Yeah, but it’s probably the engineering design trade off that will likely be adopted. Just populate the DIMM to the max data bus width and it’ll be OK.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Disrupting the desktop market
- Thread starter Bodhitree
- Start date
- Sort by reaction score
According to Wikipedia, a 16 lane PCIE 5 slot is 63 GB/s in each direction. PCIE 6 is twice that at 126 GB/s. An MPX card can use two 16 lane slots so at best if Apple updates to PCIE 6 you get 252 GB/s in each direction with MPX. I'm not sure that is enough bandwidth to keep high-end Mac Pro users happy. On the other hand, if they are clever with their architecture and figure out how to partition memory between unified memory and PCIE/MPX it may be enough. Latency is another question though.
PCIe protocol overhead will probably kill the thruput advantage, especially for random access. Probably ok for bulk transfers. So quite unlikely for memory to sit on PCIe bus.
1. The market has factors other than just processor architecture. If someone can buy an M1 desktop but can’t buy a gpu to compliment their build, then maybe they’ll buy an M1
2. Rosetta seems to be one of the things that Apple really did well.
Looking at how the M1 achieves its speed and efficiency, I highly doubt we’ll see a “traditional” box. I’m concerned about the PCI slots due to many peoples reliance on cards for different functions. It will be interesting to see if Apple takes the easy route (sealed box with no removable parts) or the difficult one.
Last edited by a moderator:
Applications do expand to use more memory when the standard amounts of ram allow them to do so, but 10 GB is rather ridiculous if you ask me. That’s five times the whole address space available on a 32 bit Windows machine. You have to be pretty profligate with memory to use all that.
IMHO, current M1 Macs are not meant for professional content creation use, which will require a lot of memory. That it is performant enough to do it just shows that the M1 is good. Probably too good for it's own good, making folks use it for applications it is not designed to handle.
I kind of disagree? Not so long ago all the big content creation packages ran comfortably on machines with 4 GB of ram, and a big chunk of that went to the OS. It’s feature creep on 64-bit machines with large amounts of ram that allow an application to grab so much memory.
You're not wrong. I still remember using Adobe Photoshop and Aldus Freehand on my PowerMac 6100 with 8 or 16 MB of RAM, haha. The difference between then and now is that the CPU power and media size has grown quite a bit. Chewing thru 4K/8K video (which seems to be the norm now) requires a lot of memory and CPU power, and also pictures are a lot bigger now, compared to only 1MP during the PowerMac 6100 time.
The M1 replaced the i3 and i5 Macs, which (the replaced Intel Macs) would struggle with requirements of media creation now. M1 chews thru it effortlessly. So in a way, this is disruption.
10GB is far from ridiculous, that's why cards like nVidia RTX3090 have 24GB of RAM
RTX 8000 has 48GB of RAM
This is dedicated video memory alone, not taking into account what the host system uses.
I use Davinci Resolve with 6GB video memory, and if fine for 4K footage, but not so with 6K footage.
I've not tried resolve on the M1, but given minimum specs of resolve is 16GB System + 4GB VRAM, well, 10GB all up is far from what I would consider unusual.
RTX 8000 has 48GB of RAM
This is dedicated video memory alone, not taking into account what the host system uses.
I use Davinci Resolve with 6GB video memory, and if fine for 4K footage, but not so with 6K footage.
I've not tried resolve on the M1, but given minimum specs of resolve is 16GB System + 4GB VRAM, well, 10GB all up is far from what I would consider unusual.
https://www.extremetech.com/computi...-but-intel-amd-are-sticking-with-ddr4-for-now
Seems like DDR5 in Octa channel mode could hit 537GB/s in theory. If you're going to be selling a $20,000 - $100,000 Mac Pro, it seems like having an octa channel wouldn't be strange.
Let's not forget that with Apple Silicon, the bandwidth has to feed both the CPU and the GPU. Vega Pro Duo II has 1TB/s of VRAM bandwidth for example. Apple can probably balance it out a bit by offering larger caches, but it still sounds like a tough balancing act.
Or maybe they will do something completely crazy and use 12 memory channels.. with DDR5-7200 (that actually exists in production) that would be close to 700GB/s, which sounds more appropriate.
A very good point. I think with the M1 Apple have committed to a unified memory architecture for all Apple Silicon machines, its going to be interesting to see how well they can scale it.
The whole thing reminds me a little of a modern games console with 8 cores and a fat channel to unified memory.
The whole thing reminds me a little of a modern games console with 8 cores and a fat channel to unified memory.
I remember when I used to be a freelance graphic designer back around 1998, we were always told that the output resolution of images should be at 300 dpi, which means that a 3.5” x 3.5” print image in an illustrated book would only come to 1050x1050 pixels. For today’s macs, working with that is peanuts.
I remember the first time I ever scanned something.. It took minutes to load, and didn't even fit on a floppy!
Expectations have definitely changed over the years.
It's easy to forget those days, not that long ago actually, where you would have to keep flattening your images in photoshop as there were too many layers to work with!
About 10 years ago I was working on a 108MP panorama and remembering how it was so damn painful, and slow.
I tweaked it only last year for a fresh canvas print, and it was like working with, well, any other picture. Nothing special.