Threadripper 9000 And Radeon AI Professional 9000: AMD’s Workstation Energy-Up Unveiled

At Computex 2025 a few weeks again, AMD revealed its newest CPUs and GPUs for top finish desktops (HEDT) and the skilled workstation market, the Ryzen Threadripper 9000 and Radeon AI Professional 9000 sequence. As you’re in all probability conscious, AMD’s Ryzen Threadripper 9000 sequence and new Radeon AI Professionals are primarily based on the Zen 5 and RDNA 4 architectures, respectively, each of which have already been in-market on the patron facet for fairly a while. As such, we’re already intimately aware of the applied sciences that make them tick. AMD’s Ryzen Threadripper 9000 and Radeon AI Professional 9000 sequence, nonetheless, additionally blaze new trails of their respective product classes and signify AMD’s strongest Excessive Finish Desktop {and professional} workstation merchandise up to now.

We’ve already coated the preliminary bulletins out of Computex and printed a few deep dives for Zen 5 and RDNA 4, so if you wish to be taught and the juicy particulars you’ve bought some studying to do. We’ve bought some new product particulars and efficiency knowledge to share right this moment, nonetheless, a little bit of evaluate might be so as. First up, let’s talk about the Ryzen Threadripper 9000 sequence…

Introducing AMD’s Ryzen Threadripper 9000 Sequence

The brand new Threadripper 9000 sequence is all about elevated knowledge throughput, efficiency, and platform bandwidth. We’re coming into considerably of a brand new period with skilled workstations the place conventional engineering, scientific, media and leisure, and design workloads are as prevalent as ever. Nascent AI and machine studying workloads, nonetheless, may also profit from the large bandwidth and parallel compute capabilities of the Threadripper 9000. With that in thoughts, Threadripper 9000 is effectively positioned to energy a recent wave of workstations that may profit considerably from its large quantities of compute, bandwidth, and IO. Quad GPUs with full PCIe bandwidth are potential on this platform. 

A abstract of what Zen 5 gives is printed within the slide above. Zen 5 gives larger IPC due to improved fetch and decode, wider dispatch and execution engines, and elevated cache bandwidth. Integer, Floating Level, and Vector math advances have been made as effectively.

Zen 5 additionally options higher department prediction with decrease latency than Zen 4. The structure options twin decode pipelines that make the most of extra refined and correct TAGE (TAgged GEometric) prediction algorithms. Instruction cache latency and bandwidth have been improved too. 

The structure options an 8-wide dispatch / retire engine — versus 6-wide in Zen 4 — with 6 ALUs (3 Multiples) and a brand new unified ALU scheduler design, whereas older Zen architectures featured a singular scheduler per ALU. Because of this, Zen 5 has a 40% bigger execution window that may help as much as 448 excellent operations. The structure can basically deal with extra directions and higher recuperate from misses.

AMD additionally enlarged the Zen 5 knowledge cache by 50% (32K vs. 48KB) and doubled the utmost bandwidth to the L1 cache and FPU. Zen 5 encompasses a 48KB 12-way L1 knowledge cache, however maintains a 4-cycle load to maintain latency low, regardless of the upper capability. The bandwidth from L2 to L1 has additionally been doubled versus Zen 4. 

In Zen 4, AVX-512 is carried out utilizing double-pumped 256-bit knowledge chunks, so 512-bit AVX-512 workloads have been damaged down into two 256-bit chunks and despatched by way of the engine. Zen 5, nonetheless, encompasses a full 512-bit huge knowledge path. The FP/vector math unit in Zen 5 options six pipelines with two-cycle latency for FADD directions, which is a discount from 3-cycles in Zen 4. They’ll additionally deal with a bigger variety of in-flight FP directions. 

As we’ve talked about quite a few instances previously, Zen 5 cores provide a median 16% IPC uplift versus Zen 4. That clearly interprets to higher single-thread efficiency at related clocks, however in mixture, multi-thread efficiency is improved as effectively, as soon as that uplift is scaled throughout many cores.

Threadripper 9000 Works In Present Motherboards

Threadripper 9000 sequence processors are drop in appropriate with present WRX90 and TRX50 platforms, which each options the sTR5 socket. There’ll in all probability be some new motherboards coming alongside the processors, however they’re technically not a requirement – present boards will work after a BIOS / UEFI replace. 

When utilized in WRX90 motherboards, Threadripper Professional 9000WX processors may be configured with as much as 8-channel reminiscence they usually function 148 complete PCIe lanes, 144 of that are usable, and 128 of that are PCIe Gen 5. The TRX50 HEDT platform will max out with 4-channel reminiscence and a complete of 92 PCIe lanes (88 usable, 48 PCIe 5). Each platforms require DDR5 RDIMMs, identical to the Threadripper 7000 sequence, however quicker speeds will probably be formally supported. 

When it comes to their reminiscence capabilities, Threadripper Professional 9000WX processors will help as much as 2TB of system reminiscence at knowledge charges as much as DDR5-6400, for peak bandwidth of as much as 410GB/s in an 8-channel configuration. For non-Professional HEDT Threadripper 9000s, peak bandwidth is halved, because of its most of 4 reminiscence channels. Larger speeds can be found by way of overclocking, nonetheless, and the platform will help AMD EXPO reminiscence profiles for fast and straightforward configuration.

The Threadripper Professional 9000WX sequence additionally differs from its customary HEDT counterparts in that it’ll provide AMD’s “Professional” safety and manageability applied sciences. The precise construct and packaging is completely different as effectively. As you possibly can see within the pictures beneath, the Threadripper Professional 9000WX have 4 further compute dies. The HEDT elements aren’t the identical elements with cores disabled — they really lack the compute dies for these cores.

These newest Threadrippers scale to 96 cores and 192 threads, identical to the earlier technology line-up; the primary distinction is that the Threadripper 9000 sequence is predicated on AMD’s Zen 5 structure.

With the HEDT elements, AMD has launched a trio of fashions. Sitting on the high is the Ryzen Threadripper 9980X, a 64-core/128-thread chip with a 3.2GHz base clock, as much as a 5.4GHz increase frequency, and a beneficiant heaping of complete cache at 320MB, together with 256MB of L3 cache. Subsequent is the Ryzen Threadripper 9970X with 32 cores, 64 threads. It encompasses a 4GHz base clock, a 5.4GHz max increase clock, 128MB of L3 cache, and 160MB of complete cache. Lastly, there’s the Ryzen Threadripper 9960X with 24 cores, 48 threads, a 4.2GHz base clock, as much as a 5.4GHz max increase clock, 128MB of L3 cache, and 152MB of complete cache.

Whereas the HEDT fashions high out at 64 cores and 128 threads, the Professional 9000 WX sequence scales a lot larger. The flagship is the Ryzen Threadripper Professional 9995WX with 96 cores, 192 threads, a 2.5GHz base clock, as much as a 5.4GHz increase frequency, an enormous 384MB of L3 cache, and 480MB of complete cache. On the opposite finish of the spectrum, the Ryzen Threadripper Professional 9945WX encompasses a tamer configuration with 12 cores, 24 threads, a 4.2GHz base clock, as much as a 5.4GHz increase clock, 64MB of L3 cache, and 152MB of complete cache. 

Threadripper 9000 Anticipated Efficiency

When it comes to their anticipated efficiency ranges, AMD offered fairly a bit of knowledge versus each previous-gen Threadrippers and Intel’s Xeon W workstation processors. 

Be aware that this comparability pits the 64-core HEDT Threadripper 9980X towards Intel’s 60-core Xeon w9-3595X. As you possibly can see, the good points range however may be large and the Threadripper can sometimes provide greater than double the efficiency. 

And now for some Threadripper Professional comparisons. Versus the previous-gen Threadripper Professional 7995WX, which has an identical core rely, the upcoming Threadripper Professional 9995WX gives  as much as 26% larger efficiency in multi-threaded workloads. We anticipate enhancements in responsiveness and single-thread workloads as effectively.

While you evaluate a 96-core Threadripper Professional 9995WX to Intel’s high Xeon W processor, the Threadripper Professional pulls even additional forward. Whether or not its for design work, media manufacturing or AI / ML workloads, based on AMD’s numbers, the brand new Threadripper Professional makes minced meat of the present top-end Xeon W.  We’ll attempt to confirm a few of these numbers, assuming we get to evaluate the brand new Threadrippers sooner or later sooner or later.

AMD Radeon AI Professional 9700: RDNA 4 For Engineers And Professional-Vis

Which brings us to the Radeon AI Professional 9700. At its core, the Radeon AI Professional 9700 is just like the consumer-class Radeon RX 9070 XT, however it encompasses a completely completely different board design with double the reminiscence, which makes it a lot better suited to Professional-Visible and AI workloads. 

The Radeon AI Professional 9700 is constructed across the RDNA 4-based Navi 48 GPU. The GPU is organized into 4 shader engines, every with 8 workgroup processors, that are geared up with two compute items every. It is a related construction to earlier gen, RDNA 3 GPUs, however there are lots of architectural enhancements, which enhance efficiency and effectivity and add new options. 

Inside every CU there are up to date third technology Ray Accelerators that provide 2x the height throughput of RDNA 3. RDNA 4’s Ray Accelerators provide double the ray intersection charges, with improved BVH compression, accelerated ray traversal and shading, and add help for a function known as Oriented Bounding Packing containers. third Era Matrix Accelerators are additionally current, which provide improved efficiency, together with help for 8-bit float knowledge varieties, with structured sparsity help as effectively. RDNA 4 is healthier suited to AI / ML workloads and it helps further knowledge varieties. 

The GPU options 2MB of mixture CU cache, together with 8MB of L2 cache, and 64MB of third technology Infinity Cache located out on the edges, between the reminiscence controllers and shader engines. The L2 cache surrounds an up to date, centrally situated Command Processor, with enhanced packet acceleration. The up to date Command Processor additionally helps some new directions and it options improved department prediction, to higher predict knowledge fetches and maintain caches higher utilized. 

On the Radeon AI Professional 9700, the GPU connects to 32GB of GDDR6 reminiscence working at an efficient knowledge price of 20Gbps, over a 256-bit interface (4, 4×16 reminiscence controllers). The reminiscence controllers additionally function enhanced reminiscence compression expertise, to make higher use of obtainable bandwidth.

The Radeon AI Professional 9700 options the complete implementation of the Navi 48 GPU, with 64 complete CUs, 64 Ray Accelerators, and 128 AI (Matrix) Accelerators. It gives as much as 1,531 peak TOPs (Int4 with Sparsity) with a lift clock of about 2.9GHz and a complete board energy of 300 watts. That’s barely decrease than the Radeon RX 9070 XT, for these preserving observe.

Twin media engines and an up to date Radiance Show Engine are current on the GPU as effectively. There’s truly quite a few updates with the media and show engines, and we’ve bought way more element relating to gaming / graphics associated updates – however you’re going to should learn our RDNA 4 deep dive for the complete scoop. There’s an excessive amount of to floor to cowl for our evaluation right this moment. 

In a significant departure from earlier Radeon Professional playing cards, AMD has determined to permit companions to convey Radeon AI Professional 9700 playing cards to market. AMD made this transfer to broaden availability and to convey its Professional {hardware} and software program to a wider vary of consumers. AMD additionally hopes to work with its companions to develop new enterprise on ROCm and Radeon.

Radeon AI Professional 9700 Anticipated Efficiency

With its quicker reminiscence and newer structure, the Radeon AI Professional 9700 ought to provide some important efficiency enhancements over previous-gen Radeon Professional playing cards as effectively.

The GPU gives as much as 4X extra TOPs with some knowledge varieties, however when it comes to FP32 efficiency the Radeon AI Professional 9700 is barely marginally extra performant than the Radeon Professional W7800.

Even with related quantities of reminiscence, nonetheless, the upcoming Radeon AI Professional 9700 can provide hige efficiency enhancements over the RDNA 3-based Radeon Professional when it comes to tokens per second. And as we noticed in our Radeon RX 9070 XT evaluate, the Navi 48 GPU gives a giant increase for rendering and gaming versus previous-gen Radeons as effectively. 

AMD Threadripper 9000 And Radeon AI Professional For Workstations: Key Take-Aways

The Threadripper 9000 and Radeon AI Professional 9000 sequence aren’t on the market simply but — they’re going to be coming to market just a little later within the quarter. Once they do arrive although, we’re anticipating excellent issues. Upgrading Threadripper with Zen 5 ought to imply elevated efficiency and effectivity throughout, and RDNA 4 has been considered one of AMDs extra thrilling GPU architectures in fairly some time.

All the issues which have made Zen 5 and RDNA 4 profitable within the shopper house are coming to skilled workstations, and we anticipate AMD’s going to get pleasure from loads of success. We’ll clearly maintain ultimate judgement till we get to truly check the merchandise, however issues are wanting fairly good at this level.