2009 MP faster than 2013nMP

[Steve Expat]

I'm not buying a 2013 nMP in 2017 (a month away). So here's what I did starting about a year ago. The result was a MP that's as fast as a 12-core nMP, and faster in disk read times and boot drive speed.

First let me say that I need extreme multithreading capabilities that only Xeons can provide.

I bought a mint used 2009 4,1 MP. $700.
Here are the modifications I made.

1) Upgraded the firmware to 5,1.

2) A pair of x5690 dual hex core (12-core) 3.46GHz CPUs and de-lidded them. $400

3) 96GBs of RAM (16GBx6). $600

4) 512GB Samsung SM951 with the Lycom PCIe card and installed it in slot 3 as my OS/apps boot. 1500MB/s reads. $425

5) Squid m.2 carrier card in slot 2 and installed (4) 256GB SM951 in a RAID 0 giving me 5700MB/s reads. $800

6) (4) 4TB HDDs internally. $500.
Graphics card with an AMD R9 280x. $200

7) USB 3 card with four independent lane controllers. $129

Geekbench score - 32000
Read speed from the Squid RAID - 5700GB/s
Boot time from SM951 - 10 seconds

Total cost = $3.500

So I bought a second 4,1 2009 MP and did the same exact upgrades.

This is a MIDI mockup rig that needs every bit of CPU cores and speed, large amount of RAM, and the fastest I/O possible for the sample drive. The two MacPros are in a host/slave environment.

These upgrades could extend the useful lives of these 8/year old MacPros.

However, I'm very unhappy with Apple's direction. I'm highly invested in OS X/macOS software and training. Some apps have Windows versions but not all.

I'm going to begin a social media campaign asking for Apple to take care of the pro user by providing bi-yearly Mac Pro updates, or demand Tim Cook's head.

I've had enough of Apple taking away products for no apparent reason. And I'm going to do something about it, and hopefully very loudly.

Until then, if you want a Xeon based Mac, try my formula above. You won't be disappointed. Even with a few weeknesses (internal SATA II,
1333MHz RAM, no TB) they're screamers.

 

[ITguy2016]

I didn't see anything other than Geekbench scores to show this system faster than a 6,1 Mac Pro. While I'm not a fan of the 6,1 Geekbench scores are next to worthless when it comes to real world application performance. If you've got some real world benchmarks then post them. Otherwise this is nothing more than sour grapes.

 

[cerberusss]

sour grapes

II thought it was a very interesting alternative to buying a 2013 MP, I just skipped over that last part

 

[h9826790]

I didn't see anything other than Geekbench scores to show this system faster than a 6,1 Mac Pro. While I'm not a fan of the 6,1 Geekbench scores are next to worthless when it comes to real world application performance. If you've got some real world benchmarks then post them. Otherwise this is nothing more than sour grapes.

Of course it depends on the work flow to determine if the maxed out cMP or nMP can be faster. Interestingly, 32000 Geekbench 3 score definitely not faster than the 12 cores nMP, it's just on par with it (at most).

However, I can see at least 3-4 items on this cMP is faster than the nMP. e.g.
4x SM91 RAID 0 sure is faster than any storage solutions that the nMP can do.
A single SM951 itself should faster than the stock Apple nMP PCIe SSD.
A single 280X should be faster than a single D700 (unless VRAM size limiting). And it can be further upgraded.
Some USB 3.0 card in cMP can provide 2A charging per port, so it can charge the phone / tablet etc faster

Apart from faster, the cMP has the option to have 96GB memory which the nMP doesn't have. So, let's say if a job is memory size limiting, 96GB RAM or even 128GB RAM may make the cMP significantly faster than the nMP on these particular job.

Of course, these are not any real world benchmark, but just my logical thinking told me that how the cMP can be faster than the nMP in those particular areas / situations.

 

[ITguy2016]

Of course it depends on the work flow to determine if the maxed out cMP or nMP can be faster. Interestingly, 32000 Geekbench 3 score definitely not faster than the 12 cores nMP, it's just on par with it (at most).

Which is why I said:

If you've got some real world benchmarks then post them.​

Geekbench is, in and of itself, fairly worthless.

However, I can see at least 3-4 items on this cMP is faster than the nMP. e.g.
4x SM91 RAID 0 sure is faster than any storage solutions that the nMP can do.
A single SM951 itself should faster than the stock Apple nMP PCIe SSD.
A single 280X should be faster than a single D700 (unless VRAM size limiting). And it can be further upgraded.
Some USB 3.0 card in cMP can provide 2A charging per port, so it can charge the phone / tablet etc faster

Yes, can be. But it depends on the application. Which is why I said, again:

If you've got some real world benchmarks then post them.​

Apart from faster, the cMP has the option to have 96GB memory which the nMP doesn't have. So, let's say if a job is memory size limiting, 96GB RAM or even 128GB RAM may make the cMP significantly faster than the nMP on these particular job.

Mac Pro 6,1 can utilize 128GB of RAM if you want to install it.

Of course, these are not any real world benchmark, but just my logical thinking told me that how the cMP can be faster than the nMP in those particular areas / situations.

Which is why I said, again:

If you've got some real world benchmarks then post them.​

 

[bernuli]

Nice going. If you want something done right, you gotta do it yourself.

I just don't see why Apple can't put out high end machines as well as midrange for the every day web browser. They may not sell as many high end, but the R&D will filter down to the rest of the fleet.

Apple has a great OS and they are the tops in build quality. It is a shame they don't go the extra half step and offer machines that will really take advantage of the OS and community.

I want to want a new Apple computer again.

 

[h9826790]

Which is why I said:

If you've got some real world benchmarks then post them.​

Geekbench is, in and of itself, fairly worthless.


Yes, can be. But it depends on the application. Which is why I said, again:

If you've got some real world benchmarks then post them.​

Mac Pro 6,1 can utilize 128GB of RAM if you want to install it.


Which is why I said, again:

If you've got some real world benchmarks then post them.​

If something is obvious, I don't think we must do anything from real world to back it up.

Obviously copying large file on a 4x SM951 RAID 0 array is faster than the nMP can do. So, why need any real world benchmark to prove it? Do I need to jump out from a building to prove that I will die if I do that? Or because there is no "real world benchmark" to prove that I will die, so it means absolutely nothing? There is something call common sense, logical thinking, etc.

Keep asking the real world benchmark but ignore the basic fact / number / spec doesn't mean it's practical. I am not saying that the pure number is more important. In some case, the number can means absolutely nothing. However, numbers usually can tell us something and help us to accurately predict the result. And we usually make purchase decision base on spec, isn't it? That's because we can expect what we roughly can get in real world.

 

[ITguy2016]

If something is obvious, I don't think we must do anything from real world to back it up.

Obviously copying large file on a 4x SM951 RAID 0 array is faster than the nMP can do. So, why need any real world benchmark to prove it? Do I need to jump out from a building to prove that I will die if I do that? Or because there is no "real world benchmark" to prove that I will die, so it means absolutely nothing? There is something call common sense, logical thinking, etc.

Yes, if you're claiming doing so will not kill you.

Keep asking the real world benchmark but ignore the basic fact / number / spec doesn't mean it's practical. I am not saying that the pure number is more important. In some case, the number can means absolutely nothing. However, numbers usually can tell us something and help us to accurately predict the result. And we usually make purchase decision base on spec, isn't it? That's because we can expect what we roughly can get in real world.

Numbers are only useful if they're relevant. The OP claimed a cMP is faster than a nMP solely based on Geekbench score. Geekbench is essentially useless. If the OP has a specific task for which the cMP is faster I'm willing to consider it.

I'm no fan of the nMP but I am realistic about its strengths and weaknesses compared to the cMP.

 

[h9826790]

Numbers are only useful if they're relevant. The OP claimed a cMP is faster than a nMP solely based on Geekbench score. Geekbench is essentially useless. If the OP has a specific task for which the cMP is faster I'm willing to consider it.

I'm no fan of the nMP but I am realistic about its strengths and weaknesses compared to the cMP.

I totally agree that we should use relevant numbers. That's why I suggest few examples that the cMP can be faster base on the relevant numbers. Those examples are solely reply to the sentence "I didn't see anything other than Geekbench scores to show this system faster than a 6,1 Mac Pro."

Of course, you are entirely correct if you specifically only talking about GB score cannot mean cMP is fast than nMP on OP's work.

 

[wonderspark]

Which is why I said:

If you've got some real world benchmarks then post them.​

Geekbench is, in and of itself, fairly worthless.


Yes, can be. But it depends on the application. Which is why I said, again:

If you've got some real world benchmarks then post them.​

Mac Pro 6,1 can utilize 128GB of RAM if you want to install it.


Which is why I said, again:

If you've got some real world benchmarks then post them.​

http://barefeats.com/hard210.html

There it is, real world benchmark. 5.5x to 6x faster than 6,1 Mac Pro, depending on which test you run.

Therefore, the sour grapes are being tasted by nMP owners, not cMP owners. It's clearly better to have PCIe slots all the way into the year 2017.

I have an Areca RAID card in my 4,1 that excels in disk read/write speeds as well, by about the same margin. Being able to attach sixteen 8TB disks for 128TB of space at over 3300MB/sec is pretty nice, but I'm using it in RAID 6... also a nice option. If one uses the current 10TB disks from Western Digital, and maxes out their ix-24 cards, you have 280TB of disk space at speeds well beyond what's possible in a nMP.

Heck, I could have 40TB in the HDD slots, and various SSDs in other PCIe slots and ODD bays, all internally. Can't put all that into a tiny little trash can.

 

[ITguy2016]

I totally agree that we should use relevant numbers. That's why I suggest few examples that the cMP can be faster base on the relevant numbers. Those examples are solely reply to the sentence "I didn't see anything other than Geekbench scores to show this system faster than a 6,1 Mac Pro."

Of course, you are entirely correct if you specifically only talking about GB score cannot mean cMP is fast than nMP on OP's work.

Then why are you challenging my request for them?
[doublepost=1480899921][/doublepost]

http://barefeats.com/hard210.html

There it is, real world benchmark. 5.5x to 6x faster than 6,1 Mac Pro, depending on which test you run.

Therefore, the sour grapes are being tasted by nMP owners, not cMP owners. It's clearly better to have PCIe slots all the way into the year 2017.

I have an Areca RAID card in my 4,1 that excels in disk read/write speeds as well, by about the same margin. Being able to attach sixteen 8TB disks for 128TB of space at over 3300MB/sec is pretty nice, but I'm using it in RAID 6... also a nice option. If one uses the current 10TB disks from Western Digital, and maxes out their ix-24 cards, you have 280TB of disk space at speeds well beyond what's possible in a nMP.

Heck, I could have 40TB in the HDD slots, and various SSDs in other PCIe slots and ODD bays, all internally. Can't put all that into a tiny little trash can.

Perhaps I missed something there but I didn't see any nMP benchmarks included in your reference.

 

[Steve Expat]

I didn't see anything other than Geekbench scores to show this system faster than a 6,1 Mac Pro. While I'm not a fan of the 6,1 Geekbench scores are next to worthless when it comes to real world application performance. If you've got some real world benchmarks then post them. Otherwise this is nothing more than sour grapes.

Don't underestimate the read speed of my m.2 RAID. It's faster than anything I ever imagined two years ago. The boot m.2, the read m.2 RAID are all faster than anything internal or external the nMP can offer. The CPU score is a wash. But together this is a faster and very balanced system. I made sure of that. It's sour grapes that I want to help pro users who are unsure how to move forward?

I'm a former employee of Apple (late 90's), and I want the best for the pro Apple user. I'm here to help, not to be negative.

If I need to I could spend another $500 or so and I could have better graphics than a 6,1 (via a trade up).

My point is this, in 2016 (three years after the release of the 6,1), I see so many posts from Mac users wondering if they should sell their 4,1 or 5,1 cMP and buy an iMac or a 6,1 nMP. The answer, if you need Xeons and more than 32GBs
of RAM, is a resounding no.

For the price of a fully loaded 12-core 6,1 I could build FOUR of these upgraded cMP monsters. For the time being, until we know what Apple's plans are, this is the best route.

It's not that I don't like the 6,1. I actually think it's brilliant. But Apple is not upgrading it annually to keep it competitive.

I'm making a video for my YouTube channel right now that shows performance.

More soon.

 

[linuxcooldude]

http://barefeats.com/hard210.html

There it is, real world benchmark. 5.5x to 6x faster than 6,1 Mac Pro, depending on which test you run.

Therefore, the sour grapes are being tasted by nMP owners, not cMP owners. It's clearly better to have PCIe slots all the way into the year 2017.

I have an Areca RAID card in my 4,1 that excels in disk read/write speeds as well, by about the same margin. Being able to attach sixteen 8TB disks for 128TB of space at over 3300MB/sec is pretty nice, but I'm using it in RAID 6... also a nice option. If one uses the current 10TB disks from Western Digital, and maxes out their ix-24 cards, you have 280TB of disk space at speeds well beyond what's possible in a nMP.

Heck, I could have 40TB in the HDD slots, and various SSDs in other PCIe slots and ODD bays, all internally. Can't put all that into a tiny little trash can.

Not going to put major bucks into an already old +7 year unsupported system. If it just stopped working, that would be sour grapes. Since I've bought my 2009 8 core Mac Pro new, I already got my money out of it.

 

[ITguy2016]

Don't underestimate the read speed of my m.2 RAID. It's faster than anything I ever imagined two years ago. The boot m.2, the read m.2 RAID are all faster than anything internal or external the nMP can offer. The CPU score is a wash. But together this is a faster and very balanced system. I made sure of that. It's sour grapes that I want to help pro users who are unsure how to move forward?

I'm a former employee of Apple (late 90's), and I want the best for the pro Apple user. I'm here to help, not to be negative.

If I need to I could spend another $500 or so and I could have better graphics than a 6,1 (via a trade up).

My point is this, in 2016 (three years after the release of the 6,1), I see so many posts from Mac users wondering if they should sell their 4,1 or 5,1 cMP and buy an iMac or a 6,1 nMP. The answer, if you need Xeons and more than 32GBs
of RAM, is a resounding no.

For the price of a fully loaded 12-core 6,1 I could build FOUR of these upgraded cMP monsters. For the time being, until we know what Apple's plans are, this is the best route.

It's not that I don't like the 6,1. I actually think it's brilliant. But Apple is not upgrading it annually to keep it competitive.

I'm making a video for my YouTube channel right now that shows performance.

More soon.

Yes, it is if you're not going to provide a use case and supporting benchmarks.

Again...I'm not pro 6,1...I'm a 2010 5,1 owner myself. I really wish Apple would continue with the 5,1 form factor. However I'm not about to generically state the cMP is universally faster than the 6,1. While I love the cMP I am realistic the 6,1 has performance advantages over it. To use a lousy benchmark, Geekbench, to claim the 5,1 is universally faster than the 6,1 is naive.

 

[Steve Expat]

Yes, it is if you're not going to provide a use case and supporting benchmarks.

Again...I'm not pro 6,1...I'm a 2010 5,1 owner myself. I really wish Apple would continue with the 5,1 form factor. However I'm not about to generically state the cMP is universally faster than the 6,1. While I love the cMP I am realistic the 6,1 has performance advantages over it. To use a lousy benchmark, Geekbench, to claim the 5,1 is universally faster than the 6,1 is naive.

Again, I remind you it wasn't just the GB score I based my statement on. It's the read and write speeds of the PCIe m.2 RAID, the faster m.2 boot SSD, and the possibility of buying a flashed graphics card that outperforms the 6,1 GPUs.

And, most importantly I have two identical upgraded 4,1s in a host/slave and distributive computing environment, which gives me 24/48 cores (audio over Ethernet), at half the price of a 12-core nMP). Latency is well compensated for thanks to the applications I use which are optimized for multiprocessing and networked machines.

Granted the E5-2697v2's bus speed is slightly faster. It's memory bandwidth is significantly more than the x5690, and Hyperthreading is v2 while the x5690 is v1. However, not all software can directly take advantage of those upgrades, especially Hyperthreading.

That being said, it's my opinion that the Westmere series received the biggest technology jumps of recent Xeons.

Luckily your 5,1 and my 4,1 (which are identical after the upgrade) had the best balence of CPU PCI lanes and PCIe slots for us to take advantage of in 2016.

Again, video coming soon. I don't have a 6,1 to test but I know 6,1 owners in my industry who can perform similar if not exact real world tests and we'll see.

I'm confident that I can keep up with anything the 6,1 can perform and best it in some areas.
[doublepost=1480910951][/doublepost]

Not going to put major bucks into an already old +7 year unsupported system. If it just stopped working, that would be sour grapes. Since I've bought my 2009 8 core Mac Pro new, I already got my money out of it.

How is your system not supported? If it's a 4,1 just apply the 5,1 firmware update and you can run macOS Sierra, install 1333MHz RAM, install Westmere CPUs, and PCIe m.2 SSDs for relatively little money.

Besides, what options from Apple do you currently have?

A MacBook Pro? Only 16GBs of RAM. Nope.

An iMac? Only 32GBs of RAM and a throttled 4-core i7. Nope.

A $10,000 12-core nMP? Ok if you can afford it. But you're stuck with the GPUs which are only useful if you're apps are Open CL based.

Or spend $2000 and have the fastest Mac currently available.

The choice is clear.

 

[wonderspark]

Perhaps I missed something there but I didn't see any nMP benchmarks included in your reference.

http://barefeats.com/hard218.html
Thought that you already knew the fastest benchmark of the nMP. It's much, much slower than the cMP despite the upgrades.

It's like running a race in clown shoes for the cylinder.

 

[Wiltonian]

Not going to put major bucks into an already old +7 year unsupported system. If it just stopped working, that would be sour grapes. Since I've bought my 2009 8 core Mac Pro new, I already got my money out of it.

But isn't that the fundamental advantage of the cMP: if any component 'stops working', anyone with the ability to create a high-power system can just swap out the failed part, keeping everything else, and getting up and running for, at most, a few hundred £/€/$?

Stuart

 

[ITguy2016]

Again, I remind you it wasn't just the GB score I based my statement on. It's the read and write speeds of the PCIe m.2 RAID, the faster m.2 boot SSD, and the possibility of buying a flashed graphics card that outperforms the 6,1 GPUs.

And, most importantly I have two identical upgraded 4,1s in a host/slave and distributive computing environment, which gives me 24/48 cores (audio over Ethernet), at half the price of a 12-core nMP). Latency is well compensated for thanks to the applications I use which are optimized for multiprocessing and networked machines.

Granted the E5-2697v2's bus speed is slightly faster. It's memory bandwidth is significantly more than the x5690, and Hyperthreading is v2 while the x5690 is v1. However, not all software can directly take advantage of those upgrades, especially Hyperthreading.

That being said, it's my opinion that the Westmere series received the biggest technology jumps of recent Xeons.

Luckily your 5,1 and my 4,1 (which are identical after the upgrade) had the best balence of CPU PCI lanes and PCIe slots for us to take advantage of in 2016.

Again, video coming soon. I don't have a 6,1 to test but I know 6,1 owners in my industry who can perform similar if not exact real world tests and we'll see.

I'm confident that I can keep up with anything the 6,1 can perform and best it in some areas.

You initial post had three references to speed:

1. Geekbench
2. Theoretical speed of a four SSD RAID
3. Boot speed

I don't see any reference to applications. The system may very well be faster than the 6,1 MP but what you've presented, at least in the post I responded to, only had Geekbench scores to support that.

My reason for responding about Geekbench scores is we tend to see this regularly. Someone uses GeekBench to show how their system is faster when in real life scenarios it's not. Case in point: There are a lot of MacBook Pros which score much higher on Geekbench compared to my cMP. However when it comes to transcoding video my cMP easily outperforms them. I suspect the cMP is able to maintain its performance lead over a "faster" laptop due to thermal throttling that occurs in the laptop.

Again: Not questioning whether the cMP can compete with the nMP. Just pointing out Geekbench scores aren't going to cut it. Real world applications are what really matter.
[doublepost=1480942206][/doublepost]

http://barefeats.com/hard218.html
Thought that you already knew the fastest benchmark of the nMP. It's much, much slower than the cMP despite the upgrades.

It's like running a race in clown shoes for the cylinder.

You're not attempting to compare the speed of a single SSD to that of a four SSD RAID configuration are you?

 

[linuxcooldude]

How is your system not supported? If it's a 4,1 just apply the 5,1 firmware update and you can run macOS Sierra, install 1333MHz RAM, install Westmere CPUs, and PCIe m.2 SSDs for relatively little money.

The choice is clear.

No Apple care support since the majority are now obsolete/vintage.

The choice isn't so clear for people who have no interest in upgrading or providing their own support if things go wrong.

Some people are more interested in getting work done rather than upgrading or or providing their own technical support.

Upgrading a 2009 Mac Pro can be a tricky procedure for de-lidded processors and have had instances of broken or bent pins. While I'm sure there have been many successful upgrades, its not quite the same as a drop in replacement on a PC.

But isn't that the fundamental advantage of the cMP: if any component 'stops working', anyone with the ability to create a high-power system can just swap out the failed part, keeping everything else, and getting up and running for, at most, a few hundred £/€/$?

The major advantage is initially cost. But if that bad component ends up being the motherboard or processor board your back to square one. Since they are already aging systems many are at the end of life.

 

[wonderspark]

You're not attempting to compare the speed of a single SSD to that of a four SSD RAID configuration are you?

Therein lies the point. How much speed and volume can be had in the tube using RAID? Less. Bottlenecks in the design. One CPU, one internal SSD and thunderbolt limits make any world, real or theoretical, slower.

But you're right. The tube is faster than the box. It's more aerodynamic, so it will hit the sidewalk sooner when tossed out a window.

 

[ITguy2016]

Therein lies the point. How much speed and volume can be had in the tube using RAID? Less. Bottlenecks in the design. One CPU, one internal SSD and thunderbolt limits make any world, real or theoretical, slower.

But you're right. The tube is faster than the box. It's more aerodynamic, so it will hit the sidewalk sooner when tossed out a window.

IOW, yes...you were comparing apple's to oranges. Having said that one should be able to build a RAID across multiple Thunderbolt ports thus matching or exceeding the speed in the example.

 

[wonderspark]

One should, should be able to, but it's not as straightforward as that. I know one major company that spent several digits to build a custom solution to aggregate two thunderbolt channels, and their test benchmarks did NOT exceed the example given. Since you are keen on real world here, we might agree that spending ten times the cost of the Mac itself to achieve less than that of a prior generation Mac isn't a real world solution at all.

 

[ITguy2016]

One should, should be able to, but it's not as straightforward as that. I know one major company that spent several digits to build a custom solution to aggregate two thunderbolt channels, and their test benchmarks did NOT exceed the example given. Since you are keen on real world here, we might agree that spending ten times the cost of the Mac itself to achieve less than that of a prior generation Mac isn't a real world solution at all.

We might if that single company and their custom solution represents the only RAID solution. As they are not we're not in agreement.

 

[h9826790]

IOW, yes...you were comparing apple's to oranges. Having said that one should be able to build a RAID across multiple Thunderbolt ports thus matching or exceeding the speed in the example.

Then how about compare Apple to Apple?

Is there anyone know how the nMP distribute the bandwidth? Each Thunderbolt 2 port and USB 3.0 port has their own dedicated bandwidth?

If yes, then the max possible RAID speed is 6x thunderbolt 2 port + 4x USB 3.0 port + 1 instal SSD.

On the cMP side, 40x PCIe 2.0 lane (from slot 1,2,3) + 6x SATA 2 port + 5x USB 2.0 port + 4x firewire 800

Anyone can work out the answer? I know it's quite meaningless in real world because no one will create a RAID 0 like that. But just want to know the theoretical number for fun

Or a more simple possible real world solution.

nMP - 6x thunderbolt 2 RAID 0 array, assume user connect the monitor via HDMI

cMP - 8x SM951 RAID 0 array from both slot 1 and 2, assume user install the graphic card in slot 3

In this case, which one is faster?

 

[ITguy2016]

Then how about compare Apple to Apple?

Is there anyone know how the nMP distribute the bandwidth? Each Thunderbolt 2 port and USB 3.0 port has their own dedicated bandwidth?

If yes, then the max possible RAID speed is 6x thunderbolt 2 port + 4x USB 3.0 port + 1 instal SSD.

On the cMP side, 40x PCIe 2.0 lane (from slot 1,2,3) + 6x SATA 2 port + 5x USB 2.0 port + 4x firewire 800

Anyone can work out the answer? I know it's quite meaningless in real world because no one will create a RAID 0 like that. But just want to know the theoretical number for fun

Or a more simple possible real world solution.

nMP - 6x thunderbolt 2 RAID 0 array, assume user connect the monitor via HDMI

cMP - 8x SM951 RAID 0 array from both slot 1 and 2, assume user install the graphic card in slot 3

In this case, which one is faster?

The one which allows you to complete your task in the shortest amount of time. That's the problem with all the theoretical discussion...it may not mean anything in the real world.