Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Why stop at that, you could run 6 4090x. Good look maintaining and powering that thing.
Other than physical size, shouldn't be much of a problem. We have several small machines with 2x and 4x GPU configuration and they work just fine. Larger machines are usually RTX 5000/6000 with single or multi GPU configuration as well. The rest (non workstation) is in our data center running in the cluster (A100, Hx00, Bx00) for the heavy lifting. Still wondering how you end up with the 50k equivalent?


And then there's the AMD Ryzen AI Max+ 395 which goes up to 128GB of unified memory. How does its iGPU stack up against Apple's?
No experience with the Ryzen. I'm hesitant to invest in new platforms. I thought about getting a few Tenstorrent machines a while ago, but those systems are never about hardware alone. It's also software. Everyone doing research not limited to very specific things is heavily invested in the Nvidia software ecosystem. And while those Tenstorrent cards are nice, it's just painful to go somewhere else (same for Ryzen). We bought Mac Studios for AI experiments (lots of cheap memory) and it backfired. Most of the machines are now running CI pipelines and the rest is used for student experiments. Our cluster with a few hundred Pis is getting more use. While most of the staff are running MacBooks as personal machines for daily work, they're complimented by Workstations with Nvidia GPUs under the desk and the rest are servers.

While 128GB or 512GB is nice to have, Apple needs to up the game with the next generation. M5 needs 2TB. If they can offer that for 20k to 30k, that would allow everyone to have a desktop machine and run it with something like Kimi K2 locally at the cost of additional overhead to maintain compatibility. Of course it's not a solution for everyone. As soon as compute is involved, there's no way around Nvidia.
 
I don’t play games, and don’t spend any money on gaming.

you are being incredibly disingenuous

you said
Doesn’t matter, it’s not all about speed. Memory matters, and it is binary.

to which I merely pointed it that gaming is an area where your claim doesn't necessarily apply. just because you don't care about gaming doesn't make it not true

my point still stands, whether it matters or not absolutely does depend on what you are doing
 
you are being incredibly disingenuous

you said


to which I merely pointed it that gaming is an area where your claim doesn't necessarily apply. just because you don't care about gaming doesn't make it not true

my point still stands, whether it matters or not absolutely does depend on what you are doing
Why edit and quote out of context. it was reply to this... We still don’t know what you are doing so there’s no way to answer that. ---> It doesn't matter what I do, for me its memory. not speed.

And yes. for me it doesn't, I don't waste money on gaming.
 
Last edited:
I was contemplating both, but I'm already up to my armpits in hardware.

The Duo is(are) about the same price as the RTX Pro 6000

Very exciting!
Yes. I am in no rush, but may be I can get Spark as an option to run linux workstation. But ultra with 512 GB does look good.
It does make your concerns irrelevant. And the person is not “wrong” for caring about GPU use outside of games.
Yep, funny he had to pick and choose and quote reply out of context.
 
that's fair, so what would be a good way to go about comparing apple silicon to Nvidia gpu?
..
Find a job that you need to do that needs a GPU and do it on both computers. Ideally, each software woiuld be otimized for the system it runs on.

Things you might try, rotate a 3D CAD model that has a few thousand parts, Render The part using ray tracing for 100 iterations. Then do a forward pass through a large neural network. And so on. Just do things that users do and see which computer makes the job faster.

What you will find is that the Mac is best at some tasks and the gamer pc better at others
 
Find a job that you need to do that needs a GPU and do it on both computers. Ideally, each software woiuld be otimized for the system it runs

Exaxtly. I’ve already done this for several game between m4 pro and amd 6800, both in macOS

In each case the 6800 massively outperforms the m4 pro
 
Last edited:
It does make your concerns irrelevant. And the person is not “wrong” for caring about GPU use outside of games.

I’m not saying they are wrong for caring about gpu use outside of games

I am saying that they are wrong to claim that no matter the application, speed doesn’t matter unless you have huge amounts of vram

That was their claim and it is plainly wrong


Edit: it appears, as is often the case, that it was their imprecise language that led to misunderstanding their point
 
Last edited:
Yes. I am in no rush, but may be I can get Spark as an option to run linux workstation. But ultra with 512 GB does look good.

Yep, funny he had to pick and choose and quote reply out of context.
In your use case memory capacity is more important then bandwidth or GPU speed? It looks like all the PC SoCs are limited to 128GB, which doesn't seem like enough.
 
What use is speed if Nvidia GPU doesn’t have enough memory, it will run out of memory and errors out. 24-32 GB GPU isn’t much, the data sets these days are much larger. Sure I can spend $$$ on cloud or just wait 20 more minute for my M4 max to finish. Either you have enough GPU memory or you don’t.
Thanks for the response. Agree that for some algorithms I suppose you need to have enough GPU memory for the whole dataset. However, batch processing is certainly a thing in a lot of cases.
 
Thanks for the response. Agree that for some algorithms I suppose you need to have enough GPU memory for the whole dataset. However, batch processing is certainly a thing in a lot of cases.
Context is more important in some use cases, and can make huge impact. Do I consider a small subset of data or larger data context when analyzing.
 
You didn’t say that

They’ve been pretty adamant that they were talking about their own use case. Not sure why you’re determined to lie about someone else’s comments in this thread even though you can just go back and look yourself. Very weird.
 
They’ve been pretty adamant that they were talking about their own use case. Not sure why you’re determined to lie about someone else’s comments in this thread even though you can just go back and look yourself. Very weird.

there's no lying. it was based on the words that they typed.

I asked what they were doing

they said "doesn't matter"

that I took to mean it didn't matter what they were doing, ie: there statement would apply to any use case

a misunderstanding on my part? sure.

a lie? nope
 
Last edited:
Yes. I am in no rush, but may be I can get Spark as an option to run linux workstation. But ultra with 512 GB does look good.

Did you secure an initial Reservation?

My Duo Reservation was verified (last week); and--as long as I pony-up the $--it will come.

I'm hoping that the next Ultra will come with 1TB+ (tongue-in-cheek: AAPL has shown a tendency to 21.8X everything they do, so this is a definite possibility).


Unfortunately, I have always associated 'MediaTek' with mediocrity.

I tend to not lavishly praise my $ on stuff that will generate regret. My M2 Max sufficiently carries my Day . . . I'd rather opt for usefulness, as I just don't have the time to chase spec. metrics.
 
It looks like all the PC SoCs are limited to 128GB, which doesn't seem like enough.

Not to deprecate your assertion, but many of the recent AM5 AmATX/ATX/xATX systems are 2DPC (two-channel memory controller sub-divided into four slots).

Most of the validated specs for AM5 state "192GB-max" RAM.

In my recent build (using a Supermicro Epyc board (H12SAE-MF)), I have proved that it runs extremely well populated with two Crucial 64GB DIMMS @ DDR5-5600. I would probably (I haven't tried evey board!) achieve the same results if I were to use two 8/16/32/48/64GB DIMMS.

If I populated the board with four 8/16/32/48/64GB DIMMS, the memory would run @ DDR5-3600.

128GB is a lot ;)
 
In your use case memory capacity is more important then bandwidth or GPU speed? It looks like all the PC SoCs are limited to 128GB, which doesn't seem like enough.

We just did a build with an MSI MAG Tomahawk X870E. This is a midrange motherboard and supports 256 GB of RAM. This motherboard gives you PCIe Gen 5 so you can make the most of video cards. Previous system had a 4060 on a Gen 3 board and moving it to this board improved video card performance by 10%. The 4060 is Gen 4 so upgrading to a Gen 5 video card like the 5060 would provide benefits just due to the increased bandwidth.

It supports two NVMe Gen 5 SSDs as well if you need 14.9 GBps read/write speeds.

If you need more RAM, you can just go with Threadripper chips where the motherboards typically support up to 2 TB.
 
We just did a build with an MSI MAG Tomahawk X870E. This is a midrange motherboard and supports 256 GB of RAM. This motherboard gives you PCIe Gen 5 so you can make the most of video cards. Previous system had a 4060 on a Gen 3 board and moving it to this board improved video card performance by 10%. The 4060 is Gen 4 so upgrading to a Gen 5 video card like the 5060 would provide benefits just due to the increased bandwidth.

It supports two NVMe Gen 5 SSDs as well if you need 14.9 GBps read/write speeds.

If you need more RAM, you can just go with Threadripper chips where the motherboards typically support up to 2 TB.
I think the point is to have unified memory. But this thread is all over the place with people jumping back and forth about setups, so it's hard to tell.
Yep. I can go much higher on RAM with threadripper or AMD EPYC chipset, but problem is it can’t be used with GPU. I have no appetite to spend money. if RAM can’t be used for my GPU tasks.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.