not enough space to install big sur, with 90GB of free space available

[stuartrozier]

Hello,

i was wondering if anyone has had this issue in regards to space for the installation,

i have 90GB of free space on my HD but big sur is still stating its needs an additional 16GB, is that a a glitch ?

im not sure if there's a work around this, ive moved the installion file to an external drive thinking that might free up some, but it shouldn't surely need over 90GB just install,

the install utility is also saying i only have 30GB free, which isn't the correct amount, im so confused, ive attached a pic

or does it ?

many thanks in advance

 

[Nimoy]

Same problem for me.

 

[casperes1996]

While macOS doesn't show it, I guess maybe it's purgeable space from APFS snapshots; Try and run
diskutil ap listSnapshots

 

[stuartrozier]

this helped me guys, deleting snapshots on the comp itself, which i didn't even know existed, ive still got time machine backs on external drive, but didn't know there was snapshots being saved on the main system itself, its a bit long an tedious, but after i deleted about 12 snapshots, it finally made room, hope this helps

https://www.reddit.com/r/MacOS/comments/c24cld [automerge]1596759986[/automerge]

Same problem for me.

this helped me guys, deleting snapshots on the comp itself, which i didn't even know existed, ive still got time machine backs on external drive, but didn't know there was snapshots being saved on the main system itself, its a bit long an tedious, but after i deleted about 12 snapshots, it finally made room, hope this helps

https://www.reddit.com/r/MacOS/comments/c24cld

 

[Nimoy]

Thanks that helped a lot. But still need to free up 6gb to install, even though my Mac has 50gb free!

 

[casperes1996]

.

re read the msg you quoted and you will find your answer

.

Sorry, my bad.

So, in the iMacs here it’s worth noting that there’s more of a difference between the GPUs than just the VRAM. You can’t get the 5700 XT without 16GB. And the speed improvement in the GPU is worth more IMO than the extra VRAM. But it depends what you do.

Games, at the moment, generally won’t ever push that far VRAM wise. If you’re doing a complex blender render or scientific compute you might, but for games and even a lot of prosumer work you’ll never use that much VRAM. - But it’s still a faster GPU than what you can otherwise get in the iMac.

You need at least a 2:1 ratio between system RAM:VRAM. Because the information in the VRAM must also reside into the System RAM. So if you have 16 Gb 5700XT you need at least 32 GB system RAM. Keep the 2x4 GB standard modules and add 2x16 GB of Crucial/Kingston/Samsung RAM so you will have a total of 40GB which will be more than enough.

That’s not strictly speaking true. It depends how you code it. It’s true that you often have replicated data but it’s actually very rare that literally everything will exist in two places and often it will not actively be using the system memory copy for most of the time, so it wouldn’t matter so much if that bit got hit by compression or even partially swapped to disk. It would only rarely need to blitz the data (flush VRAM data back to system memory) and use it on the CPU; at least depending on what work you do.
For example in a video editing context the CPU could prepare a render buffer, blitz it to VRAM and only have the GPU blitz back rendered frames once it’s rendered all of it.

(Source: I’ve done a little Metal programming )

 

[pldelisle]

.
.

Sorry, my bad.

So, in the iMacs here it’s worth noting that there’s more of a difference between the GPUs than just the VRAM. You can’t get the 5700 XT without 16GB. And the speed improvement in the GPU is worth more IMO than the extra VRAM. But it depends what you do.

Games, at the moment, generally won’t ever push that far VRAM wise. If you’re doing a complex blender render or scientific compute you might, but for games and even a lot of prosumer work you’ll never use that much VRAM. - But it’s still a faster GPU than what you can otherwise get in the iMac.



That’s not strictly speaking true. It depends how you code it. It’s true that you often have replicated data but it’s actually very rare that literally everything will exist in two places and often it will not actively be using the system memory copy for most of the time, so it wouldn’t matter so much if that bit got hit by compression or even partially swapped to disk. It would only rarely need to blitz the data (flush VRAM data back to system memory) and use it on the CPU; at least depending on what work you do.
For example in a video editing context the CPU could prepare a render buffer, blitz it to VRAM and only have the GPU blitz back rendered frames once it’s rendered all of it.

(Source: I’ve done a little Metal programming )

True. It’s a very broad rule of thumb.

My domain is more machine learning. In ML, the model reside first in the system RAM, and is then sent in the GPU (the weights) once the initialization is done. After that, only the batches are copied from system RAM to the GPU. There’s always pointers between the allocated system RAM for data batches and GPU VRAM. Video rendering might work a bit differently. That’s why I usually use 2:1 ratio. It is sufficient for 100% use case (or almost).

 

[casperes1996]

True. It’s a very broad rule of thumb.

My domain is more machine learning. In ML, the model reside first in the system RAM, and is then sent in the GPU (the weights) once the initialization is done. After that, only the batches are copied from system RAM to the GPU. There’s always pointers between the allocated system RAM for data batches and GPU VRAM. Video rendering might work a bit differently. That’s why I usually use 2:1 ratio. It is sufficient for 100% use case (or almost).

Sure that makes a lot of sense. And it is often good advice as well