Trying to understand Relative Performance Between Mobile and Desktop CPU's

gvx64

Limp Gawd
Joined
Jul 30, 2022
Messages
236
I was wondering if somebody could help me to understand why these two processors are neck and neck in terms of single thread performance benchmarks when the i5-11400F has a TDP of 65W and the i5-1135g7 has a TDP of 15W

https://cpu-benchmark.org/compare/intel-core-i5-1135g7/intel-core-i5-11400f/

To be clear, the i5-11400F is a desktop CPU while the i5-1135g7 is a mobile CPU that is often used in SurfacePro tablets and such.

What I am trying to understand more generally here is the difference in performance between mobile (eg. surface pro) CPU's and the traditional desktop CPU. It doesn't make sense to me that these two processors have similar performance, despite having similar specs and being from the same gen, when the power usage is so insanely different. I mean, SurfacePro's don't even have fans, as far as I know. My understanding is not strong here but I always thought that Mobile architecture was not on the same footing as X64 in terms of heavy computational performance even if the clock frequency, cores, mem speed, etc was similar.

For example, I remember a few years back, at work, running a Python script on a high end desktop PC and I ran the same script on a very high-end Surface pro where the script basically controlled mouse movements to have a CADD program repeatedly perform a Finite Elements Analysis (heavy computation) and I would let this run overnight. When I came back the desktop PC just killed (and I mean killed) the Surface Pro in performance. The Surface Pro would get like 1/5 as much done as the desktop and the back side of the SurfacePro was so hot that you could almost burn your hand by touching it. I was trying to explain to my boss why I needed my desktop because even though the surface pro had a similar GHZ value to the desktop, it was just extremely slow for what I needed it to do. I always thought that there was some architectural differences between mobile and desktop tech which explained the difference in performance I was seeing but I couldn't understand it well enough to explain.

In conclusion, it really confuses me when I see the benchmark results above and the two CPU's are almost 1:1, I mean if this is the case then why even bother use desktop CPU's anymore as they just chug power. I know that this is a n00b question, but I really want to understand this properly and I am very grateful for any help that you might be able to provide.
 
Last edited:
why even bother use desktop CPU's anymore
I've been wondering the same thing. In this case it's
14nm vs 10nm
65W vs 28W

I do know that laptops often have really terrible thermal solutions, sometimes. Despite similar initial performance, as things heat up, mobile devices often just can't compete. This isn't always the case, but in my experience, it happens a lot. From watching reviews, it can happen to new-ish, high-end laptops as well.
 
  • Like
Reactions: gvx64
like this
Better binning allows lower power use at the same clock speed, and if the other parts of the architecture (cache, bus lanes/gen, etc) are similar, performance will be as well, or they can be culled to fit a certain power envelope...at cost of performance. Add dedicated hardware accelerators (which may not be included on some desktop chips), and power can be reduced further (on some workloads) while maintaining similar perf and quality. And as mentioned, sometimes the process they use is more efficient as well.
 
  • Like
Reactions: gvx64
like this
I've been wondering the same thing. In this case it's
14nm vs 10nm
65W vs 28W

I do know that laptops often have really terrible thermal solutions, sometimes. Despite similar initial performance, as things heat up, mobile devices often just can't compete. This isn't always the case, but in my experience, it happens a lot. From watching reviews, it can happen to new-ish, high-end laptops as well.
Thanks for this reply. I am wondering if thermal throttling might explain at least some of the performance difference I was seeing in my SurfacePro vs PC example. The SurfacePro was extremely hot to the touch and I guess it is likely that any boost frequency the CPU had was disabled was disabled on the pro while the desktop PC was sailing on at max output.

Better binning allows lower power use at the same clock speed, and if the other parts of the architecture (cache, bus lanes/gen, etc) are similar, performance will be as well, or they can be culled to fit a certain power envelope...at cost of performance. Add dedicated hardware accelerators (which may not be included on some desktop chips), and power can be reduced further (on some workloads) while maintaining similar perf and quality. And as mentioned, sometimes the process they use is more efficient as well.
Thanks. Could the hardware acceleration that you are describing mean that the SurfacePro would be better at certain types of computation (eg. streaming, Teams chats, etc.) while the traditional desktop CPU would still be inherently dominant in others (eg. deep mathematical computations)?
 
Thanks. Could the hardware acceleration that you are describing mean that the SurfacePro would be better at certain types of computation (eg. streaming, Teams chats, etc.) while the traditional desktop CPU would still be inherently dominant in others (eg. deep mathematical computations)?
It could be -- it's entirely dependant on the CPUs in question (well, and software/driver support), though, and I'm not familiar with these particular chips.
 
  • Like
Reactions: gvx64
like this
It could be -- it's entirely dependant on the CPUs in question (well, and software/driver support), though, and I'm not familiar with these particular chips.
Thanks, that does make a bit of sense to me. I was using the word "architecture" in my original post and I realize that this is not the correct word to use. I had thought that the SurfacePro used ARM but after doing some more reading it is pretty clear that the above i5-1135g7 uses X86-64 just like the i5-11400F, so it wouldn't be an architectural difference per-say but perhaps acceleration and optimization accounts for a difference. The desktop CPU definitely has an advantage in terms of cache sizes where the mobile one above doesn't even have an L2 cache. Definitely makes me think that the mobile CPU is designed for lighter computation, I will look to see if I can find some benchmark comparisons between these two chips that focus on heavier mathematical tests.

Thanks again for your help.
 
In addition to what's already been said:
Intel's 11th gen for desktop had to be back-ported to 14nm. In doing so, they lost a lot of efficiency and some of the effectiveness of the architecture.

11th gen mobile is on a smaller, more efficient process node. And the architecture is designed for it. So, full effectiveness.

If Intel had not had to backport the desktop version of 11th gen; its probable that performance would have been even better. And they may have been able to do a 10 core part. Instead of maxing out at 8 cores.
 
  • Like
Reactions: gvx64
like this
Well, you're looking at 1t short-term performance where the thermal limitations of laptops are less relevant. Furthermore a 15W U-series processor has a PL2 of (at least) 28W, and at 4.2/4.4 GHz you're looking at around 20-30W per core so neither CPU is hitting its power limits.
The last 20% performance also throws efficiency out the window:

image-19-1.png


Taking this chart with a grain of salt because libx264 is not a particularly representative workload, we can see that Golden Cove can achieve 10 fps at about 41W, but requires *87W*, more than double, to go 20% faster. Back in the day we would have called that "overclocking", but for better or for worse competitive pressure has forced both x86 players to ship what are effectively overclocked processors since Zen2.
 
Well, you're looking at 1t short-term performance where the thermal limitations of laptops are less relevant. Furthermore a 15W U-series processor has a PL2 of (at least) 28W, and at 4.2/4.4 GHz you're looking at around 20-30W per core so neither CPU is hitting its power limits.
The last 20% performance also throws efficiency out the window:


Taking this chart with a grain of salt because libx264 is not a particularly representative workload, we can see that Golden Cove can achieve 10 fps at about 41W, but requires *87W*, more than double, to go 20% faster. Back in the day we would have called that "overclocking", but for better or for worse competitive pressure has forced both x86 players to ship what are effectively overclocked processors since Zen2.
Ok, this is very helpful, thank you very much for this. This is exactly what I was looking for.

After seeing the terminology that you used: PL2, 1t I looked up the below article which further helped me to understand what is going on:

https://www.hardwaretimes.com/intel-10th-gen-cpu-power-consumption-explained-pl1-pl2-and-tau/

Basically, these two processors have similar specs on paper but the desktop CPU will likely spend a lot more time in its boost state than say an I5-1137g7 installed in a surface pro tablet with no active cooling. Battery life and heat dissipation are not as much of an issue in a desktop and so motherboard manufacturers tend to push the turbo much further past Intel's recommended Tau values in desktops as opposed to laptops and tablets where the Tau value must be obeyed strictly.

This would help explain why my old Surface Pro at work struggled to do a fraction of what the comparably spec'd desktop computer could do for overnight multi-hour, heavy mathematical computation. In addition to the turbo boost restrictions, given how hot the Surface Pro backside was in the morning it was probably getting thermal-throttled as well which would account for the insane difference in performance (about 1/5 of the desktop performance).

Thanks again for this clarification.
 
Back
Top