which earn me advertising fees or commissions.
As an Amazon Associate I earn from qualifying purchases.
2019 Mac Pro: Memory Bandwidth
Related: 2019 iMac 5K, 2019 Mac Pro, 4K and 5K display, bandwidth, computer display, CPU cores, iMac, iMac 5K, Mac Pro, Macs, memory, memory bandwidth, video
MPG tested the $9399 Apple 2019 Mac Pro with 384GB OWC memory vs the top-end $3849 2019 iMac 5K 3.6 GHz Intel Core i9 8-core / Vega 48 with 128GB OWC memory. The memory difference is not a factor in this test.
The support of OWC / MacSales.com and B&H Photo is essential for making these reviews possible. Please order through our links via these companies and/or subscribe to diglloyd publications. Mac wish list • Storage wish list • SSD wish list • Ports wish list
This page uses the mt stress command of diglloydTools MemoryTester to evaluate available real-world memory bandwidth by a real application. Not to be confused with memory or CPU specifications—this is the real deal, what an application actually has available to it.
This test in effect forces the effects to be seen of both thermal throttling (extended intensive CPU usage), and the loss of Intel Turbo Boost (all CPU cores maxed-out). These figures represent what is possible with sustained memory access for 5+ minutes.
Memory bandwidth is especially important for more CPUs and/or fast ones because it can constrain how quickly data can be accessed by each CPU core. For example, the 2019 Mac Pro especially with 16/24/28 CPU cores needs all the bandwidth it can get.
$2398 SAVE $602 = 20.0% OWC 384GB (12 x 32GB) PC23400 DDR4 ECC 2933MHz 288-pin RDIMM Mem… in All Other Categories
$1229 SAVE $271 = 18.0% OWC 192GB (6 x 32GB) PC23400 DDR4 ECC 2933MHz 288-pin RDIMM Memo… in All Other Categories
$670 SAVE $230 = 25.0% OWC 96.0GB (6 x 16GB) PC23400 DDR4 ECC 2933MHz 288-pin RDIMM Mem… in All Other Categories
Does memory bandwidth matter for real world applications?
For memory intensive computations, memory bandwidth can have a measurable effect on performance. But many applications require only modest bandwidth, since a great deal of computation is done relative to memory access. And if disk I/o is involved then memory bandwidth falls by the wayside. So it all depends.
With only 8 CPU cores it is not likely to be an issue and not much with 12 cores. But starting at 16 cores, memory bandwidth is more and more of a factor (when all the cores are going full-tilt).
Results: Memory Bandwidth
The 2019 Mac Pro just stomps the 2019 iMac 5K for memory bandwidth. But note that the per-core memory bandwidth on the 8-core iMac 5K (16 virtual cores) is about 1.5GB.sec, whereas it is 2.5 GB/sec for the Mac Pro. This should not surprise, given 6-channel memory on the Mac Pro, versus 2-channel memory on the 2019 iMac 5K.
The more interesting 'find' here is that the Apple 48GB memory kit has about 1/3 less memory bandwidth than the OWC 6 X 32GB 192GB memory kit! Moreover, mixed sizes not matched up drop things even more (4 X 8 GB + 2 X 32GB). Anyone buying this class of machine would thus be well advised to pay attention to matched sets of memory, and not just any memory.
Why would you pay 2X to 3X as much for same amount of memory? Why would you buy a Mac Pro with overpriced and underperforming Apple 8GB modules?
Why is OWC memory substantially faster than Apple?
One answer might be that the OWC 32GB modules have many more memory chips on each module, and that memory access could be interleaved (parallelized) within the chips on the module (???). It has to be something like that for a 33% faster throughput.
See the pictures below—the Apple 8GB RDIMM modules have about 1/4 the chips on them as do the OWC 32G RDIMM modules.