Mac Pro (Late 2013) BootCamp Windows Review

[wheelhot]

I've received my MacPro about a week ago and finally has the time to write a proper review about it. As there's many reviews out there regarding its performance in OSX, I'll only be covering the Windows side of things, from installation to Windows benchmarks.

My model is the stock QuadCore Xeon 3.7GHz, 12GB RAM and D300. It is hooked up to a Dell U2711 with max resolution of 2560x1440.

Note: To achieve full resolution, there's only 2 ways for it, either getting a miniDisplayPort to Dual-Link DVI (the adapter usually has a male USB + mDP to female USB + DVI) or miniDisplayPort to DisplayPort (I opted for this option as it's cheaper and less potential problem). Some MR readers reported that they are abled to achieve full resolution (higher then 1920x1080) via HDMI cable, but from my research, this will depend if your monitor will internally output higher then 1920x1080 if connected via HDMI

Installing Windows
Installing Windows was quite straight forward except for a few things to take note of:

1. 16GB USB thumbdrive is required, either USB3 or USB2 will work fine, I'm guessing USB3 thumbdrive will install Windows 8 much faster. BootCamp Assistant will take roughly 12GB of thumbdrive space (Windows 8 disk image + Apple Windows drivers). I've read that they'll be issues with the nMP installation if it's not done via thumb drive.
2. If you bought the Windows 8 installation disk, you'll first need to convert it into a disk image, you can read about it here
3. Windows 8, including Apple drivers will take a whopping 30GB of your hard disk partition

Other then those things, the installation went on smoothly. After installing I had a few problems where I'm unable to connect to my wireless network but leaving the machine for 15mins or longer somehow fixed it, not exactly sure why. And I encountered a few times where the audio driver crashed.

Benchmarking
As you'll know, I've been asking for a specviewperf test and it'll be starting my review with it. As most of the specviewperf test is done with 1920x1080 resolution, I downscaled my monitor resolution to match it. I'm quite sure it affects the result somewhat, but I don't think it'll be far off either. I'm the kind of person where a 5fps difference is negligible. Attached is my Specviewperf 12 result with results taken from Specviewperf official site as well. I'm looking for any nMP owners with D500 and D700 to supply me the missing information. Thanks polygoo and nathan43082

Note: I've enabled CrossFireX for all applications

(right click to open the image in full size)

From the benchmark, I must say that the D300 does really well, it compares favourably to the AMD FirePro W5000 (MSRP $599.00) and Quadro K4000 (MSRP $1,249.00). The only benchmark where it does badly is Energy-01 where the driver repeatedly crashes and SNX-02 (Siemens NX) where I'm guessing is related to driver optimisation.

Conclusion
The word Apple and value doesn't go together so often, especially when it comes to their Macintosh product line. However the new redesigned MacPro is truly a value for money. There's nothing else that offers the performance in a similar form factor. If the iPad Pro 13in really comes true, I hope Apple consider allowing it to have a TargetDisplayMode (similar to the iMac), so it can be used in conjunction with the nMP making it a portable workstation. As other portable displays out there uses USB3 connection, I wonder how the portable monitor will perform if it uses Thunderbolt 2 instead?

I'm accepting to do other Windows benchmarks providing it's free.

 

[Che Castro]

Thanks for the review

 

[Anim]

Thanks for this. You finally got your specView perf bench done

Has anybody done a full windows install writeup yet? I believe there are three different ways? EFI, Bootcamp and WinClone. It would make a popular article I think.

 

[m4v3r1ck]

Thanks for your review, the more the merrier!

 

[AidenShaw]

1. Windows 8, including Apple drivers will take a whopping 30GB of your hard disk partition

Probably more than half of that is the paging file and hibernate file in the root directory.

The pagefile can be shrunk - since you might not ever even come close to using all of the RAM while in Windows. I usually set it to 256MiB:4096MiB (initial size 256 MiB, expandable to 4096 MiB). Some memory-mapping operations require reserved backing store even if they never page, so some apps might need a higher initial size.

The hibernate file (default around 75% of RAM size) can be disabled if you don't want to be able to hibernate the Windows image. Use the command "powercfg -h off" in an elevated command window.

 

[MDP]

Nice review! But what drivers were you using for the graphics cards - official Apple Bootcamp drivers or AMD Catalyst drivers from their website?

Becuase I think you can uninstall the Bootcamp-included drivers and get newer ones from AMD online...

 

[wheelhot]

Nice review! But what drivers were you using for the graphics cards - official Apple Bootcamp drivers or AMD Catalyst drivers from their website?

Becuase I think you can uninstall the Bootcamp-included drivers and get newer ones from AMD online...

Thanks! I'm using the official Apple BootCamp drivers. If I install AMD Catalyst Pro driver, how am I suppose to roll back to Apple BootCamp driver if things doesn't work as it should?

I wonder when or will AMD ever update their site to list the nMP FirePro Dx00, right now the only mention of it is in their, solutions we enable page.

The GPU is not mentioned anywhere else.

----------

Probably more than half of that is the paging file and hibernate file in the root directory.

The pagefile can be shrunk - since you might not ever even come close to using all of the RAM while in Windows. I usually set it to 256MiB:4096MiB (initial size 256 MiB, expandable to 4096 MiB). Some memory-mapping operations require reserved backing store even if they never page, so some apps might need a higher initial size.

The hibernate file (default around 75% of RAM size) can be disabled if you don't want to be able to hibernate the Windows image. Use the command "powercfg -h off" in an elevated command window.

Oh okay, that's new. Is there a folder to check the page file size or can only be configured via settings? I got 30gb from a fresh install btw.

 

[deconstruct60]

.... I'm looking for any nMP owners with D500 and D700 to supply me the missing information. ...

Didn't someone provide some of that already???

https://forums.macrumors.com/threads/1692605/

D500 results got posted there. Oops, I guess the link is dead now. (I didn't cache/copy it, but someone else might have ). From what I remember not particularly that much better/different. ( I think that one was done without Crossfire. So it was a D500 up against W5000 and the rest and it didn't look good. ).

From the benchmark, I must say that the D300 does really well, it compares favourably to the AMD FirePro W5000 (MSRP $599.00)

Eh? The Dell T3610 / W5000 entry is highly suggestive of being just a single W5000 up against two D300's. The "raw" GPU in the D300 is closer to (if not exactly the same) as the W7000. Pretty sure a top end iMac would top those K2000 (and lower) numbers too. [athough outside the official supported config windows.]

The dual D300s are cheaper than two W7000s, but the notion of a bargain is a bit of stretch.

Those benchmark far more so seem indicative there is something unoptimized on the Mac Pro side.

 

[AidenShaw]

Oh okay, that's new. Is there a folder to check the page file size or can only be configured via settings? I got 30gb from a fresh install btw.

It would default to the root of C:

From an elevated prompt (wrong click on "Command Prompt" and select "Run as administrator") type dir /a C:\*.sys. You can also use File Explorer, but you need to set "Show protected files" and "Show hidden files.

Pagefile size (and additional pagefiles) are set in "Advanced System Settings"

powercfg options are:

-HIBERNATE, -H
Enables-Disables the hibernate feature. Hibernate timeout is not
supported on all systems.

Usage: POWERCFG -H <ON|OFF>
POWERCFG -H -Size <PercentSize>
-Size Specifies the desired hiberfile size in percentage of the
total memory. The default size cannot be smaller than 50.
This switch will also enable the hiberfile automatically.

If the hiberfile is too small, the hibernate will fail and the system will halt. Exact size is hard to predict - it depends on the compressibility of the actual memory that's in use.

 

[nathan43082]

I'm accepting to do other Windows benchmarks providing it's free.

Here's one on me, which you can use to update some blank spots in your chart:

Just got my nMP 8-Cores, D700s, 16GB RAM, 1 TB PCIe Flash. I installed Windows 8.1 via Bootcamp, then installed the latest AMD Catalyst drivers.

I downloaded and ran this same SPECviewperf12 test (huge file for a test) and used these settings:

And received these results:

 

[nathan43082]

After I posted, I noticed that my numbers were not too far off from the twin D300s, which got me to thinking that perhaps the test was only using one D700. I disabled CrossFireX in the AMD Catalyst control panel and re-ran the test and received the same numbers ±2 or so.

Either the test is hard-wired to use both cards or hard-wired to only use one. I saw no obvious way to check or change it using their supplied GUI control panel which lets you start the whole thing off.

Perhaps someone can let me know the console commands to force the test one way or the other and I can retest.

 

[Anim]

After I posted, I noticed that my numbers were not too far off from the twin D300s, which got me to thinking that perhaps the test was only using one D700. I disabled CrossFireX in the AMD Catalyst control panel and re-ran the test and received the same numbers ±2 or so.

Either the test is hard-wired to use both cards or hard-wired to only use one. I saw no obvious way to check or change it using their supplied GUI control panel which lets you start the whole thing off.

Perhaps someone can let me know the console commands to force the test one way or the other and I can retest.

Install a GPU monitor and see if both cards are being used?

Never tried it but maybe this one (windows only) ?
http://www.geeks3d.com/20100517/gpu-tool-gpu-shark-0-1-0-simple-gpu-monitoring-utility/

Or this one. Again never tried it.
http://www.geeks3d.com/20131113/gpu...lity-opengl-opencl-geforce-quadro-radeon-gpu/

 

[nathan43082]

Good idea. I'll give it a try. Thanks.

 

[nathan43082]

I found a nice system monitor from AMD themselves which not only shows GPU but the CPU cores as well.

The results are surprising. The SPECviewperf12 used one GPU, regardless of how I have the CrossFireX setting configured. I even tried uninstalling the stable AMD Catalyst drive and installing the recent beta version. No change.

What's worse is that many of the tests don't even hit 100% of one GPU because the tests are often throttled by poor implementation of parallelism in the CPUs as you can see with the screen cap I've linked below (you will want to right-click it to open it in a new window).

Behold, SPECviewperf12 running on just one GPU, and not very well at that, even though CrossFireX is enabled and the extra "helper" checkbox is checked:

 

[Anim]

I found a nice system monitor from AMD themselves which not only shows GPU but the CPU cores as well.

The results are surprising. The SPECviewperf12 used one GPU, regardless of how I have the CrossFireX setting configured. I even tried uninstalling the stable AMD Catalyst drive and installing the recent beta version. No change.

What's worse is that many of the tests don't even hit 100% of one GPU because the tests are often throttled by poor implementation of parallelism in the CPUs as you can see with the screen cap I've linked below (you will want to right-click it to open it in a new window).

Behold, SPECviewperf12 running on just one GPU, and not very well at that, even though CrossFireX is enabled and the extra "helper" checkbox is checked:

Image

Just one core and low usage of one GPU in a dual GPU system Seems like it needs an update. You could email that pic to the developers and ask them for an explanation, they may say they are working on it or that the software is now unsupported, if your interested that is.

 

[nathan43082]

Just one core and low usage of one GPU in a dual GPU system Seems like it needs an update. You could email that pic to the developers and ask them for an explanation, they may say they are working on it or that the software is now unsupported, if your interested that is.

This was just the worst of the bunch. There was another like it, of a car, that moved *very* slowly on the screen. None of the tests used more than one GPU but many of them red-lined the one, or very nearly so when it wasn't dipping down to 50-60% here and there.

I am not really interested in contacting the developer. I only ran this test to see how my D700s compared to the D300s and I see now that this test is not going to provide what I need since only one GPU is being used and that the CPU is a limiting factor. We know that the 4-core is probably faster at the 8-core in these situations, so there's no longer any apples to apples. The Heaven and Valley benchmarks seem a bit more fair.

 

[V.K.]

Windows 7 on 2013 MP?

thanks for the guide. will windows 7 install work via this process? bootcamp assistant says that I need windows 8 or later. I have a a copy of windows 7 and don't really feel like buying windows 8.

 

[lesferdinand]

Probably more than half of that is the paging file and hibernate file in the root directory.

With 32GB memory, just my hibernate file was 29Gb... I was shocked to see that my clean Windows 8.1 Enterprise install took around 60GB. Once I learned about the hibernate file, I turned off that functionality and got a good chunk of the space back.

I used a USB 3.0 card reader (built into my screen) with a 45mb/s SD card as the Win install disk. I've never experienced a quicker OS install. I will never be able to go back to non-SSD systems.

 

[wheelhot]

Sorry for the late reply, was busy these past few days working on a project with the nMP and SolidEdge

From the newly updated results. The D300 performs remarkable well (considering the price point and against its 2 brothers), the D500 again is a so-so card as it loses to the D300 in almost all except for Energy and Siemens NX. The D700 performs quite well especially against the W7000 (MSRP US$1000) and K5000 (MSRP US$2500).

 

[wheelhot]

Didn't someone provide some of that already???

https://forums.macrumors.com/threads/1692605/

D500 results got posted there. Oops, I guess the link is dead now. (I didn't cache/copy it, but someone else might have ). From what I remember not particularly that much better/different. ( I think that one was done without Crossfire. So it was a D500 up against W5000 and the rest and it didn't look good. ).

Thankfully I saved it in my old computer and just has the time to retrieve it back , updated the chart already

Eh? The Dell T3610 / W5000 entry is highly suggestive of being just a single W5000 up against two D300's. The "raw" GPU in the D300 is closer to (if not exactly the same) as the W7000. Pretty sure a top end iMac would top those K2000 (and lower) numbers too. [athough outside the official supported config windows.]

Hmm, depends, on some view sets it'll perform really bad. I'm not sure really as some people has shown that their "gaming GPU" performs really well in a specviewperf test which usually only emphasis in OpenGL performance.

The dual D300s are cheaper than two W7000s, but the notion of a bargain is a bit of stretch.

Those benchmark far more so seem indicative there is something unoptimized on the Mac Pro side.

I compared it to the W5000 as it performs better and or worse then it. There's no way the D300 can be compared to W7000

And yes, looking at the result from D300, D500, D700. It seems that the drivers is unoptimised. Though it's interesting to see that it'll perform much better or worse depending on the view sets

----------

Here's one on me, which you can use to update some blank spots in your chart:

Thanks! Updated

----------

Just one core and low usage of one GPU in a dual GPU system Seems like it needs an update. You could email that pic to the developers and ask them for an explanation, they may say they are working on it or that the software is now unsupported, if your interested that is.

My guess is Spec purposely made it that way as most CAD software doesn't make use or optimised for more then 1 core or 1 GPU. It's a sad reality with these CAD vendors.

----------

This was just the worst of the bunch. There was another like it, of a car, that moved *very* slowly on the screen. None of the tests used more than one GPU but many of them red-lined the one, or very nearly so when it wasn't dipping down to 50-60% here and there.

I am not really interested in contacting the developer. I only ran this test to see how my D700s compared to the D300s and I see now that this test is not going to provide what I need since only one GPU is being used and that the CPU is a limiting factor. We know that the 4-core is probably faster at the 8-core in these situations, so there's no longer any apples to apples. The Heaven and Valley benchmarks seem a bit more fair.

Yes, each of the model represents a view set (there's a few models for 1 view set)
In most cases, a spec test will do really well on a workstation driver GPU then a gaming driver GPU. Or that's how it used to be

----------

thanks for the guide. will windows 7 install work via this process? bootcamp assistant says that I need windows 8 or later. I have a a copy of windows 7 and don't really feel like buying windows 8.

Sorry, the nMP Bootcamp Assistant only accepts Windows 8. One of the MR users is trying to get Win 7 to work with the nMP but unless you're technically savvy, the hassle might not even be worth it.

I don't feel like using Windows 8 as I prefer Windows 7, but guess we'll will have to accept that change sooner or later

 

[orbitalpunk]

was anyone able to determine if it was better to run bootcamp drivers or catalyst beta drivers?