Intel's Alder Lake is slower than AMD's Cezanne Zen 3 at 45W TDP

[Redaktion]

We are currently getting more laptops with the new Intel Alder Lake processors. A direct comparison is not always easy due to different TDP settings, but the two devices Alienware x14 and the Lenovo Yoga Slim 7 Pro 14 now give us the chance for that - with the better end for AMD.

https://www.notebookcheck.net/Intel-s-Alder-Lake-is-slower-than-AMD-s-Cezanne-Zen-3-at-45W-TDP.607902.0.html

[Wildfear]

Strange, how can Alienware X14 get only 11k CB23 here when all other reviewers got 15-16k on the same benchmark ?

[Wildfear]

Strange, how can Alienware X14 get only 11k CB23 here when all other reviewers got 15-16k on the same benchmark ?

Talking about this part :

During our initial tests, we noticed that there is no increased short-term performance limit and the processor just runs at 45W all the time.

[LL]

Bad pure performance results. If you are getting 11000 what is the advantage over a year's old 5800H that can do 13000?  wattage consumption?

[asun]

great job amd for beating intel in loooong time

[Redman]

Curious did Intel fix the heat problem with the new laptop cpu? The 10500 runs consistent 96c gaming and they said it's normal operating temp...

[hs4]

This appears to be a very different result from the article previously discussed in noteboookcheck ("The Intel Core i7-12700H dispatches the AMD Ryzen 9 5900HX in performance and efficiency benchmarks"). This article is less precise in its environmental validation and needs to be refined.

[Andreas Osthoff]

The other article was just talking about external results. The data in this article was taken by us in a direct comparison with the Alienware x14 and Yoga Slim Pro side by side on the same day with actual retail units.

[rs]

Didn't expect the single core power consumption to be so much higher. 10-20% higher performance aligns with many other tests. But the power consumption makes it look much less impressive and attractive. Intel's core architecture still has a lot to improve to be competitive again on all fronts.

[Wildfear]

It might be interesting to normalize results :

Multicore 45W CB23 :
Intel = 244,4 pts per W
AMD = 253.9 pts per W
3.8% better for AMD

And for single core CB23 (if total consumption is for CPU only ?) :
Intel = 58.9 pts per W
AMD = 81,5 pts per W
38,3% better for AMD

For both result AMD wins in efficiency.

[nabat]

future Intel will increase cache.. good for in the huge data benchmarks..

[HNH]

Why comparing ryzen 9 to core i7?
It wise to compare it to corei9.

[_MT_]

One would expect the 12900H to be a better bin and therefore it should perform better, it should be more efficient. By how much, that's a question. And to answer it properly, you'd need multiple samples and perform a statistical analysis of the results.

What I find interesting is that the Alienware cuts off at a lower temperature. Quite uncharacteristic for them. That doesn't help.

[_MT_]

And for single core CB23 (if total consumption is for CPU only ?) :

I imagine total is for the whole package, including GPU. The basic problem is that these relationships aren't linear. You have to perform the test either at set power or set performance. I'm not disputing that AMD is more efficient, but when both power and performance differ, it's hard to tell anything. It's a law od diminishing returns. As you go up, you pay more and more for less and less. What would happen if Intel was capped at the same 14/18 W? How much performance would be lost? Intel can compete in efficiency only when running at significantly reduced frequencies. That much is clear from Intel's own presentation. When you run it flat out, there is no chance in hell.

[_MT_]

"it is also an explanation why Intel only uses up to 6 performance cores in its mobile processors"

Do tell. One would think that it has to do with the die size and target audience. You can easily add more cores and it will just reduce the all-core frequency. Ten big cores should be doable. You can't go too high because of overheads (some can grow very quickly with the number of cores).

Also, in this case, the smaller cores primarily lack certain features like AVX-512 units which are rather expensive and not that needed in consumer applications. They're not like small cores in ARM designs which are optimized for energy consumption. They were designed more as economical cores as in cheaper to produce, meaning you can have more of them for given budget.

How many consumer applications can utilize even six of those big cores properly? The smaller cores are there for multithreaded performance. The more big cores you have, the fewer smaller cores you can have for the same budget. It's not exactly rocket science. Frankly, it wouldn't have surprised me if there were only four big cores. Again, it's more than enough for a typical consumer workload. Perhaps a 4+12. But then again, how many consumer workloads can meaningfully utilize 16 cores? It's a compromise between a smaller number of heavy threads and a multitude of lighter threads. Big cores are better for the first scenario, small cores should be better for the second scenario.