The chip designer says its upcoming Ryzen Threadripper 9000 CPUs feature up to 96 cores and AVX-512 support to boost AI workloads. The Radeon AI Pro R9700 graphics card, on the other hand, packs 32 GB of GDDR6 memory to handle larger AI models.
AMD is revving up the workstation PC market again, this time with the newly revealed Ryzen Threadripper 9000 CPUs along with what it’s calling the Radeon AI Pro R9700 graphics card.
Announced at Computex 2025 in Taiwan on Wednesday local time, the Santa Clara, Calif.-based company said the “ultimate workstation processor just got faster” with the Threadripper 9000 series, which will become available in July.
[Related: AMD Boosts Spending On Product Road Maps, Go-To-Market For Next ‘Growth Arc’]
Based on AMD’s Zen 5 architecture, the Threadripper 9000 series comes in two variants: the Threadripper Pro 9000 WX-series, which comes with up to 96 cores and 192 threads, and the Threadripper 9000 series for high-end desktops, which sports up to 64 cores and 128 threads. All chips come with a 350-watt thermal design power.
While the Threadripper Pro 9000 WX-Series is designed for professional workstations with support for the AMD Pro Technologies, the vanilla Threadripper 9000 chips are meant for prosumers who want maximum CPU horsepower in their desktop PC.
All chips from both lines feature a maximum boost frequency of 5.4GHz while the base clock speeds vary from 4.2GHz for a 12-core prosumer model to 2.5GHz for the flagship, 96-core professional model, the latter of which has the largest L3 cache at 384 MB.
Like the consumer-focused Ryzen 9000 processors that launched last fall, the new Threadripper chips support AVX-512. This, along with the maximum 128 lanes of PCie Gen 5 connectivity and up to eight channels of DDR-6400 memory, makes these processors “engineered to adapt to the growing needs of AI,” according to AMD.
Both Threadripper lines are supported by AMD’s sTR5 socket, which was introduced alongside the company’s Threadripper 7000 series in 2023.
AMD Reveals 32-GB Radeon Graphics Card For AI Workloads
AMD’s new Radeon AI Pro R9700 graphics card represents a departure in branding from previous Radeon Pro products due to the emphasis on accelerating AI inference workloads.
Featuring 32 GB of GDDR6 memory and 128 AI accelerators, the Radeon AI Pro R9700 is capable of up to 96 teraflops of peak 16-bit floating-point math (FP16) and as many as 1,531 tera operations per second of spare 4-bit integer math (INT4).
The thermal design power for the graphics card is 300 watts.
The product is due out in July and will be supported by several add-in board partners, including ASRock, Asus, Gigabyte, PowerColor and Sapphire.
The graphics card provides up to two times better performance than Radeon Pro W7800 from 2023 for the 8-billion-parameter DeepSeek R1 Distill Llama model, according to AMD.
Compared to Nvidia’s 16-GB GeForce RTX 5080 on average tokens per second, the Radeon AI Pro R9700 is 361 percent faster for the Phi 3.5 Mixture-of-Experts Q4 model and 437 percent faster for the Mistal Small 3.1 24B Instruct 2503 Q8 model, according to AMD.
It’s also 454 percent faster for the DeepSeek R1 Distill 32B Qwen Q6 model, 447 percent faster for the 32B Qwen 3 32B Q6 model and 496 percent faster for the same model but with a large prompt of over 3,000 tokens, the company added.
AMD said up to four Radeon AI Pro R9700 graphics cards can be paired together using PCIe 5 connectivity in the same system to handle larger models such as the 123-billion-parameter Mistral Large Instruct model.
The company said the new graphics cards features expanded support for its ROCm on Radeon software stack for advanced AI development.
