Nitro: same cores, more memory

The memory-bound middle

A surprising number of workloads hit a wall at 8 GB while four cores sit half idle. Java and Kotlin services with generous heaps. Redis holding a working set that is just too big. A search index that wants its segments in the page cache. A Rails or Django app with enough worker processes to keep up, each of them a few hundred megabytes. For all of those, more cores do nothing and more memory does everything.

Nitro exists for exactly that gap. It costs $2 more than Turbo and gives you 50% more RAM and 50% more disk. Per gigabyte of memory it is one of the better-value sizes on the price list.

Sizing a JVM on 12 GB

The mistake to avoid is setting -Xmx to nearly all of RAM. The heap is not the whole JVM: metaspace, code cache, thread stacks and direct buffers live outside it, and the kernel still needs page cache.

JAVA_OPTS="-Xms6500m -Xmx6500m -XX:MaxDirectMemorySize=512m \
  -XX:+UseG1GC -XX:MaxGCPauseMillis=100 -XX:+ExitOnOutOfMemoryError"

Setting -Xms equal to -Xmx means the heap is committed at start, so you find out immediately if the plan is too small rather than at 2 a.m. under load. ExitOnOutOfMemoryError lets systemd restart a wedged process instead of leaving it half alive.

A cache that fits

If Redis is the reason you are here, give it a hard ceiling and an eviction policy, then let the rest of the box breathe:

# /etc/redis/redis.conf
maxmemory 7gb
maxmemory-policy allkeys-lru
save ""
appendonly no

Disabling persistence is right for a pure cache; it avoids fork-time memory spikes during snapshots. If Redis is your source of truth, keep persistence on and lower maxmemory to about half of RAM so a background save has room to copy pages.

Why 120 GB matters here

Memory-heavy services tend to be data-heavy too. Search indexes, Kafka-style logs, local artefact caches and database data directories all grow. The extra 40 GB over Turbo is often what lets you keep a week of logs or a full local index instead of trimming aggressively. On RHEL-family images such as Rocky Linux 9, check where space is going with:

du -xh --max-depth=2 /var | sort -h | tail -15

Choosing between neighbours

• CPU is fine, memory is fine, disk is fine: you may be on the wrong side of the line. Turbo saves $2.
• Memory is fine but four cores are pinned: go to Warp for six vCPUs and 16 GB.
• One large in-memory data set needs far more than 12 GB: skip ahead to Hyper and its 32 GB.

The portal's CPU graph will tell you quickly which side of the box you are pushing against. Give it a week of real traffic before deciding.

Price

Nitro is $15 a month or $150 for a year. Checkout charges in US dollars; a card in another currency is converted by your issuer. No setup fee, unmetered traffic on a 1 Gbps port, and one dedicated IPv4.