Hacker News Reader

Comment by cehrlich | Hacker News Reader

Seems like for things that are: 1. Perfectly parallel 2. Not accelerated by some of the other stuff that's on the Apple Silicon SoC's ...it will be a toss-up.

Threadripper 3990X get about 25k in Geekbench Multicore [1]

M1 Max gets about 12.5k in Geekbench Multicore, so pretty much exactly half [2]

Obviously different tasks will have _vastly_ different performance profiles. For example it's likely that the M1 Ultra will blow the Threadripper out of the water for video stuff, whereas Threadripper is likely to win certain types of compiling.

There's also the upcoming 5995WX which will be even faster: [3]

[1] https://browser.geekbench.com/processors/amd-ryzen-threadrip...

[2] https://browser.geekbench.com/v5/cpu/search?utf8=%E2%9C%93&q...

[3] https://www.amd.com/en/products/cpu/amd-ryzen-threadripper-p...

View on HN

Also of note is that half of the Mac Studio's case is dedicated to cooling. Up to this point, all M1 Max benchmarks are within laptops while all Threadripper benchmarks are in desktops. The M1 Max in the Mac Studio will probably perform better than expected.

This is sound logic and probably be the case but I wonder if this effect will be less than what we have seen in the past because of the reduced TDP of the M1 processors in general.

Maybe the cooling and power delivery difference between laptop formfactors and PC formfactors will be less with these new arm based chips.

Maybe; every chip hits a point where feeding in more power doesn’t make it go any faster.

If I was to guess, the increased cooling probably helps the Studio sustain similar boost clocks as the laptops, but for longer.

Although it’s possible these are on N4x, which might increase the attainable boost.

Geekbench is extremely sensitive to the OS. Like the same CPU on Windows & Linux score wildly different on Geekbench. For example the 3990X regularly hits 35k multicore geekbench when run on Linux: https://browser.geekbench.com/v5/cpu/11237183

Something is seriously fishy about those geekbench results.

24-core scores 20k, 32-core scores 22.3k, and 64-core score 25k. Something isn't scaling there.

Probably it's the thermals that don't scale. The more the cores, the lower the the peak performance per core.

Yeah, it’s the real world tasks that GeekBench tries to simulate that don’t tend to scale linearity with processor count. A lot of software does not take good advantage of multiple cores.

> A lot of software does not take good advantage of multiple cores.

It sounds pointless to come up with synthetic benchmarks which emulate software that is not able to handle hardware, and then use said synthetic benchmarks to evaluate the hardware performance.

It has a very specific point: communicating performance to people who don't know hardware.

Most consumers are software aware, not hardware aware. They care what they will use the hardware for, not what they can use it for. To that end, benchmarks that correlate with their experience are more useful than a tuned BLAS implementation.

That's certainly true. But if that's your workload you shouldn't be buying a 64-core CPU...

I use a few 32 and 64 core machines for build servers and file servers, and while the 64-core EPYCs are not twice as fast as the 32-core ones due to lower overall frequency, they're 70% or so faster in most of the things I throw at them.

Does Geekbench actually attempt to simulate that in their multi-core score? And how?

I was under the impression that all of their multi-core tests were "run N independent copies of the single-threaded test", just like SPECrate does.

Many GB5 (and real world) tasks are memory bandwidth bottlenecked, which greatly favors M1 Max because it has over double a Threadripper's memory bandwidth.

Sort of. The CPU complex of the M1 Max can achieve ~200 GB/s, you can only hit the 400 GB/s mark by getting the GPU involved.

At the same time the Threadrippers also have a gargantuan amount of cache that can be accessed at several hundred gigabytes per second per core. Obviously not as nice as being able to hit DRAM at that speed.

That cache is not uniform time access. It costs over 100ns to cross the IO die to access another die's L3, almost as much as going to main memory. In practice you have to treat it as 8 separate 32 MB L3 caches.

Also, not everything fits into cache.

Reply on news.ycombinator.com