In-depth evaluation of Tencent Cloud AMD server and ARM architecture server, which is more cost-effective?
When buying a cloud server, the most habitual action is to close your eyes and choose Intel(x86 architecture). However, in the past two years, if you often visit the selection page of tengxunyun, you will find that two "non-mainstream" figures are becoming more and more active, and the price is often much cheaper than Intel: one is
AMD(x86 architecture)
and the other is
ARM architecture (Xinghai/Ampere instance developed by Tencent Cloud)
.
Many technical executives, architects or individual developers are very excited to see the low price, but they are beating drums in their hearts:
"Will AMD have compatibility issues? Will it be unstable?"
"Aren't ARM architectures for mobile phones? Can I run my code when I move to the server?"
"Which of these two is the real king of cost performance?"
Today's evaluation directly strips off all obscure PPT terminology, without talking nonsense, from
The bottom of the architecture, the actual business, the compatibility pit and the real ledger.
Four dimensions, help you to these two servers completely clear.
Player Disassembly: What are they?
Before entering the cost-performance PK, let's set a position for these two players.
1. AMD instance (representative model: SA3 / SA5)
AMD servers and Intel are the same, they all belong
x86 architecture (complex instruction set)
.
Big vernacular understanding: it is Intel's "brother". The software and Docker image compiled on Intel servers or even Windows computers in your own home can be seamlessly pasted directly into AMD servers without any modification.
Current situation of Tencent Cloud: AMD instances of Tencent Cloud (such as the current main SA3 or the latest SA5) usually focus on high cost performance. With the same number of cores and memory, it tends to be about 15% to 20% cheaper than Intel.
2. ARM instance (representative models: SR1/SR2/Kunpeng/Ampere)
ARM server is another dimension of the product, it belongs
reduced instruction set
.
It is an "enlarged version" of mobile phone chips (such as Apple M series, Huawei Kirin and Qualcomm Xiaolong) on the server side ". It is characterized by extremely low single-core power consumption, very large number of cores, and orderly processing of concurrent requests like queuing.
Tencent Cloud Status: Tencent Cloud has spent a lot of money on ARM in recent years (including self-developed Zixiao, Xinghai and other technologies landing). Under the same calculation power index, the official pricing of ARM instances is usually the lowest in the game, sometimes even 30% cheaper than Intel.
The first round: software ecology and compatibility PK (decide whether you can use it or not)
Conclusion: AMD lies to win, ARM has an invisible time cost.
Talk about cost, can not be separated from the "compatibility". If you want to save hundreds of dollars server fees
, let programmers work overtime to change the code for half a month, then this is called "negative optimization".
AMD:0 Migration Costs
Because AMD is the standard x86 architecture, any compiled binary, old. NET project, closed source commercial software, running on AMD will not have any feeling.
In a word: as long as Intel can run, AMD can run.
ARM: ECOLOGICAL BOOMING, BUT 'FAULT PITCH"
Many people still have the impression of ARM in the past and feel that the ecology is not good. In fact, the current mainstream open source underlying technologies: Linux system (Ubuntu/CentOS), Nginx, MySQL, Redis, Docker, Java (JVM), Go language, Python, have all been
Native perfect support ARM architecture
.
However, if you have the following scenarios, choosing ARM will step into the quagmire:
Old closed-source software: If your company uses commercial software (only compiled x86 firmware) from a supplier that you bought more than a decade ago and has gone out of business, it won't run at all on ARM.
C/C Deep Dependency: If your project contains an underlying C language encryption library or video decoding library that depends on a specific x86 instruction set (such as AVX-512), an error will be reported when migrating to ARM, and you need to find an alternative library and recompile it.
Docker Image: You cannot directly throw the Docker image packaged on the local x86 computer into the ARM server to run. You must use docker buildx to build a multi-architecture image or build it again on the ARM server.
The second round: business scenario performance PK (determines whether you use it well or not)
Conclusion: ARM is selected for high concurrency and lightweight computing; AMD is selected for single-core heavy computing and traditional heavy database.
In order to let you understand their advantage range more intuitively, we divide the business into two categories:
Scenario A: High concurrency, I/O intensive (home of ARM)
For example: Web API, microservice gateway, Redis cache server, message queue (Kafka/RabbitMQ).
These businesses are characterized:
The calculation is not complicated, but there are many requests coming in at the same time.
The physical core of the ARM instance is very solid, and because of its low power consumption, it does not play Intel's common set of "hyper-threading (Hyper-Threading)" tricks. ARM's 4 cores are often 4 real physical cores. When handling such highly concurrent and queued I/O requests, ARM's multi-core efficiency is extremely high, and the delay performance can even lift x86 at the same price.
Scenario B: Single-core recalculation, complex data analysis (AMD's home)
For example: large file compression, complex SQL multi-table off.
Joint query, audio and video transcoding, scientific computing.
What these businesses need is
Pure single-core explosive power (single-core dominant frequency is high, single accounting power is strong)
.
At this point, AMD with a powerful complex instruction set and high frequency advantage, can still maintain the absolute suppression of ARM. AMD may run a complex computing task in 1 second, while ARM may take 1.5 seconds because it needs to be disassembled into more steps because of the reduced instruction set.
Round 3: Real Ledger vs. Price-Performance Actuarial (Deciding Who Saves More Money)
In order to let you see the "real routine" of tengxunyun, we directly take the official standard pricing of tengxunyun (in the classic
4 core 8G specifications
For example) to calculate an account:
Schema/Instance Type
Tencent Cloud Representative Model
Price Factor (100% Intel basis)
Advantage interval
Intel Architecture
Standard S6
100% (most expensive)
Companies with closed-source old systems and extreme beliefs in stability
AMD Architecture
Standard SA3 / SA5
Approximately 80% - 85%
The vast majority of mainstream Internet services, MySQL, Java backend
ARM architecture
Universal SR1 / SR2
Approximately 65% - 70%
Microservices, Redis, front-end lightweight applications, large-scale clusters
From the books,
ARM's price advantage is rolled
. If your enterprise needs to deploy 100 servers in batches to run distributed microservices, replace these servers from Intel to ARM, and the expenses saved in one year can directly recruit one more senior engineer.
However, AMD by virtue
20% off price +100% compatibility
, became the most worry-free one in the price-performance map.
The Ultimate Selection Guide: What Should You Choose?
There is no need to struggle, just check the following "check-in" list:
Don't hesitate to choose [AMD] scenario:
You don't want to toss: you are a personal webmaster, or the company's technical team has only two or three people, and you don't want to spend any time troubleshooting the architecture problem of "why this software reports an error on the server.
Core database: Your server is mainly used to run complex traditional relational databases (such as Oracle, large SQL Server, complex enterprise MySQL).
Refactoring of old projects: The old code passed down a few years ago is maintained. No one in the team can guarantee that there will be metaphysical bugs if this pile of code is replaced by ARM.
The scene of decisively embracing [ARM]:
Modern cloud native applications: your business is a microservice developed in Go language, Java, Node.js, Python, and runs all the way in Kubernetes(K8s) or Docker containers.
Large-scale access layer: used for Nginx load balancing, API gateway, or front-end static resource distribution.
Large factory or budget-sensitive team: the team has enough research and development capabilities to do multi-architecture compilation (CI/CD has run through multi-architecture packaging), and server costs have become a major pain point in the company's finances.
Summary
In the ecology of Tengxun Cloud,
ARM architecture server is a well-deserved "king of absolute cost performance"
, it provides extremely horrible multi-core concurrent performance at a price of nearly 60%, which is suitable for teams with certain technical tossing ability and modern development route.
And
AMD server is "the king of cost performance with your eyes closed"
. Although it is not as cheap as ARM, it reduces the "compatibility risk" to absolute zero with a 20% discount.
My final suggestion is: if your project is just starting, buy one first.
Tencent Cloud AMD instance (such as SA3)
Let the business go online quickly and take the 20% hard currency bonus. When the company's business grows and the servers become dozens or hundreds of units, the micro-services and cache layers at the edge will be gradually moved forward.
ARM instance
On the migration, to squeeze the last 30% of the cost limit.
