Trying to understand the iMac's SSD performance
@danj Do you have some evidences to prove that T2 have similar performance gain as RAID 0? I’d like to know as well.
Is this a good question?
Score
0
519282
1 Answer
Rep: 479.1k
35
329
971
The performance is similar to RAID 0. Unlike a SATA or PCIe SSD which has a controller the controller logic is within the T2 chip as the storage is raw flash which is directly accessed removing a layer.
Was this answer helpful?
Score
0
did the system just turned my comment into a question? I was asking about the bandwidth of x4 before.
Restored comment thread (1/3):
@isesame - Its faster than a normal SATA/AHCI or PCIe/NVMe HDD or SSD RAID! Look over the reviews that have been published.
Normal HDD/SSD’s use a controller on each drive. So the R/W needs to be driven the the correct drive. Here the T2 is the controller and the raw flash is directly accessible to the T2 controller as a flat storage space.
Basically, removing a process step.
Dan - 4 days ago
@danj what’s the difference between a T2 and traditional M.2? It sounds like to me that you can still build similar technology in a single SSD. Also, both iMac Pro and the new MacBook Pro has faster flash chip to begin with.
Xizhi Ma - 4 days ago
@isesame - Very different devices!
An M.2 SSD is no different from a 2.5” SSD device. It has a controller (ARM processor in most) RAM and then the flash chips them selves.
The T2 is the controller with its processor and RAM. Here the flash chips are on independent modules with direct I/O to the controller. Look at Step 8 for the details
Now lets look at a RAID config with any SATA or PCIe drive setup the data move goes into the PCI bus (one or two lanes) to the controller of one of the drives then to the flash. Here the T2 is directly tied to the logic board PCI bus (no buffering) and its thought to be using four lanes. But since it’s a direct connection the CPU does not need to know which drive the data move goes to as the flash is flat to the T2. The best way to visualize this is the difference between an hourglass and a direct tie-in design so there is no bottleneck.
Both the iMac Pro & the newer MacBook Pro 13” & 15” use this technology.
Dan - 4 days ago
@danj Looking at the block diagram, I did noticed that T2 only using PCIe x4, while two flash storages also using x4. I wonder if this is a bottleneck.
https://egpu.io/forums/desktop-computing...
Xizhi Ma - 3 days ago
The PCIe width is x4 to the discrete flash modules but these are mux'ed so the effective width is x8. Keep in mind these are direct feeds so they don't need any driver logic only buffer logic.
As for the feed for the PCH to the T2 logic while the connection appears to be PCIe x4 there is also a high speed direct feed eSPI which is a modern VME bus.
