The RTX 5050 has already been tipped to feature GDDR6 memory rather than GDDR7 found in the RTX 5060 through RTX 5090. However, what was still unknown was the speed bin Nvidia would choose. Leaker MEGAsizeGPU on X reports that the RTX 5050 will allegedly utilize the fastest version of GDDR6 in mass production — 20 Gbps modules, which are used in AMD's RX 9070 series GPUs and Nvidia's GDDR6 variant of the RTX 4070.
If the X leak is accurate, the RTX 5050 will feature 320GB/s of memory bandwidth, 40% of the bandwidth of the RTX 5060 (which boasts 448GB/s of memory bandwidth). The massive drop-off in memory bandwidth goes hand in hand with the RTX 5050's 50% drop in CUDA cores from the RTX 5060. It remains to be seen what the RTX 5050 would be capable of with GDDR7, but it is likely the GPU's significantly cut-down compute power will prevent the 20 Gbps GDDR6 modules from being a major bottleneck.
The 20 Gbps speed bin is the fastest version of GDDR6 used by any GPU currently. This latest variant is nearly as fast as Micron's 21 Gbps GDDR6X memory modules used in Nvidia's RTX 30 series and RTX 40 series. Micron's G6X is just 5% quicker than the aforementioned GDDR6 counterpart. Technically, Samsung has developed 24 Gbps GDDR6 memory modules, but those are not in mass production (and may never be).
Latest Videos FromTom's Hardware
At least for the Samsung version, the main advantage of GDDR6 is its power consumption; Samsung's 20 GBps GDDR6 requires just 1.1V to operate. By contrast, older versions of GDDR6 and Micron's GDDR6X memory require 1.35V (or 23% more voltage) to function. Though strangely, SK hynix's does not have any GDDR6 variants with a voltage rating below 1.35V, suggesting its 20 Gbps counterpart runs at 1.35V.
One interesting side effect the RTX 5050 might have to endure is a potential 1-2% performance deviation depending on which models get Samsung modules and which ones get SK Hynix modules. The RX 9070 XT allegedly boasts a 1-2% performance delta based on which manufacturer its GDDR6 modules come from, with the SK Hynix-equipped models running 1-2% faster than their Samsung counterparts.
The disparity in performance allegedly comes from looser timings on the Samsung modules; it is likely that Samsung prioritized power efficiency with its 20 Gbps variants, resulting in slightly lower performance but significantly less power draw than SK Hynix's counterpart (as we mentioned already).
The RTX 5050's specs have all but been confirmed, featuring 2,560 CUDA cores, 20 SMs, and a 128-bit memory interface based on the GB207 die. The only part of the RTX 5050's specs that has been absolutely confirmed is the fact that it will be using GDDR6 memory of some kind (even though the RTX 5050 laptop GPU uses GDDR7). Other leaks suggest the RTX 5050 will run at a peak boost clock of 2,520MHz. A release date has not been leaked or confirmed by Nvidia, but the RTX 5050 (desktop version) will likely launch sometime this year.
Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.
