Ubuntu 26.04 Early Benchmarks on AMD EPYC Turin: Linux 6.18 and GCC 15.2 Show Modest Gains Over 24.04 LTS

Ubuntu 26.04 Development Snapshot: Early Performance on AMD EPYC Turin

With Ubuntu 26.04 LTS still months away from its April 2026 release, we took an early development build for a spin on AMD's EPYC 9655P 96-core server to see how the next Ubuntu LTS is shaping up. This isn't a final verdict—Canonical has yet to implement the feature freeze, and the kernel will likely jump from the current Linux 6.18 to 6.20 or even 7.0 before launch—but it gives us a preview of the performance trajectory for server deployments.

Test Configuration and Baseline

Our test platform is the same Supermicro server used in recent CachyOS and Arch Linux comparisons:

• CPU: AMD EPYC 9655P (96 cores, 192 threads)
• Memory: 256GB DDR5-4800
• Storage: 2TB NVMe SSD
• GPU: None (headless server testing)

We compared three Ubuntu releases on this hardware:

Release	Kernel	GCC	Status
Ubuntu 24.04.3 LTS	Linux 6.14 (HWE)	13.2	Stable
Ubuntu 25.10	Linux 6.17	15.2	Current stable
Ubuntu 26.04 (dev)	Linux 6.18	15.2	Development

The key difference between 26.04 and 25.10 is the kernel version bump, while compiler versions remain identical for now. This gives us a clean view of kernel-level improvements.

Kernel 6.18: What's Actually Different?

Linux 6.18 brings several AMD-specific optimizations that matter for EPYC Turin:

Scheduler Improvements: The EEVDF scheduler refinements from 6.17 continue to mature, with better handling of high-core-count workloads. In our HPC-style benchmarks, this translated to 1-2% reduction in task scheduling latency.

Memory Management: Transparent Huge Page (THP) optimizations for 1GB pages show measurable benefits in memory-intensive workloads. The kernel now more aggressively promotes 2MB pages to 1GB when appropriate on EPYC's 1GB page support.

Power Management: AMD P-State driver updates improve idle power consumption on Turin's enhanced power gating. In our idle measurements, 26.04 dev dropped 3.2W average compared to 24.04 LTS.

Benchmark Results

Compute-Intensive Workloads

H.264 Video Encoding (x264):

• Ubuntu 24.04.3 LTS: 184.2 fps
• Ubuntu 25.10: 188.7 fps (+2.4%)
• Ubuntu 26.04 dev: 190.1 fps (+3.2%)

The GCC 15.2 compiler in 25.10 and 26.04 already provides most of the encoding improvement. The kernel jump to 6.18 adds another 0.8% on top.

Blender BMW Render:

• Ubuntu 24.04.3 LTS: 8:42
• Ubuntu 25.10: 8:31 (-2.1%)
• Ubuntu 26.04 dev: 8:26 (-2.9%)

Faster render times indicate better thread scheduling and memory allocation patterns.

OpenSSL RSA-4096 Signing:

• Ubuntu 24.04.3 LTS: 12,847 ops/sec
• Ubuntu 25.10: 13,201 ops/sec (+2.8%)
• Ubuntu 26.04 dev: 13,384 ops/sec (+4.2%)

Database and I/O

PostgreSQL pgbench (TPC-B, 100 concurrent clients):

• Ubuntu 24.04.3 LTS: 18,421 TPS
• Ubuntu 25.10: 18,893 TPS (+2.6%)
• Ubuntu 26.04 dev: 19,102 TPS (+3.7%)

The kernel 6.18's improved I/O scheduler and page cache management show up here.

Redis SET/GET operations (1M keys):

• Ubuntu 24.04.3 LTS: 482,000 ops/sec
• Ubuntu 25.10: 495,000 ops/sec (+2.7%)
• Ubuntu 26.04 dev: 501,000 ops/sec (+3.9%)

Compilation Performance

Linux Kernel Build (defconfig, 96 parallel jobs):

• Ubuntu 24.04.3 LTS: 4m 32s
• Ubuntu 25.10: 4m 21s (-4.0%)
• Ubuntu 26.04 dev: 4m 18s (-5.1%)

This is where GCC 15.2's improved code generation for Zen 4/Turin architecture really shines.

Power Consumption Analysis

We measured power at the wall during idle and under full load (all cores):

Idle (system at prompt):

• Ubuntu 24.04.3 LTS: 87W
• Ubuntu 25.10: 84W
• Ubuntu 26.04 dev: 81W

Full Load (stress-ng --cpu 96 --cpu-method matrixprod):

• Ubuntu 24.04.3 LTS: 423W
• Ubuntu 25.10: 419W
• Ubuntu 26.04 dev: 416W

The 7W reduction in full load power for 26.04 dev is notable for data center operators running at scale—multiply that across a rack of EPYC servers and you're looking at meaningful cooling and electricity savings.

Compiler Impact: GCC 15.2 vs 13.2

The jump from GCC 13.2 in Ubuntu 24.04 LTS to GCC 15.2 in 25.10/26.04 accounts for roughly 60-70% of the total performance improvement we observed. Key optimizations in GCC 15 that benefit EPYC Turin:

• Improved auto-vectorization for AVX-512 instructions
• Better branch prediction hints for high-core-count scenarios
• Enhanced register allocation reducing memory spills

For server workloads compiled from source, this compiler upgrade alone justifies considering the newer Ubuntu release.

What This Means for Production Deployments

For Ubuntu 24.04 LTS users: The performance gap isn't large enough to warrant immediate upgrades if you're running stable workloads. The 3-5% gains are real but not transformative.

For new EPYC Turin deployments: Ubuntu 25.10 offers most of the benefits today with a stable kernel. The 26.04 dev branch shows where we're headed, but isn't ready for production.

For homelab builders: If you're running bleeding-edge AMD hardware, the kernel 6.18 + GCC 15.2 combination in 26.04 dev is already solid. Just remember it's pre-release code.

Looking Ahead to Final 26.04 LTS

Canonical will likely backport additional drivers and optimizations between now and April. The real question is whether Linux 6.20 or 7.0 will deliver more substantial gains for EPYC Turin. AMD has been actively upstreaming Turin-specific optimizations throughout the 6.x series.

We'll revisit this comparison when Ubuntu 26.04 reaches beta status. For now, the early numbers suggest a healthy incremental update—not revolutionary, but meaningful for those pushing the limits of server performance.

Current recommendation: Ubuntu 25.10 for EPYC Turin today, with confidence that 26.04 LTS will be a solid upgrade path when it stabilizes.

---

All benchmarks conducted January 14, 2026. Results are averages of three runs. Standard deviation <1.5% across all tests.