Xeon 678X vs. Threadripper 9980X: SMT Efficiency Tested
Simultaneous Multithreading (SMT) can increase CPU throughput by allowing a physical core to execute work from multiple software threads, but the performance gain depends heavily on workload characteristics and microarchitectural design.
On September 17, 2026, Phoronix published a specialized workstation CPU test examining this question across Intel and AMD’s high-end platforms. Instead of focusing primarily on absolute processor performance, the benchmark series investigates the practical efficiency of Intel Hyper-Threading and AMD SMT.
The comparison centers on the Intel Xeon 678X “Granite Rapids WS” and AMD Ryzen Threadripper 9980X, with additional configurations designed to separate the effects of physical core count from SMT itself.
The test follows more than 400 benchmarks previously conducted on Intel Xeon 600-series and AMD Ryzen Threadripper 9000-series processors. By enabling and disabling logical threads and normalizing the physical core count, Phoronix aimed to provide a more architecture-focused view of SMT scaling.
🧪 Test Design and Hardware Configuration #
The benchmark uses two high-end workstation processors based on different CPU architectures:
- Intel Xeon 678X: 48 physical cores and 96 threads, based on Granite Rapids WS and Redwood Cove cores.
- AMD Ryzen Threadripper 9980X: 64 physical cores and 128 threads, based on the Zen 5 architecture.
Rather than comparing only their default configurations, Phoronix created several control groups.
Intel Xeon 678X configurations #
The Xeon 678X was tested in two configurations:
- 48 cores / 96 threads: Default configuration with Hyper-Threading enabled.
- 48 cores / 48 threads: Hyper-Threading disabled.
This makes it possible to measure the incremental contribution of Intel’s logical threads while keeping the physical core count constant.
AMD Ryzen Threadripper 9980X configurations #
The Threadripper 9980X was similarly tested with SMT enabled and disabled:
- 64 cores / 128 threads: Default configuration with SMT enabled.
- 64 cores / 64 threads: SMT disabled.
This provides an equivalent before-and-after comparison for AMD’s implementation.
48-core normalized configuration #
The most important control group reduces the physical core-count advantage of the Threadripper 9980X.
Through BIOS configuration, the 9980X was limited to 6 active cores per CCD, producing a total of 48 physical cores. Both SMT-enabled and SMT-disabled configurations were then tested.
This creates a direct 48-core-versus-48-core comparison between the Threadripper 9980X and Xeon 678X.
The approach is particularly useful for analyzing SMT efficiency because the comparison is no longer dominated by the Threadripper platform’s higher physical core count.
🖥️ Workstation Platforms and Software Environment #
The processors were tested on complete workstation systems rather than isolated CPU test benches.
The Xeon 678X was installed in an HP Z4 G6i workstation, while the Threadripper 9980X was tested in a System76 Thelio Major.
The software environment consisted of:
- Operating system: Fedora Workstation 44
- Kernel: Linux 7.1
- Compiler: GCC 16.2
The benchmark suite covered a range of workstation-oriented workloads designed to expose differences in CPU throughput and SMT scaling.
This workload diversity is important because SMT rarely produces a uniform performance improvement. Highly parallel workloads with available execution resources can benefit substantially, while workloads that already saturate core resources may see smaller gains.
Phoronix also recorded CPU power consumption and temperatures alongside performance results, providing additional context for the efficiency characteristics of the different configurations.
⚙️ Why SMT Efficiency Matters #
SMT does not double the computational resources of a physical CPU core.
Instead, two logical threads share the underlying execution resources of the same physical core. Depending on the workload, the second thread can make use of execution capacity that would otherwise remain idle.
The potential benefit depends on factors such as:
- Instruction-level parallelism
- Memory latency and bandwidth
- Execution-unit utilization
- Cache behavior
- Branch behavior
- Thread scheduling
- Workload synchronization
- Resource contention between sibling threads
This makes SMT efficiency particularly relevant to workstation CPUs.
Professional workloads can range from compilation and rendering to scientific computing, software development, simulation, compression, and other heavily threaded applications. Some scale effectively across logical threads, while others gain little or can even experience contention.
Consequently, simply comparing processors by advertised thread count does not fully describe their computational behavior.
🔬 Separating Core Count From SMT Scaling #
The 48-core normalized Threadripper configuration is particularly useful from an architectural-analysis perspective.
At their default configurations, the processors have different physical core counts:
| Processor | Physical Cores | Threads | SMT/HT |
|---|---|---|---|
| Intel Xeon 678X | 48 | 96 | Hyper-Threading |
| AMD Threadripper 9980X | 64 | 128 | SMT |
| Threadripper 9980X normalized | 48 | 96 | SMT |
The normalized configuration allows the test to ask a more specific question: how does each architecture scale when logical threads are added to the same number of physical cores?
That is different from asking which processor is faster overall.
A 64-core Threadripper configuration can naturally outperform a 48-core processor in workloads that scale efficiently with additional physical cores. By reducing the Threadripper to 48 physical cores, the benchmark can instead expose differences in SMT behavior while minimizing the influence of core-count scaling.
This methodology is valuable when evaluating CPU architecture rather than simply selecting the fastest workstation processor.
🌡️ Performance, Power, and Thermal Behavior #
Performance alone does not fully characterize SMT efficiency.
Adding logical threads can increase processor utilization, but it can also increase power consumption and thermal output depending on the workload and processor implementation.
For workstation deployments, this creates a practical trade-off between:
- Additional throughput
- Energy consumption
- Sustained clock behavior
- Cooling requirements
- Performance per watt
- System-level operating costs
Phoronix therefore measured CPU power consumption and temperatures alongside benchmark performance.
These measurements can help distinguish a raw performance improvement from a more efficient improvement in workload throughput.
For professional systems that operate under sustained workloads, this distinction can be more meaningful than a short benchmark run.
🧭 Final Benchmark in the Xeon 678X Workstation Series #
According to Phoronix, this benchmark represents the final batch of testing completed in its lab using the HP Z4 G6i workstation equipped with the Xeon 678X.
The platform had previously been used for multiple comparisons with AMD’s Threadripper 9000 series, including:
- Overall workstation performance
- Core-count scaling
- Hyper-Threading performance
- Intel AMX performance
- Processor architecture comparisons
The latest test extends that work by concentrating specifically on logical multithreading.
This is an important distinction because physical core count, SMT behavior, vector extensions, memory subsystems, and software scheduling can all influence workstation performance independently.
🧩 What the Test Is Designed to Reveal #
The Xeon 678X versus Threadripper 9980X comparison is less about declaring a universally superior processor and more about examining how two high-end workstation architectures extract additional throughput from their physical cores.
The test design provides three useful perspectives:
- Default platform behavior: How the Xeon 678X and Threadripper 9980X perform with their standard core and thread configurations.
- SMT/Hyper-Threading scaling: How much additional performance each architecture obtains when logical threads are enabled.
- Normalized architecture comparison: How the two processors behave when the Threadripper 9980X is restricted to the same 48 physical cores as the Xeon 678X.
By combining these configurations with workstation workloads, power measurements, and thermal data, the benchmark provides a more detailed view of SMT efficiency on modern high-core-count CPUs.
For developers, engineers, and workstation users, the key lesson is that core count and thread count should not be treated as interchangeable measures of CPU capability. The real value of SMT depends on how effectively the underlying architecture, operating system, scheduler, compiler, and workload can exploit the additional logical execution contexts.