You can run memtest from a linux stick's bootloader, should be as little ram as possible used by this minimal system.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 1,1-5,1 cMP & nMP {MacPro 4,1 & 5,1} Memory Upgrade Compatibility and FAQ
- Thread starter n8ballz
- WikiPost WikiPost
- Start date
- Sort by reaction score
- Status
You can run memtest from a linux stick's bootloader, should be as little ram as possible used by this minimal system.
I bought Qty (8of) Samsung M393B4G70BM0-YH9 2R*4 @16GB on ebay for US$225 (accepted best offer) plus US$25 shipping to Australia. I'm only using Qty (6 of) to give me 96GB of total Ram, better bandwidth than using all 8 apparently. Works great. There's 2 more lots left if anybody is interested, although the seller has now jacked up the 'buy it now' price to $279.99 from $249.99
https://www.ebay.com/itm/Samsung-12...e=STRK:MEBIDX:IT&_trksid=p2057872.m2749.l2649
I am using the 8gb version of those samsung chips. I have another 32gb laying around, waiting to find a decent deal on a dual tray for my 2009 mac pro so I can have a total of 64gb in the machine.
Would 8 of these sticks work in a Mac Pro 5,1? Would that be considered the optimal configuration?
https://www.ebay.com/itm/Samsung-M3...h=item54698ead27:g:CpQAAOSw~sJcJp3R:rk:1:pf:0
https://www.ebay.com/itm/Samsung-M3...h=item54698ead27:g:CpQAAOSw~sJcJp3R:rk:1:pf:0
Should work, but not considered optimal generally. Unless you really need more than 96GB RAM.
Wow. Fast responses. Thanks guys!
Curious why they wouldn't be considered optimal. In terms of speed? Or are you referring to maxing out every slot? (i.e. I'd be better off with 6 of them).
I mainly use After Effects, Premiere, Resolve etc. I don't believe I need 128 GB.
Curious why they wouldn't be considered optimal. In terms of speed? Or are you referring to maxing out every slot? (i.e. I'd be better off with 6 of them).
I mainly use After Effects, Premiere, Resolve etc. I don't believe I need 128 GB.
I'm using 128GB right now in 5,1. Modules are M393B2G70DB0-YK0. Unsure the exact spec difference from the link you posted, but here's all the specs of the RAM I'm using:
https://forums.macrumors.com/threads/upgrading-a-mac-pro-4-1.2117587/page-2#post-26063061
AE CC 2019 will use more than 96GB RAM at times depending on which plugins are being used, stacked effects, masks, etc. The slight performance drop for more available RAM is worth the tradeoff in my situation. Would really suggest you monitor your own situation. For $50 difference, I'd probably buy 8 modules and test with all 8 then with just 6 to see if it makes a "REAL WORLD" difference instead of a benchmark difference.
I rarely see Premiere or Media Encoder "need" more than 96GB.
If you're not already on fast(er) PCIe boot drive via SATA or NVMe, take a look at that. Adobe CC 2019 start times will speed up quite a bit and may be a better spend of your upgrade dollars.
When you purely measure memory performance, the memory can preform quite a bit better if you can optimise the configuration to fit that triple channel architecture.
That's why quite a few cMP users consider using all 8 slots are not optimum.
However, modern OS nowadays can utilise the excess memory to serve as cache. Therefore, may able to speed up the whole system (even pure memory performance is lower).
Also, in real world, I don't know any software (except benchmarks) will intentionally disable system cache (I mean the real hardware cache here. Not the "cache" above which store data from hard drive). Therefore, the actually memory performance is very negligible in most case.
So, in the end, it still depends on your usage.
If your workflow require load the same data repeatedly from the hard drives. Then go all the way to 128GB usually is a better choice.
But if your workflow only need to read the data once, but very memory bandwidth limiting. Then 96GB should work better.
OK perfect. Thanks for the reply! Just ordered 128 GB of RAM for only $200. Even if I end up only using 6 sticks, it's still a hell of a bargain. Thanks again for all the advice!
I've been thinking about upgrading to either 48GB or 96GB RAM. It'll be ECC and 1333MHz obviously, but I am wondering about a few other parameters ...
1.
A few weeks ago a friend of mine, and an HP Enterprise server technician, suggested that I should go with RDIMMs not UDIMMs if I would go with 96GB to not run the risk of overloading the memory controller of the Mac Pro. When I read up on the differences between RDIMM vs UDIMM I learnt that RDIMMs in introduce some kind of latency into every RAM access cycle though, because of a built in buffer. This worries me as what I do with my computer is sensitive to single core performance and hardware latency specifically.
When I see comparisons between RDIMM and UDIMM I see people comparing RAM in GB/s transfers, not so much RAM access time. Do any of you know how RAM access time is affected by choosing RDIMMs?
If it is, I might as well go with 48GB to be able to go with UDIMMs.
2.
What is the preferred CL of 1333MHz RAM for cMP 5,1? 9? If so, will the machine utilise CL 7 if given?
3.
Is 2x rank the recommended configuration for 16GB modules?
4.
I've seen a mention of RAM with an on board "Apple Specified Thermal Sensor". What is this? Or, how do I look for this on eBay? Is it called something else too, that makes it easier to find among generic server RAM?
1.
A few weeks ago a friend of mine, and an HP Enterprise server technician, suggested that I should go with RDIMMs not UDIMMs if I would go with 96GB to not run the risk of overloading the memory controller of the Mac Pro. When I read up on the differences between RDIMM vs UDIMM I learnt that RDIMMs in introduce some kind of latency into every RAM access cycle though, because of a built in buffer. This worries me as what I do with my computer is sensitive to single core performance and hardware latency specifically.
When I see comparisons between RDIMM and UDIMM I see people comparing RAM in GB/s transfers, not so much RAM access time. Do any of you know how RAM access time is affected by choosing RDIMMs?
If it is, I might as well go with 48GB to be able to go with UDIMMs.
2.
What is the preferred CL of 1333MHz RAM for cMP 5,1? 9? If so, will the machine utilise CL 7 if given?
3.
Is 2x rank the recommended configuration for 16GB modules?
4.
I've seen a mention of RAM with an on board "Apple Specified Thermal Sensor". What is this? Or, how do I look for this on eBay? Is it called something else too, that makes it easier to find among generic server RAM?
Last edited:
I also work on server class stuff for a living. To be quiet blunt the difference isn't enough to quantify. There has been numerous tests done with memory and the difference is marginal. The biggest leap in technology that makes a big difference is the processor architecture changes. There are some boards out there that accept either ddr3 or ddr4 and will run 7th gen processors as an example. The difference between ddr3 and ddr4 with the same processor was so minor it wasn't even worth testing further. Udimm vs RDIMM is basically just registered ram and unregistered. UDIMM is faster if you are talking a single channel, when you go to 2 or 3 channels UDIMM is not faster. The way RDIMM handles parity and correction is better too. It may not actually fix the write issue but the memory knows the error occurred and reports it to the memory controller, UDIMM does not. You can also use x4 with RDIMM, you can only use x8 with UDIMM. What's the difference? Well with x4 it allows it to correct all possible device dram errors.
In short RDIMM x4 is much better. Only scenario UDIMM may shine is in single channel mode.
In short RDIMM x4 is much better. Only scenario UDIMM may shine is in single channel mode.
Hi. Thanks for chiming in.
How much would you say is "To be quiet blunt the difference isn't enough to quantify"? The performance issue I'm having is not a human being judging the performance (like when a human is interacting with a GUI or a virtualised machine on a remote server) but a software failing / stopping when the hardware takes too long to respond.
My machine is struggling to keep up although I've set my software to the most forgiving setting. I'd like to adjust that setting again to make the software feel responsive again - and most importantly; I'd like the machine to keep up and not fail on me.
If I may run some numbers to keep things clear:
- Todays setting amount to the following numbers: 94 accesses per second to its dedicated hardware, including up to 2KB of IO per access, with a delay of <11ms. Or else it fails / stops.
- I need to get as close as for it to handle: 1500 accesses per second to its dedicated hardware, including up to 2KB of IO per access, with a delay of <0,067ms.*
... all while keeping up with the rest of macOS, drivers and third party software.
I'm just asking, it may still well be that UDIMM vs RDIMM won't matter in my case. I know were talking about differences in micro seconds when dealing with memory speeds. But then again, I don't know how many of these micro second delays of access add up each millisecond.
* I may settle with 750 accesses, up to 1024KB of IO at maximum delay of ~0,33 ms if the cMP is to old to keep up with my highest wishes. Hoping to buy me some time before I need to go for a 7,1. (You know how costly maxed out modern Apple hardware is.)
Ah, well ... Because I want 48 or 96GB RAM I need more than a stick per processor. RDIMM might be the only way I can move. I may have to adjust my hopes / settings to what ever that leads to ...
The more I research this topic the more I believe my performance troubles are not as much dependent on the speed of my RAM, rather the CPU. I might just need a CPU with a single core speed of 4000+, GeekBench units, to be able to achieve what I want. So .. maybe specs like RDIMM instead of UDIMM, 2xRANK+ vs 1xRANK and CL9 vs CL7, things I've been unsure about, are perfectly okay and will not be the bottle neck ...
You won't get past ~3000 with a MP5,1, the best single core score that you can get using the two fastest CPUs that work with a MP5,1, X5677 and X5690.
I know tsialex, that's a sad part of this.
The route I was thinking of going this coming week was 2xX5690 + 96GB RAM + RX580 + USB3.1g2. Now I'm not so sure that will help me that much or help me get through 2019/20. It would only give me about a 50% increase of single-core speed going from my 2x2.4GHz to 2x3.46GHz. My multi-core speed is already more than I need though.
It's so ironic. In the room next to this I'm building a gaming machine for a client, using the X470 chipset and a Ryzen 2700X. In this room where I write this and usually do my work, using the machine in the signature, I'm barely even able to boot the latest macOS.
I know, this is the memory topic. I should continue my CPU concerns in the CPU topic ...
The route I was thinking of going this coming week was 2xX5690 + 96GB RAM + RX580 + USB3.1g2. Now I'm not so sure that will help me that much or help me get through 2019/20. It would only give me about a 50% increase of single-core speed going from my 2x2.4GHz to 2x3.46GHz. My multi-core speed is already more than I need though.
It's so ironic. In the room next to this I'm building a gaming machine for a client, using the X470 chipset and a Ryzen 2700X. In this room where I write this and usually do my work, using the machine in the signature, I'm barely even able to boot the latest macOS.
I know, this is the memory topic. I should continue my CPU concerns in the CPU topic ...
I've just ordered Samsung 128GB kit (4 x 32GB) PC3L-10600R 4Rx4 M393B4G70BM0-YH9 modules for my Mac Pro
It will take 2 - 3 weeks to arrive as theyre coming from the USA to Australia
It will take 2 - 3 weeks to arrive as theyre coming from the USA to Australia
Mine are still coming in from China as soon as it arrives i will test them out.
1) I never see any 16GB UDIMM ECC that can work on cMP. You have no option, only RDIMM if you want 96GB.
For 48GB (6x8), it’s your own choice, there should be no noticeable difference.
2) 1333 with CL9, 1066 with CL7, no choice for cMP user.
3) Yes
4) You can ignore this.
[doublepost=1549984854][/doublepost]
RDIMM works better (faster) when have more than one DIMM per channel, not per CPU.
Hi!
I'm looking to get 64GB of ram for a single hexacore cMP.
I could get 4 of these HP sticks for a good price:
HP 672612-081 M393B2G70BH0-CK0Q9, DDR3 16GB 1600 MHz REG ECC 2Rx4
The second part of the serial suggests Samsung modules.
I read the guides and many topics on this forum. Apart from the 1600 MHz, it all looks fine.
Will the downclock to 1333 MHz have any problems/benefits? Should I just look for 1333 MHz RAM?
Update: another option for 4x16GB:
KINGSTON KTH-PL313LV/16G 16GB 2RX4 PC3L-10600R DDR3
Which is 1333 MHz, but low voltage by default.
Thanks!
I'm looking to get 64GB of ram for a single hexacore cMP.
I could get 4 of these HP sticks for a good price:
HP 672612-081 M393B2G70BH0-CK0Q9, DDR3 16GB 1600 MHz REG ECC 2Rx4
The second part of the serial suggests Samsung modules.
I read the guides and many topics on this forum. Apart from the 1600 MHz, it all looks fine.
Will the downclock to 1333 MHz have any problems/benefits? Should I just look for 1333 MHz RAM?
Update: another option for 4x16GB:
KINGSTON KTH-PL313LV/16G 16GB 2RX4 PC3L-10600R DDR3
Which is 1333 MHz, but low voltage by default.
Thanks!
Last edited:
Both should work.
1) 1600MHz CL11 DIMM can down clock to 1333MHz CL9, that's the standard. Shouldn't be an issue.
2) PC3L support BOTH PC3L and PC3, which means on cMP, will automatically run at 1.5V. Again, this is the standard.
Thank you. Very helpful.
What about ECC vs non-ECC? That should make a difference in speed shouldn't it?
Or should one expect crashes and nothing but trouble if going for non-ECC?
(My office is about 30 meters above sea level at roughly 59° latitude - if that matters to anyone familiar with bit flipping caused by cosmic rays.)
Thanks! I would assume chosing either PC3L 1333Mhz or PC3 1600Mhz memory will not be different from getting the normal PC3 1333 Mhz: the first one will run at 1.5V consuming the same power as a PC3, the other will run at 1333 Mhz (in CL9), so practically all three routes yield the same results in the cMP.
Can the low voltage be better silicon to begin with?
Last edited: