Graviton 4 (r8g) is already available for Aurora in the Tokyo region! Here's a performance comparison with each generation of Graviton.
This is Yuta Kikai (@fat47) from the Service Reliability Group (SRG) of the Media Division.
#SRGThe Service Reliability Group primarily provides comprehensive support for the infrastructure surrounding our media services, focusing on improving existing services, launching new ones, and contributing to open-source software (OSS).
This article summarizes the availability of Graviton 4 (r8g) in the Tokyo region of AWS Aurora, as well as the results of a performance comparison test.
I hope this is of some help.
In a surprising turn of events, Graviton 4 (r8g) was released incredibly quickly in the Tokyo region.What are the performance differences compared to the previous generation (r7g) and the generation before that (r6g)?Regarding feesPerformance comparison using sysbenchIn conclusion
In a surprising turn of events, Graviton 4 (r8g) was released incredibly quickly in the Tokyo region.
Aurora's Graviton 4 (r8g) support was released in November 2024.
At that time, it was only available in a limited number of regions, including the US East (Northern Virginia, Ohio) and US West (Oregon).
With Graviton3 (r7g), it took over a year for it to reach the Tokyo region, so I was caught off guard, but Graviton4 was released in the Tokyo region on March 14, 2025.
I'm so happy that it's coming to the Tokyo region in just four months!
What are the performance differences compared to the previous generation (r7g) and the generation before that (r6g)?
I will quote from the press release.
Compared to equivalent-sized Graviton3-based instances of Amazon Aurora databases, you can achieve up to 40% performance improvements and up to 29% price/performance improvements at on-demand pricing.
Up to 40% performance improvement from R7G to R8G.It seems there is.
Incidentally, the release article for r7g stated the following:
Compared to Amazon Aurora Graviton2 instances, it delivers up to 30% better performance and up to 20% better price/performance.
Up to 30% increase from r6g to r7gThey said there would be an improvement in performance.
In other words…
Does this mean that switching from Graviton2 (r6g) to Graviton4 (r8g) results in an 82% performance improvement?
This needs to be investigated...!
Regarding fees
Let's compare the on-demand standard rates for the same size files, from r6g.large to r8g.large.
| Instance type | Hourly rate (on-demand) | Price difference compared to the previous generation |
|---|---|---|
| db.r6g.large | USD 0.313 | 0% |
| db.r7g.large | USD 0.333 | +6.39% |
| db.r8g.large | USD 0.333 | +0% |
Surprisingly, Graviton3 (r7g) and Graviton4 (r8g) were the same price. I compared other sizes as well, and they were almost the same price.
In addition,Reserved instances for Graviton4 (r8g) are not yet available in the United States.It seems so.
Performance comparison using sysbench
Just like when Graviton3 (r7g) was released last time, we'll use sysbench to check its performance.
Please refer to this article for information on installing sysbench and its commands.
r8g can createAurora MySQL v3.08 or higherTherefore, this time we standardized the engine version to v3.08.1 for all instance types and recalculated the performance check values.
| Engine Version | v3.08.1 (MySQL 8.0.39 compatible) |
| Load environment | c7a.xlarge (same Availability Zone) |
I ran the process three times on each instance. I didn't consider the difficulty of clearing the cache in Aurora.
The average of the three trials is summarized below.
| read | write | other | total | transactions | queries | |
|---|---|---|---|---|---|---|
| r6g | 273,014 | 78,194 | 39,097 | 390,973 | 19,548 (325per sec) | 390,973 (6514per sec) |
| r7g | 326,778 | 93,841 | 46,587 | 465,873 | 23,294 (388per sec) | 465,873 (7763per sec) |
| r8g | 351,008 | 100,288 | 50,144 | 501,440 | 25,072 (417per sec) | 501,440 (8356per sec) |
The performance ratios of the transactions were summarized as follows:
| Performance difference compared to r6g | Performance difference compared to the previous generation | |
|---|---|---|
| r6g | - | - |
| r7g | + 19.16% | + 19.16% |
| r8g | + 28.26% | + 7.63% |
Hmm...? This isn't what I expected...?
In a previous blog post, I tested it and found that upgrading from R6G to R7G resulted in a performance improvement of about 34%.
Previously, I was using engine version v3.04 (compatible with MySQL 8.0.28),
I'll create a separate cluster using v3.04.3 and re-run the scores for r6g and r7g.
We also took three measurements and recorded the average value.
| read | write | other | total | transactions | queries | |
| r6g(v3.04.3) | 235,456 | 67,273 | 33,636 | 336,366 | 16,818 (280per sec) | 336,366 (5605per sec) |
| r7g(v3.04.3) | 284,624 | 81,321 | 40,660 | 406,606 | 20,330 (338per sec) | 406,606 (6775per sec) |
| Difference from r6g (v3.04.3) | + 20.88% | + 20.88% | + 20.88% | + 20.88% | + 20.88% | + 20.88% |
For some reason, the improvement is only about 20%. I wonder why...
In conclusion
Graviton4 (r8g) is now available in the Tokyo region.
It certainly seems like a good deal, as it offers improved performance while maintaining the same price as the previous generation.
On the other hand, since reserved instances are not yet available, those who have purchased Graviton2 (r6g) will need to consider other options.
Also, since Graviton4 cannot be used in the LTS version v3.04, that's something that needs to be taken into consideration.
I'm writing this article based on the breaking news, but the verification results don't quite sit right with me,
I would love for you all to try out performance comparison tests and let me know your results!
SRG is looking for new team members.
If you are interested, please contact us here.