Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Benchmarking the Apple M1 Max

tlkh.dev

Post reply on HN
Print view View original

191–200 of 217 posts

Re: Benchmarking the Apple M1 Max

#191

Post by vlovich123 »

opencl wrote: post #121
ls612 wrote: post #109

Earlier quoted context omitted.

The rumors are that the M1 supersized version they are putting in the new Mac Pro is 20 CPU cores and 128 GPU cores, which would place it at a little under twice the performance of a 3090. Not saying that Nvidia can't catch up with a hypothetical 4090 next year but it'll be a tall order.

The M1 Max 32 core GPU is ~1/8 the performance of the 3090. 4x the core count should put it half of 3090 performance, not double.

If they improve the architecture at the same time, perhaps feasible?

Re: Benchmarking the Apple M1 Max

#192

Post by GeekyBear »

rgbrenner wrote: post #181
nwienert wrote: post #169

Earlier quoted context omitted.

Aren’t you assuming they won’t scale the memory bandwidth?

Yes, that's the way that works. To scale memory bandwidth, you would need a new chip. It wouldn't be an M1 Max anymore. If rumors of them using multiple m1 maxs is true, then they have 400mb/sec of memory bandwidth.

> they have 400mb/sec of memory bandwidth

Per chiplet.

Re: Benchmarking the Apple M1 Max

#193

Post by protoax »

jessriedel wrote: post #114

> We already know that the M1 Max CPU should have really strong matrix multiplication performance due to Apple's "hidden"/undocumented AMX co-processor embedded in the CPU complex, and that it is leveraged when you use Apple's Accelerate framework Does this hold for the M1 Pro?

I see no reason why not considering the core CPU component of the SOCs are very similar.

Re: Benchmarking the Apple M1 Max

#194

Post by grishka »

Behemoth66 wrote: post #164
grishka wrote: post #78

Earlier quoted context omitted.

BTC mining isn't memory-bound, it's compute-bound. The more sha256 hashes you can compute per second, the better. And I'm highly doubtful that any general-purpose hardware at all could even begin to compete with mining ASICs.

Would you or someone happen to know/guess the hash rate of the M1 Max?

I don't think there's a definite answer to this. There's no existing bitcoin miner software optimized for Apple hardware. So, do you use the CPU? The GPU? Both? If you do use the GPU, which API do you use to program it?

Re: Benchmarking the Apple M1 Max

#195

Post by ksec »

Thaxll wrote: post #37
PragmaticPulp wrote: post #3

Great detailed benchmarking. This mirrors my experience with my M1 Max: Absolutely amazing battery life and performance in a laptop. I’m thrilled to have it. Huge step up from last gen Apple laptops. But at the same time, it feels like some of the rhetoric around the performance claims got a little out of hand in the wake of the launch. It’s fast, but it’s not actually crushing my AMD/nVidia desktop like a lot of new…

I don't understand why people keep talking about power performance per watt, do we know for a fact that a 90w Apple GPU would perform 4x faster if it was running at 500w? Does the performance scales linearly? The other thing is price, the m1 max cost $5k here. You know what kind of PC you can get for that price? A 3080 MSRP is $700 and beat any mac GPU right ( and that's a GPU from last year ). 5900x cost $480.

>do we know for a fact that a 90w Apple GPU would perform 4x faster if it was running at 500w? Does the performance scales linearly?

The answer is yes. Scaling with more cores running at the same frequency giving unlimited bandwidth. Yes it is pretty much linear for GPU.

I guess that is why most of the discussions is so tiring. Because this fact is not known.

Re: Benchmarking the Apple M1 Max

#196

Post by ksec »

JohnBooty wrote: post #128
tambourine_man wrote: post #88

Earlier quoted context omitted.

> Does the performance scales linearly? On GPUs, pretty much. It’s even called embarrassingly parallel for this reason.

At some point you hit bandwidth limitations, right? Unless each core has local, dedicated memory they will need to access the GPU's shared memory for various things right?

>At some point you hit bandwidth limitations

Yes, with HBM3 you could go up to 5TB/s and in the future 6.4TB/s. To give a comparison a 3980 RTX Ti has ~900 GB/s with GDDR6x.

But just with everything we hit TDP or cost ceiling well before any technical barriers.

Re: Benchmarking the Apple M1 Max

#198

Post by newaccount74 »

setpatchaddress wrote: post #84
newaccount74 wrote: post #25

Earlier quoted context omitted.

> XPC is not asynchronous It's been some time since I dug into the internals of XPC, but my assumption was that the underlying protocol is asynchronous, and if you do sync calls the wrappers just do the waiting for you. The problem is that it has a tendency to get stuck in some rare cases, where services just don't reply for some reason. Then the sync calls are the worst -- the UI of the app is completely frozen and…

XPC sync is not in fact async underneath in recent macOS versions. It’s a severe performance pessimization to use async in many cases, because sync propagates thread priority and async often can’t. You don’t seem to have a full grip on the reasons for the intermittent hangs you’re experiencing. Can I suggest two things? 1 Grab a sysdiagnose during one of the hangs and file a feedback report with Apple 2 Use the `samp…

Yeah, I've recently done exactly this for one of the hangs. Unfortunately Apple was extremely unhelpful, it took a few back-and-forths with DTS to get a response from someone who even bothered to actually read my bug report. (I just got a few generic replies "please submit an Xcode project" first). But even then they just kept asking me for a way to reproduce the issue, which I couldn't, since it happened on a customers machine, and disappeared after a restart. Even though the issue was reported multiple times over a couple of months by various customers, it took me a lot of time to actually get a usable sample from a customer. But apparently even a sample of the process showing exactly where Apple's frameworks are hanging isn't enough for them to start investigating.

I try to report all of the issues I see to Apple, but at some point there's nothing I can do except complain that Apple's frameworks are buggy.

The biggest problem with these bugs is that customers always seem to think it's an app bug, and there's nothing the app developer can do except hope that Apple fixes the issue. One early sandbox bug took Apple about 3 years to fix. To be honest I don't even know if they fixed it, it was never mentioned in a changelist, it's just that I stopped getting reports of the issue at some point.

Re: Benchmarking the Apple M1 Max

#199

Post by JohnBooty »

edgyquant wrote: post #186
FireBeyond wrote: post #141

Earlier quoted context omitted.

macOS has a bunch of energy management and performance-related "private" APIs and libraries that Safari is allowed to call, but Chrome is not.

I would also like to see some documentation on these private apis. Sounds fascinating I wonder what they’re doing differently? Are they exposing the same kind of abstractions but the public api is purposely slower? Why would they not want all apps to use less battery life?

Yeah, that's my question.

Other commenters are suggesting that Apple's intent is to keep Chrome and other third-party apps from matching Safari's energy efficiency.

But as you say, I'm not sure that makes strategic sense. If that is true, Apple is essentially crippling most Mac use cases just to benefit Safari. That really would not seem to be in Apple's best interests.

I am a developer, but not a iOS/MacOS developer. So I have followed this sort of thing only very loosely over the years.

But whenever I have heard about grumblings about private Apple-only APIs being used by blessed first-party Apple apps on iOS, it seems to me that the explanation was always that the private APIs either:

1. posed some sort of security issue

2. were simply not yet stable enough to expose to third-party apps

As every developer knows, publicly exposing any API represents a maintenance burden: you're committing to support that API and keep it stable for X number of years. In general you see this kind of pattern a lot: a platform developer dogfoods APIs internally for some period of time before they're stable enough to expose publicly.

Re: Benchmarking the Apple M1 Max

#200

Post by sudosysgen »

menmob wrote: post #179
sudosysgen wrote: post #167

Earlier quoted context omitted.

Hmm, technically no, an R7 5700G system and 2x 32GB sticks would be around a fifth of the price. That's if you don't need a lot of compute. It's also right around 1/8th the performance of a 3080 but it doesn't have Tensor cores which is big downside in ML. In theory you could even push the 5700G to 128GB if you figured out a way to get ECC to work on it.

But that’s system memory. Not GPU memory. M1 shares that memory, so it’s addressable by both directly, but with ryzen (and almost every other consumer platform) the cpu and gpu memory are separate.

No, you misunderstand. The 5700G is an APU. Memory is shared between it's GPU and CPU. Hence the "G".
Post reply on HN