AMD RX 9070 & 9070 XT GPU Prices, Specs, & Release Date
We go over the RX 9070 and RX 9070 XT’s price, specs, and AMD’s performance claims
The Highlights
- Both the RX 9070 and RX 9070 XT will have 16GB of VRAM
- AMD did not compare its new cards against NVIDIA’s 40 or 50-series GPUs
- The 9070 and 9070 XT will only be separated by $50
- Original MSRP: $550 (RX 9070), $600 (RX 9070 XT)
- Release Date: March 6, 2025
Table of Contents
- AutoTOC
Intro
AMD's RX 9070 XT MSRP is $600, with the 9070 at $550. They are releasing on March 6th, which is a day after the RTX 5070 launches.
That’s a weirdly small $50 gap between these, but AMD isn’t a stranger to smothering one product with another. It did that with the 7900 XT and the 7900 XTX before.
AMD today announced its RX 9070 series GPUs -- again. It announced them back in January, but then decided it didn’t want to announce them, so its announcement shriveled up and receded back whence it came. But now, AMD is proud to re-announce its announced RX 9070 series.
Editor's note: This was originally published on February 28, 2025 as a video. This content has been adapted to written format for this article and is unchanged from the original publication.
Credits
Host, Writing
Steve Burke
Video Editing
Vitalii Makhnovets
Tim Phetdara
Camera
Andrew Coleman
Writing, Web Editing
Jimmy Thang
The RX 9070 is positioned to fight the RTX 5070, with the 9070 XT up against the 5070 Ti. AMD has decided to steal NVIDIA’s naming this time, growing tired of the naming scheme that brought us the R9 270, R7 370, RX 480 (watch our review), Fury, Fury X (watch our review), RX 580 (watch our review), Vega Frontier Edition (watch our review), Vega 56, Vega 64 (watch our revisit), back to RX 590 (watch our review), Radeon VII (watch our review), RX 5700 XT (watch our review), RX 6950 XT (watch our review), and RX 7900 XTX.
Maybe leaving the old names behind is for the best. We look forward to seeing what they do when they hit the dreaded “10.” Maybe AMD will be the first to bring us the 10,080 Ti…XT.
Overview
First up, the price: At $50 apart, it seems like these cards will smother each other. Until we can publish test data, we won’t know which one will cannibalize the other -- but it seems likely, as happened with the 7900 XT and 7900 XTX when they were about $100 apart at launch.
A quick pricing recap:
AMD’s RX 7800 XT is currently priced at around $500 to $530 depending on where or if you can get it. The RX 7900 GRE has been around $550 (though has been mostly out of stock lately). The RX 7900 XT is basically gone, but was around $630 to $640 at the end of last year and more commonly around $680. The 7900 XTX was as low as $800 to $820 in November and was commonly $930 in January. It launched at $1,000.
The 7900 XT received a review from us entitled “AMD’s Greedy Upsell” in December of 2022, knocking it for the $900 price point. We revisited it in October of 2023, so less than a year later, because you could get some models for $720. AMD dropped the price by 20% or so in less than a year.
The 7900 XT became an awesome value after the price drops at least for a window there, especially in raster performance.
We’ll talk more about the pricing once we have benchmark numbers in. That’ll include NVIDIA’s lineup.
With that context, let’s get into the specs of the 9070 and 9070 XT.
Specs
AMD provided this spec sheet for its RX 9070 and RX 9070 XT. The RX 9070 will have 56 Compute Units, or CUs, against the 64 on the 9070 XT (a 14% increase in CUs). AMD has done splits like this before, like Vega 56 and 64, but the architecture has changed dramatically since then.
In a call with the press, AMD claimed that the reason it didn’t go up to something like 80 CUs is because it didn’t want to make something super expensive. That definitely would have been very expensive. It would have been a larger die so that is accurate that the costs would go up. We also think that the architecture might just struggle to compete with a 5090 at the cost they would need to hit to do so and that’s probably a large reason for this as well. AMD is focusing on the mid-range, where the architecture is perhaps better tuned to compete.
The hardware ray tracing accelerators match the CU count at 56 and 64, with AMD’s so-called “AI accelerators” at 112 and 128 units.
Boost clocks are significantly different between them: The 9070 XT is able to make more use of its extra 84W of power budget to hit an advertised boost of 2.97 GHz, with the RX 9070 non-XT at 2.52 GHz (the original slides said 2.54, but AMD brought that down before announcement). Memory capacity is identical between them at 16 GB. Both cards will utilize GDDR6 at 20 Gbps. Board power is advertised at 220W and 304W, hopefully with room for board partners to scale up with overclocking support. NVIDIA’s OC support has been relatively lackluster this generation, despite what the company claimed. Hopefully AMD can make it exciting.
Finally, the cards technically are on PCIe 5.0 x16 slots, but we already showed that there’s no real impact to performance here in regards to the 5090. There almost certainly won’t be any impact with the 90 series either with the exception if they were to cut down the slot on a lower-end model or something, but for this, it won’t matter. DisplayPort is up to 2.1a and HDMI at 2.1b.
Block Diagram
Here’s the block diagram for the RDNA 4 die and architecture. AMD noted that this same die will be used for the 9070 and 9070 XT. The company says this will be 356.5mm^2, contain 53.9 billion transistors at most, and run on a 4nm process node. And then the 9070 would have some of the CUs basically turned off. AMD says that this is monolithic silicon and not a chiplet design.
This variation of the GPU has 4 Shader Engines, within each is contained 8 dual-compute units. L2 Cache is centralized and located towards the middle of the logical block, totaling 8MB max of L2, with 64MB of infinity cache at the outer edges and closer to the memory controllers.
AMD’s big claim here is a renewed focus on ray tracing performance, where it says it has doubled ray intersection rates and improved ray traversal, alongside changes to how it’s handling bounding boxes for BVH probing. This is a big focus for AMD this generation and it is somewhere the company needs to focus because they’re pretty competitive in raster compared to NVIDIA a lot of the time, but there’s instances where they get completely crushed in ray tracing.
Here’s a closer look internally. The area we’ll focus is on RT. Within what AMD calls the compute engine, AMD now has 2x ray accelerators to handle box and triangle intersections. AMD noted that the second intersection engine within the accelerator “doubles the performance for both ray-box and ray-triangle testing” over RDNA 3, although remember that this doesn’t mean a clean doubling in performance in an actual ray tracing game scenario. AMD noted that RT processing takes advantage of a 128KB shared memory block, also shown here.
This slide was dedicated to RT improvements specifically and highlights the addition of AMD’s dedicated ray transform block, which it says will “offload transformation as you transition from the top-level RT structure to the bottom level, where there may be many instances of a particular geometry.” Previously, shader instructions handled this task and, AMD says, added overhead to ray traversal which it claims to now have eliminated.
The ray accelerators have the ray transform block and two intersection engines, which themselves required changes to BVH handling to be fully leveraged.
AMD said it is moving to an 8-wide BVH solution from a 4-wide option on earlier hardware, which it couples with its new oriented bounding box approach that attempts to reduce wasted, empty space in bounding boxes by better conforming to the geometry in the scene, theoretically reducing false positives and also reducing performance overhead and loss during geometry intersection probing.
The slide claims that traversal performance improves by 10%. Again, this is not a literal 10% gain in the final framerate in an RT game, but is a building block in a series of others to contribute to AMD’s claimed uplift.
This slide above was pretty cool. The right side shows AMD’s register allocation, with the top-right showing RDNA 3 and the bottom-right showing RDNA 4. Between the two, there’s a change from static allocation in RDNA 3 to dynamic in RDNA 4. AMD highlights that RDNA 3 would reserve registers which may not be put to work, so it’d hold them in case they were needed, potentially not need them, and end up with inefficiency and unavailable resources. RDNA 4 is trying to resolve this. The bullets on the left make all of this pretty clear, stating that the improvement is in efficient utilization, largely because registers can be released or requested as needed.
AMD is claiming that its RDNA 4 CUs improve traversal by 2x over RDNA 3 when iso clock and bandwidth. The 3D block in the image is supposed to roughly illustrate where AMD thinks it’s finding most of its performance: It appears that the 2x intersectors and BVH8 change (from 4) are the largest contributions.
RDNA 4 also introduces more out-of-order queuing and aims to reduce latency of memory requests, which AMD claims further benefit RT performance. Further out of order execution allows work to complete even while longer latency requests are processing or queuing. AMD makes an example out of an uncached leaf node on the slide above, which would contain the lowest level of detail in an RT workload and could otherwise hold up a scene.
First-Party Claims
Grab a GN15 Large Anti-Static Modmat to celebrate our 15th Anniversary and for a high-quality PC building work surface. The Modmat features useful PC building diagrams and is anti-static conductive. Purchases directly fund our work! (or consider a direct donation or a Patreon contribution!)First-Party Claims: RX 9070
Let’s get into the first-party benchmark claims next. We won’t spend a ton of time on these since you’ll be able to find plenty of third-party reviews soon enough, but it will help set expectations for what AMD is targeting.
AMD’s quick reference slide shows a claimed 26% uplift against the RTX 3080 and 38% uplift against the 6800 XT. AMD didn’t show anything from NVIDIA’s 40 or 50 series here.
At 4K/Ultra and without upscaling, AMD is marketing the RX 9070 non-XT as an average of 21% improved over the RX 7900 GRE. The 7900 GRE was originally a $550 card. AMD is showing non-RT performance as improving up to 28% on baseline 100%, with the ray tracing performance showing a disproportionately favorable gain to the new architecture at up to 34%. This is good for AMD, as it was weakest in ray tracing historically. This disproportionate gain won’t wipe-out AMD’s deficit in something like Cyberpunk 2077 or possibly Black Myth: Wukong, but the key will be whether it can close the gap with better value.
At 1440p/Ultra and native, AMD claims the 9070 will be 20% faster on average, with the peak at 38% improved for ray tracing and 26% improved for raster. AMD observed a slightly larger improvement in RT at 1440p for F1, which is interesting, despite overall losses in scaling in raster. 4K diverging from 1440p isn’t abnormal, though. One thing we do want to call attention to and give AMD credit for here though is that they’re showing native performance. Even if they show FSR, that’s fine if they kept it locked to the same FSR options between their older and newer gen cards if they’re comparing their own products to each other; showing native is a better step than that. This is a massive improvement over what we’ve been complaining about NVIDIA doing, which was comparing its 50 series to its 40 series and enabling MFG 4X on the 50 series but not the 40 series and then just making it look like they are wildly better than they actually are. So we do want to call attention to and give AMD credit for making a more fair head-to-head comparison between its own products here rather than enabling some special multiplier on one and not the other.
First-Party Claims: RX 9070 XT
As for the RX 9070 XT, AMD compared it against its 6900 XT and NVIDIA’s RTX 3090, again lacking in any 40 or 50 series comparisons here. It claims a 51% uplift over the 6900 XT and 26% average uplift over the RTX 3090.
Against the same 7900 GRE, AMD claims an average uplift of 42%, or 66% in F1 24 with ray tracing at the high end. AMD claims it saw the same uplift in Cyberpunk with RT. In raster performance, the gains max-out at 48% over baseline.
At 1440p, AMD is seeing lower overall average uplift, with a slight uptick in F1 24 with ray tracing.
Conclusion
Visit our Patreon page to contribute a few dollars toward this website's operation (or consider a direct donation or buying something from our GN Store!) Additionally, when you purchase through links to retailers on our site, we may earn a small affiliate commission.We could run these percent scaling improvements against our own numbers to approximate where AMD would land since we have all the details we need here to calculate the expected performance. Launch is only a few days away though, and we’d rather test the cards than extrapolate on data that we have no control over or insight into. We’ll have third-party numbers in our own reviews soon enough.
Pricing is going to remain the key concern for these cards. AMD is doing some interesting things architecturally. We’ve covered a bit of that in this article. AMD has plenty of opportunities to have…ROPs and cables that…are not going to burn. The field is set for AMD to have a victory here. It is up to the company to execute on it. We just made a video called “AMD, Don’t Screw This Up” that talks primarily about the pricing, but the reason AMD has this amazing opportunity is because of NVIDIA’s screw-ups.
Pricing is the key concern, but we’ll withhold judgment on it until we review it about a week from now.
From what we understand, it sounds like supply will be okay, but it’s hard to know what that really means.
AMD really needs to gain market share. According to Jon Peddie Research, AMD is close to the lowest they’ve ever been in the GPU market at around 10 percent while NVIDIA is down in terms of reputation and trust in their brand, so now is the time for AMD to strike. We’ll let you know if they execute on that soon.