苹果M4性能分析:尽力了,但芯片工艺快到头了!
Summary
TLDR意外的是,M4在M3发布后仅半年就已到来,并首先在iPad上发布。虽然外观相似,但经过测试,M4显示了其作为新一代处理器的实力。M4采用了TSMC的N3E工艺,性能和能效显著提升,CPU峰值频率高达4.5GHz,小核心频率也达到了2.88GHz。通过SPEC2017和GeekBench的测试,M4在单核和多核性能上大幅超越M3和现有的14900K处理器。尽管如此,M4的能效改进有限,但在游戏和日常使用中的表现依然令人印象深刻。
Takeaways
- 🚀 M4 已在 M3 发布仅半年后推出,并首先在 iPad 上发布。
- 🔍 M4 采用台积电第二代 3nm 工艺 (N3E),表现出色,证明值得拥有新名称。
- 💡 新架构和高频率使 M4 的单核频率可达 4.5GHz,小核频率可达 2.88GHz。
- ⚡ M4 的 CPU 部分在架构上有显著改进,前端解码能力更强,指令吞吐量提升。
- 🧠 M4 引入 SME 单元,进一步提高 AI 应用性能。
- 📈 实测显示,M4 的大核性能比 M3 提升了约 20%,但功耗也相应增加了 60%。
- 🔥 在单核性能方面,M4 的分数比 14900K 高 23%,表明其强大性能。
- 🕹 M4 在 iPad Pro 上的游戏性能优异,如《原神》可以稳定在 60fps,并且功耗较低。
- 🧊 M4 的散热能力有所提升,尤其是 13 英寸版的散热表现显著改善。
- 🔋 在电池续航方面,M4 iPad Pro 的表现令人惊喜,尤其是 13 英寸版的电池续航时间几乎是 M2 的两倍。
Q & A
为什么M4处理器仅在M3推出半年后就发布了?
-M4处理器在M3推出仅半年后发布,可能是为了展示苹果在研发效率上的提升,同时展示其在处理器技术上的快速进步。
M4处理器首次在哪个设备上发布?
-M4处理器首次在iPad上发布。
M4处理器相比M3有何主要改进?
-M4处理器在架构上进行了调整,CPU大核心解码单元从M3的9宽度扩展到10宽度,提升了指令并行处理能力。此外,M4还引入了SME单元,提高了在支持SME程序中的性能。
M4处理器的最高频率是多少?
-M4处理器的单核负载下最高频率可以达到4.5GHz,而小核心的频率也达到了2.88GHz。
M4处理器在NPU性能方面有什么显著优势?
-苹果在发布会上强调M4的NPU性能优于目前市场上的所有“AIPC”,并且首次广泛提及“AI”,暗示其在AI应用方面有显著优势。
M4处理器的内存选择是什么?
-M4处理器使用了更高频率的LPDDR5-7500内存,而不是预期的LPDDR5X,这主要是为了提升内存带宽,支持AI应用和GPU性能。
M4处理器的微架构有什么变化?
-M4处理器在微架构上进一步扩展了解码单元的宽度,并增加了Dispatch Buffer和浮点单元的调度队列,提高了核心的并行处理能力和指令吞吐量。
在SPEC2017测试中,M4处理器的性能提升了多少?
-在SPEC2017测试中,M4处理器的大核心性能相比M3提升了约20%,单核性能提升约为8.5%。
M4处理器在实际使用中的能效表现如何?
-M4处理器在实际使用中,相比M3能效提升有限,主要在高频率下有显著性能提升,但相应的功耗也大幅增加。
M4处理器在游戏性能和电池续航方面表现如何?
-在游戏性能方面,M4处理器的iPad Pro在高分辨率下能够保持较高的帧率,并且在能效方面表现优于M2。电池续航也有显著改善,尤其是13英寸版本的续航时间几乎翻倍。
Outlines
🚀 M4处理器的突破性进展
M4处理器在M3发布半年后意外推出,并首次搭载在iPad上。经过测试,M4不仅在名字上有所更新,还在性能上展现了显著提升。尽管局限于iPad Pro的体积,但其架构、性能和能效比依然值得关注。本次测试旨在探究M4的新架构以及苹果今年的性能极限,同时作为全球首款使用N3E工艺的芯片,其能耗表现将为年底的手机芯片发展提供预期。视频提供了大量实际测量数据,包括M4架构和新特性的详细分析、能耗性能、基准测试、游戏性能以及电池寿命测试。
🔍 M4架构和性能的深入分析
M4采用TSMC第二代3nm工艺,即N3E工艺,理论上应比M3的N3B工艺有更高的性能极限。CPU部分,苹果在发布会上提到了大核心架构的改进,M4的大核心解码单元宽度从M3的9增至10,意味着每个时钟周期可以解码10条指令。此外,M4还增加了Dispatch Buffer和浮点单元的调度队列,提高了内核的并行性。M4的ROB和PRRT深度也是所有当前处理器架构中最大的。GPU方面,M4基本是M3 GPU的小幅优化版本,频率从1.34GHz提升至1.47GHz。NPU性能大幅提升,内存也升级为更高频率的LPDDR5-7500,主要是为AI应用做准备。
📈 M4性能测试与能效比探究
M4的性能测试显示,其大核心在4.5GHz时的性能比M3高出近20%,但这也带来了更高的能耗。在被动散热条件下,M4的性能和能耗比仅略有提升。通过Xcode模拟的轻度降频条件下,M4的性能比M3高出约8.5%,能耗高出约5%。新工艺主要提升了高频运行能力,但并未显著提升能效。在相同频率下,M4的新架构相较于M3的大核心,整数IPC提高了约7.3%,浮点性能提高了约8.6%,平均IPC提升接近8%。
🎮 M4在游戏性能上的表现
M4在游戏性能上表现出色,尤其是在Genshin Impact和Honkai Impact: Star Rail等游戏中,M4 iPad Pro能够以高分辨率稳定运行,帧率和能效表现均优于M2。在Genshin Impact中,M4能够以60fps的帧率稳定运行,而在Honkai Impact: Star Rail中,尽管面临更高压力,M4的性能仍然强劲。此外,M4在3A游戏如Resident Evil 8中也展现出接近50帧的运行能力,尽管未能达到60fps。
🔥 M4 iPad Pro的散热和电池续航测试
M4 iPad Pro在散热方面进行了改进,11英寸和13英寸版本的持续散热能力分别提升至约13W和14W,相比前代产品有显著提高。在电池续航测试中,M4 iPad Pro的表现出乎意料,尤其是13英寸版本,电池续航几乎翻倍,11英寸版本也有显著提升。OLED屏幕的使用大幅降低了整机功耗,加上芯片的能效改进,使得M4 iPad Pro的电池续航得到显著提升。
🤔 对M4及未来iPad Pro发展的思考
尽管M4在技术上达到了行业最先进水平,但随着半导体技术的进步,性能提升的步伐似乎正在放缓。M4追求极致性能的同时,也让人思考苹果的M系列处理器是否会重蹈Intel的覆辙,从高能效比走向高功耗。M4对于Mac平台而言,其性能足以推动未来MacBook Pro达到强劲水平,但对于iPad Pro,M4可能只是意味着“多用几年”。作者对苹果未来在iPad散热和性能优化方面的举措表示期待。
📅 对WWDC及iPad Pro未来的展望
随着WWDC的临近,作者希望iPad Pro能够充分利用M4的性能,而不仅仅是为了延长使用年限。节目制作不易,呼吁观众点赞、关注频道,并预告了百万粉丝福利。
Mindmap
Keywords
💡M4处理器
💡N3E工艺
💡大核架构
💡LPDDR5-7500内存
💡能效表现
💡SPEC2017测试
💡解码单元
💡前端和后端处理能力
💡SME单元
💡游戏性能
Highlights
M4处理器在发布仅半年后就推出,并首先在iPad上发布。
M4的架构、性能和能效得到了显著提升,值得一个新名称。
M4的峰值频率在被动冷却设备中达到了前所未有的高度,单核负载下甚至能运行到4.5GHz。
小核心频率达到了2.88GHz,达到了英特尔当前一代Core Ultra 5的水平。
M4引入了SME单元,这可以显著提升支持SME的程序性能。
与M3相比,M4在SPEC2017测试中的整数和浮点性能分别提高了18%和26%。
M4在能效方面表现出色,尽管高峰时的功耗较高,但其整体能效仍有所提升。
M4的内存延迟控制达到了88ns,相比前代M3的96ns有了显著改善。
M4在GeekBench 6.3中的单核得分接近4000分,比M3高出26%。
M4的多核得分在GeekBench 5中接近15000分,显示出其多核性能的强劲。
M4的能效提升主要来自新增的小核架构和技术。
M4的GPU性能提升了10%,但能效改进不明显。
M4 iPad Pro在游戏《原神》中表现出色,平均帧率达到了60fps,能效优于M2。
M4 iPad Pro的续航时间显著提高,13英寸版本的续航时间几乎翻倍。
M4 iPad Pro的散热性能得到了改进,机身表面温度比上一代显著降低。
Transcripts
Unexpectedly,
the M4 has already arrived just half a year after the launch of the M3!
And the most outrageous thing is that it was first launched on the iPad.
I thought that as a new processor launched only half a year ago,
it might just be a version of M3 with enhanced NPU and N3E process.
But in fact, after these few days of testing,
M4 has already shown us It proves that it deserves a new name.
Although it is still confined to the small body of iPad Pro, this
does not affect us through various methods to understand its architecture, performance and energy efficiency.
This time we test M4, we mainly want to understand two The
first thing is to test its new architecture
to see what Apple's performance limit is this year,
including the potential of new Macs and even iPhones in the future.
The second thing is as the world's first N3E process chip,
how will its energy consumption
perform ?
All mobile phone chips will be manufactured using the N3E process,
which can also provide an expectation for the progress of mobile phone chips at the end of this year.
Today we have brought you a large number of actual measurements,
not only a detailed analysis of the M4's architecture and new features
, but also detailed energy consumption. Performance, benchmark tests
, and gaming performance and battery life tests of the iPad equipped with the M4 chip.
The amount of information in this video is absolutely explosive.
Trust me, you will never be late for a good meal!
So without further ado, let’s get started!
First of all, let me briefly understand what the new M4 is.
Apple said at the press conference that this processor
is manufactured using TSMC’s second-generation 3nm process.
This is actually the N3E process,
and it has also become the first processor to use N3E. In theory , the processor manufactured by the process
should have a higher performance limit than the N3B process used in the M3. In
the CPU part,
Apple once again mentioned the large-core architecture improvement at the press conference.
I thought it was just 80% wider than the M2.
It is the vest of the M3.
However, you will see from the subsequent architectural analysis
that this is not the case.
The M4 has made another adjustment to the micro-architecture.
Considering that it is only half a year apart from the launch of the M3,
the R&D efficiency is still something.
And this time The peak frequency is also unprecedentedly high among passive cooling devices.
We actually measured that it can even run up to 4.5GHz under single-core load!
What’s even more outrageous is that the small core actually ran to 2.88GHz.
This frequency has reached the level of Intel’s current generation Core Ultra 5.
It is a bit outrageous that this CPU appears on the passively cooled iPad Pro.
This frequency is indeed unprecedented
for the iPad. In terms of
GPU,
it is basically a slightly optimized version of the M3 GPU.
The frequency is increased from 1.34GHz to 1.47GHz.
Maybe Apple will make some small optimizations to it
, but in theory there should be no obvious difference in the same-frequency performance.
On the other hand, the NPU This is the first time that
Apple has mentioned "AI" extensively in its press conference,
and it is full of gunpowder, claiming that its NPU performance
is better than all the "AIPC" currently on the market.
This is a mockery of who should not use me more. Having said that
, there is one more thing that was not mentioned at the press conference.
The M4 has been replaced with higher frequency LPDDR5-7500 memory.
This should be mainly to prepare for AI applications
. After all, memory bandwidth is important for both NPU and GPU,
but it is worth it. What I mentioned is that
it uses not the LPDDR5X memory I guessed after the press conference,
but a higher frequency LPDDR5 memory.
This is most likely for latency considerations.
We will also test whether its memory latency performance has improved.
Next, we will Let’s talk about the micro-architecture of the CPU.
This time I still want to thank @Junjie1475 for his help.
This testing process can be said to be quite tortuous.
Next, let’s compare the micro-architecture of the three generations of M2, M3 and M4
to see what has been changed this time. What?
We talked about A17 Pro and M3 before. It can be said that
Apple’s micro-architecture has changed the most after A14 and M1.
It is also the most powerful one on the market.
This time, the large core of M4
is based on the previous excellent architecture. Another evolution has been made.
Compared with the large core of M3,
the large core decoding unit of M4 has been widened again.
This time, the width of the decoding unit has been expanded from 9 width of M3 to 10 width, which
means that 10 instructions can be decoded in each clock cycle.
We know that Apple has maintained the 8-wide decoding unit design from the A14 to the A16 three-generation architecture.
As a result, from the A17 Pro to the M4 in just over half a year,
the strategy of widening the decoding unit architecture
twice in succession is really quite radical this time
. A wide front-end decoding unit generally corresponds to a stronger back-end processing capability.
We measured that
the number of back-end execution units has not actually increased.
Perhaps the front-end improvements of the previous generation are not enough to feed the significantly increased back-end scale
this time M4 The new architecture significantly increases the Dispatch Buffer
and the Scheduler scheduling queue of the floating point unit.
Such improvements can further improve the parallelism of the kernel.
The increase in the scheduling queue means that more instructions can be executed in parallel
, which means the utilization efficiency of the processor back-end. It can be even higher. Together
with the improvements in the front-end decoding part mentioned earlier,
the instruction throughput of the entire processor can be further improved.
The further broadening of the new architecture is also reflected in the ROB and PRRT depth of
Apple’s super-large core. It can be said to be the largest
among all current processor architectures.
This also makes me look forward to its actual performance.
In addition to the wider micro-architecture,
this time M4 also introduces the SME unit.
You can understand it simply. For the Arm version of AVX512,
M4's P core and E core cluster each have an SME unit,
which shares the L2 cache with the CPU core.
M4 will have a very large performance improvement
in programs that support SME.
You have also seen that
GeekBench 6.3 introduced support for SME. After that,
the running scores of M4 increased significantly.
We also briefly tested the instruction throughput of P core and E core respectively. The
addition of SME support should be to
accelerate future AI applications.
The ARM public version kernel at the end of the year will also be like M4.
To provide support for SME 2,
we also measured the branch prediction error rate.
From the results,
there should be no improvement in the branch prediction ability of M4
. Generally speaking,
the larger core of M4 once again improved the front-end decoding ability
on the basis of M3. You can say that the back-end parallelism
further taps into the potential of the micro-architecture design of the A17 Pro and M3 generations.
Its design goal should be to make
the back-end execution units added in the A17 Pro and M3 as powerful as possible.
We also tested it. From the memory and cache performance of M4
, we can see that the memory delay of M4 is significantly lower than that of M3.
The delay control has reached a level of 88ns,
which is improved compared to about 96ns of the previous generation M3.
This should be the main reason for using LPDDR5-7500 instead of LPDDR5X.
As for the small core,
the small core of the M4 still uses the micro-architecture of the A17 Pro and the M3 small core
, but the frequency is higher.
This is the complete architecture diagram of the M4 small core.
Interested students can also take screenshots to study and study by themselves.
After understanding the architecture,
we will It’s time to invite the industry-recognized cross-platform test standard SPEC2017
to visually see the performance improvement.
In this SPEC2017,
we still measured the accurate core power consumption and real-time core frequency
. M4 will inevitably be affected by the impact of placing it in an iPad. The impact of passive cooling and frequency reduction
, but M4 is not essentially a mobile SoC.
Its packaging method, memory structure, frequency strategy, etc. are
all quite different from mobile phone processors.
Obviously, the peak frequency of 4.5GHz is not for the iPad at all. So
first of all, in order to measure its peak performance,
we will use liquid nitrogen cooling
to reduce the chip surface temperature below 0 degrees
to speculate on how it will perform on the actively cooled Mac platform.
This will also give us a preview of the M4. Pro and M4 Max
. Obviously, our channel is
more concerned about the technical level of the M4 chip itself than the iPad itself
. In order to have a more comprehensive view, the
extreme peak value must be measured
, but we will also show the performance of M4 on the iPad
and To compare the energy efficiency with M3
, we also introduced a new test this time
we
can
use
The high-frequency performance is naturally indistinguishable.
In terms of absolute performance,
the peak performance of the large core of M4 is almost 20% higher than that of the large core of M3!
An absolute performance improvement of 20% in half a year!
This CPU should have
the largest single-core performance improvement since the launch of M1 until today.
However, the super performance also brings significantly higher power consumption.
When both run at peak performance,
the large core of M4 is only the core function. The consumption is already 60% higher than that of M3!
Using 60% of the peak power consumption for 20% of the peak performance
is a bit like Intel's,
or it's a bit like AMD's PBO idea
of greatly increasing the power budget to increase the processor's ultimate single-core performance.
This is similar to the logic of Apple's M1 M2 It's completely different
. This kind of gameplay is reasonable for PCs
, but obviously not suitable for mobile phones
. So if you want to compare the M4 and mobile phones to compare the single-core power consumption,
it may be difficult to find a reason,
because the competitors of this thing are actually
Rather than finding a sweet spot for a single-core frequency PC , Core Ultra and Ryzen
actually need to maximize the single-threaded performance within the power budget.
To be honest, this seems to be contrary to Apple’s philosophy in the past two years.
Do you still remember? In 2022, we interviewed the big guys who developed Apple Silicon
and asked them why they didn’t overclock?
Why not overclock?
Did they know how to talk at that time?
The implication is that we are different from Intel and AMD.
We don’t overclock and have the
highest energy efficiency.
The result is now really good, right?
Considering the performance requirements on the Mac,
it is definitely good to push the extreme performance up,
as long as it does not affect the mid- and low-frequency energy efficiency.
Of course, the extreme performance of the M4 now installed on the iPad is
not easy to measure.
In fact, if you use it at room temperature If tested,
its frequency strategy will be much more conservative.
The turbo frequency strategy of the M4 we actually tested is very similar to the strategies of Intel and AMD in every sense.
The maximum frequency can reach 4.5GHz in single thread at low temperature
, while it is 4.4GHz at normal temperature or dual thread.
If there are more than 3 threads, it will be reduced to the full-core 3.94GHz
, which is a frequency near 4GHz.
Like other PC processors,
each large core can run to the highest single-core turbo frequency
, instead of having only one core to run relatively fast like a mobile phone CPU. Other high-frequency
cores can only run in a relatively low-frequency state. So what
if we consider the performance and power consumption of M4 at room temperature
?
It just so happens that we can simulate a mild downclocking condition for the iPad Pro through Xcode.
In this case, its frequency will reach a level similar to that of the M3.
Basically, it will run at an average frequency of around 4GHz. At
this time, the performance of the M4 It is roughly 8.5% higher than M3,
and the power consumption is about 5% higher than M3.
From this result, it can be seen that
the energy consumption ratio of M4's large core has only a very small improvement. It may
be that the performance improvement brought by the architecture offsets the energy efficiency advantage of the new process.
The new process mainly improves the ability to run high frequencies
but does not significantly improve energy efficiency.
This will have to be verified later when we measure the GPU power consumption.
Then we will compare the same-frequency performance of the new architecture with
the large core of M4. The IPC is about 7.3% higher than the M3 integer
and the floating point is about 8.6% higher.
On average, the IPC increase this time is close to 8%.
Although there is still no double-digit increase
, this is already the increase in Apple's IPC
since the launch of the A14 (M1). The biggest one is that
it once again consolidated Apple's leading position in processor architecture. However
, the M4 using N3E did not surprise us
in terms of large-core energy efficiency.
Has the energy efficiency of N3E really improved?
It seems that there is still a lot of suspense about the A18 Pro at the end of the year. So
what is the peak single-core performance of M4's new large core compared with the current strongest 14900K in the x86 camp ?
The M4's int score at 4.5GHz
is about 18% higher than the 14900K SPEC2017 int score
at 6GHz
, and the fp score is even 26% higher!
Taken together,
the single-core performance of M4 is already 23% higher than that of 14900K!
If the mission of M3 is to equal the single-core performance of 14900K,
then the mission of M4 this time is to completely crush 14900K.
I am really curious about what Intel will do this year
to improve single-core performance by 23%
to regain the top spot. The throne of strong single-core performance.
When I tested the M1 back then, I said that
Apple’s architecture was very strong,
but it couldn’t run high frequencies
, so the single-core performance was still outperformed by the 12900K back
then. However, times have changed
and the IPC is higher and it can run high frequencies. The opponents that M4 faces are the 14900K, which
is stuck in the same place as IPC
and can only keep overclocking to maintain its appearance.
Even Intel recently was forced to downclock in the latest firmware
due to excessive overclocking.
To be honest, I am looking forward to AMD Zen5 now,
which is simply an x86 village. There is hope,
come on AMD!
After talking about the super-powerful big core
, let’s take a look at the performance of the small core.
Since the small core architecture has not changed
, its performance only comes from the increase in frequency.
Small cores with this frequency are unlikely to appear on mobile phones
. It is a bold guess that the use of 6 small cores in the M4 this time may mean that
the core clusters of 6 small cores introduced by Apple starting from the M3 Pro
may become the mainstream of Apple processor design in the future.
Does this mean that we will see it in the future? What about the A-series processors with 2+6 core design?
We will wait and see
to summarize the results of SPEC2017.
M4, a new architecture and high-frequency large core, has run a very outrageous performance.
The peak performance is 23% higher than 14900K.
It makes people very curious what will happen if M4 gets rid of the iPad body
and comes to the Mac platform. What kind of performance has exploded?
It is quite unexpected that the IPC has increased by 8.6% only half a year after the launch of M3.
This also makes people look forward to how powerful the future M4 Pro, M4 Max
and even M4 Ultra will be.
Obviously Mac is the M4. The real stage of the family,
the M4 on the iPad Pro, can only be regarded as an appetizer.
However, I
actually have no confidence in the performance of this new core on the A-series processors of future mobile phones
because of the same-frequency energy consumption ratio of N3E. It is only about 3% stronger than the large core of M3.
It seems that the energy consumption ratio of N3E is not as great as expected.
Is human technology really locked by sophons?
A larger core will definitely bring higher power consumption
, so for mobile phones
, adding more small cores to improve the multi-core energy consumption ratio is the real solution.
Dimensity 9300 and Snapdragon 8Gen3 have already proved this.
In fact, the M series has been tested I have always resisted using GeekBench for processors
. After all, we have already tested single-core SPEC and
multi-core. You can easily compare performance and power consumption by running Cinebench directly on Mac. It is
also very convenient to compare Intel and AMD CPUs.
However, Apple This time I actually put M4 into the iPad first.
Well, let’s run GeekBench to see
if M4 is added to the Cyber Fighting Cricket package of mobile processors
to see the difference in performance and power consumption between PC processors and mobile phone processors. How big!
This time we not only tested the full-blooded M4
but also the 3+6-core three-cylinder Youth Edition M4 in the low-end version.
I must see how much performance Apple has cut off with this move
and whether the gap in energy efficiency between them is big.
First, let’s take a look at the peak performance of Geekbench 6 under forced cooling.
The single-core of M4 ran an outrageous score of nearly 4,000 points.
This score seems to be 26% higher than that of M3.
It is also better than the Qualcomm X Elite, which has not been released yet. The official score
is about 1000 points higher
, but such a high score is actually because
GeekBench 6.3 has added SME support
, so M4 will have obvious advantages in the new version of GeekBench 6.
I think the score of GeekBench 6 can only be said to be for reference only.
Maybe the SPEC2017 results are more universal
in terms of multi-core.
The full-blooded version of the M4 reached close to 15,000 points, with a score of
3+6. The residual-blooded version of the M4 is a little lower,
around 13,500.
But even the residual-blooded version of the M4
is still better than the M3. The performance of
the full-blooded version is even higher than that of the M3 Pro.
In GeekBench 5, which has not yet been optimized for SME,
the performance of the M4 is still very strong.
The single-core score of about 2750 points is about 17% higher than that of the M3.
This improvement is actually also It is very close to the previous improvement in SPEC2017.
In terms of multi-core, the full-blooded version of M4 easily surpasses M3 by 25% with its higher specifications,
higher frequency, and higher IPC.
Even the residual-blooded version can surpass M3 by 10%.
If you want to further compare with the previous generation. The gap between single-core and multi-core iPad Pro M2
has widened to nearly 50%.
This improvement is indeed a bit large.
If we compare the GeekBench 5 results
with mainstream notebook processors in the Windows camp
, the multi-core results of M4 are closer to the power consumption. The fully-operated Core Ultra 9 185H
and i9 13900H
are a little better than the Ryzen 9 8945HS,
not to mention its single-core performance.
There is currently no opponent that can come close.
Of course, there are limitations when using GeekBench to run PC chips.
Wait until the M4 is really After getting a Mac,
we will also run Cinebench 2024
to see which Windows notebook processor it is closer to.
The peak performance of the M4 looks really strong.
What about its power consumption?
Hehe, in the peak state,
the full-blooded version of the M4 has a power consumption of 30W in GeekBench 5. This
power consumption is slightly higher than that of the M3 and M2. The
residual-health version has a power consumption of 25W.
This power consumption is closer to that of the M3 and M2.
Although the power consumption is high, we have already said that we
should not just look at the limit but also run the power saving mode
to judge the approximate energy efficiency performance.
In the power saving mode,
the performance and power consumption of the full-blooded M4 are better than those of the M3. It is higher
, but the performance of the residual version of the M4 is a little lower than that of the M3
, and the power consumption is a little
higher. The frequency set by the power saving mode of the M2 is more aggressive. The
overall energy efficiency is probably like this.
There is still some improvement in the energy efficiency of the M4
, but it still needs to be Thanks for adding these two small cores.
If it were not for the size of the heap,
there might not really be any improvement in multi-core energy efficiency.
The 3-cylinder version of the M4 can already illustrate this problem.
Of course, we can also add the energy efficiency of current flagship phones to compare
this. The trend of the curve is completely different from that of mobile phones.
The energy efficiency of M4 is really different from the performance of mobile phones. The high
-performance M series will undoubtedly beat mobile phones
, but in the power saving mode, the energy efficiency will be close to that of mobile phones.
You can see 3+6 The M4 has almost reached the 8Gen3 curve
in the power saving mode .
If the energy efficiency goes to low frequencies, it may be overtaken by mobile phone chips.
After all, the basic power consumption of the M series is higher than that of mobile phones.
But there is a saying that more of them run at low frequencies. The core of low and high IPC
is indeed the only way to improve multi-core energy efficiency.
To summarize the performance of the M4 CPU part,
the 30W peak motherboard power consumption
is exchanged for
multi-core performance at the same level as Intel Ultra 9 AMD Ryzen 9 notebooks.
In other words, the current iPad
The
peak CPU performance of the Pro
is even close to
that of the ROG Magic
To make a specification of 12 large cores and 6 small cores
, how strong is its performance?
Can the x86 camp really come up with a notebook processor that can cope with it
? But on the other hand,
energy efficiency is what these mobile processors should care about most. But
according to the current test,
the energy efficiency improvement of M4 basically comes from the two newly added small core
architectures and technology.
It seems that they have not brought better energy efficiency.
Let’s talk about the GPU.
I also talked about
the M4 earlier. The GPU does not bring anything new.
The 10% frequency increase also brings a 10% performance increase.
There is nothing worth talking about.
If you want to know more about the details and performance of this GPU,
you can check out our review of the M3. At the time of the detailed evaluation,
I had done enough tests on this GPU.
Compared with performance
GPUs, maintaining the same architecture allows us to see more intuitively
whether the N3E process energy consumption ratio will be better than the previous generation.
In fact, the peak power of the M4 GPU Compared with the M3, the power consumption has indeed increased significantly.
Similar to the CPU,
it has also increased from about 25W to about 30W.
If we control the M4 to around 8300 points, which is close to the M3,
their power consumption will basically maintain the same level.
There are no surprises when it comes to the high-frequency energy efficiency of N3E.
What about the low-frequency energy efficiency?
Both GPUs run in power-saving mode.
The M4 GPU’s performance is slightly higher than that of the M3 and its
power consumption is slightly lower
, but the two points are actually very close.
It can be said that
there is indeed a slight improvement in energy efficiency
at low frequencies , but it is really only Just a little bit
, so the answer is very clear now that
the N3E process has not brought any significant improvement in energy consumption compared to the N3B process. Abstractly,
the
previous generation of N3B has not actually improved the energy consumption of the previous generation of N4P.
In fact, this situation is really a loss for Apple.
It spends a lot of money to use the most advanced process
but cannot get the energy efficiency buff as before.
If you look forward to what new developments the N3E process
will bring to mobile phone processors at the end of the year. If so
, you may be disappointed
. It is more realistic to look forward to the architectural improvements of each company.
Of course, I also tested a very interesting thing this time.
If the M4 GPU is downclocked to the performance level of a mobile phone,
its power consumption will be How about it?
The power consumption of the M4 GPU running WLE is only a little over 6W,
which is close to the 4400 points of the iPhone.
Obviously, this power consumption is much lower than that of the A17 Pro.
In this case, it is very easy to beat all mobile phone chips in terms of energy efficiency.
It seems that the A17 Pro The size of the GPU is obviously too small.
How can the energy consumption performance be better than that of Qualcomm and Fa Ge?
Everyone is really piling up specifications,
so can Apple
build a larger-scale GPU for the A-series processors?
On the mobile side, a large-scale GPU running at low frequency
is a good civilization
. By now,
you should have a basic understanding of the performance of the M4
, but we have not talked about the iPad so far,
so we have to focus on the next test. Back to the iPad Pro
, let’s start with the game
. First, let’s take a look at Genshin Impact. It’s very rare that
Genshin Impact was adapted on the day the new iPad was released
. We actually counted
the resolution of the 13-inch model as high as 2292x1718
and the 11-inch model. The rendering pressure
is about 2292x1600 , which is the same as 2560x1440.
The 13-inch version is slightly higher.
Can the iPad Pro of
Xumimap M4 hold up at such a high resolution ?
It turns out that
this pressure is too easy for the M4 iPad Pro.
No matter which one of the three M4 iPad Pros
ran a straight line average frame of 60fps,
combined with the power consumption,
the energy efficiency performance of this game is obviously better than The iPad is much more powerful than the M2
, and the most important thing is that
there is no difference in the frame rate and power consumption between the three-cylinder M4 and the four-cylinder M4 in actual games.
In contrast,
the M2 can only run about 50fps in the second half of the test.
Is it possible? Is Miha lobbying for increased performance pressure in Genshin Impact 5.0
because they have already played the M4 in advance?
In addition, the 13-inch Air of M2
is much better than the previous generation 12.9-inch iPad Pro that uses a mini LED screen.
The frame rate is higher and the average power consumption is much lower.
It seems that the previous generation of mini LED consumes too much power,
including this generation of M4 iPad. The advantage in gaming power consumption
may be largely due to the heat generated by the OLED screen.
It is also full of surprises.
After the test, the surface temperature of several M4 iPads was significantly lower than that of the M2 model.
Looking at the heat distribution,
the heat dissipation of the body has indeed improved.
If you still think it is too hot,
you can turn on the rendering accuracy. At the lowest,
this is almost mobile phone-level picture quality.
In this case, the power consumption of the M4 iPad Pro including the screen will drop to more than 7W.
The maximum temperature is even less than 40 degrees.
It feels like
my hand is heating it when I hold it.
The performance of Genshin Impact in a straight line at 60 frames is too boring
, so we directly used the highest image quality to unlock 120fps
to see if the M4 can still hold up under this situation.
Facts have proved that
the pressure of Genshin Impact at this resolution is still It’s very scary that
the unlocked M4 iPad Pro
can only sustain 120 frames for about 1 minute
, and then it will drop to 80 frames for a period of
15 minutes. After 15 minutes, as the body heat gradually accumulates,
the frame number will slowly drop to 60-80 frames. It fluctuates between them.
It turns out that even M4
is difficult to run Genshin Impact at 120fps at desktop-level resolutions exceeding 2K.
Let’s take a look at the performance of Honkai Impact: Star Rail.
The resolution of Star Rail is the same as Genshin Impact. But
obviously the pressure of Star Rail is still higher.
This time, the three iPad Pros of M4 could not resist the destruction.
The two 11-inch machines started to drop frames about 10 minutes after the test started.
The 13-inch one lasted a little longer,
but it got there. After about 15 minutes, you still have to lower the frequency.
The test results of Star Rail are very easy to see
the improvement of the GPU of M4 compared to M2 in actual games.
You can see that the power consumption of the whole machine is similar to that of
M4 iPad Pro. The average frame rate is basically about 50% higher than that of M2.
It is still very strong.
I also tried Bengtie to unlock 120 frames. The performance
is worthy of Bengtie. The pressure is really ridiculous.
I can only hold it for a few seconds at 120 frames
and then It quickly dropped to 70-80 frames.
Then after the body warmed up,
it couldn't maintain 60 frames in the second half.
The pressure was still too difficult for the M4.
In addition to these two mobile games,
there are already some 3A games ported to PC. Has it been launched on the iPad platform?
Should we also test it?
It is a pity that 3A games such as Resident Evil 4, Death Stranding
and COD War Zone
have not been adapted to the M4 iPad Pro.
The current picture quality is very blurry.
The only one that can run is Resident Evil 8
because Resident Evil 8 can be easily changed to For several other games with the highest quality and native resolution,
even if you change the file, it won't actually take effect.
However, even with Resident Evil 8, which has the highest quality,
we seem to have encountered a card power consumption wall
, and the frame rate has not even reached 60fps
with power consumption. But it is locked at around 10w of the whole machine.
It may be that the game developers have set a power consumption limit
, but this
can ensure that it will not heat up for a long time when playing 3A.
Considering this passive cooling body
and such a low power consumption,
the M4 can It's quite impressive to run to close to 50 frames at a resolution of 2K or above.
Playing these 3A games on the iPad is honestly more like a console than a PC.
Once you download it, you just play with
the image quality options. Normally , these 3A games
are not open. It squeezes out the performance of the iPad.
This is something I didn’t expect before the test.
At the iPad Pro launch conference in London,
Apple rarely mentioned the heat dissipation performance of the iPad.
In the past two days, various disassembly videos have also been released.
You should have also seen that the new iPad
has indeed made some heat dissipation improvements in the body.
So has the actual heat dissipation capability been improved?
For the 11-inch iPad Pro,
the M4 iPad Pro, whether it is a 3+6 or 4+6 model,
basically has a continuous heat dissipation capacity of around 13W.
Compared with the previous generation’s 10W, it is indeed a significant improvement.
The 13-inch version has a huge body.
Therefore, the heat dissipation capacity is not a problem.
The new 13-inch M4 iPad Pro has a heat dissipation capacity of 14W, which
is slightly improved compared to the 13.5W of the previous generation.
The surface temperature of the fuselage is more than one degree lower.
In general,
the M4 iPad Pro does have good heat dissipation.
It’s not as wide-ranging as the iPhone.
So here’s the question:
When will Apple upgrade the heat dissipation for the iPhone?
Aren’t you anxious when you see competing products stacked one after another with vapor chambers
?
So the last and most important question is
how long is the battery life of the M4 iPad Pro?
You must know that in the past, the battery life of the iPad Pro was almost the aspect that users complained about the most.
So the biggest test for the iPad Pro
is actually the battery life
. Considering that the application scenarios we use the tablet are relatively simple but last a long time,
this iPad battery life test It is a cycle every two hours
, including browsing IT Home for 20 minutes,
playing Genshin Impact for 40 minutes, and watching Bilibili videos for 1 hour.
The cycle is measured from full power to no power,
and the screen brightness remains at 300nit.
The test results are very surprising for
the M4 iPad Pro. The battery life has been greatly improved compared to the previous generation!
Especially
the battery life of the 13-inch version is nearly doubled compared to the M2 iPad Pro!
The 11-inch version also has a lot of improvements.
As for the M2 iPad Air,
its battery life is between the M4 iPad Pro and the M2 iPad Pro.
The battery life of this generation of iPad Pro is completely beyond my expectation,
because the battery life calibrated by Apple’s official website has not even changed
. How can it be so much better in actual application testing?
The energy consumption gap between their chips is not that big, right?
In fact, for tablets,
in addition to the energy efficiency of the chip,
the screen power consumption is the biggest factor
that affects battery life
. Let’s do a very simple test. Play Bilibili videos
at 300nit screen brightness to measure the average power consumption.
The results are obvious.
The 12.9-inch iPad of the previous generation M2 The power consumption of Pro for watching videos can be said to be outstanding.
This is the biggest problem of mini LED.
Its power consumption is too high
and because its backlight is dynamically adjusted,
if you often browse the web or type,
there will be a lot of white display area.
Its power consumption will only be higher.
This generation of OLED screens has indeed significantly reduced the power consumption
of the entire machine. Coupled with the energy efficiency improvements of the chip,
the battery life of the M4 iPad Pro has been greatly improved.
After completing the review of the M4, my biggest The feeling is that
Apple has indeed worked hard
, but it seems that our human semiconductor technology is indeed coming to an end.
I still remember that in the 7nm era,
A12, A12X, 865, Zen2, and Zen3
all had huge improvements compared to their previous generations.
A15 and A15 in the 5nm era. 8Gen2 and 4090 also left a deep impression on us.
But after TSMC’s process finally evolved to the 3nm node,
it seems that the two generations of processes left us only disappointed.
As the first N3E processor,
M4 has the most advanced technology in the industry. The processor architecture
of N3E actually optimized the architecture of M3 again after only half a year.
It does allow the new processor to obtain a higher frequency limit
, but this comes at a cost.
You will see that M4 seems to be pursuing extreme performance. With the shadow of Intel
, I still remember that Core 2 swept that era with its unparalleled energy efficiency advantage.
A few years ago, Apple conquered thin and light notebooks with the energy efficiency ratio of M1,
which changed the thinking of the PC industry.
But as time goes
by, The Core 2 with excellent energy consumption ratio
eventually evolved into today's big stove Core i9 14900K.
Will Apple's M series processors follow this path in the future?
I can't predict the future
but I know there is an end to the road to performance improvements through more aggressive power budgets
As far as the M4 is concerned,
I think it reaches the sweet spot frequency of the Mac platform.
It should be enough to push future MacBook Pros to a very strong performance level.
But the power of the M4 may mean the most to the current iPad Pro
just "a few more years of use." "Well,
we are very close to WWDC now.
I hope WWDC can see that the iPad Pro can really take advantage of the performance of the M4.
If it is just for "a few more years of use",
then I don't think the future of the iPad will be brighter with the arrival of the M4. Okay
. That’s all the content of this program.
It’s really not easy for us to do this program.
If you like our program,
don’t forget to press and hold the like button to give us a threesome.
Don’t forget to follow our channel.
We will 400 points immediately. If you are a million fans,
there will be lucky draw benefits for everyone.
See you in the next episode.
Weitere ähnliche Videos ansehen
这才是真AIPC: M3 Max MacBook Pro体验
Transformers: The best idea in AI | Andrej Karpathy and Lex Fridman
一招让你的ChatGPT变聪明|context window原理讲解
Googles GEMINI 1.5 Just Surprised EVERYONE! (GPT-4 Beaten Again) Finally RELEASED!
[ML News] Jamba, CMD-R+, and other new models (yes, I know this is like a week behind 🙃)
OpenAI shocks the world yet again… Sora first look
5.0 / 5 (0 votes)