Without digging into it, one possible broad interpretation would be that M1 might have more accelerators for specific, real-world tasks. While TR has more raw power. And that would be par for course for ARM. x86 processors have more raw power and can perform such tasks adequately without acceleration at the expense of efficiency. The result is that in these specific tasks, those processors can punch above their weight. But in general computing and good old heavy lifting, you can't get around the limited raw power.

It really is foolish to expect 20 cores without SMT to compete with 64 cores with SMT. Although you're well into diminishing returns with that processor in many workloads. The 3970X is the better all-around choice. Even the 3970X, 32 SMT cores against 20 non-SMT should be no-brainer. And don't forget that only 16 of them are performance cores. Really foolish. As good as Firestorm is, it's nowhere near being four times as good as Zen 3.

Quote from: hs4 on March 10, 2022, 08:11:13
This is a matter of the type of benchmark: Geekbench is a synthetic index that is sometimes used to evaluate smartphones, and while it is consumer-oriented, it is not suitable for evaluating workstations.

Cinebench is just one workload. Workloads have different characteristics. Geekbench is a suite of workloads. It might not have the reputation of SPECint, but it's similar in nature. If you want to interpret the numbers, you have to look at individual benchmarks and understand their characteristics. Then you know which benchmarks matter for your workload. E.g. M1 Ultra posted roughly twice as high score as 3990X running Linux in AES encryption while 3990X posted four times as good a score in text compression. And so on and so forth.

This is a matter of the type of benchmark: Geekbench is a synthetic index that is sometimes used to evaluate smartphones, and while it is consumer-oriented, it is not suitable for evaluating workstations. For example, accroding to the Notebookcheck database:

3975WX: 42744 @ Cinebench MT, 28798 @ Geekbench MT
5950X : 26017 @ Cinebench MT, 16248 @ Geekbench MT
12900K: 27005 @ Cinebench MT, 17854 @ Geekbench MT
M1 Max: 12340 @ Cinebench MT, 12636 @ Geekbench MT

M1 ultra would be about twice as good as M1 max, but would be inferior to both 5950X and 12900K on Cinebench R23 (which natively supports Apple Silicon).

This article is incorrect. The 3990x gets well over 37000 in geekbench. The M1 isn't even remotely close.

We are not there yet. Some tasks will be faster on either system. Also note that Apple chose second tier consumer chips for comparisons. Efficiency and noise await tests.

If true, this is the death of Intel and Amd. They basically caught up to last gen flagships in a single generation. Sait until M2 rolls around. If these numbers are true, everyone will want these for datacenters. It's lightyears ahead of any other ARM processor out there.

Who cares about benchmarks numbers ?

The M1 line might be the king of benchmarks, but when it comes to real life experience and performance, that's another matter.

I have a Macbook Pro 14 and a HP Spectre 14.
The M1 in the MB is supposed to be something like twice as powerfull as my HP laptop.

In the real world,the loading time of identical applications on both systems takes several seconds longer on the Mac.

eg. Foxit PDF, 0.5s on HP laptop, 5s (!!) on MB

Quote from: DS2 on March 09, 2022, 12:15:07
Just look at the MT score of the 12900k compared to the threadripper to see that something is really really wrong lol

Just to add, in the real world, the number of cores matters because of the scalability of a workload. Many real workloads have scalability issues and you run into a wall (adding more cores doesn't help or even degrades performance). So, the unspoken assumption was that the workload scales ideally. Then it's down to efficiency.

Quote from: DS2 on March 09, 2022, 12:15:07
Just look at the MT score of the 12900k compared to the threadripper to see that something is really really wrong lol

It's Windows. 3990X consistently underperforms on Windows compared to Linux. Also, do realize that the difference is largely down to efficiency. You have to keep in mind the power. It takes power to perform calculations. If you have two processors that both take the same 200 W and one outperforms the other, it's because it's more efficient. The 12900K has a huge power budget given the relatively low core count and therefore can give a lot more power to each core compared to the 3990X. When you add cores, you also want to add power. Otherwise, you're just relying on higher efficiency resulting from lower operating frequencies because you don't have enough power to maintain the higher frequencies. Fortunately, Zen cores don't scale that well so you're not losing that much but you also don't gain that much by giving it more juice. Good for efficiency, not awesome for all-out performance.

I dont like fake news. delete it.

Just look at the MT score of the 12900k compared to the threadripper to see that something is really really wrong lol

For some reason Geekbench seems to drastically favour M1 systems. When comparing benchmarks in other tools such as cinebench the M1 relative performance is not that far ahead. I suspect the 12900k will still faster in cinebench than the M1 Ultra. Definitely in single-core and maybe also in multi-core. It will use significantly more power though. Kudos to Apple for pushing the industry froward and hopefully given Intel a kick in the a** it really needs.