The 9060 XT 16GB can be $350 with the 8GB mannequin at $300. They are going to launch on June fifth. The GPU die is identical for each fashions, however new from the 9070 (learn our overview) and 9070 XT (learn our overview) (which additionally shared a GPU die). The brand new die is Navi 44 for the 9060 XTs and sized at 199mm^2, down from 357mm^2 on the 9070-class playing cards.
The corporate didn’t present as many first-party testing outcomes as they sometimes have up to now. That’s not essentially a nasty factor, as a result of all of these outcomes should be taken with a grain of salt anyway, however we’ll primarily simply be sticking to the specs at present. We plan to overview these playing cards as soon as they launch. Our understanding is that, in contrast to the 5060, AMD plans to proceed sampling GPUs as regular.
Only a heads-up: The knowledge on this article is from a pre-briefing, so that is based mostly on conversations with AMD and never the reside presentation itself.
AMD Radeon RX 9060 XT
The 9060 XT will are available in 16GB and 8GB variations.
As for the options that can be shared between the 2: Every can have 32 compute models, 32 {hardware} RT accelerators, 64 of what AMD calls its AI accelerators, and a 3.13 GHz enhance clock. Each fashions will run PCIe gen 5.0 x16 slots, DisplayPort 2.1a, and HDMI 2.1b. AMD additionally lists a spread of 150W-182W for board energy, which explains the only PCIe 8-pin connector pictured within the rendering of the GPU. In talking with AMD, the decrease finish of the vary is for 8GB fashions.
For reference, AMD’s 9070 XT has 64 compute models, 64 RT accelerators, and 128 AI accelerators, or double the quantity of the 9060 XT’s CUs and accelerators. The 16GB 9060 XT matches the reminiscence capability of each the 9070 and 9070 XT, however with a weaker core.
These 9060 XTs can be direct opponents to NVIDIA’s 5060 Ti playing cards, even mirroring the identical VRAM configurations. Regardless of equivalent reminiscence sizes, NVIDIA’s 50 sequence playing cards make the most of a more moderen GDDR7 reminiscence in comparison with AMD’s GDDR6. As for the way a lot that issues, that is determined by the structure and the way a lot it’s going to depend on the reminiscence bandwidth and that further pace. We’ll have a look at it in testing and see the way it performs in the true world.
Extra variations embody the 9060 XT’s use of PCIe 5.0 x16 versus the 5060 Ti’s (learn our overview) PCIe 5.0 x8 interface. In benchmarking at x8 vs x16 on gen 5, it’s not going to matter. The place the place it’d matter is socketing it into an older board the place reducing the lane depend in half goes to turn out to be a restriction in some configurations.
AMD’s first-party benchmarks in contrast it in opposition to the 8GB RTX 5060 Ti, which we expect is honest because it’s a price-parity comparability. We’ll do our personal benchmarking fairly quickly in our overview.
AMD Ryzen Threadripper 9000 and PRO 9000 WX-Sequence CPUs
AMD additionally introduced its latest Threadripper 9000 Zen 5 CPUs, together with the 9980X, 9970X, and 9960X, codenamed “Shimada Peak.” These have been upgraded with elevated reminiscence assist and enhanced AVX-512 for extra demanding duties.
We’ll begin with the usual AMD Ryzen Threadripper 9000 CPUs. This sequence contains the 9980X, 9970X, and 9960X.
The 9980X is a 64C/128T CPU at 3.2 GHz base clock and with 256 MB of L3 cache. The 9970X has 32 cores, 64 threads, a 4.0 GHz base clock, and 128 MB of L3 cache. And at last, the 9960X will include 24 cores, 48 threads, a 4.2 GHz base clock, and in addition 128 MB of L3 cache. All of those CPUs can even function an as much as 5.4 GHz max enhance clock, PCIe 5.0, the identical sTR5 socket, and a 350W TDP.
For AMD’s Ryzen Threadripper PRO 9000 WX-Sequence: The corporate introduced six new CPUs, which embody the 9945WX, 9955WX, 9965WX, 9975WX, 9985WX, and the flagship 9995WX. Beginning with the 9945WX and dealing our manner up, these chips will include core counts of 12, 16, 24, 32, 64, and eventually 96 cores for the 9995WX. This mirrors the present and prior 7000 sequence CPU configurations simply now on Zen 5.
Each the PRO and non-PRO Threadripper CPUs appear to resemble the identical primary specs because the 7000 sequence of Threadripper processors. In these spec sheets, with a better max enhance frequency for the 9000 sequence CPUs, however loads of the remainder is acquainted.
One notable distinction between the PRO WX and the non-PRO sequence of Threadrippers is that the workstation sequence presents “AMD PRO applied sciences,” which AMD describes as, “a sturdy suite of enterprise-grade options together with multilayered safety, superior distant manageability, and long-term platform stability.” Moreover, not less than up to now, the PRO WX-series playing cards supported the WRX90 chipset along with the TRX50 chipset.
AMD hasn’t introduced any costs presently, however the press-brief lists availability for July 2025, so we ought to be anticipating to see these quickly.
AMD Radeon AI Professional R9700
Lastly, AMD launched its newest AI Professional workstation GPU. Intel additionally simply introduced its new Professional GPUs this previous week and we have already got a tear-down up of the B60.
For specs, this RDNA 4 card will include 128 AI accelerators, 32GB of GDDR6 reminiscence, as much as 1531 TOPS claimed, and a 300W TDP.
In comparison with one among its predecessor, the Radeon Professional W7700, the brand new R9700 will increase TFLOPS (FP16) from 56.54 to 96, will increase AI accelerators from 96 to 128, upgrades to PCIe Gen 5 from Gen 4, and doubles the reminiscence measurement from 16 to 32GB. Sadly, AMD’s press-brief didn’t embody any CU, stream processors, or reminiscence bandwidth data for the R9700, so we’ll have to attend to see these precise specs.
Because of the R9700’s noticeable configuration enhancements over its predecessor, the brand new GPU finally ends up being barely extra corresponding to the Radeon Professional W7800 which has 140 AI accelerators, 32GB of GDDR6 reminiscence, and 90.5 (FP16) TFLOPS.
In its press-brief, AMD included a slide as an example how 32GB of VRAM provides customers extra choices of their capability to load bigger AI fashions by highlighting 4 fashions that might exceed 16GB of VRAM, however can be utilized with 32GB of VRAM as a substitute. Moreover, as a result of GPU’s capability to load fashions with bigger parameters or which might be much less quantized, the GPU can also see an uplift within the accuracy of the mannequin’s responses.
To broaden upon that time, AMD contains one other chart labeled “Giant AI Fashions Efficiency” the place it compares an RTX 5080 (learn our overview) to its AI Professional R9700. As soon as once more, this chart demonstrates how 32GB presents entry to run bigger fashions that 16GB simply can’t deal with. These outcomes are anticipated. We predict a extra significant comparability may’ve been utilizing the RTX 5090 that additionally has 32GB of VRAM. This could symbolize a extra like-for-like state of affairs however we don’t do loads of ML testing so we’ll depart that for another person.
AMD additionally exhibits off the cardboard’s “Multi-GPU PCIe 5 platform” that permits customers to attach 4 AI PRO R9700s for some extraordinarily demanding fashions that want as much as 128GB of mixed VRAM and theoretically 4x the computing energy.
We didn’t obtain a value for this card, however AMD lists an availability of July 2025.
Conclusion
That’ll wrap it up for AMD’s bulletins.
Sadly, it’s a bit troublesome to get an thought for efficiency based mostly on the specs alone, and even more durable to get an thought for the worth for one thing with out a value.
Ideally, we’d be capable to get our fingers on a few of these as soon as they’re publicly accessible, which ought to be quickly in accordance with AMD’s press-brief.
