AMD’s RX 6950XT, 6750XT, and 6650XT GPUs are now on sale for $399 and up


Enlarge / Three new, yet largely familiar, GPUs are now available from AMD.

AMD

Starting today, AMD and a number of GPU makers will begin rolling out three new graphics cards in the Radeon 6000-series lineup. These new products add a “50” suffix to existing model numbers, and if you’ve followed AMD over the past few years, you can probably guess what that means.

The RX 6950XT, RX 6750XT, and RX 6660XT are heading to retailers today at suggested minimum prices of $1,099, $549, and $399, respectively. Each GPU comes with nearly identical specs to their “00” suffix predecessors, with the only difference between each being a factory-level boost to clock speeds and memory speeds.

Infinity Cache again—but also, very finite clock tweaks

In a press briefing ahead of today’s announcements, AMD reps suggested that the gains are in part thanks to touch-ups on both the silicon and software levels, though for two of the GPUs, a wattage jump factors in, as well. The 6950XT now demands 335W TDP, up 35 W from its 6900XT predecessor, and we’re at 180 W TDP for the 6650XT, which is a 20 W jump from the 6600XT.

Examine AMD's three newest GPUs from a slightly different angle.
Enlarge / Examine AMD’s three newest GPUs from a slightly different angle.

AMD

The jump from the RX 6900XT to the RX 6950XT nets 2.6 percent higher boost clocks and 7.8 percent higher memory clocks; all other GPU elements, particularly AMD’s beefy allocations of L3 “Infinity Cache,” remain identical between the two models. A similar story plays out for the other 50-series jumps: less than 1 percent higher boost clocks for the 6750XT, and 1.8 percent higher boost clocks for the 6650XT.

The scant jump in clocks suggests that AMD might be introducing new models to replace the old ones outright, but AMD told Ars Technica that the RX 6900XT and RX 6700XT will continue to be sold, with the RX 6600XT being phased out once its existing stock is cleared. (For a step down at that level, AMD will continue selling the RX 6600.) AMD is manufacturing its own direct versions of the RX 6950XT and RX 6750XT, but not the 6650XT; all three will be made and sold by the usual roster of popular Add-In Board (AIB) GPU partners.

Without review hardware in our hands, we’re left wondering exactly how much these slight spec bumps jump ahead of comparable “00” models. For some modern GPU customers, that kind of comparison may be moot; the older models, like most modern GPUs, sold out incredibly quickly amidst a crypto boom, yet recently stabilizing GPU prices mean that you might actually buy a “50” model for less in the end.

Team Red vs. Team Green, once again

At that point, then, it’s the usual tale of the tape between AMD and Nvidia. AMD’s general leads in clock speeds, VRAM allocations, and L3 cache per price tier give its RX 6000 series cards a lead over comparatively priced Nvidia GPUs in certain games, while Nvidia’s cards either exceed AMD in select raw benchmarks or pull ahead when they can lean on superior ray-tracing performance or the silicon-fueled wizardry of Nvidia’s RTX-exclusive Deep Learning Super-Sampling (DLSS).

AMD is using today’s rollout of new GPUs to push back on the latter point, however, as it has also confirmed that its FidelityFX Super Resolution (FSR) system will finally be upgraded to a “2.0” version starting on Thursday, May 12. As previously hinted, Bethesda’s 2021 shooter Deathloop will exclusively get the newest flavor of FSR in a patch this week, at which point eagle-eyed gamers will be able to directly compare DLSS and FSR 2.0, as the game supports both standards. AMD says to expect patches in the “coming months” to add FSR 2.0 to PC games like Grounded, EVE Online, and Microsoft Flight Simulator (which is slated to get its own DLSS-specific patch, as well).

All RX 6000 series GPUs can leverage any existing or upcoming version of FSR, while Nvidia GPUs dating back to the GTX 1070 and the GTX 16 series are compatible with the standard, as well. We look forward to the Deathloop patch coming later this week and are curious to see how its temporal-based approach to resolution upsampling has been improved over the noticeably blurry FSR 1.0 version thus far—and how the update compares to the machine-learning model that Nvidia uses for its own proprietary standard.



Source link

Share

Written by bourbiza

bourbiza is an entertainment reporter for iltuoiphone News and is based in Los Angeles.

Leave a Reply