According to the die shot pasted at bottom (source: https://twitter.com/highyieldYT/status/1719379808516280716/photo/1 ), the M3 max has 4 TB controllers, so the Ultra would have 8
And yes, the 8x pool does offer more bandwidth. Four of those lanes go to the I/O card, which offers 2 x HDMI and 2 x USB-A, which gives bandwidth beyond what's available from the Studio. But, with 8 TB4 ports on the MP, you probably don't need the HDMI anyways. Besides, it's cleaner to do the comparison without the I/O card. So let's proceed on that basis....
According to
https://softron.zendesk.com/hc/en-u...-Ultra-Limits-and-possibilities-of-PCIe-slots (nice comparative table from that article pasted below), all the non-TB built-in I/O on the MP together uses 2` PCIe lanes (10 lanes – 8 lanes for SSD = 2):
2 x 10 Gb/s ethernet + 2 x SATA6 (2 x 6 Gb/s) + 1 x 5 Gb/s USB-A + Wi-Fi 6E (9.6 Gb/s ?) + BT 5.3 (0.05 Gb/s).
This leaves a total of 32 – 8 (for the SSD) –2 (for the built-in I/O) = 22 open PCIe4 lanes (the same figure is shown in the table below)
Thus we have:
I/O in the Ultra Studio not in MP: HDMI 2.1 + 1 x USB-A (5 Gb/s) + 1 x headphone + 1 x UHS-II SDXC (0.156 Gb/s)
I/O in the MP not in the Ultra Studio: 22 x PCIe4 + 2 x TB4 + 1 x 10 Gb ethernet + 2 x SATA6 (2 x 6 Gb/s)
Subtracting these, we have that the difference is:
22 x PCIe4 + 2 x TB4 + 1 x 10 Gb ethernet + 2 x SATA6 (2 x 6 Gb/s) – HDMI 2.1 – 1 x USB-A (5 Gb/s) – 1 x headphone – 1 x UHS-II SDXC (0.156 Gb/s)
I don't know how to compare the bandwidth from TB4 and HDMI 2.1, since TB4 is 40 Gb/s duplex, while HDMI 2.1 is 48 Gb/s simplex (with a small return channel, for control), but let's coarse-grain it and say :
1 TB4 + 2 x SATA6 (2 x 6 Gb/s) ≈ HDMI 2.1 + 1 x USB-A (5 Gb/s) + 1 x headphone + 1 x UHS-II SDXC (0.156 Gb/s)
Then the additional I/O bandwidth from the MP is about:
22 x PCIe4 + 1 x TB4 + 1 x 10 Gb ethernet ≈ 25 x PCIe4 equivalent.
Here's the table from the softron article. Even though the 2019 Intel MP had many more PCIe lanes, because many of those lanes were used for the TB ports and GPU, and because they were Gen3, the net I/O bandwidth of the AS MP is substantially higher than that of the Intel MP.
In assessing the I/O bandwith of the current Intel chips, I assume we'd also need to subtract the PCIe reserved for their internal GPU and TB ports (as well as the SSD). Intel says their current Sapphire Rapids Xeon chips have 112 PCIe 5.0 lanes. After subtracting the lanes used for the SSD, GPU, and ports from that 112 figure, that chip, and a hypothetical PCIe 5.0 AS Extreme, should be in the same I/O ballpark.
Last edited:
