Notifications
Clear all
SoftRAID Volumes
6
Posts
2
Users
0
Reactions
3,289
Views
Topic starter
I have a new Mac Pro with the following setup:
Amfeltec Squid Carrier Card 6 NVME
4 WD BLACK SN750 1tb NVME's
2 Barracuda 1tb NVME's
Softraid Raid 0
All NVME's are advertised to run at 3000mb/s+ and the Amfeltec card is designed for and uses the 16x PCI-E slot.
When I created a 6 disk RAID 0 with Softraid I was generally getting 6000-7000mb/s R/W which was less than I expected. I considered that the Barracuda's were bottlenecking me so I tested each blade individually and it seems to be the case as they were testing erratically at 500-2000mb/s each. The WD Black's consistently tested at 2500-3000mb/s each which I found acceptable.
1. If I understand the math here shouldn't the 4 WD Black blades alone in RAID 0 be testing at 8-10000mb/s? I removed the Barracuda's and setup only the 4 WD Blacks in RAID 0 and I get roughly the same speed as with all 6.
2. Do the blades individually suffer a performance loss when in RAID 0 and if so does that explain the loss of performance overall?
The card has been tested at 12-14000mb/s using the Samsung 970 Pro blades(According to Amfeltec) which I believe are considered top of the line nvme's. I understand there will be some overhead that I have to account for but certainly I should be getting better speeds? And If I were to replace the Barracuda's with WD Blacks and setup 6 of them in RAID 0 then certainly I should be able to max out the card's performance, correct(Assuming the blades individually are performing as expected)?
What am I missing?
Posted : 28/04/2020 8:11 am
Go back to Amfeltec. 12GBs is not possible.
They may be getting 12-14GB using a PC, but not a Mac pro. The Mac Pro bottlenecks at about 7GB/s
I proved this by using 4 Accelsior cards, which "theoretically" could get up to 25GB/s, in a test to try to "peg the meter" and find the mac throughput on the Mac Pro.
7GB/s is the max you can transfer data on the Mac Pro. I have confirmed this with other brands of NVMe cards also.
Posted : 28/04/2020 11:41 am
Topic starter
Go back to Amfeltec. 12GBs is not possible.
They may be getting 12-14GB using a PC, but not a Mac pro. The Mac Pro bottlenecks at about 7GB/sI proved this by using 4 Accelsior cards, which "theoretically" could get up to 25GB/s, in a test to try to "peg the meter" and find the mac throughput on the Mac Pro.
7GB/s is the max you can transfer data on the Mac Pro. I have confirmed this with other brands of NVMe cards also.
Argh! That’s terrible! With the four accelsior cards your total throughput is maxed at 7 gb/s?
Posted : 28/04/2020 5:17 pm
Topic starter
Go back to Amfeltec. 12GBs is not possible.
They may be getting 12-14GB using a PC, but not a Mac pro. The Mac Pro bottlenecks at about 7GB/sI proved this by using 4 Accelsior cards, which "theoretically" could get up to 25GB/s, in a test to try to "peg the meter" and find the mac throughput on the Mac Pro.
7GB/s is the max you can transfer data on the Mac Pro. I have confirmed this with other brands of NVMe cards also.
Can you take a look at here and let me know what you think? It appears in their test results that they can reach 12gb/s using the Samsung 970 Pro blades.
Posted : 28/04/2020 5:30 pm
Topic starter
Go back to Amfeltec. 12GBs is not possible.
They may be getting 12-14GB using a PC, but not a Mac pro. The Mac Pro bottlenecks at about 7GB/sI proved this by using 4 Accelsior cards, which "theoretically" could get up to 25GB/s, in a test to try to "peg the meter" and find the mac throughput on the Mac Pro.
7GB/s is the max you can transfer data on the Mac Pro. I have confirmed this with other brands of NVMe cards also.
I dug into this and found that no matter what NVME's in any configuration I create I cannot get passed 7000mb/s, as you suggested(I actually have gotten 7500mb/s to be technical). But there is something definitely throttling the bandwidth and I'm pretty disappointed by it. I'd be curious to hear from Apple why that is.
Anyhoo, I checked back with Amfeltec and noticed that they used the speed test app called ATTO. As does the tester in that link I posted before. So I went ahead and downloaded ATTO and to my surprise I was getting roughly 10000mb/s with my config. Hmmm, so I tested again with Blackmagic and AJA but was still showing the same bottlenecked speeds. Tested again with ATTO and got 10000mb/s.
ATTO tells me what I expected the speeds to be but the others do not. I suspected ATTO was being generous with its test results but I tested another config that I KNOW the max speed and ATTO checked out correctly. So I'm not sure which is telling the truth and which is lying. I spent too much money on this setup to not get my expected speeds.
Argggh!
Posted : 29/04/2020 5:46 am
ATTO's test is indeed generous. We do not use it, as it is not "real world". AJA is a much better indicator of performance, as it is writing to the file system.
I will look into the Rob Morgan (barefeets) article and post back
Posted : 29/04/2020 1:38 pm
