
AMD
Beginning at this time, AMD and numerous GPU makers will start rolling out three new graphics playing cards within the Radeon 6000-series lineup. These new merchandise add a “50” suffix to current mannequin numbers, and for those who’ve adopted AMD over the previous few years, you may most likely guess what which means.
The RX 6950XT, RX 6750XT, and RX 6660XT are heading to retailers at this time at advised minimal costs of $1,099, $549, and $399, respectively. Every GPU comes with almost equivalent specs to their “00” suffix predecessors, with the one distinction between every being a factory-level enhance to clock speeds and reminiscence speeds.
Infinity Cache once more—but additionally, very finite clock tweaks
In a press briefing forward of at this time’s bulletins, AMD reps advised that the features are partially due to touch-ups on each the silicon and software program ranges, although for 2 of the GPUs, a wattage leap elements in, as effectively. The 6950XT now calls for 335W TDP, up 35 W from its 6900XT predecessor, and we’re at 180 W TDP for the 6650XT, which is a 20 W leap from the 6600XT.

AMD
The leap from the RX 6900XT to the RX 6950XT nets 2.6 p.c increased enhance clocks and seven.8 p.c increased reminiscence clocks; all different GPU components, significantly AMD’s beefy allocations of L3 “Infinity Cache,” stay equivalent between the 2 fashions. The same story performs out for the opposite 50-series jumps: lower than 1 p.c increased enhance clocks for the 6750XT, and 1.8 p.c increased enhance clocks for the 6650XT.
The scant leap in clocks means that AMD may be introducing new fashions to interchange the previous ones outright, however AMD advised Ars Technica that the RX 6900XT and RX 6700XT will proceed to be offered, with the RX 6600XT being phased out as soon as its current inventory is cleared. (For a step down at that stage, AMD will proceed promoting the RX 6600.) AMD is manufacturing its personal direct variations of the RX 6950XT and RX 6750XT, however not the 6650XT; all three might be made and offered by the standard roster of in style Add-In Board (AIB) GPU companions.
With out evaluate {hardware} in our palms, we’re left questioning precisely how a lot these slight spec bumps leap forward of comparable “00” fashions. For some trendy GPU prospects, that form of comparability could also be moot; the older fashions, like most trendy GPUs, offered out extremely shortly amidst a crypto boom, but recently stabilizing GPU prices imply that you just may truly purchase a “50” mannequin for much less in the long run.
Staff Purple vs. Staff Inexperienced, as soon as once more
At that time, then, it is the standard story of the tape between AMD and Nvidia. AMD’s normal leads in clock speeds, VRAM allocations, and L3 cache per worth tier give its RX 6000 collection playing cards a lead over comparatively priced Nvidia GPUs in sure video games, whereas Nvidia’s playing cards both exceed AMD in choose uncooked benchmarks or pull forward once they can lean on superior ray-tracing efficiency or the silicon-fueled wizardry of Nvidia’s RTX-exclusive Deep Studying Tremendous-Sampling (DLSS).
AMD is utilizing at this time’s rollout of latest GPUs to push again on the latter level, nevertheless, because it has additionally confirmed that its FidelityFX Tremendous Decision (FSR) system will lastly be upgraded to a “2.0” model beginning on Thursday, Could 12. As previously hinted, Bethesda’s 2021 shooter Deathloop will completely get the most recent taste of FSR in a patch this week, at which level eagle-eyed avid gamers will be capable to immediately evaluate DLSS and FSR 2.0, as the sport helps each requirements. AMD says to anticipate patches within the “coming months” so as to add FSR 2.0 to PC video games like Grounded, EVE On-line, and Microsoft Flight Simulator (which is slated to get its personal DLSS-specific patch, as effectively).
All RX 6000 collection GPUs can leverage any current or upcoming model of FSR, whereas Nvidia GPUs relationship again to the GTX 1070 and the GTX 16 collection are appropriate with the usual, as effectively. We sit up for the Deathloop patch coming later this week and are curious to see how its temporal-based strategy to decision upsampling has been improved over the noticeably blurry FSR 1.0 model so far—and the way the replace compares to the machine-learning mannequin that Nvidia makes use of for its personal proprietary normal.