHz (should be x5650's). It currently has 26 Gig of Ram 6x4GB Sticks and two 1GB Sticks. The stock graphics card is a ATI Radeon HD 5770. Right now, because I had them on the shelf I have two WD Green drives (1/2 TBs) one WD Red (3TB). (Machine had no drives because they purged it when they turned over the equipment). Also has the wireless upgrade from Apple.

Currently my upgrade MacBook Retina is running as the home iTunes server, the DVD ripper, the goto Mac on my desk, my network manager, my Xcode builder, etc. I serve it using a small thunderbolt raid, which backed up to a USB 3.0 Drobo raid weekly. I would like to switch all of those duties to the Mac Pro to regain the portability of my retina when I need power on the go.

Here is my plan (based on what I had on the shelf, what was cheap given my budget at hand).

I have an Apple 256GB SSD from my MacBook Retina (2012), a Seagate Momentus 750 GB Drive, a Diamond Sapphire R9 280X DDR5 3GB (PC Card).

Here's is what I have ordered so far, my plans for them, and an open to suggestion to modify the plan, change the design, or ways to optimize what I am doing. I will also outlay the things I eventually want to do as well, money, budget, time permitting. Have combed though a lot of the more recent threads in the forum, but if you have forum posts that you suggests I read, I will do that to.

Initial Plans
SSD and Momentus Fusion Drive.
OWC is running their garage sale right now. I picked up an Accelsior S PCI-e card and a converter for the MacBook Pro Retina SSD drive. My intent is to populate that Accelsior card, with the SSD and then create a fusion drive with the Seagate Momentus Drive. I haven't seen a lot of posts about fusion drive setups. Should I just skip it, run the drives separately? Not wedded to this idea either way, more of a science experiment.

Diamond Sapphire Vapor X
I know there are other cards out there, but this is what I had on my shelf. In reading the forums, it seems like this chipset is fairly well supported (7950?) though only a few forum posts about this particular card. I have ordered the extra 6 pin cable and the two 6 to 8 pin connectors necessary to power it. (Apparently this card pin senses to check all 8 pins are connected). Any known worries about this card? Support in Sierra?

Future Plans
I would like to do a CPU upgrade. I have done a review of the CPU thread, but I didn't see anyone doing benchmarks between the various levels, which makes price / performance comparison difficult. I was looking at the X5675 since they seem to run near $100 a pair rather than the X5690s at $500 or so a pair, or am a missing a source of cheap processors.

I would also eventually like the OSXWifi upgrade to get continuity, handoff, etc. but that is a lower priority than performance.

The last upgrade I was thinking would be USB 3.0 / 3.1 card. I have looked at the caldigit card, but at $169 it is steep. I have seen some other usb 3.0/3.1 cards without the esata, but haven't seen if anyone has tried them yet. I would like a USB-C connector for future proofing, but understand that the performance looks a little dicey on those cards right now, so was also looking for recommendations on the best price/performance USB 3.0 card.

These are plans, know this is a long thread, looking to learn as well as contribute back after I try these things.

 

[usna92]

First upgrade is complete, though I am beginning to wonder about the effectiveness, or that I need a better way to measure it.

I installed the Sapphire Vapor X R9 280X. As expected the Mac did not have boot screen, etc., but did come up for log in. It seems to operate pretty stable and the desktop and animations seem snappier, though that might be anecdotal. I re-ran all the benchmarks I ran before I started screwing with anything (black magic, cinebench, and GeekBench 4). All of the benchmarks were slightly down (within 2% so might be some process in the background) with the exception of GeekBench 4 OpenCL test which showed a 10x improvement in capability. Should I be seeing improvement anywhere else other than that benchmark. Is there a better way I should be measuring all this stuff?

 

[h9826790]

First upgrade is complete, though I am beginning to wonder about the effectiveness, or that I need a better way to measure it.

I installed the Sapphire Vapor X R9 280X. As expected the Mac did not have boot screen, etc., but did come up for log in. It seems to operate pretty stable and the desktop and animations seem snappier, though that might be anecdotal. I re-ran all the benchmarks I ran before I started screwing with anything (black magic, cinebench, and GeekBench 4). All of the benchmarks were slightly down (within 2% so might be some process in the background) with the exception of GeekBench 4 OpenCL test which showed a 10x improvement in capability. Should I be seeing improvement anywhere else other than that benchmark. Is there a better way I should be measuring all this stuff?

It really depends on your workflow / usage. If your workflow is not GPU demanding, then it's hard to see the real world difference other than benchmark.

Anyway, Cinebench GPU test is a well known CPU limiting benchmark, it's basically measuring your CPU single thread performance. Since you haven't upgrade your CPU yet, that's why you didn't see any difference.

Blackmagic sure has nothing to do with GPU.

If you want more benchmark, or stress testing the system (mainly the GPU at this moment). Furmark is a good choice.

 

[usna92]

It really depends on your workflow / usage. If your workflow is not GPU demanding, then it's hard to see the real world difference other than benchmark.

Anyway, Cinebench GPU test is a well known CPU limiting benchmark, it's basically measuring your CPU single thread performance. Since you haven't upgrade your CPU yet, that's why you didn't see any difference.

Blackmagic sure has nothing to do with GPU.

If you want more benchmark, or stress testing the system (mainly the GPU at this moment). Furmark is a good choice.

Thanks for the feedback. I will look at Furmark and see where it pegs the improvements. Hopefully, my plan for the SSD can be implemented next week when more parts arrive. I am thinking I might scrap the Fusion drive idea, but I think I am going to try it, benchmark it, and then make a decision.

 

[orph]

why fusion drive?
if the HD fails you will lose all your data, seems simpler to have a nice boot drive (SSD) and a storage drive with permissions disabled so it's easier to get your data.
that might also make it easier to backup your data depending on workflow.

 

[usna92]

I am not wedded to the idea to but the SSD is smallish (256GB) since it is a re-use drive from my laptop and with a FD I can span that with a relatively speedy (for an HD) drive that will allow for some seamless room to grow until I figure out what this machine's final purpose will be. Right now its main purpose is to replace my iTunes server / handbrake machine. The boot drive's data is not critical and should one of the drives in the configuration fail, I will just restore it from a CCC backup that I plan on keeping of the drive. It does seem like FD options have fallen out of vogue because of decreasing SD costs and increased overhead in setting one up. I am also interested in seeing how benchmarks treat it over time, because it has seemed to me based on reading when FD's were first implemented that no one has figured out a great way to measure them. This might be a good way to give back a little over time if I can. If it screws up I wipe it and go with a two solution.

 

[orph]

i gess, the drive will only be as big as the HD ie 750GB the 256GB of SSD will just be a catch of files on the HD. So less room to grow in to from my point of view.

maybe it's different user needs, workflow. Im always happy with my 250GB SSD with apps and some user data and HD with files on, apps ect always at the speed of the SSD and less I/O as each drive is focused at a job.

 

[usna92]

If it turns out I hate it, then I haven't really lost anything and I'll go from there. I would like to benchmark it a bit and see what is does though. Curiosity is a terrible thing sometimes.

 

[kschendel]

I don't have numbers written down, but my recollection is that a CPU upgrade from stock 2.66 W3520 to a 3.33 W3680 got me about a 1/4 to 1/3 improvement in run times for the sort of work I do (compiling and DBMS loads). Replacing the original drives with a drop in SATA SSD (I used a Mushkin Reactor at 1 Tb) made a similar improvement, plus made some disk-access things like booting up a LOT snappier. I also installed a couple WD Blacks which are a marginal improvement over the original stock drives. Overall, the machine is not quite twice as fast as what I started with. I don't do any heavy graphics so the original GT120 is still in there.

Your notion of making a Fusion drive out of the 256 SSD and the Seagate sounds reasonable, and I think you're going at it the right way.

 

[Slash-2CPU]

Removing those 1GB RAM sticks and running 3x 4GB on each CPU in proper triple-channel mode will give a performance boost.

Those CPU's have 3 memory channels each. When you're not running 3 matched sticks per CPU, it's less efficient. Zero-cost upgrade.

Look into X5675(3.06), X5679(3.20), and X5680(3.33). X5679's are sometimes cheaper than X5675's.

If your work is lightly threaded, quad-core CPU's with higher clocks will feel as fast as 6-cores. X5677(3.40) and X5687(3.60).

USB 3.0 card: KT4004 (330MB/s max) (requires 10.6.3 or later.)
USB 3.1 card: Ableconn PU31-1A1C (700MB/s max) (requires 10.12 or later)

Neither one requires extra power connectors.

 

[usna92]

Those are all great suggestions. The fusion drive stuff comes tonight so I will be benchmarking for performance improvements after that tonight. I will be interested to see how that varies over time as I use it more. There is a lot of discussion about whether a FD actually migrates files back to the SSD based on use, so I will be interested to see if it actually does or not. I will probably be picking up the USB card next. I only actually need the 3.0, but it would be nice to future proof a little with the 3.1 card, and the prices are about the same. The CalDigit card really looks interesting, but I'd end up spending more on that then just about anything I have put into this computer to date. Price balance just not there yet. Maybe soon.
[doublepost=1485798396][/doublepost]

Removing those 1GB RAM sticks and running 3x 4GB on each CPU in proper triple-channel mode will give a performance boost.

Those CPU's have 3 memory channels each. When you're not running 3 matched sticks per CPU, it's less efficient. Zero-cost upgrade.

Look into X5675(3.06), X5679(3.20), and X5680(3.33). X5679's are sometimes cheaper than X5675's.

If your work is lightly threaded, quad-core CPU's with higher clocks will feel as fast as 6-cores. X5677(3.40) and X5687(3.60).

USB 3.0 card: KT4004 (330MB/s max) (requires 10.6.3 or later.)
USB 3.1 card: Ableconn PU31-1A1C (700MB/s max) (requires 10.12 or later)

Neither one requires extra power connectors.

This will become my primary handbrake machine, which means I should be able to get some use out of the multiple cores, but I have looked at the quad cores as well. It will depend on what is cheapest when the funds to do arrive. I ordered a t-handle and some thermal paste, so I would have all the tools ready when it is time. It will just have to wait until my next budget authority comes in.

 

[Slash-2CPU]

Caldigit only makes sense if you need to squeeze USB3.1 and eSATA into one slot.

Fusion, due to caching not usually immediately benefiting in-flight data like a benchmark, will probably show no difference.

For the triple-channel upgrade, you want to use the three slots each farthest from the CPU heatsinks; the ones closest to the outside of the tray. Compared to asymmetric dual-channel that you're currently running, it improves memory bandwidth by 15-25%.

 

[usna92]

I have the extra RAM coming as well to take the machine to 32 GB. Is the performance 3x channel better than the RAM space overhead?

 

[Slash-2CPU]

Yes, with the exception being if you are using apps that need more than 24GB.

I'm running 3x8GB with a single CPU and have yet to come close to maxing out, except when doing dumb things with torrent caching.

 

[usna92]

Fair enough. I will try it both ways tonight to see how the numbers play out and if it ends up being faster, I'll try it until I see some RAM pressure. I might end up being to be able to get another one of these machines, so if I will have a test mule to try all this stuff out on....

 

[h9826790]

I have the extra RAM coming as well to take the machine to 32 GB. Is the performance 3x channel better than the RAM space overhead?

If your main use is handbrake, and you are encoding some large videos. RAM size should gives you more advantage than optimising the RAM speed. I heavily use Handbrake as well. Most of the videos are within 10GB in size. Even my 48GB RAM is more than enough for encoding (memory pressure always green), but I can clearly see that my Mac can fully utilise all the RAM. When running Handbrake, all the remaining RAM can be used as cache to speed up the process. (Most of the time, ~24GB real usage and ~24GB work as cache, plus <2GB memory compression and <2GB swap)

In real world, optimising tripple channel architecture can hardly gives you any benefit, especially the process that are not memory bandwidth limiting. But more cache are always good. Unless you only use Handbrake to encode some small videos, and the overall memory usage (including cache) can never go above 24GB. Then install only 3 sticks may be actually the better option for you .

 

[usna92]

If your main use is handbrake, and you are encoding some large videos. RAM size should gives you more advantage than optimising the RAM speed. I heavily use Handbrake as well. Most of the videos are within 10GB in size. Even my 48GB RAM is more than enough for encoding (memory pressure always green), but I can clearly see that my Mac can fully utilise all the RAM. When running Handbrake, all the remaining RAM can be used as cache to speed up the process.

In real world, optimising tripple channel architecture can hardly gives you any benefit, especially the process that are not memory bandwidth limiting. But more cache are always good. Unless you only use Handbrake to encode some small videos, and the overall memory usage (including cache) can never go above 24GB. Then install only 3 sticks may be actually the better option for you .

Thanks, that's great information. I remember reading in the CPU post that there was a knee in the curve between RAM channel and amount, but wasn't sure where it was. I do plan on a fair amount of video encoding, at least in the near future. I want to clear out old video cameras, so I am going to digitize a bunch of old family videos we have around the house, so I can purge all that old equipment.

 

[Slash-2CPU]

If your main use is handbrake, and you are encoding some large videos. RAM size should gives you more advantage than optimising the RAM speed. I heavily use Handbrake as well. Most of the videos are within 10GB in size. Even my 48GB RAM is more than enough for encoding (memory pressure always green), but I can clearly see that my Mac can fully utilise all the RAM. When running Handbrake, all the remaining RAM can be used as cache to speed up the process. (Most of the time, ~24GB real usage and ~24GB work as cache, plus <2GB memory compression and <2GB swap)

In real world, optimising tripple channel architecture can hardly gives you any benefit, especially the process that are not memory bandwidth limiting. But more cache are always good. Unless you only use Handbrake to encode some small videos, and the overall memory usage (including cache) can never go above 24GB. Then install only 3 sticks may be actually the better option for you .

Sorry. You don't understand how caching works. Mac OS caches the file, assuming Handbrake will look at it again. Handbrake only reads the file once. The cache gets huge, but has 0 hits. Try it yourself. You are correct that Handbrake doesn't care much about memory bandwidth, it's almost 100% CPU-limited. In this case, more cache does nothing.

For Handbrake, dual-channel, triple-channel, 8GB or 128GB doesn't matter. It's all about more cores, higher CPU clocks. SSD provides a *tiny* benefit here; maybe 1-4% over HDD. I'm really not sure on Fusion setup.

If you look at avg FPS on a very large file, you may see +1-3% on triple channel. Barely anything.

 

[usna92]

Will know more tonight.

 

[usna92]

Well the performance, at least right now is pretty much SSD speeds for the fusion drive. I haven't started writing the hard drive yet, so it is performing as an SSD. 399 Write speeds with 450 read speeds. On the platters I was at 76 and 78. So that is a market performance improvement. However, I found a 240GB SSD in an old computer that I forgot I threw it in there. It is getting ready to be recycled, so what is going to happen next is a JBOD between the 256 Retina SSD and the 240 SSD I just dug up. I think that will be a sufficiently sized boot drive, so that I no longer need the fusion drive on boot. It looks like I will get another one of these machines from work, so I will have second machine to measure the performance of the long term average.