DoIT benchmarks N4A Arm VMs topping C4A and M8g in CPU price-perf

TL;DR

DoIT engineer benchmarks Google Cloud's new N4A Arm VMs against N4 x86, C4A Arm, and AWS M8g Graviton4 instances.[[1]](https://engineering.doit.com/first-look-at-google-cloud-n4a-vms-benchmarked-against-n4-c4a-and-aws-m8g-aba8017ea927)[[2]](https://www.linkedin.com/posts/nathanabaum_first-look-at-google-cloud-n4a-vms-benchmarked-activity-7422399459298447360-bWUj)
N4A tops Sysbench CPU tests by 13.7% over C4A and 21.7% over M8g; offers 33% better price-performance on compute workloads.[[2]](https://www.linkedin.com/posts/nathanabaum_first-look-at-google-cloud-n4a-vms-benchmarked-activity-7422399459298447360-bWUj)
N4A suits cost-sensitive general-purpose Arm workloads, pairing with pricier C4A for high-performance needs to cut TCO.[[1]](https://engineering.doit.com/first-look-at-google-cloud-n4a-vms-benchmarked-against-n4-c4a-and-aws-m8g-aba8017ea927)

The story at a glance

DoIT's Alex Gkiouros tests Google Cloud N4A VMs, powered by Axion on Arm Neoverse N3, pitting them against N4 (Intel Emerald Rapids), C4A (Axion Neoverse V2), and AWS M8g (Graviton4 Neoverse V2). The article comes out right after N4A hits general availability in late January 2026, as DoiT gets early access via its Google Cloud Premier Partner status.[[1]](https://engineering.doit.com/first-look-at-google-cloud-n4a-vms-benchmarked-against-n4-c4a-and-aws-m8g-aba8017ea927)[[3]](https://cloud.google.com/blog/products/compute/axion-based-n4a-vms-now-in-preview) N4A positions as the cheaper general-purpose option at $0.0385 per vCPU-hour (1v/4GB), versus C4A's $0.0449.[[1]](https://engineering.doit.com/first-look-at-google-cloud-n4a-vms-benchmarked-against-n4-c4a-and-aws-m8g-aba8017ea927)

Key points

Tests use n4-standard-8, n4a-standard-8, c4a-standard-8, and AWS m8g.2xlarge, all on Debian 13 (ARM64 or x86_64).[[1]](https://engineering.doit.com/first-look-at-google-cloud-n4a-vms-benchmarked-against-n4-c4a-and-aws-m8g-aba8017ea927)
In Sysbench CPU, N4A shows highest raw performance, beating C4A by 13.7% and M8g by 21.7%.[[2]](https://www.linkedin.com/posts/nathanabaum_first-look-at-google-cloud-n4a-vms-benchmarked-activity-7422399459298447360-bWUj)
N4A delivers about 33% better price-performance for compute workloads, based on on-demand pricing.[[2]](https://www.linkedin.com/posts/nathanabaum_first-look-at-google-cloud-n4a-vms-benchmarked-activity-7422399459298447360-bWUj)
C4A leads AES-256-GCM crypto (hardware-accelerated); N4 (x86) second at 44.5 GB/s, ahead of M8g and N4A.[[1]](https://engineering.doit.com/first-look-at-google-cloud-n4a-vms-benchmarked-against-n4-c4a-and-aws-m8g-aba8017ea927)
Google frames N4A and C4A as "better together": C4A for latency-sensitive tasks, N4A for scale-out general compute.[[1]](https://engineering.doit.com/first-look-at-google-cloud-n4a-vms-benchmarked-against-n4-c4a-and-aws-m8g-aba8017ea927)

Details and context

N4A VMs support up to 64 vCPUs and 512 GB DDR5 memory in standard (4 GB/vCPU), high-cpu (2 GB/vCPU), and high-mem (8 GB/vCPU) shapes, with 50 Gbps networking and Hyperdisk storage. They target microservices, databases, analytics, and dev/test environments where cost matters more than peak speed.[[3]](https://cloud.google.com/blog/products/compute/axion-based-n4a-vms-now-in-preview)

Google claims N4A gives up to 2x better price-performance versus current x86 VMs like N4, plus 80% better perf-per-watt, though DoIT's head-to-head focuses on Arm peers. C4A offers higher specs (72 vCPUs, 576 GB RAM, 100 Gbps) for demanding jobs; N4A trades some power for lower cost. AWS M8g, on Graviton4, competes in general-purpose Arm but trails N4A in these Sysbench results.[[3]](https://cloud.google.com/blog/products/compute/axion-based-n4a-vms-now-in-preview)[[4]](https://aws.amazon.com/ec2/instance-types/m8g)

Benchmarks cover CPU (Sysbench), compression, and crypto, highlighting trade-offs: N4A wins raw compute value, but x86 or C4A better for crypto-heavy tasks.

Key quotes

"If your Arm workloads don’t need the C4A performance premium, consider the new N4A which can drastically reduce your TCO."

— Alex Gkiouros, DoiT[[1]](https://engineering.doit.com/first-look-at-google-cloud-n4a-vms-benchmarked-against-n4-c4a-and-aws-m8g-aba8017ea927)

Why it matters

Arm instances like N4A push cloud providers to compete on efficiency, challenging x86 dominance in general compute.

Users running scale-out workloads can swap to N4A for 30%+ price-perf gains over rivals, lowering bills without code changes on compatible apps.

Watch real-world adoption on GKE or Batch, plus C4A.metal preview, though results vary by workload.[[3]](https://cloud.google.com/blog/products/compute/axion-based-n4a-vms-now-in-preview)