A t3.large held at 100% CPU pays more in credit overage than for the instance itself
T8i arrives with up to 70% more compute than T3 and, by one AWS Builders account, a 20% higher hourly rate. Whether that pays back depends on how much Unlimited-mode credit overage the T3 is already accruing.
The Engineer · Build desk
What happened
- AWS made T8i generally available in four sizes only, nano through medium, each with two vCPUs presented as a single core, on custom sixth-generation Intel Xeon Granite Rapids processors and the Nitro system.
- Against T3, AWS claims up to 30% better price performance, up to 70% higher compute performance, up to 1.25x network bandwidth and up to 2.4x EBS bandwidth.
- T8i can be bought On-Demand or as Spot with a Savings Plan option listed as coming soon, supports shared tenancy only, and launched in 16 Regions across the US, Europe, Asia Pacific and Canada.
- A post in the AWS Builders channel on dev.to reports that T8i costs 20% more than T3, a figure the launch announcement does not mention.
Why it matters
- Who has to decide decision For each T3 instance the deciding input is its own Unlimited-mode overage line, and only the account holder can see that line.
- What it costs cost Anyone holding committed T3 spend is comparing a discounted rate against an On-Demand T8i rate, so the like-for-like comparison does not exist until Savings Plan coverage lands.
- What it limits constraint Because the family stops at medium, upgrading t3.large and above means leaving the burstable credit model. That is a capacity-planning exercise.
Next story · Build · Sep 15
Delos Data raises more than $100M to keep inference running when a link dropsConfirmed2 publishers
In Standard mode, a T3 that runs out of credits is clamped back to baseline, which on a t3.large is 30% per vCPU, according to the AWS Builders post on dev.to [13]. Unlimited mode bills instead of clamping: if the 24-hour average sits above baseline, AWS adds $0.05 per vCPU-hour [18]. Unlimited is the default on T8i as well [5], so overage accrues from launch.
The same post works an example. A t3.large has two vCPUs, so an instance held at 100% CPU around the clock accrues $0.10 an hour in surcharge, taking the effective rate from about $0.0835 to about $0.1835 [19]. The surcharge alone is larger than the base rate [28], and the effective price is about 2.2 times the sticker [27].
Set the reported 20% premium against that. Twenty percent of $0.0835 is about $0.017 an hour, roughly a sixth of the $0.10 surcharge [30]. The dev.to author makes the point directly: on a T3 burning credits in Unlimited mode, the faster CPU lowers credit consumption, and that saving can absorb the 20% [21]. One limit on the example is that there is no t8i.large, so a t3.large has no same-size replacement in the new family [4].
Both AWS figures are "up to". A 70% compute gain that nets out to 30% better price-performance needs an hourly rate 1.70/1.30 higher, or about 31% [20]. For that to be the real premium, both "up to" numbers would have to come from the same workload at the same instance size. The implied 31% sits above the 20% the dev.to post reports [29].
For anything bigger, AWS's own recommendation leaves the T family. Channy wrote in the launch post that workloads needing instance sizes larger than T8i's nano, micro, small and medium should go to M8i Flex, which offers up to 30% better price performance than equivalent previous generation T3 instances along with the flexibility to scale up to 16xlarge [10].
The dev.to post sorts the rest by usage pattern. Sustained-CPU workloads go to dedicated families such as c6a, m6a or c7i-flex, where credit ceilings stop producing either overage or bottlenecks [25]. A move to the c family changes the ratio to 2 GB of RAM per vCPU, and the post rates that valid only where the current T3 uses less than half its memory [22]. Keeping the ratio on m6a is rated worthwhile only where the T3 is already paying Unlimited overage [23]. For a t3.xlarge that needs its memory but not its cores, it suggests r8i-flex.large: half the vCPU count on a CPU it describes as 70% stronger than T3's, with CPU monitored after the change to confirm the result is acceptable [24].
t8i.micro and t8i.small are in the AWS Free Tier [7]. A test box is a cheap place to measure the 70% claim against your own binary. For per-Region availability and upcoming expansion, AWS points to the CloudFormation resources tab of AWS Capabilities by Region [11].
What to watch
- Whether Savings Plan coverage for T8i arrives before your current T3 commitment term ends.
- Whether AWS adds a T8i size above medium or keeps routing larger burstable workloads to M8i Flex.
- Whether the launch Region list grows enough for multi-Region fleets to standardise on T8i.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence62
- Adoption32
- Hype gap+22
- Incentives74
- Confidence58
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Amazon EC2 T8i instances are generally available, powered by the AWS Nitro System and custom sixth generation Intel Xeon Scalable Processors (Granite Rapids), available only on AWS.
- [2]
AWS says T8i instances deliver up to 30% better price performance over previous generation T3 instances.
- [3]
AWS says T8i instances deliver up to 70% higher compute performance, up to 1.25x higher network bandwidth and up to 2.4x higher EBS bandwidth compared to T3 instances.
- [4]
T8i offers four sizes only: nano, micro, small and medium, each with two vCPU offered as a single core.
- [5]
Like T3, T8i uses the CPU credit system with Standard and Unlimited credit configuration modes, and Unlimited mode is the default on T8i.
- [6]
T8i instances can be purchased via On-Demand and Spot instances, with a Savings Plan option coming soon.
- [7]
t8i.micro and t8i.small instances are available under the AWS Free Tier.
- [8]
T8i is available in US East (N. Virginia, Ohio), US West (Oregon, N. California), Asia Pacific (Hyderabad, Malaysia, Mumbai, Seoul, Singapore, Sydney, Tokyo), Canada (Central), and Europe (Frankfurt, Ireland, London, Paris).
- [9]
The What's New post says T8i instances are powered by custom sixth generation Intel Xeon 6 processors and built on the latest sixth generation AWS Nitro cards.
- [10]
"For workloads that need larger instance sizes above T8i offerings (nano, micro, small, and medium), I recommend M8i Flex instances that offer up to 30% better price performance than equivalent previous generation T3 instances along with the flexibility to scale up to 16xlarge"
- [11]
For Regional availability and upcoming Region expansion, AWS directs readers to search the instance type in the CloudFormation resources tab of AWS Capabilities by Region.
- [12]
The dev.to post notes T8i is not yet covered by Savings Plans and may not be available in the reader's Region.
- [13]
In Standard mode, if accumulated credits are exhausted, CPU performance is restricted to baseline, for example 30% per vCPU on a t3.large.
- [14]
Direct replacement with T8i carries no performance risk in the dev.to post's assessment, because vCPU and memory amounts are unchanged for nano, micro, small and medium.
- [15]
Because T8i does not offer a large size, AWS recommends migrating larger sizes to Flex instances.
- [16]
T8i instances support shared tenancy only and do not support Dedicated tenancy or Dedicated Hosts.
- [17]
The launch Region list contains 16 AWS Regions.
- [18]
In Unlimited mode, which is enabled by default, if usage averaged over 24 hours exceeds the baseline AWS charges an additional $0.05 per vCPU-hour.
- [19]
A t3.large with 2 vCPUs run at 100% CPU continuously accrues a $0.10 per hour surcharge, raising the real total cost from about $0.0835 per hour to about $0.1835 per hour.
- [20]
Taken together, a 70% compute gain yielding 30% better price-performance implies an hourly price about 31% higher.
- [21]
The dev.to post says that if the T3 instance is consuming credits in Unlimited mode, the CPU improvement will reduce credit consumption, which can offset the 20% cost increase.
- [22]
Migrating to a c-family instance changes the CPU-to-memory ratio to 2 GB of RAM per vCPU, and the dev.to post says this option is only valid if less than 50% of memory is used on the current T3.
- [23]
Keeping the CPU-to-memory ratio means choosing m6a, which the dev.to post says is only valid if the T3 instance is consuming credits in Unlimited mode and incurring extra costs.
- [24]
The dev.to post suggests replacing t3.xlarge with r8i-flex.large where CPU use is not intensive and memory must be kept, arguing the R8i processor is 70% more powerful than T3 CPUs so halving vCPUs may be compensated, and says CPU must be monitored afterwards to confirm performance is acceptable.
- [25]
The dev.to post recommends dedicated-performance families without credits (c6a, m6a, c7i-flex) for applications with continuous or sustained CPU use where T-series credit limits generate extra cost or bottlenecks.
- [26]
T3 instances launched in 2018.
- [27]
The effective rate in that example is about 2.2 times the base On-Demand rate.
- [28]
In the worked t3.large example the $0.10 hourly credit surcharge exceeds the roughly $0.0835 hourly instance rate.
- [29]
The AWS Builders post on dev.to says T8i instances are 20% more expensive than T3 and that the launch announcement does not mention this.
- [30]
A 20% premium on the $0.0835 base rate is about $0.017 per hour, roughly one sixth of the $0.10 hourly credit surcharge.
Sources
2 independent publishers whose own reporting we read for this story.
- aws.amazon.comIntroducing Amazon EC2 T8i instances
2 articles · September 17, 2026
- dev.toNuevas instancias T8i, ¿hora de jubilar tus T3i?
1 article · September 19, 2026
Topics and entities
Topics
- Cloud Cost Optimization
- Cloud instance pricing
- Burstable compute
Entities
- Amazon Web Services
- AWS Savings Plans
- Amazon EC2
- AWS Nitro System
- Intel