Menu
Sign In Search Podcasts Charts People & Topics Add Podcast API Blog Pricing

John Siracusa

๐Ÿ‘ค Speaker
11321 total appearances

Appearances Over Time

Podcast Appearances

Accidental Tech Podcast
686: Write Two Letters

They were optimized for moving huge amounts of data into and out of the GPU.

Accidental Tech Podcast
686: Write Two Letters

And how did that data get into VRAM?

Accidental Tech Podcast
686: Write Two Letters

Well, the computer would ship it over, probably out of its regular RAM, and it would say, here, video card, you're going to need these textures and this geometry to do your work.

Accidental Tech Podcast
686: Write Two Letters

It would send it over the bus, PCI bus or whatever, AGP bus, whatever bus was the bus of the day, and it would go into VRAM, and the GPU would read and write to VRAM and do all of its stuff.

Accidental Tech Podcast
686: Write Two Letters

So you had these two pools of memory, regular RAM that the CPU used, and then VRAM that the GPU used.

Accidental Tech Podcast
686: Write Two Letters

Unified memory means not having a separate pool of VRAM and regular RAM.

Accidental Tech Podcast
686: Write Two Letters

Macs and the Apple Silicon age have one pool of RAM that is used by the GPU as it's quote unquote VRAM and by the CPU as it's regular memory.

Accidental Tech Podcast
686: Write Two Letters

And it's any part of the RAM can be used by either one.

Accidental Tech Podcast
686: Write Two Letters

And they both have a really fast, really high bandwidth connection to that RAM.

Accidental Tech Podcast
686: Write Two Letters

That's how it works, right?

Accidental Tech Podcast
686: Write Two Letters

If it was just like, you know, just have a GPU and a CPU and just one pool of RAM, that one pool of RAM would be too slow and too low bandwidth.

Accidental Tech Podcast
686: Write Two Letters

That's why they had VRAM, special VRAM with really high bandwidth to the GPU.

Accidental Tech Podcast
686: Write Two Letters

But what Apple did is they said, we're going to take our RAM and we're going to make it a huge, wide, fast, low latency bus and solder the chips right next to the SoC.

Accidental Tech Podcast
686: Write Two Letters

And now that pool of memory can be used by both.

Accidental Tech Podcast
686: Write Two Letters

And this is one of the reasons why the early days of people saying like,

Accidental Tech Podcast
686: Write Two Letters

oh, it's okay for a Mac, Apple Silicon Mac to have a smaller amount of RAM because it has unified memory and that makes it use less RAM.

Accidental Tech Podcast
686: Write Two Letters

The opposite is true.

Accidental Tech Podcast
686: Write Two Letters

Back in the day, if you had like, you know, one gigabyte of regular RAM and 512 megabytes of VRAM, you had a total of 1.5 gigabytes of RAM in that machine.

Accidental Tech Podcast
686: Write Two Letters

500 gigs for the video card and one or 500 megs of video card and one gig for the ram right if you unified that you'd want it to be combined you say okay well i'm just going to throw away the vram and leaving you with just the one gig now you have less ram so it seems like you have less memory it's like now i have the the gpu fighting with the cpu for who's going to get to use what part of the one single pool of ram and in apple's case that one single pool of ram isn't any bigger and

Accidental Tech Podcast
686: Write Two Letters

So yeah, it seems like you're wasting more memory unless you always have the same amount you have.