Originally from the User Slack
@stewart: Hi all,
I discovered an interesting phenomenon while benchmarking ScyllaDB.
I provisioned separate ScyllaDB clusters using two different instance types: r6i.2xlarge and r6g.2xlarge, and ran benchmarks on each.
Interestingly, the r6i (Intel) instance type showed better performance. (throughput, latency)
I’m not sure why the Intel-based instance outperforms the ARM-based one.
Could someone help explain the possible reasons for this?
and also tested with r6gd.2xlarge and i4i.2xlarge,
i4i.2xlarge got better performance 
@dor: Yes, it’s consistent with what we know. Intel is stronger than arm
@avi: The ARM instances used by AWS have relatively small caches
@stewart: you mean small L1 cache?
Can you explain the difference in more detail?
@Robert: which distribution and version of linux did You take for benchmarking?
@stewart: scylladb version : 5.4.9
distribution : 5.10.224-212.876.amzn2.x86_64
NAME="Amazon Linux"
VERSION="2"
ID="amzn"
ID_LIKE="centos rhel fedora"
VERSION_ID="2"
PRETTY_NAME="Amazon Linux 2"
ANSI_COLOR="0;33"
CPE_NAME="cpe:2.3:o:amazon:amazon_linux:2"
tested on kubernetes v1.29.
cc. @Robert
@Robert: uch, need to figure out which kernel version it uses, because on ubuntu20 vs ubuntu22 there is a cache issue on graviton. U20 doesn’t use all available L1…
if You have a time to play with, simple execute lscpu on Amazon linux vs Ubuntu22 vs Ubuntu20. On the other hand it will be double weird if amazon distro doesn’t fully operate with amazon cpus 
@stewart: i’ll check it
on i4i.4xlarge
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 16
On-line CPU(s) list: 0-15
Thread(s) per core: 2
Core(s) per socket: 8
Socket(s): 1
NUMA node(s): 1
Vendor ID: GenuineIntel
CPU family: 6
Model: 106
Model name: Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz
Stepping: 6
CPU MHz: 3500.636
BogoMIPS: 5799.90
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 48K
L1i cache: 32K
L2 cache: 1280K
L3 cache: 55296K
NUMA node0 CPU(s): 0-15
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch invpcid_single ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves wbnoinvd ida arat avx512vbmi pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq rdpid md_clear flush_l1d arch_capabilities
@Robert: and how about graviton?