Alder Lake

If the 12900HK has a TDP of 85W (I assume that is some sort of median) and the RTX 3080 Ti has a TDP of 350W, does that mean that the Raider could be briefly drawing close to half a KW under heavy loads?
 
Frankly if I was Intel I'd be embarrassed (and worried) by these results and what it took to get them. Performance per watt is terrible compared to the competition and that is before we factor in that M1 architecture is coming up on year 2 right now, with another M2 rumoured to be dropping this year. The multi-core score in particular required a greater number of cores, higher clock speed, multiple times the power requirement connected to the wall, a lot more noise, to achieve .... 4% on an artificial benchmark that likely has not been optimized for Apple Silicon (due to the nature of benchmarks).

I don't use Cinebench daily or stock crabs chess benchmarks or whatever the latest troll is over at the mothership. I like the ports, sound , screen, build materials and quality of the MBP.
I don't value loud fans, I don't want to have to tether to a wall to access full performance. I don't like go-faster neon lights etc... I'm not the target audience of the Raider laptop.

That being said I do use Xcode, Final Cut Pro, Logic and Capture One. On the first 3 apps my new m1 pro max is multiple times faster in quantifiable performance over my 2019 i9 MBP. When I compare performance (not strictly apples:apples) between a similar video project export on Resolve on a Ryzen 5950 desktop with a 3080 and a similar project export on M1 Max (with LUTS applied, some color grading, exposure adjustments etc...) FCP on the M1 Max; the MBP is quantifiably faster for what I care about .... a lot faster in operation.

Even comparing to a desktop xeon Mac Pro, Xcode compile times are faster and Final Cut exports are significantly faster... so for my use cases, on real world applications that ARE optimized for Apple Silicon (before accounting the other benefits that I care about - battery life, build quality, sound, screen quality, ports, noise, heat).

For anything else that runs better on a WinTel I'll wait for a Thinkpad P1 ( my laptop of choice on Wintel), sacrifice a little performance over the Raider, sacrifice the go-fast lights, and get a more portable solution.
 
Andropov said:
Yeah, I think it's been pretty obvious that the i9 12900HK was specifically designed and tuned to beat the M1 Max, no matter what.
Click to expand...
It's the Weyland-Yutani philosophy of CPU architecture. Special Order 386: "Priority one. Ensure defeat of M1 in synthetic benchmarks. All other considerations secondary. Design team expendable."
 
Cmaier said:
That said, how can that graph be right given what we now know?
Click to expand...
The numbers are likely real, just unrealistic in their setup. I believe (since they didn't say anything other than that they compiled for M1 Max using Xcode 13) that they used the default compiler flags for LLVM on Xcode 13 (that is, -Os), which leaves a lot of the most performance-critical optimizations on the table. It's a reasonable default flag for an app, but not for a compute-intensive benchmark. And on the other hand they used ICC for the Intel compilations, which already introduces very aggressive optimizations, and then they (likely) hand-picked the best compile settings for each of the subtests in SPECint. That, along with crippling the M1 Max performance by deliberately using suboptimal compiler options gave them a massive advantage. Had they compiled for M1 Max using -O3 or something like it I suspect the difference would be much, much smaller, and the graph wouldn't exist.

So not made up, just comically bad cherry-picking.
 
Andropov said:
The numbers are likely real, just unrealistic in their setup. I believe (since they didn't say anything other than that they compiled for M1 Max using Xcode 13) that they used the default compiler flags for LLVM on Xcode 13 (that is, -Os), which leaves a lot of the most performance-critical optimizations on the table. It's a reasonable default flag for an app, but not for a compute-intensive benchmark. And on the other hand they used ICC for the Intel compilations, which already introduces very aggressive optimizations, and then they (likely) hand-picked the best compile settings for each of the subtests in SPECint. That, along with crippling the M1 Max performance by deliberately using suboptimal compiler options gave them a massive advantage. Had they compiled for M1 Max using -O3 or something like it I suspect the difference would be much, much smaller, and the graph wouldn't exist.

So not made up, just comically bad cherry-picking.
Click to expand...

Someone should redraw that graph now that we have information about actual power consumption. :-). At 30W it must be awful.
 
Cmaier said:
Someone should redraw that graph now that we have information about actual power consumption. :). At 30W it must be awful.
Click to expand...
Ah, I totally forgot about the power consumption part. I have no explanation for that. Maybe it is pure fantasy after all.
 
Cmaier said:
Someone should redraw that graph now that we have information about actual power consumption. :-). At 30W it must be awful.
Click to expand...
Maybe they have a sophisticated COP that can measure per-core draw and they are comparing the draw of one core to than of an entire M1?
 
Cmaier said:
Someone should redraw that graph now that we have information about actual power consumption. :). At 30W it must be awful.
Click to expand...

The only source I've really looked at for 12900HK performance so for is Anandtech, but they put the Raider laptop in "Battery saver mode" or whatever that power profile in Windows is called and noted that it locked the power draw of the CPU at 30W, i.e. no turbo above either. Just 30W. Unfortunately the only test they ran at 30W was Cinebench R20 (not R23 for some reason), making it hard to compare against any Apple Silicon since R20 is older and not updated for ARM. But comparing it to itself works as a proxy. In the performance mode which is 85W sustained 110W turbo (PL2). It hits 6,869. In the battery saver mode at 30W, it hits 3,494. So it's ≈ half. Intel might be able to undervolt and set up a different configuration of E cores only but loads of them that does really well in multi-threaded benchmarks at somewhat low wattage, if we allow them as many marketing tricks as possible to fudge the numbers in place? :P
 

Similar threads

Cmaier
Alder Lake bios source leak
Replies
0
Views
227
Cmaier
Cmaier
Back
Top