Drive enclosure performance results - nMP vs iMac

[analog guy]

as most folks who hang around here know, i received my nMP and have been putting it through the paces as i figure out if it will meet my needs (still photography; audio; at least one VM at all times; only casual video use). i've been evaluating it vs a 2013 iMac.

as part of my move from my older Mac Pro to either solution, i decided to commit to a mostly external storage system via thunderbolt. i evaluated a number of enclosures (sonnet echo express iii-d, pegasus2 r4, OWC mercury elite pro dual, promise j4, and more), and i settled on the sonnet as the winner for my setup. it provided faster results in my testing than any other enclosure, and it more than doubled the performance of some (*cough* OWC *cough*).

tonight i did a test where i compared the very same drive setup (2x840EVO1TB on a Tempo Pro in the Sonnet) connected to the iMac, and then a few minutes later connected to the nMP. I figured I'd get the same results -- nothing to report, right?

Wrong.

I'll let the images speak for themselves, but on QB4.0 I was seeing the following for reads/writes for nMP vs iMac:

4k-1024K randoms: 317/339 vs 289/289
4k-1024k sequentials: 391/343 vs 363/305
2-10MB: 801/737 vs 785/645
20-100MB: 828/767 vs 830/678

via Blackmagic (5GB), I was seeing 814/737 on the nMP and 790/640.

the nMP is TB2, but i have not yet upgraded the card in the sonnet enclosure to the new TB2 version! (i will do that next week. i imagine the results may be a bit better.

as a point of reference, the iMac PCIe SSD's (Samsung 512) random reads/writes are 207/372 vs the nMP PCIe SSD's (Samsung 512) 255/419. for larger transfers the iMac hits 781/722 while the nMP hits 1222/1061. those tests were all run when each machine was new.

you can see more detail in the attached screenshots (results for 4k-1024k & 2-10MB in this message; the rest in the next one).

hope this is helpful to some folks.

 

[analog guy]

more results....20-100MB & Blackmagic (5GB).

 

[VirtualRain]

as most folks who hang around here know, i received my nMP and have been putting it through the paces as i figure out if it will meet my needs (still photography; audio; at least one VM at all times; only casual video use). i've been evaluating it vs a 2013 iMac.

as part of my move from my older Mac Pro to either solution, i decided to commit to a mostly external storage system via thunderbolt. i evaluated a number of enclosures (sonnet echo express iii-d, pegasus2 r4, OWC mercury elite pro dual, promise j4, and more), and i settled on the sonnet as the winner for my setup. it provided faster results in my testing than any other enclosure, and it more than doubled the performance of some (*cough* OWC *cough*).

tonight i did a test where i compared the very same drive setup (2x840EVO1TB on a Tempo Pro in the Sonnet) connected to the iMac, and then a few minutes later connected to the nMP. I figured I'd get the same results -- nothing to report, right?

Wrong.

I'll let the images speak for themselves, but on QB4.0 I was seeing the following for reads/writes for nMP vs iMac:

4k-1024K randoms: 317/339 vs 289/289
4k-1024k sequentials: 391/343 vs 363/305
2-10MB: 801/737 vs 785/645
20-100MB: 828/767 vs 830/678

via Blackmagic (5GB), I was seeing 814/737 on the nMP and 790/640.

the nMP is TB2, but i have not yet upgraded the card in the sonnet enclosure to the new TB2 version! (i will do that next week. i imagine the results may be a bit better.

as a point of reference, the iMac PCIe SSD's (Samsung 512) random reads/writes are 207/372 vs the nMP PCIe SSD's (Samsung 512) 255/419. for larger transfers the iMac hits 781/722 while the nMP hits 1222/1061. those tests were all run when each machine was new.

you can see more detail in the attached screenshots (results for 4k-1024k & 2-10MB in this message; the rest in the next one).

hope this is helpful to some folks.

Thanks. This is very interesting. You're getting almost 100MB/s faster writes via the nMP than the iMac. Bizarre. I wonder if the PCIe allocation in the iMac has the TB controller sharing lanes with something else?

Ps... Sad the OWC enclosure performed so poorly. But isn't there another thread on here in the last few days where someone else was complaining about RAID0 performance in an OWC enclosure?

 

[g4cube]

4 lanes for nMP vs. 2 lanes on the iMac, and different Thunderbolt controller.

 

[deconstruct60]

4 lanes for nMP vs. 2 lanes on the iMac, and different Thunderbolt controller.

For the internal PCIe SSDs, yes x4 versus x2. The second half not so much. If attach just a TB v1 external device directly to the Mac Pro 2013 only going to see TB v1 speed out of it. It is a different controller but it is running in the older's ones primary mode.

More likely the gap is in part that the TB devices is on a CPU PCIe bus and the internal PCIe SSD is sharing bandwidth with the TB controller in the iMac on the PCH chipset bus. ( unless Apple hobbled the dGPU in the iMac, the x16 lanes from the CPU are dedicated to the GPU. )

If want to maximize TB and internal PCIe SSD performance you move them away from the same PCIe lane source. Even more so if going to use both at the same time.