MacPro 5,1 Upgrade Questions

[AndreeOnline]

If you have a file to download I can test it on my 12c 3.46GHz.

I'm not familiar with Matlab, but I've got the latest R2017b installed as a trail. I managed to just type 'bench' and run that, but I get the feeling you're running something else?

 

[whg]

If you have a file to download I can test it on my 12c 3.46GHz.

I'm not familiar with Matlab, but I've got the latest R2017b installed as a trail. I managed to just type 'bench' and run that, but I get the feeling you're running something else?

Thank you for your offer. Unfortunately most of my code is proprietary and cannot be shared. I will switch to the latest Matlab version, but for now I have version 2017a installed on all systems I use (and everything works as expected). 2017b will be needed for High Sierra, but I just upgraded my MacBook to Sierra, which seems very stable now (in contrast to what I hear from High Sierra). And my Paragon HFS+ driver on Windows would now longer work with the new AFS file system.

For benchmarking, it is important to use the same version of Matlab, as they made significant speed improvements from version to version recently. There were significant differences between Windows and Mac versions in the past, but 2017a performs equally on macOS and BootCamp Windows on my Macbook (at least for my test setup).

Last but not least: I really like my "new" Mac Pro. What a beautiful machine! Hopefully Apple will come back to something like this with the upcoming version.

 

[mp2017]

Thank you for the link, but I'm confused now because this is where I got the reference to the XFX graphics card. In this thread it is also said that MSI should be omitted because those don't have a double ROM (or something like this).

Today I got the Mac Pro 5,1 with the 2.8 GHz 4 core CPU and 7 GB RAM. I upgraded to El Capitan, installed Matlab 2017a, and performed a quick benchmark with 4 threads. To my disappointment, the run time was 3 (three) times of what my 2015 MacBook Pro 15" needed (with the 2.5GHz CPU).

OK, this is a much newer computer, but when I tested the same software against my old Dell, also end of 2010, with 2.93 GHz i7 CPU and 8GB RAM, I found that the Dell is also almost twice as fast as the Mac Pro. How can this be possible? Are the Xeon CPUs so that much older technology?

I still have some hope for when I'll receive the "new" X5690 and 64GB RAM.

Would it be possible to have you list the exact processor models from each of the systems mentioned? If not that then the specific model number for each one? That may help us identify a pattern. Off the top of my head I'm thinking single thread versus multithread differences.

 

[whg]

Would it be possible to have you list the exact processor models from each of the systems mentioned? If not that then the specific model number for each one? That may help us identify a pattern. Off the top of my head I'm thinking single thread versus multithread differences.

While investigating possible Windows PCs for my project, I stumbled upon this very issue of relatively low single thread performance with higher core counts. As only parts of my Matlab code is parallelized, the single thread performance might be more important than I anticipated. The exact CPU models of the computers I mentioned before are currently unknown to me. I'm aware that there are tools to find out, but this is probably not really helping a lot, anyway. I'm no IT specialist or computer scientist, I just want to calculate some physical models. At the end, my Matlab algorithms will be re-written in C/C++ for retail applications.

 

[kschendel]

I think my main interest is why the Dell, which should have the same generation CPU, is so much faster. I suppose the Dell was running Windows? It's possible that (some of) the difference is simply windows matlab vs mac OS matlab, whether because of compilers, different parallelization, different porting-layer code, or whatever.

 

[whg]

I think my main interest is why the Dell, which should have the same generation CPU, is so much faster. I suppose the Dell was running Windows? It's possible that (some of) the difference is simply windows matlab vs mac OS matlab, whether because of compilers, different parallelization, different porting-layer code, or whatever.

I started my current job on 2 January 2011. The Dell was already there at that time. It still runs Windows 7 x64, but Matlab was upgraded to version 2017a. This is the same version I have on the BootCamp partition of my MacBook Pro 15" 2015. This BootCamp Matlab has practically the same speed as the genuine Mac version. The Dell has an i7 2.93GHz quad core processor, very probably the Sandy Bridge version. I had the Sandy Bridge on my early 2011 iMac and MacBook Pro.

Sandy Bridge was a significant step in Intel's processor development. The Xeon 2.8GHz in my MP 5,1 is probably based on the architecture before Sandy Bridge. Sorry if this doesn't make sense, but I'm not an expert for these things.

 

[whg]

If you're going from a 2.8 Ghz Nehalem to a 3.46 Ghz Westmere, you can expect very (very!) roughly 50% better performance single threaded, and you get the extra 2 cores.

Just wanted to give an update: I firstly received the RAM (4x16GB) and found that only 3 of the rdimms can be used with the original CPU. Today I upgraded the CPU (X5690) and now my Mac Pro can use all 64 GB RAM.

The performance gain for my parallel processing Matlab benchmark is a full factor of 2 (30" down to 15" runtime).

The CPU upgrade was a very easy job, after viewing a video and instructions I found in this forum.

Overall, I'm very happy with the upgraded Mac Pro. Next step will be the SSD.

 

[whg]

I now made a longer test (30 minutes), and the 50% advantage of my Macbook Pro 2015 with 4-core 2.5GHz CPU shrinks down to only 8%.