My nMP Review (vs my old 4,1 Mac Pro)

[milo]

Applications load nearly instantaneously, including Aperture and Logic.

Hey, if you would be willing to try out the Logic benchmark session and report back, that would be very much appreciated. It's a simple test, download a small zipped Logic session and see how many tracks can be unmuted before it gives an error. Lots of people are dying to know how Logic runs on that machine.

http://www.gearspace.com/board/music-computers/371545-logic-pro-multicore-benchmarktest.html

 

[VirtualRain]

I use Aperture and RAWs for my photography hobby and I'm looking to do pretty much your same upgrade.
Do you think the D500 was worth it, versus the base D300? Also, any suggestion on 12GB Ram (that I currently have on a MP 5,1 6-core) versus 16/32GB?

Hey, as I mentioned in the review, I was fully intending to just get the 6-core with D300 GPUs until I got to the order page and realized the D500s were only a few hundred dollars more. For the added VRAM and potential compute power, I decided that was a worthy way to spend a few hundred since it's very unlikely these GPUs will ever be upgradable. However, the D500s are purely potential at this point. I don't think there's any software I run that can really leverage that power at the moment.

As for RAM, I think 16GB is probably fine. It seems the more RAM you have, the more Aperture will use as a file cache for your library which may make jumping around in your library faster. I also got along well with 12GB in my oMP. I went for 32GB on this nMP right out of the gate because I knew if I didn't need it now, I would end up needing it down the road, and again, it's not a big investment ($400 BTO option).

 

[analog guy]

Actually, it is a problem because Time Machine will happily (by design) completely fill up whatever volume you place it on, so once that happens you can't put any more other files on the partition. There's no way to limit Time Machine so that it leaves free space for other use on the drive. But, it's not an issue if you're using an Airport, because you can just partition the drive before you plug it in. Might be an issue, as you speculate, if there's no way to partition the drive in a Time Capsule.

this is a valid point; TM will eat up all space it is permitted.

the TC's disk does not appear in disk utility, but if you access it via airport utility the title of a field is "partitions" (plural), and it clearly has space for more than one. i am not willing to erase my disk right now to test it out, but i do believe that one sets up partitions from airport utility.

afterwards, each partition should be mountable in finder.

 

[analog guy]

forgot to attach the screenshot.

 

[Dark Goob]

That's the same I've seen in another thread. Pretty sure it's accurate. Keep in mind, it's only utilizing one of the GPUs in that test. BTW, what are you comparing that to in concluding it sucks?

Well my 2009 Mac Pro (2x 2.26ghz) with EVGA NVIDIA 570 GTX (base model, 1280 MB with mini-HDMI and 2x DVI) scored a 731 on Heaven 4.0 Extreme with max settings at 1080P (stereoscopic off). Newer cards are getting well over 1000.

 

[Macshroomer]

No doubt, and I don't think the OP was hiding that at all. In fact the OP posted this in the start of his review:
The nMP:
3.5GHz 6-Core/D500s configuration with BTO upgrades to the RAM (32GB) and SSD (1TB).

The oMP:
2.93GHz 4-Core/GT120s with DIY 12GB RAM and 256GB M4 SSD on a PCIe Velocity Solo x2. It also has a RAID0 array of SSDs for my photo library running on a Highpoint 2720 PCIe RAID card (thread here).

Frankly if you are running a Mac Pro 5,1 from 2010 or 2012 you probably aren't ready to make the jump to a nMP. There hasn't been a dramatic change in processing power and we are probably a year or two away from OpenCL really taking the world by storm (unless you use FCPX solely). Any of us rocking 4,1's (especially stock) and older cMP's are the ones more likely to make the jump.

So for those really needing to upgrade, this was a perfect comparison/review.

Yeah, pretty awesome review, but I agree with the above in terms of upgrade cycle / cost benefit. If I were going to continue to use digital cameras in my career as a photographer, I might consider the new machine next revision but since I am not ( Film & darkroom only soon ), I would just be blowing capitol that I could use elsewhere.

I am encouraged by his numbers on the "Macrumors Photoshop" test though, while a big improvement over his old machine, he beat my score of 9.5 seconds by a hair.

Beautiful machine for those who need it and a *great* review, thanks!

 

[Killerbob]

No need to get so defensive.

1. I posted the GTX 680 Heaven scores, using the same settings as OP, as it's a more modern GPU than the miserable GT 120, and hence a little more representative of what the new machine is up against in the real world.
2. The LuxMark scores, again using the same settings as OP, are from an OpenCL benchmark and do show the nMP's pair of GPUs having a clear advantage over the nMP's single card. It's 2 v 1 though, so "crushing" would be a bit of overstatement.

It's no surprise that the nMP is more powerful than the old machines but what this does show is that, with thoughtful upgrades, the 2009-on Mac Pros can still be eminently usable workhorses and that the performance gains of the new one aren't as large as some folks are trumpeting!

I fully agree... I just ran the Luxmark/Heaven benchmarks on my ol' trusty MP3,1/2.8GHz Octo/16GB/Accelsior PCIe/GTX680/USB3, and the results were interesting in comparison with the super duper nMP:

* Luxmark, CPU - 322 - so a lot worse than the MP5,1 and the MP6,1 (as expected).
* Luxmark, GPU - 726 - so better than flashed GTX680 in MP5,1 (???), but a LOT worse than D500 in nMP.
* Luxmark, both - 1031 - worse than both other combinations (as expected due to CPU power).

* Heaven, Extreme, FPS - 34,2 - slightly worse than MP5,1, but SIGNIFICANTLY better than nMP/D500s.
* Heaven, Extreme, Score - 862 - again worse than MP5,1, but over 50% better than nMP/D500s.

With the Accelsior getting comparable scores to the nMP in read/write, it is "only" CPU power and TB2 that would be interesting to me. hmmmmmmm.

 

[wildmac]

It's seeming more and more that it's a heaven-or-hell situation with the Dx00 cards, where, ironically, Heaven is hell for the Dx00 series.

So, with that all the benchmark-fu in mind, where in the real world, do the Dx00 cards that Apples says is the future shine, and who should run for the hills?

 

[Dark Goob]

I fully agree... I just ran the Luxmark/Heaven benchmarks on my ol' trusty MP3,1/2.8GHz Octo/16GB/Accelsior PCIe/GTX680/USB3, and the results were interesting in comparison with the super duper nMP:

* Luxmark, CPU - 322 - so a lot worse than the MP5,1 and the MP6,1 (as expected).
* Luxmark, GPU - 726 - so better than flashed GTX680 in MP5,1 (???), but a LOT worse than D500 in nMP.
* Luxmark, both - 1031 - worse than both other combinations (as expected due to CPU power).

* Heaven, Extreme, FPS - 34,2 - slightly worse than MP5,1, but SIGNIFICANTLY better than nMP/D500s.
* Heaven, Extreme, Score - 862 - again worse than MP5,1, but over 50% better than nMP/D500s.

With the Accelsior getting comparable scores to the nMP in read/write, it is "only" CPU power and TB2 that would be interesting to me. hmmmmmmm.

Perhaps it might be that Heaven is a gaming benchmark, while FirePros are not optimized for gaming. Rather they are optimized for pure number crunching power. They are workstation GPUs not gaming PC GPUs. So for mining bitcoins and raytracing 3D scenes, the nMP is an absolute beast, but I doubt it's the best for gaming.

Now my 570 GTX scores:

With CPU:

Clearly the CPU is behind the times but the 570 GTX holds its own still. I think I'm going to wait a couple more years until the nMP has an 8TB internal SSD option so I don't have to use two external RAIDs (just one for time machine backup).

Note that I don't rely on oMP for performance-critical applications. The primary reason I bought it was for the internal storage capacity and upgradeable GPU. Otherwise I'd just have gotten a Mac Mini or a second MBP.

 

[cal6n]

*snip*

* Luxmark, CPU - 322 - so a lot worse than the MP5,1 and the MP6,1 (as expected).
* Luxmark, GPU - 726 - so better than flashed GTX680 in MP5,1 (???), but a LOT worse than D500 in nMP.
* Luxmark, both - 1031 - worse than both other combinations (as expected due to CPU power).

*snip*

That's somewhat puzzling. I even tried running LuxMark from a cleanly logged-in test account with virtually no difference in score. I'd be interested to know how you're getting around a 7% performance boost in OpenCL.

What GTX 680 are you running? Any overclocking? Apple's drivers? Mine's originally a 2GB PNY card flashed with the original EVGA ROM, running on Apple's stock drivers in 10.9.1.

 

[Killerbob]

That's somewhat puzzling. I even tried running LuxMark from a cleanly logged-in test account with virtually no difference in score. I'd be interested to know how you're getting around a 7% performance boost in OpenCL.

What GTX 680 are you running? Any overclocking? Apple's drivers? Mine's originally a 2GB PNY card flashed with the original EVGA ROM, running on Apple's stock drivers in 10.9.1.

I am running a flashed EVGA GeForce GTX680 Superclocked videocard. It is identified as an Nvidia GeForce GTX 680 2048 MB as per "About This Mac". It's in Slot-2, x16, and the Link Speed is 5.0 GT/s.

Using Cuda-Z I can see the clock rate is 1124MHz, and the Memory clock rate is 3104MHz.

Since I am using Mavericks (10.9.1) I am on standard Apple drivers.

 

[Raddock]

An Apogee Quartet: http://apogeedigital.com/products/quartet.php



Ha, me too! Lacie TBolt SSD to be exact.

That's funny, I ordered the exact same MP as you, and have the same Lacie drive.

 

[cal6n]

I am running a flashed EVGA GeForce GTX680 Superclocked videocard. It is identified as an Nvidia GeForce GTX 680 2048 MB as per "About This Mac". It's in Slot-2, x16, and the Link Speed is 5.0 GT/s.

Using Cuda-Z I can see the clock rate is 1124MHz, and the Memory clock rate is 3104MHz.

Since I am using Mavericks (10.9.1) I am on standard Apple drivers.

Well, that explains some of it. Mine's clocked at 1058 and 3004 MHz.

 

[rtrueman]

Congratulations, VirtualRain and thank you for the great review. As I've mentioned before, I use Aperture often along with PS and plugins. If there were any doubts before I'm ready to purchase the exact same system... Well, except for the bank account! I've got an upcoming wedding show and I'm hoping to book enough business to take the plunge. I look forward to reading the rest of the thread and hearing about any new information you discover.

Best!

 

[spaz8]

It seems like Mavericks is one of the big issues atm with the Dxxx series performing extra poorly in gaming situations.

You are not going to see a Dxxx under a different operating system than Mavericks.

I'd be curious to see how the 7950 is doing in mavericks on the heaven test. I'd also be curious to see a 2nd "gaming" test. Just because I know some 3d app's can heavily favor Nvidia or ATI. The 680 I expect to leave everything but the D700 in the dust.

I know one 3d software that I use has had bugs since osx 10.9.0, still there in 10.9.1 but are on apples side to solve. Here is hoping 10.9.2 fixes the opengl issues.

Its also pretty well known that Nvidia neutered the compute performance on their Geforce line to make opengl performance the best it could be. So a Titan or a Quadro is needed for good cuda and any attention by nvidia towards opencl.

The Amd cards are more balanced, if anything favoring compute (opencl) over opengl.

Rose coloured glasses, but I expect the nMP's to get faster and faster over the next year as more software is written for them, and the OSX drivers and OS function mature to leverage the gpu.

 

[VoR]

Perhaps it might be that Heaven is a gaming benchmark, while FirePros are not optimized for gaming. Rather they are optimized for pure number crunching power. They are workstation GPUs not gaming PC GPUs. So for mining bitcoins and raytracing 3D scenes, the nMP is an absolute beast, but I doubt it's the best for gaming.

You won't get anywhere near paying for the electricity, let alone a ROI by mining bitcoins on the nMP

If you're number crunching sha256 or scrypt, midrange consumer cards are an infinitely better choice - They're faster and cheaper.
Things have changed (and a workload like this is great for my argument), you should be looking at the hardware, not the silly 'firepro' word that's printed on the box.

 

[VirtualRain]

I redid Heaven 4 OpenGL using these settings although I could not find anyway to set AF (Anisotropic Filtering?)...

It would seem the D500 is performing as expected... given it's lower clocks.

 

[VirtualRain]

I also ran the Valley OpenGL benchmark on Extreme and got results consistent with Barefeats.

FWIW, I much prefer watching the Heaven Benchmark... I could watch that all day

 

[bcuzawd]

I redid Heaven 4 OpenGL using these settings although I could not find anyway to set AF (Anisotropic Filtering?)...

Image

It would seem the D500 is performing as expected... given it's lower clocks.

So, theoretically...the D700's should fall-in around the 75fps range?

 

[wheelhot]

I redid Heaven 4 OpenGL using these settings although I could not find anyway to set AF (Anisotropic Filtering?)...

Image

It would seem the D500 is performing as expected... given it's lower clocks.

That looks about right, judging from those benchmark and your results, it seems like the FirePros are indeed running on some sort of FirePro driver. Notice that gaming GPU dominated the top of the benchmark results whereas followed by workstation GPUs

Thanks

Now, anyone else have nMP with D300 and D700 to test this with?

 

[VirtualRain]

So, theoretically...the D700's should fall-in around the 75fps range?

I'm not sure. The D700 has about 33% more cores and a boost clock thats around 15% higher so I could see it hitting 60-65 but I doubt it will do 75. Remember these Dxxx GPUs are clocked lower than their Wxxxx counterparts.

 

[mikepj]

Yeah, iStat only shows the fan speed... no CPU temps. I've emailed them to let them know.

Can you try out my monitoring app, XRG, to see if that will show the temperatures for you?

http://www.gauchosoft.com/xrg/

Thanks for the nice write-up. Can't wait for my own nMP to be delivered…

 

[VirtualRain]

Can you try out my monitoring app, XRG, to see if that will show the temperatures for you?

http://www.gauchosoft.com/xrg/

Thanks for the nice write-up. Can't wait for my own nMP to be delivered…

Sure... I'll give that a try tomorrow. Thanks!

 

[mrkennethtong]

You beat me in Single Core, but not in Multicore.

 

[N19h7m4r3]

I redid Heaven 4 OpenGL using these settings although I could not find anyway to set AF (Anisotropic Filtering?)...

Image

It would seem the D500 is performing as expected... given it's lower clocks.

Please run that in fullscreen, it does often make a difference.

It's also worth noting that he Windows Version of Heaven supports OpenGL 4.0, while the OS X version is still only OpenGL 3.2 as far as I know.
Could also make up a difference seen in Tom's Review here.
http://www.tomshardware.com/reviews/best-workstation-graphics-card,3493-15.html