Comment by munro | Hacker News Reader

Comment by munro | Hacker News Reader

Here's the link to the MacBookPro18,2 OpenCL benchmark:

* M1 Max OpenCL https://browser.geekbench.com/v5/compute/3551790 [60,167 OpenCL Score]

Comparing to my current MacBook Pro (16-inch Late 2019) & my Hetzner AX101 server:

* MacBook Pro (16-inch Late 2019) vs M1 Max - CPU https://browser.geekbench.com/v5/cpu/compare/10496766?baseli... [single 163.6%, multi 188.8%]

* MacBook Pro (16-inch Late 2019) vs M1 Max - OpenCL https://browser.geekbench.com/v5/compute/compare/3551790?bas... [180.8%]

* Hetzner AX101 vs M1 Max - CPU https://browser.geekbench.com/v5/cpu/compare/10496766?baseli... [single 105.0%, multi 86.4%]

* NVIDIA GeForce RTX 2060 vs M1 Max - OpenCL https://browser.geekbench.com/v5/compute/compare/3551790?bas... [80.7%]

* NVIDIA GeForce RTX 3090 vs M1 Max - OpenCL https://browser.geekbench.com/v5/compute/compare/3551790?bas... [29.0%, boo]

I'm surprised the MacBook holds its own against the Hetzner AX101's AMD Ryzen 9 5950X 16-core CPU! The multi-core SQLite performance surprises me, I would think the M1 Max's NVMe is faster than my server's SAMSUNG MZQL23T8HCLS-00A07.

> * NVIDIA GeForce RTX 3090 vs M1 Max - OpenCL https://browser.geekbench.com/v5/compute/compare/3551790?bas... [29.0%]

30% performance for ~15% of power use in laptop form factor? that's not boo, that seems like a clear win for apple.

Performance doesn't go down linearly with power. I don't have that card to try, but maybe it would do even better at 15% of power.

i doubt it'll even initialize.

These cards can run idle (e.g. desktop 2D mode) at about 15W, so it might run just fine.

Actual power draw depends on monitor resolution, number of attached displays, and refresh rate, though.

These cards idle at 15w while m1 max is yielding like half the performance at that wattage.

Sorry, Apple is clearly lightyears ahead. All they need to do is put out a chip with 48 of their gpu cores and they will beat the most expensive top of the line best ever performance gpu on the market… with a laptop-grade SoC.

The moment of truth will be Mac Pro. There they'll have all the freedom to spend as much electricity as possible and push computing to the bounds.

There is a floor below which it will go straight to 0 (ie, nonfunctional).

The M1 Max is well below that floor at max wattage.

That floor is really, really low when you try to optimize for it. But the issue is that the M1 Max is at 5nm TSMC while the 3080 is on 8nm Samsung, so this isn't actually that impressive.

Also, the 3090 costs almost like the whole macbook

Hm, as far as I can tell a M1 Max Macbook pro costs like $3k+ (the config used in the link is like $4k). how do you figure that?

Easy: you're not getting a 3090 at MSRP unless you're lucky.

Best buy has been doing physical drops on a monthly basis. I got one in the last drop @ MSRP, only waited in line for 30 minutes. It's worth checking out if you are looking for a card.

I will say though if you are in a large city this doesn't work as well, the lines are longer.

They've not done these in my area, but indeed I am in a "big city," or at least the suburbs.

the truly amazing thing is that the M1 Max Macbook Pro will have twice the transistor count of a 3090 in its SOC (not counting memory stacks). That's a laptop CPU now.

It also has hexadecimal-channel RAM. 16 channels of DDR5. Instead of putting the SOC on GDDR6 and gimping the CPU side with higher latency, they just stacked in DDR5 channels until they had enough. Absolute meme tier design, Apple just does not give a single fuck about cost.

(note that DDR5 channels are half the width of DDR4 - you get two channels per stick. But, the burst length is longer to compensate (so you get the same amount per burst), and you get higher MT/s. But either way, it's conceptually like octochannel DDR4, "but better", it's a server-class memory configuration and then they stacked it all on the package so you could put it in a laptop.)

We don't know exactly of course, but I wouldn't be surprised, if Apple doesn't even save costs with their own chips vs. buying from Intel/AMD. In any case, they wouldn't be able to reach that compute power in a laptop of that size at all without the new chips.

Just wait until miners will start buying macbooks.

It's probably possible to build a 3090 system with a 750 watt power supply though.

A laptop?

Since we're comparing against M1 Max, it's more fair to use a fully maxed out Macbook Pro 16-inch for comparison:

* MacBook Pro (16-inch Late 2019) vs M1 Max - CPU https://browser.geekbench.com/v5/cpu/compare/10496766?baseli... [single 145.8%, multi 176.3%]

* MacBook Pro (16-inch Late 2019) vs M1 Max - OpenCL https://browser.geekbench.com/v5/compute/compare/3551790?bas... [153.0%]

* MacBook Pro (16-inch Late 2019) vs M1 Max - Metal https://browser.geekbench.com/v5/compute/compare/3557857?bas... [167.2%]

The M1 Max still wins by a large margin, but there's a slightly smaller gap. The CPU difference is negligible but the upgraded Radeon Pro 5600M graphics card narrows the gap a bit more. It's comparable to a Radeon RX 580. The default 5500M is pretty only about 60% as powerful and surprisingly runs hotter despite being slower. (TDP 85W vs 60W it seems)

Apple compared the M1 Max to the 3080 in the Razer Blade 15 Advanced, saying it outperforms the Nvdia’s GPU [0], but Geekbench seems to disagree [1].

[0]: https://www.apple.com/newsroom/2021/10/introducing-m1-pro-an...

[1]: https://browser.geekbench.com/v5/compute/compare/3551790?bas...

I wonder what they were comparing. The Geekbench result is for an OpenCL benchmark, if they were using Metal Compute on the M1 that may explain the different results

Apple never claimed the M1 Max was faster than the 3080, they claimed the M1 Max was almost as fast while using 100W less power and while being dramaticly faster on battery.

The 2060 and 3090 on these compares are the desktop chips, right?

The 3090 is a high end desktop machine. There is no mobile version of the 3090 (https://en.m.wikipedia.org/wiki/List_of_Nvidia_graphics_proc...).

It’s telling that we need comparisons with 300 watt desktop GPUs to make M1 look mediocre.

Thanks - interesting on OpenCL - presumably running on GPU? All that memory opens up some interesting possibilities.

Also I thought Apple was deprecating OpenCL in favour of Metal?

OpenCL and OpenGL have been deprecated in favor of Metal. Geekbench also has a Metal compute benchmark.

Is there a metal benchmark? This is the score I’ve been most interested in.

Nay, there isn't one for the new M1 Max, but FWIW it's pretty comparable, but OpenCL is a bit faster than Metal.

* MacBook Pro (16-inch Late 2019) - Metal https://browser.geekbench.com/v5/compute/3139776 [31,937 metal score]

* MacBook Pro (16-inch Late 2019) - OpenCL https://browser.geekbench.com/v5/compute/3139756 [33,280 OpenCL score]

Huh. I was expecting it to be over 80k. This is the blackmagic pro Vega 56 egpu metal score.

How to compare Metal vs non-mac hardware then?

Anyone know if that “Device Memory 42.7 GB” is fixed or can be increased?

The MacBook Pro M1 Max can be ordered with either 32 GB or 64 GB of unified memory. The geekbench report shows 64 GB of memory (maxed), not sure why only 42.7 GB is usable by OpenCL--so I guess we have to assume that's the max, unless there's some software fix to get it up to 64 GB.

It reports the currently free memory as device memory.

> unless there's some software fix to get it up to 64 GB.

I very much doubt that this would even be possible. The OS and other on-chip devices require dedicated memory as well and need to reserve address space.

You simply cannot max out physical RAM with a single device in a UMA configuration.

Reply on news.ycombinator.com