Bus speed comparison with Intel iMac

[GrassyAssAmigo]

I'm trying to compare the memory bus of my current 27" iMac i7 2017 with the M1 Max/Pros, but I'm sure I have the specs wrong because the figures are so extreme...

27" iMac 2017 256-bit bus speed = 400 MB/s = 0.4 GB/s
(https://www.bhphotovideo.com/c/product/1380904-REG/apple_z_8c11_bh_27_imac_pro_with.html)

M1 Pro 256-bit bus speed = 204 GB/s
M1 Max 512-bit bus speed = 408 GB/s
(https://www.anandtech.com/show/17024/apple-m1-max-performance-review)

So the M1 Max memory bus is more than 1000x faster than a standard iMac from 4 years ago? Is that correct??

I understand it's partly because of the bus width doubling, and also that M1s have unified RAM/VRAM, so in some ways I'm comparing apples and oranges. Am I confusing memory bus with another bus?

I wondered if someone could help me understand what I'm looking at and correct my maths...

For context: I'm a video editor so the latency and responsiveness of the UI and scrubbing long 4K timelines are tangibly bottlenecked by the bus speed of my iMac.

Thanks

 

[wilberforce]

My understanding:
2400MHz DDR x 64-bit data channel x dual channel
divide by 8 bits/byte
= 38 GB/s

whereas M1 Pro is:
6400MHz RAM x 256bit bus
=6400 x 256/8 = 204,800 = 205 GB/s

(EDIT: corrected first calc above, I was a factor of 2 out.)

 

[leman]

I'm trying to compare the memory bus of my current 27" iMac i7 2017 with the M1 Max/Pros, but I'm sure I have the specs wrong because the figures are so extreme...

27" iMac 2017 256-bit bus speed = 400 MB/s = 0.4 GB/s
(https://www.bhphotovideo.com/c/product/1380904-REG/apple_z_8c11_bh_27_imac_pro_with.html)

M1 Pro 256-bit bus speed = 204 GB/s
M1 Max 512-bit bus speed = 408 GB/s
(https://www.anandtech.com/show/17024/apple-m1-max-performance-review)

So the M1 Max memory bus is more than 1000x faster than a standard iMac from 4 years ago? Is that correct??

I understand it's partly because of the bus width doubling, and also that M1s have unified RAM/VRAM, so in some ways I'm comparing apples and oranges. Am I confusing memory bus with another bus?

I wondered if someone could help me understand what I'm looking at and correct my maths...

For context: I'm a video editor so the latency and responsiveness of the UI and scrubbing long 4K timelines are tangibly bottlenecked by the bus speed of my iMac.

Thanks

There is no simple answer to your question, because there are multiple buses involved and the relationship between them is complex. The 2017 retina iMac uses dual-channel DDR4-2400 for its system RAM, its the bandwidth is close to 40GB/s. The GPU has its own RAM, with the bandwidth of somewhere around 220GB/s. The CPU and the GPU are connected by what is likely 16 lanes of PCIe 3.0, which offers around 16GB/s in each direction.

So yeah, if scrubbing timelines involves GPU fetching the data from the system memory, it will have to go through multiple hoops. But it’s probably not as simple as lack of bus bandwidth - after all, 16x PCIe has more than enough bandwidth to deliver multiple streams of 4K video at 120 frames. I think it’s more that there a lot of moving parts (driver overhead etc.), and many steps that have to be completed until the data is ready for GPU consumption, and all this machinery just can’t keep up when you are jumping through the timeline quickly. Also probably depends on how the app is sending the data to the GPU in the first place.

Apple Silicon makes all of this simpler since there is only one pool of memory, so the process of getting Theodora into the GPU is much simpler.