Redis vs Memcached vs Valkey: 2026 Benchmark Data, Sourced
Memcached wins simple key-value caching at high concurrency. Redis wins structured operations. Valkey 8.x wins memory efficiency vs Redis 8.x. Numbers below, sourced.
memtier_benchmark or redis-benchmark on your real data before committing.Benchmark 1: Simple key-value caching under production load
IndependentHardware: 8-core AWS instances (type not named), ~300-byte objects, 150,000 req/s peak, 80% reads | Source: Dev Genius, 12 March 2026 (one team's production test)
Caveat: One team's production test, published as a blog post. The author does not name the instance type and reports no ops/sec ceiling. Treat it as one data point, not a settled figure; run memtier_benchmark on your own data.
Benchmark 2: Structured data (shopping carts)
IndependentHardware: Same 8-core AWS setup; carts held as lists, recently viewed as sorted sets | Source: Dev Genius, 12 March 2026 (same production test)
Caveat: The author states the 5x figure without publishing the raw numbers behind it. The direction (native data structures beat whole-object rewrites) is the durable lesson; the multiple will vary with object size.
Benchmark 3: Valkey 8.0 RPS gain over Valkey 7.2
VendorHardware: AWS EC2 c7g.16xlarge, 8 I/O threads, 512-byte values, 650 clients, SET | Source: Valkey blog: Unlock 1 Million RPS
Caveat: Vendor-sourced, on a 64-vCPU instance with 8 I/O threads. On small instances or low concurrency the gain is smaller.
Benchmark 4: Valkey 8.1 vs Redis 8.2 memory efficiency
VendorHardware: AWS c8g.2xlarge Graviton4, 8 vCPU, 16 GB, 50M sorted-set entries | Source: Momento: Valkey vs Redis Memory Efficiency at Hyperscale
Caveat: Momento sells Valkey-compatible managed services, so it has a product interest; the benchmark code is published for anyone to rerun. Sorted sets only; other data types will differ.
Benchmark 5: Redis 8.0 vs Redis 7.2.5
VendorHardware: Vendor hardware; io-threads test on a multi-core Intel CPU | Source: Redis blog: Redis 8 is now GA
Caveat: Vendor-sourced, Redis 8 vs Redis 7.2.5 only (not vs Valkey or Memcached). 'Up to' figures are best cases; the median command gains far less.
Synthesis table
| Workload | Winner | Source type | Approx delta |
|---|---|---|---|
| Simple key-value, production load | Memcached | Independent | P99 1.1 vs 3.4 ms; 8 vs 12 GB (one team) |
| Structured data (carts: lists, sorted sets) | Redis | Independent | ~5x (author's figure) |
| Memory at 50M sorted-set entries | Valkey 8.1 | Vendor | 28% lower vs Redis 8.2 (Momento) |
| Valkey 8.0 vs 7.2 RPS | Valkey 8.0 | Vendor | ~230% (vendor claim) |
| Redis 8.0 vs 7.2.5 p50 latency | Redis 8.0 | Vendor | Up to 87.4% (vendor claim) |
| Redis 8.0 io-threads=8 throughput | Redis 8.0 | Vendor | Up to 112% (vendor claim) |
What the numbers do not tell you
Every benchmark is workload-shaped. A 300-byte key-value test on 8-core instances says nothing about your 4KB session blobs, your sorted-set leaderboard under 50k concurrent users, or your pub/sub fan-out to 200 subscribers. The performance delta between Redis and Memcached is real but smaller than the license delta for many 2026 teams. Run your own benchmark with memtier_benchmark before deciding.