a new challenger appears to be like —

RDNA 3 GPUs will spend many chiplets to counter Nvidia’s grand RTX 4000 sequence.

Andrew Cunningham
– Nov 3, 2022 9: 00 pm UTC

Create bigger / AMD’s Radeon RX 7900 XTX, floating in a crimson-tinged room somewhere.


AMD is gearing up to launch its next-abilities Radeon RX 7000-sequence GPUs next month, and recently the firm shared extra crucial aspects about the playing cards’ pricing, performance ranges, and the brand new RDNA 3 GPU structure that will vitality all of its graphics playing cards for the next couple of years.

The launch begins at the high dwell, with the Radeon RX 7900 XTX and RX 7900 XT. AMD will launch both of those GPUs on December 13, with the 7900 XTX starting up at $999 and the XT starting up at $899 (playing cards made by AMD’s companions will for sure push these prices upward just a exiguous). Each of those label tags undercut Nvidia’s RTX 4000 sequence, which begins at $1,599 for the dwell-tier GeForce RTX 4090 and $1,199 for the RTX 4080.

Launch MSRP
Sport clock
RAM/bus width

Radeon RX 7900 XTX
96 (RDNA 3)
2300 MHz
24GB GDDR6 (384-bit)
355 W

Radeon RX 6950 XT
80 (RDNA 2)
2100 MHz
16GB GDDR6 (256-bit)
335 W

Radeon RX 7900 XT
84 (RDNA 3)
2000 MHz
20GB GDDR6 (320-bit)
300 W

Radeon RX 6800 XT
72 (RDNA 2)
2015 MHz
16GB GDDR6 (256-bit)
300 W

Nvidia had planned to subject a 12GB model of the RTX 4080 on this label vary, but a shopper backlash to its doubtlessly advanced branding and lower performance degree precipitated Nvidia to “unlaunch” the cardboard, with no discover on when it will doubtless be re-introduced, what it will doubtless be known as, or what it will label. The 16GB model of the 4080 launches November 16, beating AMD’s new GPUs to market by round a month.

AMD didn’t employ any time evaluating the brand new Radeon playing cards to any Nvidia GPUs, opting as some other to evaluate them to their predecessors.


The 7900 XTX is the brand new flagship of the sequence.


The 7900 XT is a little step wait on in memory capability, bus width, clock flee, CUs, and vitality consumption for $100 less.


When it came to evaluating its playing cards against Nvidia’s, AMD performed coy throughout its presentation, fastidiously evaluating the RX 7900 XTX sequence to the RX 6950 XT without pointing out both the RTX 3000 sequence or the RTX 4000 sequence. The 7900 XTX is supposed to be “up to 1.7x” quicker than the 6950 XT whereas taking part in games in 4K—a 1.5x enchancment used to be extra traditional among the games AMD tested.

Referring to raw compute performance, AMD says the 7900 XTX is in a position to up to 61 teraflops of performance, whereas Nvidia says the RTX 4090 is in a position to 83 teraflops and the RTX 4080 of 49 teraflops. It is some distance a huge high-degree performance number that doesn’t inform us plenty about how the playing cards will evaluate in true games (the RTX 4090 is now not 70 p.c quicker than the RTX 4080, per Nvidia’s own benchmarks). Nonetheless along with the pricing and lack of head-to-head 4090 comparisons, this vague number does imply that AMD is aiming at Nvidia’s 4080 tier with these playing cards rather than attempting to outperform the RTX 4090. This is the identical relative space that AMD has been in for a few generations now.

A receive of chiplets

A fully enabled RDNA 3 GPU, as archaic in the RX 7900 XTX.


AMD is utilizing chiplets in the RX 7000 sequence, but all of the compute property are serene in a single gigantic die.


The GCD is where most of the hardware lives, along with the present controller and media encoding engine.


The 6nm MCDs are for “up to” 16MB of Infinity Cache and memory controllers, which might originate it somewhat straightforward for AMD to scale down for lower-dwell products.


Animated from a 7nm job to 5nm has upped RDNA 3’s transistor density.


Poke processor improvements.


RDNA 3 entails two AI accelerator units per CU.


As expected, the RDNA 3 GPUs will grab a page out of Ryzen’s guide, utilizing a community of chiplets rather than a single orderly slab of silicon care for Nvidia is utilizing. By making a GPU out of many smaller interconnected silicon dies, AMD can doubtlessly reduce the label and beef up the yields of its GPUs by guaranteeing that less silicon is thrown out or binned ensuing from little imperfections or defects.

No longer like in Ryzen, where the true CPU cores are spoil up among a few chiplets, AMD is serene placing all of the GPU’s true processing vitality exact into a single portion of silicon known as the Graphics Compute Die, or GCD. Manufactured utilizing TSMC’s 5 nm job, the GCD contains all of the true RDNA 3 compute units, plus the present engine (for controlling video output) and the media encoding and decoding engine.

The other chiplets in every GPU are smaller Reminiscence Cache Dies, or MCDs; these every own a 64-bit memory controller plus Infinity Cache memory. The 7900 XTX has six of those MCDs lit up, whereas the 7900 XT has 5 of them (you would inform since the XTX has a 384-bit memory bus, whereas the XT has a 320-bit bus). AMD is constructing the MCDs on a 6 nm manufacturing job, care for the I/O die in the Ryzen 7000 CPUs.

AMD taking into account enhancing RDNA 3’s ray-tracing performance, and every CU wants to be in a position to up to 50 p.c extra performance than in RDNA 2.


AMD’s ray-tracing improvements as when put next in true games.


Besides for enhancing framerates, RDNA 3 makes a speciality of rising RDNA 2’s mediocre ray-tracing performance. In comparison with Nvidia’s RTX 3000-sequence and Intel’s Arc GPUs, AMD’s RDNA 2 ray-tracing accelerators handled fewer steps of the ray-tracing pipeline—particularly, they offloaded something known as “tree traversal” to the overall-cause movement processors, whereas Nvidia’s and Intel’s ray-tracing units handled traversal themselves. RDNA 3’s ray-tracing accelerators make handle this traversal, which, along with other improvements, wants to be stunning for “up to 50% extra [ray-tracing] performance per CU,” per AMD.

The GPU’s media encoding capabilities are moreover getting a enhance. Most particularly, they’ll earn up to the RTX 4000-sequence and Intel’s Arc playing cards by supporting hardware-accelerated encoding and decoding of AV1 video streams (up to 8K at 60 FPS, identical to what Nvidia and Intel provide). RDNA 2 GPUs easiest toughen hardware-accelerated decoding of AV1 movies. And AMD valuable that the brand new GPUs toughen the DisplayPort 2.1 spec, which might enable increased refresh rates on DisplayPort 2.1-succesful monitors when they’re extra widely on hand.

Create bigger / Would possibly perhaps this be a snarky reference to the laughably grand dimension of the RTX 4090 sequence? Most likely!


Though its announcement used to be light on performance comparisons, AMD took potshots at the sheer dimension and vitality requirements of Nvidia’s RTX 4090, demonstrating that AMD’s 7900 XTX reference own might maybe perhaps moreover without train change an RX 6950 XT without requiring an upgraded vitality offer, a vitality cable adapter, or a bigger case. Peak vitality consumption goes up from 330 W to 355 W—serene an originate bigger but round 100 W lower than the RTX 4090’s peak vitality consumption and serene in a position to being powered by a pair of 8-pin PCIe vitality connectors.

On the tool facet, AMD teased a next-abilities model of its FidelityFX Broad Choice (FSR) upscaling abilities, which it says is “coming in 2023” and will lift “up to 2x extra FPS vs FSR 2” at 4K resolutions. FSR 2 helped to shut the visible quality gap between the principle model of FSR and Nvidia’s identical DLSS upscaling tech whereas moreover working on extra GPUs, along with older Radeon GPUs and many Intel and Nvidia GPUs. We do now not but know whether FSR 3 will moreover be GPU-agnostic or if AMD will limit it to the latest-abilities Radeon GPUs as Nvidia does with DLSS 3 and the RTX 4000-sequence.

We’ll be taking a more in-depth count on at AMD’s Radeon RX 7000-sequence playing cards when they’re just a exiguous nearer to launch.