Since released the performance of MACBOOK Pro equipped with M4 Pro and M4 MAX in October, the performance of the M4 series chipsThe performance has attracted a lot of attention.However, friends who know the Apple product line know that the M4 series also has a hidden BOSS -M4 Ultra has not yet been released.
Although the Ultra model is absent from the previous generation M3 series, the performance of M2 Ultra has left a deep impression on everyone.Make people look forward to the performance of this Apples true flagship chip.
In the calculation of overseas media, if M4 Ultra uses a architectural design similar to M2 Ultra, that is, the two MAX chips are connected through UltraFusion packaging technology, then M4 UltraThe GPU performance should be about 1.7 times the M4 MAX.
combined with the 197754 points that M4 Max scored the GPU test in Geekbench 6, M4 Ultras GPU test score will be around 330000. For comparison, the current consumer level willThe strongest graphics card RTX 4090 score is only 310000-320000.
Source: MAX Tech
M4 Ultra, kill RTX 4090?I am afraid that Lao Huang has to have nightmares when he watched it, but some people have asked: Can the M4 Ultras GPU really have such high performance?Is it really comparable to RTX 4090 in the actual use scenario?
M4 Ultra running score exposure, how strong is it?
M4 Ultra has not yet been released, but we can glimpse the performance of this series of chips from the M4 Pro and M4 MAXs that have been released.If you just look at the test results of GeekBench 6, you will find that the score of M4 Pro is real horrible. In the single -core test, it surpasses the latest desktop flagship processor of AMD and Intel in 3925.It is also obviously leading Ryzen 99950X, only losing slightly in the Core Ultra 9 285K.
Source: MAX Tech
As for the M4 MAX with higher performance positioning, Ryzen 9 9950X and Core Ultra 9 285K are crushed in single -core and multi -core testing, which makes the entire PC market feel the power of Apple.In addition, the core frequency and core number of the M4 series chips in the test are significantly lower than two opponents, only 4.5GHz, but shows higher single -core and multi -core properties.
Stream: Max Tech
Is it terrible?Its terrible.You know, the Core Ultra 9 285K and Ryzen 9950X are desktop processors, with full load power consumption of more than 200W. However, M4 Pro and M4 Max are MacBook Pros mobile chip, which shows that its energy efficiency ratio is higher.
Based on the current performance of the M4 MAX, the M4 Ultra combined with two M4 MAX chips may reach 45,000 points in multi -core performance, which is double the Core Ultra 9 285KEssenceSingle -core performance generally does not improve much, but it will still be higher than M4 MAX, becoming the strongest processor in the current consumer -level market.
M4 Ultra GPU all surpassed Nvidida RTX 4090
Take a look at GPU performance, exposure is called the running branchMore than RTX 4090, and the basis is the GPU test of GeekBench 6. This test involves the five modules of GPU image processing, computing camera, machine learning, physical simulation and computing finance.M4 Ultra more scoring space.
So, combined with the current measured M4 MAX GPU performance data, this speculation is reasonable. In the measurement of Geek Bay, the AI big model of M4 MAXThe computing performance has reached 60%of RTX 4090.Even when the RTX 4090 has plummeted due to explosive existence, it can continue to run a larger number of parameters with the advantages of the 128G unified memory.
Source: Geek Bay
This also means that as long as M4Ultras architecture does not have fundamental changes. At least the performance in AI computing can surpass RTX 4090 and become the most consumer -grade AI GPU.So what about the game?Judging from the current measurement results, the game performance of M4 Ultra must not be compared with the RTX 4090, because even if it is running the native Mac game "The Gate of Bade 3", the number of frames at high resolution is only the mobile version of the RTX 4070.30%higher, it has not yet reached the level of the RTX 4080 mobile version.
In the test of 3DMark, the performance of the RTX 4090 desktop end is about twice the RTX 4080 mobile terminal.It can only reach about 75%of RTX 4090. However, if the RTX 4090 on the mobile terminal is compared, then there is a chance to overtake. After all, the mobile RTX 4090 is essentially the RTX 4080 desktop end, and the gap is still very obvious.
In addition, if it is a non -native game, the performance of the GPU will still be obviously weakened.%, Dont say the challenge RTX 4090 at this time, you can directly consider sitting with RTX 4060.
For friends who want to play games with M4 Ultra, personal suggestions are not as good as using a 10,000 -stage game computer. The experience will be much better than M4 Ultra.But if your needs are used to run local AI models, video editing, model rendering, etc., then the performance of M4 Ultra may exceed the current strongest consumer -level PC.In words, this is not a good news.
"Apple Core" running a record high, ARM is the key
Apples self -developed M seriesThe chip running scores have repeatedly reached a record high, which really made traditional semiconductor manufacturers such as Intel and AMD take a cold sweat, and even had to join forces to set up the X86 architecture ecological group to discuss the future of the X86 architecture.How pressure is ARM.
In fact, many people are also curious, Why can the energy efficiency ratio of the ARM architecture chip be so amazing? In fact, the key is that in the instruction set, ARMs instruction set is much more streamlined than X86.
You can see X86 as a complex multi -functional truck, which can process a large amount of data and tasks at one time, but when facing simple tasks, it is complex and huge.The instruction set also makes it difficult for it to reduce its own resource usage.ARM is a motorcycle. Although only a single task and a small amount of data can be performed at a time, it can distribute the best vehicle of the task through flexible vehicle scheduling, so that the ARM chip can have better energy efficiency at runtime.
Based on the characteristics of the ARM architecture, Apple and other companies soon realized that by increasing the core number (preparing more motorcycles) and optimization instruction sets (special car improvement efficiency), Can make the ARM chip make breakthroughs in performance , the high energy ratio ratio of the ARM instruction set also allows the ARM chip to easily pile a large number of cores without considering the power consumption problem without too much consideration.
In addition, the flexibility of the ARM architecture allows Apple to customize the design of the corresponding task according to the needs, which can further improve the efficiency of the chip, plus the unified memory architecture brought by the unified memory architecture.Low latency characteristics, allow the overall execution efficiency of the M series chip to remote traditional chips.
Source: Apple
Thanks to multiple optimizations, the amount changes eventually caused qualitative changesThe success of the ARM chip has surpassed the traditional X86 chip in terms of single -core and multi -core performance, becoming an emerging force in the PC market, and no longer adheres to the mobile device market.Moreover, Apple quickly found that this routine is equally feasible on the GPU, and the GPU performance also entered the explosive growth by padding the core number.Take M4 Pro and M4 MAX as an example.The above is twice the former.Theoretically, as long as Apple is willing, it can be infinitely stacked with the core of the chips internal delay to create a stronger GPU.
Source: Apple
But Apple itself is not a semiconductor enterprise, and its chip design is more design.It is to serve the product, so it will not design a product that consumes power and performance to "explode" like Nvidia and AMD for selling chips.But Apple does not do so, it does not mean that other companies think so. For example, Qualcomm has been paying attention to the ARM GPU and CPU. In addition to the consumption side, servers and data centers are also the target market of Qualcomm.
In 2024, Qualcomm announced that it returned to the server chip market and launched a server chip based on NUVIA kernel design.The proportion of chips has reached 10%.Many of the AI computing power centers have plans to increase the ARM GPU computing power server, because the high energy efficiency of ARM can significantly reduce the cost of electricity during AI reasoning, thereby reducing the cost of unit computing power.
It is precisely because of the huge threat brought by the ARM architecture. When Intel and AMD signed a cooperation agreement some time ago, Nvidia also arrived, and was also one of the cooperative companies in the ecological group.At present, the best news for companies such as Nivineda is that Apple has no intention of participating in the semiconductor chip market, and bad news is that Qualcomms NUVIA core performance is good, and I am afraid it will become a dark horse in the GPU market.
It is foreseeable that in the future PC market, ARM will inevitably have a place, and it is expected to completely challenge the X86 architecture and Nvida market position.
