Intel's Core i7-12700H Benchmarked Against Apple's M1 Max

[Lightwreather]

Summary

Alleged Geekbench 5 results obtained on Gigabyte's Aero 5 XE and HP's Omen 17 laptops with Intel's Core i7-12700H inside were added to the benchmark's database on Friday, revealing the performance of the upcoming mobile Alder Lake-P in this popular synthetic benchmark.

 

Quotes

Quote

Alleged Geekbench 5 results obtained on Gigabyte's Aero 5 XE and HP's Omen 17 laptops with Intel's Core i7-12700H inside were added to the benchmark's database on Friday, revealing the performance of the upcoming mobile Alder Lake-P in this popular synthetic benchmark. The numbers demonstrated by both machines are really close, so we can assume that the readings are more or less accurate. Meanwhile, it looks like both machines are equipped with DDR4-3200 memory, which might have limited their performance in single-threaded workloads.

Quite surprisingly, Intel's Core i7-12700H 'Alder Lake-P' fails to beat the Core i7-11800H 'Tiger Lake-H' as well as AMD's Ryzen 7 5800H 'Cezanne' (Zen 3) in single-threaded workloads. Furthermore, it is considerably behind Apple's M1-series system-on-chips that have been single-thread performance leaders for about a year now. Perhaps, usage of faster memory would have given Intel's ADL-P an advantage, but we will need to run our own tests to find out.

In multi-threaded Geekbench 5 workloads, everything looks much better for the Core i7-12700H as it leaves behind both the Core i7-11800H and the Ryzen 7 5800H. Still, Apple's M1 Max with its 10 cores (eight high-performance cores, two energy-efficient cores) beats Intel's upcoming offerings in all GB5 multi-threaded tests.

 

My thoughts

Here come the fans. /s But seriously, do take these benchmarks with a grain of salt. These are still pre-announcement benchmarks not to mention that performance comparisons in laptops is needlessly complicated. However, this is pretty interesting, Apple is still beating the next-gen mobile CPUs of Intel. However, this doesn’t take into account power draw, or thermals which we'll still have to wait until actual reviews. So, final statement: Take this with a heap of salt and wait.

Cue the needless fanboyish arguments.

Sources

Tom's hardware
Geekbench-1

Geekbench-2

[SorryBella]

8 minutes ago, J-from-Nucleon said:

do take these benchmarks with a grain of salt.

Wait, people actually take geekbench seriously?

[leadeater]

1 hour ago, J-from-Nucleon said:

These are still pre-announcement benchmarks not to mention that performance comparisons in laptops is needlessly complicated.

Freakin right, I hate how much this is actually a thing.

[JLO64]

Honestly the number one thing I’m most interested about for this generation of mobile chips isn’t performance but efficiency. I really hope Intel’s efficiency cores can help laptops with x86 CPUs compete with the insane battery life Apple has managed to squeeze out of their ARM SOCs.

[Arika]

16 hours ago, SorryClaire said:

Wait, people actually take geekbench seriously?

a certain group of people do, but they shouldn't

[InstantNewt]

18 hours ago, SorryClaire said:

Wait, people actually take geekbench seriously?

How else am I supposed to dunk on someone else’s choice of smartphone?

 

I didn’t even realize they were running “benchmarks” on PC hardware, I figured they were mobile phones only. 

[LAwLz]

On 11/13/2021 at 10:28 AM, SorryClaire said:

Wait, people actually take geekbench seriously?

 

7 hours ago, Arika S said:

a certain group of people do, but they shouldn't

There is nothing wrong with geekbench as a benchmark. If you think there is then you are ignorant.

[RejZoR]

8 minutes ago, LAwLz said:

 

There is nothing wrong with geekbench as a benchmark. If you think there is then you are ignorant.

Usually people call it "irrelevant" when Apple is destroying everyone in it, be it iPhones or in this case Macs. They keep raving about it otherwise when everything else is benched in it like Snapdragon powered devices or whatever.

[mr moose]

34 minutes ago, LAwLz said:

 

There is nothing wrong with geekbench as a benchmark. If you think there is then you are ignorant.

I think the problem is we have to do too much qualifying to make sure some comparisons are accurate (especially when they are laptop comparisons), which is beyond the scope of many.  Too many people just take the end number then use it as a blanket "see this processor is better".  

 

I Stopped looking at passmark and GB and started looking at specific benchmarks for the workload I was interested in,  only because some benchmarks favor 2T1C while others work better on 1T1C and things like that.  Also scaling isn't always linear and I rarely have the same RAM as those used in tested hardware (I usually cheap out a bit there). 

Edited November 14, 2021 by leadeater
Removed all the empty extra lines

[LAwLz]

2 minutes ago, mr moose said:

I think the problem is we have to do too much qualifying to make sure some comparisons are accurate (especially when they are laptop comparisons), which is beyond the scope of many.  Too many people just take the end number then use it as a blanket "see this processor is better".  

 

I Stopped looking at passmark and GB and started looking at specific benchmarks for the workload I was interested in,  only because some benchmarks favor 2T1C while others work better on 1T1C and things like that.  Also scaling isn't always linear and I rarely have the same RAM as those used in tested hardware (I usually cheap out a bit there). 

I agree, but none of that is a geekbench issue. It's a "laptops are tough to compare" issue because of different thermals, memory, power state, etc. The same applies to basically any benchmark, including real world ones. 

 

Comparing laptop CPUs against one another basically always end up being "how does this particular laptop perform vs how does this other laptop perform".

[RejZoR]

59 minutes ago, mr moose said:

I think the problem is we have to do too much qualifying to make sure some comparisons are accurate (especially when they are laptop comparisons), which is beyond the scope of many.  Too many people just take the end number then use it as a blanket "see this processor is better".  

 

I Stopped looking at passmark and GB and started looking at specific benchmarks for the workload I was interested in,  only because some benchmarks favor 2T1C while others work better on 1T1C and things like that.  Also scaling isn't always linear and I rarely have the same RAM as those used in tested hardware (I usually cheap out a bit there). 

The thing is, every single Macbook Pro (or Mac Air) you'll pick, it'll perform identically within that group. Where with any other laptop, it can have identical components and entirely different performance because of bunch of stupid arbitrary limitations or parameters, be it clocks, power limits, thermal designs. Or just because of literal "reasons". And lets don't forget that every model then has 300 variants for some reason. Which is the usual habit of all brands like ASUS, MSI, Gigabyte, Dell, HP etc. You often have to go down to subversion/submodel numbers to see what's what.

[mr moose]

1 hour ago, RejZoR said:

The thing is, every single Macbook Pro (or Mac Air) you'll pick, it'll perform identically within that group. Where with any other laptop, it can have identical components and entirely different performance because of bunch of stupid arbitrary limitations or parameters, be it clocks, power limits, thermal designs. Or just because of literal "reasons". And lets don't forget that every model then has 300 variants for some reason. Which is the usual habit of all brands like ASUS, MSI, Gigabyte, Dell, HP etc. You often have to go down to subversion/submodel numbers to see what's what.

Even when you know exactly what hardware is in the laptops being compared, you still have to understand the what's and how's of it all in order to drill down to an adequate comparison.

 

But it goes deeper than that too, Intel/AMD have put a lot of work into HT/SMT so their cores are each optimized for 2 threads, the M1 is a single thread per core processor (0x86 v ARM).  Depending on how the synthetic test weights threads and cores (single thread workload versus single core workload), a better result doesn't always mean better in general.   So even if we could ignore all other aspects of laptops and concentrate solely on the CPU there are still going to be issues in making comparisons.  Hence why workload specific benchmarks can be better.   It gets to a point when it just becomes nitpicking, especially when the real world difference between two devices is beyond the users requirements. 

[LAwLz]

1 hour ago, mr moose said:

But it goes deeper than that too, Intel/AMD have put a lot of work into HT/SMT so their cores are each optimized for 2 threads, the M1 is a single thread per core processor (0x86 v ARM).  Depending on how the synthetic test weights threads and cores (single thread workload versus single core workload), a better result doesn't always mean better in general.   So even if we could ignore all other aspects of laptops and concentrate solely on the CPU there are still going to be issues in making comparisons.  Hence why workload specific benchmarks can be better.   It gets to a point when it just becomes nitpicking, especially when the real world difference between two devices is beyond the users requirements. 

I couldn't disagree more with you here. 

The distinction between "per core" performance and "per thread" performance is just nitpicking to win some imaginary dick measuring contest.

 

When people (and programs) say "per core performance" they mean per thread performance, and that's what's actually important. No program will run on "one core". It will either run with one thread, or more threads. 

 

I agree that workload specific benchmarks CAN be better, but if we are making generalized statements about a processor then specific workload benchmarks becomes fairly useless. That's when benchmark suits like SPEC shines, because they give us a good overview of how the processor performs in general. 

In the case of both SPEC and geekbench, it is actually possible to view the individual benchmarks to determine how it performs for some specific task. 

 

I don't agree that "the real world difference between the two devices are beyond the users requirements" either. People have been saying that processors have been "fast enough for people" since the 90's. When Nvidia launched the dual core Tegra 2 processor people went "lol who needs a dual core processor in a phone". As it turns out, just a few years later everyone needed it. 

We shouldn't dismiss differences in performance as "well its more than enough anyway" because we don't know how things will change in the near or far future. Besides, why settle for less when you can have more? It's not like in the case of the M1 you get higher performance at the cost of something. It seems like you'll get higher performance AND lower power consumption, AND faster response times. Those are things people definently notice.

[mr moose]

10 hours ago, LAwLz said:

I couldn't disagree more with you here. 

The distinction between "per core" performance and "per thread" performance is just nitpicking to win some imaginary dick measuring contest.

 

That's exactly what I said,  the difference is nitpicking because the difference isn't always a straight up comparable measurement and in the real world might be a bit different anyway.

10 hours ago, LAwLz said:

When people (and programs) say "per core performance" they mean per thread performance, and that's what's actually important. No program will run on "one core". It will either run with one thread, or more threads. 

Yes, but x86 CPUs are designed to run 2 Threads per core, so running a single thread is sub optimal by design, the M1 will not have this issue.  The only time I can see where a benchmark like this actually showing anything usable (beyond notable performance differences) is when the number of cores available is higher than thread demand and the efficiency per core is so large you almost don't need the benchmark to know one will be better.    I think it's going to be pretty obvious that the M1 will be more efficient at many tasks,  but I don't think all synthetic benchmarks are going to reflect this adequately.  

 

10 hours ago, LAwLz said:

 

I don't agree that "the real world difference between the two devices are beyond the users requirements" either. People have been saying that processors have been "fast enough for people" since the 90's. When Nvidia launched the dual core Tegra 2 processor people went "lol who needs a dual core processor in a phone". As it turns out, just a few years later everyone needed it. 

 

 

I didn't say the differences are always beyond the users requirements, I said it is nitpicking when the differences are beyond the users requirements,  I.E if the user needs a 4 hour battery and a program to complete X operation in 15 seconds and the difference in testing puts both machines at 8+hours battery and 10 seconds or  so with only 1-2 seconds between them, then the score difference is moot.    I personally don't need a laptop beyond my current celeron lenovo, I only use it for emails and basic web browsing (in fact I have been using my phone more and more for that and it's starting to collect dust), so an M1 or 12700H is not going to improve that experience for me.

 

We always will need faster and more optimized machines as time goes on and software becomes more intensive, but when looking at a benchmark for purchasing purposes,  it really only applies to the those machines in that one instance. 

[mecarry30]

So would you say that x86 processors without SMT are not optimally designed x86 processors?

Quote

Yes, but x86 CPUs are designed to run 2 Threads per core, so running a single thread is sub optimal by design, the M1 will not have this issue. 

 

[da na]

On 11/13/2021 at 7:59 PM, Arika S said:

a certain group of people do, but they shouldn't

Same people who use UBM results for actual comparison...

[Lightwreather]

1 hour ago, mecarry30 said:

So would you say that x86 processors without SMT are not optimally designed x86 processors?

I don't think that's what he meant. But I could be wrong

[RejZoR]

1 hour ago, mecarry30 said:

So would you say that x86 processors without SMT are not optimally designed x86 processors?

 

Non SMT CPU's are less prone to idiocies like running 2 threads on same physical core instead on 2 separate physical cores. In general scheduler should make sure this doesn't happen, but no one really knows when stuff is working in the background if that is in fact true. Because in such situation you'll lose massive performance as SMT threads are some 20-30% efficiency of a full physical core. It is relatively cheap to do die size wise when you just want even more threads creating a 8c 16t CPU than having full 16 cores. Performance obviously won't be the same, but cost to threads will with higher performance than just 8 non-SMT cores..

[leadeater]

6 minutes ago, RejZoR said:

Non SMT CPU's are less prone to idiocies like running 2 threads on same physical core instead on 2 separate physical cores. In general scheduler should make sure this doesn't happen, but no one really knows when stuff is working in the background if that is in fact true. Because in such situation you'll lose massive performance as SMT threads are some 20-30% efficiency of a full physical core. It is relatively cheap to do die size wise when you just want even more threads creating a 8c 16t CPU than having full 16 cores. Performance obviously won't be the same, but cost to threads will with higher performance than just 8 non-SMT cores..

Add to that the classic argument of who's core/arch is more performant/faster than another. By measure of single thread performance or total performance of a single core i.e. 2 threads running on a single core. And then how much does it really matter because as you point to it depends on thread scheduling and also what the app that is running is doing in regards to threads in the first place.

[RejZoR]

1 minute ago, leadeater said:

Add to that the classic argument of who's core/arch is more performant/faster than another. By measure of single thread performance or total performance of a single core i.e. 2 threads running on a single core. And then how much does it really matter because as you point to it depends on thread scheduling and also what the app that is running is doing in regards to threads in the first place.

Single thread is a single thread. We say "single core performance" but we really mean single thread. Because you issue a single thread test on a core. SMT doesn't break up single thread into two parts and runs it on one core. That was much hyped for AMD's ZEN processors before they arrived. Remember the "reverse SMT" talks? Would be cool, but it never happened. That would actually measure single core performance. Or somehow forcing 2 threads to a single core (so it would be 1 physical and 1 SMT) and measure full single core potential that way. Haven't seen any bench that does that specifically.

[leadeater]

19 minutes ago, RejZoR said:

Single thread is a single thread. We say "single core performance" but we really mean single thread.

Problem is not everyone uses it or thinks of it correctly. I very often see single core performance wording uses and rarely correctly.

[mr moose]

4 hours ago, mecarry30 said:

So would you say that x86 processors without SMT are not optimally designed x86 processors?

 

I wouldn't say that. As Rejzor said,  HT and SMT are there because it's more cost efficient to make more threads with HT/SMT than to make more full cores.  I dare say  we'd need an engineer to explain why. 

 

My point was not so much to get into the technicalities of CPU design though (apologies if it wasn't worded all that well), but simply to highlight a general issue single threaded synthetics tests will face.  Or more importantly issue that exists that we end users can't completely allow for as we don't know how many threads the scheduler is throwing into the core alongside said benchmark thus effecting the outcome.   We could probably argue that it's not much different than real world, but then again, the M1 won't have this issue so one would expect the single thread/core test to be more accurate on it.  However real world tests should quickly show us what's what.  

 

 

Just as a note of interest, HT was introduced by Intel on (going from memory) northwood processors (xeons first I believe but for desktop the P4),  Which we are told was a rush to combat the awesome performance of the AMD lineup at the time.  So it has been around for a very long time.

 

EDIT: @leadeater it doesn't help that GB calls them "single core scores", or that I use single core/thread interchangeably and I probably shouldn't but I just don't care.

 

 

[LAwLz]

5 hours ago, mr moose said:

That's exactly what I said,  the difference is nitpicking because the difference isn't always a straight up comparable measurement and in the real world might be a bit different anyway.

I don't think you understood what I said, because the other part of your post goes against what I said.

 

 

5 hours ago, mr moose said:

Yes, but x86 CPUs are designed to run 2 Threads per core

They aren't really designed to run 2 threads per core. It's more like SMT was tacked on to make up for poor design decisions. It's not really anything inherent to x86 either. Apple could implement SMT too if they wanted in their ARM cores. 

 

 

5 hours ago, mr moose said:

so running a single thread is sub optimal by design

1) It isn't.

2) Even if it was, programs don't care. It's not like a single threaded program will go "oh sorry, I'll become a multi-threaded program" when executed on an x86 SMT enabled core. 

 

 

5 hours ago, mr moose said:

The only time I can see where a benchmark like this actually showing anything usable (beyond notable performance differences) is when the number of cores available is higher than thread demand and the efficiency per core is so large you almost don't need the benchmark to know one will be better.

I assume that when you say "a benchmark like this" you're talking about the "single core" score in Geekbench. That part is extremely relevant pretty much all the time because a huge portion of programs actually do run into the "there are more cores than threads" scenario. Single-threaded programs, or programs that rely on just a few number of cores (2-4) are very, very common. If they weren't then CPU manufacturers wouldn't bother making their cores larger and larger. They would just have glued 32 Atom cores together a long time ago.

And that's where I disagree with the whole "x86 cores are underutilized in single threaded tests so it doesn't matter" argument you tried to pull. Even if we ignore the whole nitpicking of what is "single core" vs "single thread" performance (spoiler: single core means single thread) it still wouldn't matter because programs don't care. If a program can only use 1 thread efficiently then it's not like running on an SMT enabled core will have an effect on performance, despite the core maybe being "underutilized".

 

I've seen plenty of people make the same argument as you have when talking about Firestorm's amazing single core performance, where they go "b-b-but the x86 core is underutilized because it has SMT so therefore single core performance is higher!". It's like listening to some fanboy of some car complain about how a race was "unfair" because it had a lot of curves and "if the track was more straight then my favorite car would have won for sure!".

 

 

 

2 hours ago, leadeater said:

Add to that the classic argument of who's core/arch is more performant/faster than another. By measure of single thread performance or total performance of a single core i.e. 2 threads running on a single core. And then how much does it really matter because as you point to it depends on thread scheduling and also what the app that is running is doing in regards to threads in the first place.

The whole "total performance of a single core i.e. 2 threads running on a single core" is a completely and utterly pointless metric though. That's what I said about it being an imaginary dick measuring context.

"total performance of a single core" is never, ever relevant. All that matters is per thread performance.

 

 

1 hour ago, leadeater said:

Problem is not everyone uses it or thinks of it correctly. I very often see single core performance wording uses and rarely correctly.

Stop reading words literally. Saying "single core performance" when referring to "single thread performance" instead of "total performance when loaded with all possible threads" is a perfectly valid use of the word. It's not "incorrect".

"Single core performance" means "single thread performance". That's how it is always used.

 

"Total performance when loaded with all possible threads on a single core" is a useless metric anyway, hence why it is not used.

[leadeater]

10 minutes ago, LAwLz said:

The whole "total performance of a single core i.e. 2 threads running on a single core" is a completely and utterly pointless metric though. That's what I said about it being an imaginary dick measuring context.

"total performance of a single core" is never, ever relevant. All that matters is per thread performance.

Don't have a cry at me when others are literally having arguments about which core and core arch is better and has more performance than another. Total performance is certainly relevant, just outside of a single thread benchmark. I've also yet to really see many applications today be well and truly only single thread as well, this is actually where HT/SMT can have effects that go both ways performance wise.

 

10 minutes ago, LAwLz said:

Stop reading words literally. Saying "single core performance" when referring to "single thread performance" instead of "total performance when loaded with all possible threads" is a perfectly valid use of the word. It's not "incorrect".

As above it is when used how I said I've seen it being used, not how you are using it right now.

 

10 minutes ago, LAwLz said:

"Total performance when loaded with all possible threads on a single core" is a useless metric anyway, hence why it is not used.

No it really is not, a thread should be scheduled on to the same core where it make sense for cache reasons or simply because it doesn't need the extra performance or to tie up CPU time on another core.

 

For me this metric is hugely important for VM hosting, if not one of the most important.

Edited November 15, 2021 by leadeater

[mr moose]

5 minutes ago, LAwLz said:

I don't think you understood what I said, because the other part of your post goes against what I said.

 

 

They aren't really designed to run 2 threads per core. It's more like SMT was tacked on to make up for poor design decisions. It's not really anything inherent to x86 either. Apple could implement SMT too if they wanted in their ARM cores. 

 

 

1) It isn't.

2) Even if it was, programs don't care. It's not like a single threaded program will go "oh sorry, I'll become a multi-threaded program" when executed on an x86 SMT enabled core. 

 

 

I assume that when you say "a benchmark like this" you're talking about the "single core" score in Geekbench. That part is extremely relevant pretty much all the time because a huge portion of programs actually do run into the "there are more cores than threads" scenario. Single-threaded programs, or programs that rely on just a few number of cores (2-4) are very, very common. If they weren't then CPU manufacturers wouldn't bother making their cores larger and larger. They would just have glued 32 Atom cores together a long time ago.

And that's where I disagree with the whole "x86 cores are underutilized in single threaded tests so it doesn't matter" argument you tried to pull. Even if we ignore the whole nitpicking of what is "single core" vs "single thread" performance (spoiler: single core means single thread) it still wouldn't matter because programs don't care. If a program can only use 1 thread efficiently then it's not like running on an SMT enabled core will have an effect on performance, despite the core maybe being "underutilized".

 

I've seen plenty of people make the same argument as you have when talking about Firestorm's amazing single core performance, where they go "b-b-but the x86 core is underutilized because it has SMT so therefore single core performance is higher!". It's like listening to some fanboy of some car complain about how a race was "unfair" because it had a lot of curves and "if the track was more straight then my favorite car would have won for sure!".

 

 

 

The whole "total performance of a single core i.e. 2 threads running on a single core" is a completely and utterly pointless metric though. That's what I said about it being an imaginary dick measuring context.

"total performance of a single core" is never, ever relevant. All that matters is per thread performance.

 

 

Stop reading words literally. Saying "single core performance" when referring to "single thread performance" instead of "total performance when loaded with all possible threads" is a perfectly valid use of the word. It's not "incorrect".

"Single core performance" means "single thread performance". That's how it is always used.

 

"Total performance when loaded with all possible threads on a single core" is a useless metric anyway, hence why it is not used.

I think you might be taking what I am saying way to far,  can you compare a CPU that is optimized for HT (2T1C) to a CPU that only has 1Tper core?  I don't thinks so, not in a "this product is better than that product" kind of way,  God knows what other threads are being seeded in the HT core that effect the test and power usage.  

 

Also the tremont Atoms had 24 cores.