Hacker News Reader

Comment by norman784 | Hacker News Reader

I know that a lot of people hates apple, but I love to see their hardware and I think they are innovating in certain areas that I'd hope amd/intel start copying, like the unified memory, if they start working in that direction it will change the PC world forever (also they need to add more memory channels to their processors, 2 is just not enough for this feature).

Also the PCB design looks cool, but they are not the only ones, if you buy a mid/premium gaming motherboard they are also beautiful.

View on HN

> unified memory

Please dear god no. First, it already exists (shared memory with GPU). Second, implementing it in the same way as Apple basically requires an SoC.

Yes unified memory already exists it isn't some newfangled thing. That said Apple is the only modern consumer brand doing it for a Laptop/Desktop. It definitely has some downsides, but it also does have some upsides.

Yes, but you need to setup in the bios how much memory you want to dedicate to the gpu, while apples approach afaik doesn't gpu/cpu can access the same memory, so if you for example want to game and the game requires more memory than the cpu you can get as much as ram you have.

I'm on having the cpu/gpu integrated, but more powerful, if apple can do it, then amd/intel can do it too, but I suspect amd will lose their dedicated gpu market, while intel could gain more with this kind of move, but I suspect that intel doesn't care too much about the enthusiast or end users as apple/amd (I mean that apple market are end user and not server/enterprise like intel).

There are bigger problems with that.

Bundling CPU and GPU is not feasible in the present day for high end systems.

There are use cases where you need say HBM2 memory. Or you need 4 outputs (most motherboards have just HDMI+VGA).

It’s easy for Apple to do, because they have 0 skin in the game in gaming or AI or even GPU compute markets. I don’t think it would make sense for AMD to do it.

It would make sense for Intel, to compete in entry level gaming market, but that’s it.

IMHO it’s just not feasible if your product is a general purpose GPU, because there’s a lot more to it than just the processing unit.

The M1 Ultra gets fully half of the performance of a desktop RTX 3090 in both synthetic and real world benchmarks.

That’s extremely impressive. Sure, no one has CUDA except NVidia but it looks like Apples GPU architecture really truly does work quite well at this stage.

I wouldn’t be shocked to see them catch up to the highest performance GPUs in the next 5 years or so. I mean I also would t be shocked if they didn’t, but it seems like a reasonable possibility at this point.

Which is crazy for a SOC, especially one borne from mobile phone architectures.

I also feel like M1 rackmount servers could be amazing if Linux ran on them.

So the m1 ultra is like a 2070S or a little bit slower than a 3060ti? A ryzen 5950 also has a similar amount of cores, and if most of the power efficiency is coming from being a process node ahead due to special tsmc contracts and covid, its not a durable advantage either because it is not due to technical innovation from apple.

The m1 ultra and max basically targeted for video editing as their target market. Same with the m1 max pro with it’s xdr screen to become a great hdr editing laptop. M1 studio similarly is targeted for the video editors that want a mac mini made for them.

You won’t take advantage of the GPU cores, but the M1 is also a fantastic chip for compiling code. The core count and memory bandwidth are delightful.

Your paying a lot for all of that GPU compute with the amount of die space it takes up. Apple's Intel CPU options were also pretty old and low power, so it feels like a much bigger difference if you are stuck on apple platforms to do your work, like iOS.

If apple kept up with x86 CPUs the difference on macOS wouldn't have been as wide. When you compare a Ryzen 5000 series with the same core count, they are fairly comparable in compile times AFAIK.

> they are fairly comparable in compile times AFAIK.

Yeah I can confirm that. I have an 8 core ryzen 5800, and its performance when compiling rust code is almost identical to my 8+2 core macbook pro.

The macbook pulls ahead slightly in single core performance, but I think the ryzen takes longer before thermally throttling - which makes sense given its a desktop workstation.

You’re looking at it from a perspective of “this instant in time”.

I’m assuming that some of the phenomenal, market-beating growth of the Apple Silicon architecture over the past 5-10 years will continue.

Also if Apple merely creates a quad M1 Pro, as some prognosticate, that would likely match the 3090 in performance.

> M1 rackmount servers could be amazing

With only 128 GB RAM max? No.

Unified Memory is not something Apple just introduced and nobody else has.

I know isn't something new (amd announced it iirc back in 2013 with their apu line) but afaik no one implemented the same way as apple, the gpu and cpu are in the same DIE (I'm not sure if this is the word or how is spelled) but they are not a unit, in apple design both can access all the memory and share between without copying or reserve specific memory from the bios for the gpu.

> both can access all the memory and share between without copying or reserve specific memory from the bios for the gpu.

That's pretty much the definition of Unified Memory, i.e. something the other implementations can also do.

Other implementations being mainly mobile Snapdragon SoCs, which don’t directly compete with M1 (yet).

And as already referenced AMD APUs, including the chips for XBox and Playstation. Hence the entire "hope that AMD starts copying" from the original comment doesn't really fit for unified memory.

I do hope these internals and designs "trickle down" into actual builds and products someday. I would love some small beefy desktop workstation similar to the XBox Series X, kinda like a PC take on the old trashcan Mac Pro.

Did you take a look at the intel NUC line? I would probably buy one if I needed a new PC today.

kinda like a PC take on the old trashcan Mac Pro

Several companies are trying that. The Corsair One series is XBox/trashcan Mac sized and plenty powerful. HP has their Z Mini series of Mac Mini sized workstations. Intel sell their NUC and NUC Extereme kits for people who are more DIY inclined.

I'm not an expert on this by any means, but I believe the XBox still uses dedicated memory for the GPU, which is slightly faster than the one reserved for the CPU. The Playstation does use a single block of shared memory.

Invention is different from innovation. The way Apple has the memory in the same silicon as the CPU and GPU all in a laptop and now desktop... has anyone else done that? It looks like an innovation.

Apple doesn't have the memory in the same silicon as CPU and GPU.

They use CoTS LPDDR memory chips running at the same JEDEC timings as everyone else. Yes, they glue these chips to an interposer board like a GPU with HBM, but unlike those GPUs they aren't doing it for any performance requirements. Keeping it on package just makes things more compact so they can shrink the motherboard even more.

> Unified Memory is not something Apple just introduced and nobody else has.

The GP did not say that.

Reply on news.ycombinator.com