Easy 4,1 dual CPU upgrade?

[crjackson2134]

Might sound silly but is the logic board battery any good? I ve had PC boards before that had no idea what they were doing until I changed the battery in them.

Don’t matter. Battery only for clock on cMP. Different technology.

 

[handheldgames]

Been following your thread and progress closely and this was not the finale I was hoping you'd have! Makes me rethink the used X5690s I was eyeing on Ebay.

By the way - which store did you source the cpus from?

I've had great luck with CPU's on ebay. Here is the listing for the x5680's I picked up.
https://www.ebay.com/itm/PAIR-Intel...e=STRK:MEBIDX:IT&_trksid=p2057872.m2749.l2648

 

[tsialex]

Might sound silly but is the logic board battery any good? I ve had PC boards before that had no idea what they were doing until I changed the battery in them.

Mac Pro and mini use BR2032, I never found that battery down here.

CR2032 works fine until the day that you cook it in the Mac Pro/mini.

 

[handheldgames]

A successful boot of 10.9 from an hdd partition. Installing an older OS from High Sierra is blocked.

I forgot the most important rule. Expect the unexpected.

But as always.. A quitter never succeeds. 🙂

 

[tsialex]

A successful boot of 10.9 from an hdd partition. Installing an older OS from High Sierra is blocked.

I forgot the most important rule. Expect the unexpected.

But as always.. A quitter never succeeds. 🙂
View attachment 780291
View attachment 780292

You just have to flash the reconstructed 138.0.0.0.0 then you can boot from a createinstallmedia 10.13.6 USB key. I’ll do it later today.

 

[Matty_TypeR]

A successful boot of 10.9 from an hdd partition. Installing an older OS from High Sierra is blocked.

I forgot the most important rule. Expect the unexpected.

But as always.. A quitter never succeeds. 🙂
View attachment 780291
View attachment 780292

Well done, great success there, your patience has paid off.

 

[crjackson2134]

Awesomeness!

Freaking firmware...

Truly people need to keep it up to date.

 

[handheldgames]

Putting everything back together. I have 2 WiFi connectors on the card and 3 cables. Is the card missing a connector?

 

[tsialex]

Putting everything back together. I have 2 WiFi connectors on the card and 3 cables. Is the card missing a connector?

View attachment 780334

It's normal, BCM4322MC and some Atheros only have two SPX connectors.

 

[handheldgames]

It's normal, BCM4322MC and some Atheros only have two SPX connectors.

Thanks. I hooked up 2 out of 3 connectors.
[doublepost=1536453802][/doublepost]Sharing additional pics from swapping the logic board.

5,1 Logic board installed

CD-Rom Bay removed. Back panel removed. Exposing logic board power connector.

Intake fan popped out of the cpu tray, pulled towards the center, allowing removal of cpu tray

 

[handheldgames]

Intro to the CPU showdown
With the 5,1 logic board install completed and 3 5,1 trays in hand. I’ve started testing a suite of CPU and RAM combinations with Geekbench, identifying performance variances between configurations.

• x5677 - 4 cores 3.46Ghz max boost to 3.6Ghz on 3+ core activity
• w3680 - 6 cores 3.33Ghz - max boost to 3.46Ghz with 3+ core activity
• x5680 - 6 cores 3.33Ghz - max boost to 3.46Ghz with 3+ core activity
• 990x - 6 cores 3.46Ghz - max boost to 3.6Ghz on 3+ core activity

Sampling the multi-Core MacOS over the past week with dual x5677's with 8 cores and dual x5680's with 12 cores, Grand Central Dispatch effectively keeps all available cores busy almost all of the time with asynchronous threads and multi-threading. With this in mind, any CPU installed in the Mac Pro will never hit max turbo boost in MacOS. Irrespective if you use 1 or 2 cpu's, MacOS keeps the workload effectively spread across available cores.

Stacking up the RAM
As noted earlier, I questioned the difference between using ram that's designed to match the Mac Pro memory specification vs A faster/newer part that down clocks to Mac Pro Speeds

• Samsung DDR3L PC3-10600R 16GB ECC x 3 YHQ95
• Samsung DDR3L PC3-10600R 16GB ECC x 6 YHQ98
• Samsung DDR3L PC3-12800R 16GB ECC x 6 YKO

CPU Tray Configurations with referenced Geekbench 4.0 Results

• x5677 x 2 96gb 16x16x16x0. 16x16x16x0
• x5677 x2 48gb 16x16x16 0x0x0x0
• x5677 x 1 48gb 16x16x16x0
• w3680 x1 48gb 16x16x16x0
• 990x x1 48gb 16x16x16x0
• x5680 x 2 96gb 16x16x16x0 16x16x16x0
• x5680 x 2 48gb 16x16x16x0 0x0x0x0

In addition to synthetic results from Geekbench, I’ve amassed a stack of screenshots illustrating how MacOS takes advantage of 8 and 12 core systems that I’ll be sharing in the upcoming week along with what I hope is a real-world translation of the some of the Math behind Geekbench’s numbers. Perhaps touching in the myth of single core clock boost in MacOS.

 

[mindcomputing]

i wasnt sure enough about delidding the X5670 I bought also I planed ahead and bought brass washer to increase the distance and sure enough a thick silicone pad

• eXtreme-Cool 360 4mm Pad 10cm x 10cm just cut two stripes of it could upgrade another 3 dual cpu configuartion cMPs with it EUR 11,70
• Washer M4 = 8,9 x 4,3 x 0,8 mm 40 pieces EUR 4,38
• CPU 2x X5670 EUR 84,90
• Artic Cooling MX-4 EUR 5,38
• HAZET 828-3 Hex Tool EUR 6,31
• TOTAL spend 112,67 EUR (done nov. 2017)

 

[handheldgames]

Dual x5677 vs Dual x5680 Showdown
Over the last week I've been testing with the x5677 and the x5680 on my 4,1 that was upgraded to a 5,1. During that time I've identified a consistent workflow. While most of my tasks don't push the Dual Mac Pro to 100% CPU utilization, the additive nature of multi-tasking on a Mac with 96GB of ram can quickly add up to 100%. While I initially thought that an upgrade from a single w3680 to dual x5677's would suffice. CPU Bandwidth on the Dual x5677 is more limited than I had expected in comparison to what can be accomplished with Dual 5680's. Long story short, the 4 additional cores on the x5680 keep the Mac running 'smoother', allowing more operations to run simultaneously than the same OS running on dual x5677's.

Overview of daily Mac Pro usage
The following lists represents the high-level workflow tasks of a Mobile Enterprise Engineer. While I'm not generally doing all tasks at the same time, there is a lot of jumping between apps to take on tasks, attend meetings, develop code and write documentation.

• Using off the shelf Web Applications in Chrome / Safari
• Messaging / Video Conferencing - Google Hangouts & Webex
• Building Mac & iOS Apps with Xcode
• Running Apps the iOS Simulator
• Building Web Apps with Visual Studio Code
• Running Apache Web Server / Testing in Chrome
• Using 3D visualization Apps in Parallels / windows 10

Left image
Completing similar tasks on the dual x5680(left) and the dual x5677(right). Running a windows VM, along with Chrome, and Video Conferencing the x5680 has more 'available bandwidth' for taking on additional tasks.

Right image
Compiling projects with Xcode, quickly pushes all available CPU's to the max. Taking advantage of the additional bandwidth across all cores. x5680(right) and the dual x5677(left).



Initial Observations
While I thought the x5677 would hold it's own in a workflow that I initially thought was not that multi-threaded. It turns out every one of the Apps I run, except for some web operations, leverages multi-threading and asynchronous operations across as many cores as possible. Basically, MacOS easily scales to as many cores as you can throw at it.

Without compiling code, the dual x5677's cores get filled up to 2/3 utilization quickly.


So is the dual x5680 worth the $50 increase in cost compared to a dual x5677? In my usage pattern, apparently so. Even though both tray's have 96GB of RAM, my inclination is to grab the dual x5680 tray.

Unless there is something I missed? I'll be using the stopwatch for some definitive testing on idle and busy cores.

 

[w1z]

So is the dual x5680 worth the $50 increase in cost compared to a dual x5677? In my usage pattern, apparently so. Even though both tray's have 96GB of RAM, my inclination is to grab the dual x5680 tray.

Unless there is something I missed? I'll be using the stopwatch for some definitive testing on idle and busy cores.

Thanks for sharing your results with us.

I am in the same boat and am planning on upgrading the cpus, ram, usb and bluetooth/wifi card over the next couple of weeks. My new enclosure is also arriving tomorrow and I already have the cleaning supplies ie. isopropyl alcohol 99.9%, anti-static brushes and IC/Fujipoly thermal pads and thermal grizzly compound on hand.

Looking at the benchmarks and results you shared, I am surprised to see the X5680 scoring higher than the X5690 under the summary scores including other tests even though it should perform better on paper and as per the specs sheet.

What's the reason for this?

Also, the 5,1 I recently bought came with 48GB of OWC 1333 memory. I am torn between selling them on ebay and buying 1 lot of 96GB samsung / micron CL9 10600R memory or buying 48GB of OWC memory for $260 to complete my 96GB memory upgrade which comes with lifetime warranty.

I was also looking at the 16GB PC3-1600R rated sticks but could not find any with confirmed CL9 speeds.

As for being able to read memory speeds, specs, serials and other granular SPD details, you can use TechTool Pro. The option is available under memory tests.

Thank you.

Edited: TechTool Pro only displays granular SPD data on pre-4,1/5,1 systems.

 

[handheldgames]

Using the dual x5677's at home... Importing content from an ipho

Thanks for sharing your results with us.

I am in the same boat and am planning on upgrading the cpus, ram, usb and bluetooth/wifi card over the next couple of weeks. My new enclosure is also arriving tomorrow and I already have the cleaning supplies ie. isopropyl alcohol 99.9%, anti-static brushes and IC/Fujipoly thermal pads and thermal grizzly compound on hand.

Looking at the benchmarks and results you shared, I am surprised to see the X5680 scoring higher than the X5690 under the summary scores including other tests even though it should perform better on paper and as per the specs sheet.

What's the reason for this?

Also, the 5,1 I recently bought came with 48GB of OWC 1333 memory. I am torn between selling them on ebay and buying 1 lot of 96GB samsung / micron CL9 10600R memory or buying 48GB of OWC memory for $260 to complete my 96GB memory upgrade which comes with lifetime warranty.

I was also looking at the 16GB PC3-1600R rated sticks but could not find any with confirmed CL9 speeds.

As for being able to read memory speeds, specs, serials and other granular SPD details, you can use TechTool Pro. The option is available under memory tests.

Thank you.

Edited: TechTool Pro only displays granular SPD data on pre-4,1/5,1 systems.

The benchmarks scores achieved with the Dual x5680 is directly attributed to the Samsung DDR3L PC3-12800R 16GB ECC YKO DIMMS that I'm using. I'm sure if I picked up a pair of x5690's, I would hit the top of the established benchmarks.
Personally, I would expect that any set of Micron / Hynix would perform slower, with relation to Geekbench Numbers, than the suggested Samsung part. For the naysayers... Please... Prove me wrong.

Be sure to buy all 6 sticks from one vendor and do not mix modules. I.E. All should be one type like - Samsung DDR3L PC3-12800R 16GB ECC YKO. Don't mix YKO with YKH... etc. There will be a hit to performance.