With the new processor family consisting of variants for professional workstations and high-end desktops, AMD says the Threadripper 9000 series can beat Intel’s competing products by as much as 119 percent in rendering and by up to 49 percent in AI workloads.
AMD claimed that its forthcoming Ryzen Threadripper 9000 CPUs can beat Intel’s competing products by double-digit and even triple-digit percentages across various workstation applications, including rendering and generative AI workloads.
On Tuesday, the Santa Clara, Calif.-based company chip designer revealed more details, including competitive comparisons, about the Threadripper Pro 9000 W-X series for professional workstations and the Threadripper 9000 series for high-end desktops.
[Related: The Biggest AMD Advancing AI News: From MI500 GPU TO ROCm Enterprise AI]
While Threadripper Pro 9000 W-X chips feature up to 96 cores and 192 threads along with support for the enterprise-grade AMD Pro Technologies security and management features, the Threadripper 9000 chips sport up to 64 cores and 128 threads for prosumers. All chips come with a maximum 350-watt thermal design power.
The processors are set to launch next month alongside AMD’s new Radeon AI Pro R9700 graphics card that is meant to accelerate AI inference workloads. The company did not disclose the pricing for either product set.
AMD’s Threadripper 9000 Vs. Xeon W-3500 Claims
In a presentation with journalists last week, AMD presented a series of performance comparisons between AMD’s flagship, 96-core Threadripper Pro 9995WX and Intel’s flagship, 60-core Xeon W9-3595X, which launched last year as part of the Xeon W-3500 series.
For CPU-centric design and manufacturing workloads, AMD said the Threadripper chip is up to 70 percent faster in the SPECapc Solidworks modeling benchmark and up to 119 percent faster in the Keyshot rendering test. For workloads that rely on both the CPU and GPU, the chip is up to 56 percent faster in the SPECapc PTC Creo modeling benchmark and up to 22 percent faster in the SPECviewperf Cati modeling test, according to the company.
For media and entertainment, AMD said it beat Intel by up to 145 percent and up to 118 percent for the CPU-based Chaos V-Ray and Cinebench R24 (nT) rendering benchmarks, respectively. The company claimed the chip is also faster by up to 78 percent for Adobe After Effects and up to 83 percent for Autodesk Maya, the latter two of which rely on both a CPU and GPU.
For architecture, engineering and construction workloads, AMD claimed to best Intel by up to 40 percent in the CPU-bound Autodesk Revit for model creation and by up to 118 percent in the CPU-bound Corona Render software. For GPU-based applications, the company said the chip is up to 40 percent faster for AutoCAD 3-D graphics creation and up to 34 percent faster for Autodesk Revit graphics editing.
In other CPU-centric workloads, AMD said the Threadripper chip is faster by up to 52 percent for Chromium compilation work, up to 73 percent for Unreal Engine compilation work and up to 75 percent for MATLAB computations.
In AI workloads reliant on a CPU and GPU, the company claimed to have up to 49 percent faster tokens per second for the 32-billion-parameter DeepSeek R1 model, up to 34 percent faster text to image for the ComfyUI + Flux.1 diffusion model and up to 28 percent faster for the PugetBench Davinci Resolve studio AI tests.
AMD Boasts Of 16 Percent IPC Uplift
AMD said the Zen 5 architecture underlying the Threadripper 9000 series delivers a 16 percent uplift on average for instructions per second across 10 workstation benchmarks in comparison to Zen 4, which powers the preceding Threadripper 7000 series from 2023.
When it comes to the SPEC workstation benchmark for AI and machine learning, Zen 5 provides a 25 percent improvement in instructions per clock over Zen 4, according to AMD.
The chip designer said Zen 5 brings improves performance-per-watt thanks to things like better branch prediction, continued multi-threading support, continued power gating improvements as well as reduced power state entry and exit times.
Among what AMD called “significant upgrades” for the Threadripper 9000 series are the eight-channel maximum and 6400 megatransfers per second throughput for DDR5 with ECC memory as well as the 410 GBps of peak theoretical memory bandwidth enabled by such capacity.
The chips are also drop-in compatible with AMD’s existing sTR5 platforms and feature AIM-T wireless LAN support as part of its AMD Pro Technologies capability. In addition, they have improved aggregate PCIe bandwidth due to internal system-on-chip topology changes.
The features that haven’t changed from the previous generation include up to 128 I/O lanes of PCIe Gen 5, RDIMM support, a 2 TB of memory capacity enabled by up to 1 DIMM per channel, up to 32 I/O lanes for SATA-based storage and AVX-512 support.
In comparison to AMD’s previous 96-core flagship Threadripper Pro 7995WX, the company said the new 96-core Threadripper Pro 9995WX is anywhere from 13 percent to 26 percent faster across the benchmarks it used to compare against Intel.
